Literature

Water Body Detection and
Delineation
with Landsat TM Data
Paul Shane Frazier and Kenneth John Page
Abstract
The aim of this project was to determine the accuracy of using
simple digital image processing techniques to map riverine
water bodies with Landsat 5TM data. This paper quantifies
the classification accuracy of single band density slicing of
Landsat 5 TM data to delineate water bodies on riverine
floodplains. The results of these analyses are then compared
to a 6-band maximum likelihood classification over the same
area. The water boundaries delineated by each of these digital
classification procedures were compared to water boundaries
delineated from colour aerial photography acquired on the
same day as the TM data. These comparisons show that Landsat
TMdata can be used to map water bodies accurately. Density
slicing of the single mid-infrared band 5 proved as successful
as multispectral classification achieving an overall accuracy
of 96.9%, a producer's accuracy for water bodies of 81.7% and
a user's accuracy for water bodies of 64.5%.
Introduction
Accurate information on the extent of water bodies is important
for flood prediction, monitoring, and relief (Smith, 1997;
Tholey et al., 1997; Baumann, 1999); production of wetland
inventories (Bennett, 1987; Johnston and Barson, 1993; Blackman
et al., 1995; Shaikh et al., 1998; Phim et al., 1999); and the
evaluation of water resources (Morse et al., 1990; Manavalan et
al., 1993). Often this information is difficult to produce using
traditional survey techniques because water bodies can be fast
moving as in floods, tides, and storm surges or may be inaccessible.
Remotely sensed data provide a means of delineating
water boundaries over a large area at a given point in time. To
capture fast moving hydrological features, the data need to be
either of a high temporal resolution or in a substantial archive
to cover a range of hydrological conditions. Landsat MSS and
TM provide high spatial resolution data at 16-day intervals over
a long archival history, exceeding 25 years in most locations.
The long archive period and repetitive capture make the data
useful for mapping water bodies at a regional scale over a range
of hydrological conditions.
Since Landsat data became available in 1972, they have
been used to map water extent. Smith (1997) cites several early
studies where Landsat MSS data, in particular Band 7, were
used to distinguish water bodies from surrounding dry soil or
vegetation. Comparison with aerial photography gave error
estimates of less than 5 percent.
Bennett (1987) used density slicing of Landsat MSS band 7
to map water bodies to the west of Griffith, New South Wales,
Australia (NsW). He compared the area of MSS-derived water
bodies with that derived from digitized aerial photography and

found that the MSS data underestimated the area of water by
around 40 percent.
School of Science and Technology, Charles Sturt University,

Locked Bag 588, Wagga Wagga, NSW 2678, Australia
(pfrazier@csu.edu.au).
Johnston and Barson (1993) evaluated the usefulness of
Landsat TM imagery for mapping lakelpond wetland extent in
western Victoria and riverine wetland extent in central Victoria.
The satellite information was compared to manually
mapped ground truth. They found that simple density slicing
of the TM5 (mid-infrared) successfully detected the lakelpond
wetland areas achieving a classification accuracy of 95 percent
but failed to map the riverine wetlands adequately. The failure
of the technique with riverine wetlands was attributed to the
narrow width of the oxbow lakes and a poor timing of data
selection.
Many other studies have reported the successful use of
density slicing of Landsat TM or MSS data to delineate water
bodies; however, no quantitative accuracy assessments were
conducted. Manavalan et al. (1993) used Landsat TMband 4 to
map the extent of the Bhadra Reservoir, India. Overton (1997)
used density slicing of Landsat TM band 5and a high flood spatial
mask to map water bodies on the Murray River between
Blanchetown and Wentworth, South Australia. Shaikh et al.
(1997) timed the acquisition of Landsat MSS band 7 to coincide
with selected significant hydrological events to determine
wetland inundation for the Cumbung Swamp, NSw, Australia.
Baumann (1999) used band 4 of Landsat TMto map flood extent
on the Mississippi River but experienced problems separating
water from certain urban features without the inclusion of an
additional band.
Other studies have successfully used both supervised and
unsupervised multispectral classification of optical remote
sensing data to delineate water boundaries (Manavalan et al.,
1993; Lee and Lunetta, 1995; Blackman et al., 1995; Kingsford
et al., 1997; Brady et al., 1999). Kingsford et al. (1997) used
supervised maximum-likelihood classification of Landsat MSS
data to map wetlands over a part of the Murray-Darling Basin.
Checking of 200 random water bodies with aerial photography
gave an accuracy of 90 percent. They then extended the project
to the entire basin using 59 images and an unsupervised classification
procedure.
Both single-band density slicing and multispectral classification
(supervised and unsupervised) have been used to map
water boundaries. Researchers typically have relied upon
visual comparisons between the classification and the raw
data to estimate accuracy. Few studies have used a formal accuracy
assessment procedure. Where statistical accuracy assessment
procedures have been implemented, they have either
compared the area or number of wetlands identified by ground
truth with that determined from the satellite image, or they
Photogrammetric Engineering & Remote Sensing

Vol. 66, No. 12, December 2000, pp. 1461-1467.
0099-1112I0OI6612-1461$3.00/0
O 2000 American Society for Photogrammetry
and Remote Sensing

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING Decernher 2000 1461
have used a random sampling procedure to check the classification

accuracy. These methods of accuracy assessment do
allow the production of a formal estimate of the error in the
classification but provide no information on where the classification
performed badly or where it performed well. To select
useful image bands and determine the optimum classification
procedure for water body delineation, a comparison of the different
techniques using a quantitative spatial accuracy assess- - ment is needed.
This project set out to determine the classification accuracy
of density slicing and multispectral image classification
of Landsat 5 TM data to map the water bodies of the Murrumbidgee
River and floodplain near Wagga Wagga, Australia (Figure
1). Landsat TMdata were classified using single-band
density slicing and multispectral maximum-likelihood classification.
These classified data were then compared to high resolution
aerial photography to determine how well the Landsat
data could be used to detect and delineate water bodies.
Specifically this project set out to
Quantify the accuracy of simple digital techniques for the detection
and delineation of riverine floodplain water bodies using
Landsat TMdata,
Recommend a suitable technique for mapping riverine flood@)
Y1.E
PI'%
(C)
Figure 1. The location of (A) the City of Wagga Wagga in Australia,

(B) the Murrumbidgee River within the southern MurrayDarling Basin, and (C) the study reach near Wagga Wagga.
plain water bodies, and
Determine the best band or set of bands required for water
boundary mapping.
The Study Area

The study examines the channel and floodplain of the M u m bidgee River near the city of Wagga Wagga, in a reach extending
15.3 krn by 5.6 km(Figure 1). The Murrumbidgee River forms a
part of the Murray-Darling basin and drains an area of approximately
82 000 km2(Olive et a]., 1994). In the study reach it is a
large meandering river with an average bankfull channel width
of 80 m, a mean depth of 6 m, a sinuosity of 2.3, and has a predominantly
suspended load (Page, 1988).
The river floodplain is 1 to 2 krnwide [Olive et a]., 1994)
and consists of 6 to 7 m of mud and sand overlying basal gravels.
There are several types of natural water bodies on the floodplain,
including lagoons (meander cut-offs or oxbow lakes),
interscroll swales, and backswamp depressions (Table 1). The
lagoons are large water bodies, often greater than 6 hectares in
size, with a width similar to that of the active channel and with
considerable variability in turbidity and depth. The interscroll
swales are smaller features, often less than 10 m wide, that contain
water for variable but usually short periods after floods or

heavy rain. Backswamps are shallower depressions that hold
water for an extended period. They are not common features in
the study reach. Several artificial dams are also found in this
area. The largest of these is the effluent storage pond for the
Wagga Wagga sewerage works (Figure 2).
Figure 3 shows the discharge hydrograph for the Wagga
Wagga gauging station horn July 1990 to November 1990. At
the time of data capture (26 October 1990), the floodplain
retained water from a flood event that occurred on 08 July
1990. This event had a maximum discharge of 63,000 Mlld, sufficient
to over top the river bank in this reach and fill many of
the floodplain wetlands (Page, 1988). Flood events of this size
can carry very turbid water with a sediment concentration of
over 400 mg/l (Olive and Olley, 1997). On the day of data capture
the river at Wagga Wagga had a flow of 24,000 Mlld (Pinneena,
1998) with a turbidity of 90 mgll (Southern Riverina
County Council, water quality archive, 26 October 1990).
Methods
In order to assess the accuracy of using Landsat TM data to accurately
locate and delineate water bodies, classified Landsat TM
data were compared with manually mapped aerial photography
of the same area. Luckily for this study, both the Landsat TM
data and the aerial photography were captured on the same date
(26 October 1990), thus ensuring identical hydrological
conditions.
Materials
The color aerial photography was acquired by the NSW Central
TABLE1.CLASSESA ND DESCRIPTIONSO F DIGITIZEDW ATERB ODIES.
Polygon label Description

River Active channel of Murrumbidgee River
(single linear feature)
Lagoon Abandoned reaches of Murrumbidgee
channel (seven features)
Small Pool
Dam
Smaller water bodies on floodplain
(closed depression in swales and backswamps)
(105 features)
Artificial water bodies
Sewerage ponds, one gravel pit (3
features)
Dimensions Vegetation
Water surface width 50 to 100 m
Max depth 8 to 10 m
Water surface width 50 to 100m
Max depth 1to 4m
Area 6 to 18 ha
Variable shape
Max depth 0.2 to 2 m
Area 0.005 to 3.2 ha
Variable shape
Max depth 1 to 8 m
Area 2.8 to 14.5 ha
Marginal trees
< 5% crown cover
Marginal trees
< 5% crown cover
Marginal trees
< 10% crown cover
Nil
1462 December 2000 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
Figure 2. Aerial photograph of the Wagga Wagga region

(print 146) showing water body types.
Mapping Authority at a nominal scale of 1:25,000. For this proj- tek ScanMaker 96OOXLA3 flatbed scanner. The
resulting images
ect, prints 142-148 of run 3 were used. A subset of the Landsat had an approximate ground pixel size of 2.1 m. The
digital
TMscene (PathIRow 92/84, level 4 processing), covering the images were then registered to the TM imagery using
eight
same area as the selected aerial photography, was excised from ground control points (GCPS) per photograph. The
GCPS were
the original data. This procedure resulted in the creation of an selected as close as possible to the margins of the
water bodies
image of 510 pixel by 187 lines with bands 1,2,3,4,5, and 7 (Fig- on each photograph in order to maximize the
likelihood of a
ure 1). ERMapper software Version 5.7 was used for all image good match to the TMdata.
processing, polygon digitizing, and polygon manipulation. The first attempt at registration gave an average rootmeansquare
(RMS)error of 10 pixels (21 m). The images were then
Aedal photograph lnterpretatlon warped using a linear geometric transformation, nearest-neighThe aerial photographs were scanned at 300 DPI using a Micro- borhood intensity interpolation, and a pixel
resampling size of
3 m. Examination of the fit between the photographs and the TM
data indicated that there were significant registration errors.
Each photograph was then re-registered using an additional
seven GCPs on the margins of the water bodies, achieving an
average RMSerror of 23 pixels (9 m). This time overlaying the
photograph onto the TMimage indicated a good fit.
All water bodies greater than 625 m2 (lmm by lmm) in size
were then highlighted directly on the hardcopy prints of photographs
143,145, and 147 using magnified stereoscopic examination,
On the photographs, water bodies appear as uniform
regions of dark green to bright khakibrown in the low lying
channels and depressions. Each water body was then digitized
directly onto the scanned, registered aerial photograph using
the ERMapper regions tool. The polygons were then labeled as
River, Lagoon, Small Pool, or Dam (Table 1, Figure 2). The polygons
were then converted into a five-class raster image for comparison
with the Landsat TM classification results.
Landsat Image CIassHkation
Single-band density slicing and the multispectral maximumlikelihood

algorithm were used to classify the satellite data.
The density slicing attempted to quantify the accuracy of a simple
technique using a single band to map water bodies. The
more advanced multispectral classification was completed to
give a benchmark result using all reflective bands for comparison
with the density slicing results.
Interrogation of the data was carried out through the selection
of 12 training areas over various water bodies. These training
areas were selected from a range of river, lagoon, and dam
sites to account for the full range in water reflectance resulting
from different water depths and turbidities. Figure 2 shows that
even within the lagoon class of water bodies there is considermwo
20WX)
1-
~ ~ z ~ q ~ ~ g ~ ~ ~ ~ s~~s$>~g2>~- ~- -$
s~:a:~~
t-4
Figure 3. Discharge hydrograph for the Murrumbidgee River at Wagga Wagga

July to November 1990.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING / I 1 1 1463
J
able variation in water brightness.

The statistics of each training site were analyzed to determine
which band or bands best discriminated water bodies from
the surrounding landscape (Table 2). There was considerable
variation in the mean water values between training sites in
each of the visible bands. Band 4 shows a reduced variation
between mean values with bands 5 and 7 showing even less
variation.
A simple density slice classification of each band was used
to determine which bands were most useful for water boundary
discrimination. The minimum and maximum value of
water from the combined training site values (Table 3) were
used to create the boundaries of the density slicing procedure.
The maximum-likelihood classification was completed
using all 12 "water" training areas (Table 2). Classified pixels
that were less than 2 percent typical of the training statistic
were eliminated fram the classification. The resulting 12 water
classes were then combined into one super "water" class.
The density slice classification of each band and the maximumlikelihood classification were then compared with the
raster ground-truth data on a per pixel basis.
Results and Discussion

Density Slicing
Figure 4 shows a section of the density slicing results for each

TM band. Clearly, the classifications of the three visible bands
substantially overestimate the area of water on the image. There
is considerable spectral overlap between the brightness values
found in the water bodies and those in the surrounding urban
and rural landscape (Figure 5).
The infrared band classifications all give a much better representation
of the water bodies than do the visible bands. Band
4 is able to identify most of the major water bodies, but there is
still some confusion with the urban area, hill shadows, and
poor pasture paddocks. The density slicing of band 7 also gives
Mean Pixel Value
Training Site Band1 Band2 Band3 Band4 Band5 Band7
Lagoon 1
Lagoon 2
Lagoon 3
Lagoon 4
Lagoon 5
Lagoon 6
River 1
River 2
River 3
River 4
Dam 1
Dam 2
Standard Deviation
TABL3E. MAXIMUMAN D MINIMUMVA LUES FOR ALL WATEBRO DYT RAINING
AREAS.
Minimum Water Maximum Water

Band Value Value
1464 December 2000
a reasonable representation of the water boundaries but

includes some very lush crop paddocks that may be undergoing
irrigation.
Density slicing of band 5 gave the best visual approximation
of the ground-truth image without including other landcover
types. Based on aerial photograph interpretation and
local knowledge, the only major error of commission appears
to be very moist or waterlogged soil areas on the flood plain.
The overall classification accuracy along with producer's
and user's accuracies (Jensen, 1996) for the water class from
each of the six bands are presented in Table 4. As expected from
the visual interpretation, the visible bands perform badly
when it comes to differentiating water bodies from other landcover
types. The overall classification accuracies for each of
these bands were all below 8 percent. Because the number of
pixels classified as water is very high for the visible bands, the
producer's accuracies were all very high, being above 96 percent
in each case. However, these results are offset by the very
poor user's accuracies, of less than 5 percent.
The three infrared bands all perform significantly better
than the visible bands, achieving high overall accuracies and
greatly increased user's accuracies. The result from band 5
achieved the best) combination of accuracy results with the
highest overall accuracy (96.9 percent), a producer's accuracy
of 81.7 percent, and the highest user's accuracy at 64.5 percent
(Table 4).
Band 1
Band 2 Band 7
Band 3
Key to densty sl~c~ng
Water Not water
Ground truth Image
Band 4
I
I
Key to ground truth Image
R~ver Dam Not Water
Lagoon Small
Figure 4. Results of density slicing each Landsat TM band.

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
Band 2
9 00 5 ; K i
Q5"m
.I 0 h
0 255 0 255
Band 4
g5;m
95q-J
.I P
rn
0 255 0255
Band 6
5 0 255
9yrl
8 0 255
-0
8K
Range ofwater training pixels
Figure 5. Individual band histograms and water training

pixel ranges.
TABLE4 . RESULTS OF ~ N D S A TT M CLASSIFICATION ACCURACYA SSESSMENT.
Overall Water Class Water Class

Classification Producer's User's
Band Accuracy (%) Accuracy (% 1 Accuracy (%)
17.799.65.0
26.496.74.8
38.096.54.9
493.677.142.6
I!
596.981.764.5
794.474.445.4
Lagoon, and Dam classes) with results above 80percent. The

smaller water bodies are not as well represented, with an accuracy
of only 48.2 percent. Many of these smaller water bodies
were often less than a single pixel in size, and most of them
were less than a pixel in width. The size of these water bodies
ensured that the classification was trying to map a mixed pixel.
Even the strong absorbance characteristics of water on infrared
electromagnetic radiation did not prevent tbe mixed pixel
being mapped as a non-water area.
The reason for the failure of the density slicing of the visible
bands to accurately classify the water boundaries is shown
in F i p e 5. The range of the density slicing determined from
the water training areas covers almost the entire range of data
in each of the visible bands. Therefore, the resulting classification
covers almost the entire image, as indicated on Figure 4.
Each of the visible bands displays a uni-modal histogram with
no indication of a separate group of data for water pixels.
The infrared bands all display a bi-modal histogram with a
darker mode consisting mostly of water pixels. The density
slice ranges for band 5 and band 7 start at near zero, meaning
that water pixels are the darkest in the image. The range for
each of these bands slightly overlaps with the largest mode in
the histogram. The density slice range of 14 to 66 for band 4
does not account for all of the pixels in the smaller darker mode
and significantly overlaps with the darker edge of the main
pixel mode.
Maxlmumlikelihood Classification
The supervised maximum-likelihood classification produced
the best overall accuracy of 97.4 percent, a producer's accuracy
of 59.6 percent, and a user's accuracy of 81.4percent. The error
breakdown, presented in Table 6, indicates that the maximumlikelihood
classification, like the band 5 density slice, was very
good at accurately mapping areas of not water.
The supervised classification proved to be more sensitive
than the density slicing method for detecting water bodies and
produced fewer errors of commission. However, this sensitivity
excluded more atypical water pixels, creating a larger error of
omission (ie. lower producer's accuracy). For each of the four
water body types the maximum likelihood classification performed
significantly worse than the band 5 density slice. The
percentage accuracy for the Small Pool, Dam, River, and
Lagoon water bodies is 21.4 percent, 56.3 percent, 66.3 percent,
and 57.5 percent, respectively, for the maximum-likelihood
classification as opposed to 48.2 percent, 85.0percent, 86.2
percent, and 81.2 percent for the band 5 density slice (Table 6
and Table 5, respectively).
Figure 6 illustrates the differences between the results of
the band 5density slice and the maximum-likelihood classification.
The density slice result gives a less speckled representation
of the water bodies but suffers from inclusion of some
pixels that clearly are not water bodies but are dark enough to
fall into the same brightness range.
Conversely, the supervised maximum-likelihood classifiA more detailed analysis of the results from the band 5 classification
is presented in Table 5. This table shows that the

classification is good at mapping areas that are not water, a
result that is not suprising given that the ground-truth mapping
indicates that the area consists of 95.1 percent non-water
land cover. However, it also indicates that the classification
did not contain major errors of commission.
Of more interest is the ability of the classification to detect
water bodies. Clearly, the classification does a reasonable job
of detecting and delineating the larger water bodies (River,
TABLE 5. ERROR MATRIX FROM BAND 5 CLASSIFICATION.
Classification Data
Percentage
Band 5Error Matrix Water Not Water Total Correct
Reference Small Pool 21222844048.2%
Dam 1642919385.0%
River 2626419304586.2%
Lagoon 77717395081.8%
Not Water 2079886639074297.7%
Total 58588951295370Percentage Correct 64.5%99.1%- 96.9%
TABLE6 . SUPERVMISAEXIDM UML IKELIHOOCD LASSIFICATIEORNR ORM ATRIX.
Classification Data
Maximum-Likelihood Percentage
Classification Error Matrix Water Not Water Total Correct
Reference Small Pool
Dam
River
Lagoon
Not Water
Total
Percentage Correct
lrr~gatedP asture
Band 5 dens~tys lice Max~mumL ~kehhcodc lassficstion
II
mm
Figure 6. Water bodies identified by the band 5 density slice

and the maximum-likelihood classification methods.
r
Small unclassi
polygons
Figure 7. The affect of misregistration
and small polygons on the accuracy of

Landsat TM classification.
cation presents a more speckled representation of the water
bodies. The greater sensitivity of this classifier appears to have
excluded more pixels that are atypical of the main training data
but were mapped as water on the ground truth.
The Effectsof Reglstratlon Errors
In any situation where long slender polygons are being
matched to an image, registration errors or distortions in either
the image or the polygons can contribute to the overall error
count. A closer viewing of the band 5density slice classification
with the digitized vector overlay shows that aerial photograph
to Landsat TM image registration error has not been
insignificant in this study (Figure 7).
Figure 7 highlights two problems'with the comparison of
the classified TMdata and the digitized ground-truth polygons.
Clearly, when comparing high resolution ground-truth data
with lower resolution test data, registration errors between the
two datasets will be a problem. In this case the aerial photography
was registered to the Landsat TM data until low RMS errors
were achieved and a good visual fit was established. Even with
this close checking of the registration, the best possible result
is a registration error slightly less than the largest pixel size (30
m). This error can amount to a significant reported error in the
classification, especially when the main features of interest are
long, thin, and meandering.
The second feature highlighted by Figure 7 is the inclusion
of sub-pixel sized polygons. These small polygons were
included to determine whether Landsat TMdata could be used
to detect water bodies smaller than a pixel. Figure 7 shows a
number of these features that are less than a pixel in width and
even less than a pixel in total area. With these smaller features
mis-registration severely reduced the possibility of an accurate
result being reported, and the effect of mixed pixels further
reduced the likelihood of an accurate classification.
Conclusion
From this study it is clear that Landsat TMdata can be used to
map water bodies associated with the Murrumbidgee River
and its floodplain in the Wagga Wagga region. A multispectral
maximum-likelihood classification was able to produce an
overall classification accuracy of 97.4percent. This classification
was able to locate all of the major water bodies but underestimated
the number of water pixels present on the image,
achieving a producer's accuracy of only 59.6percent.
Importantly, a simple density slice classification of the
mid-infrared band 5produced a classification accuracy (96.9
percent) similar to that of the maximum-likelihood classification.
The density slice gave a better estimate of water pixels but
tended to include more pixels that were not water bodies (user's
accuracy 64.5 percent).
The other infrared bands, band 4and band 7, were also useful
for locating water bodies but tended to include more errors
of commission. Lush crop areas were included in the band 7
density slice classification, and poor pasture, hill shadow, and
parts of the urban area were included in the band 4density
slice classification.
All of the visible bands proved to be inadequate for successful
density slice classification. The brightness of the turbid
water showed a range of pixel values similar to that of the
majority of the surrounding land cover, resulting in gross overestimation
of the water area.
The infrared wavelengths are the most useful that optical
scanners can offer for water mapping. The strong absorption of
infrared light by water bodies gives them a distinctively low
spectral response in this range. The mid-infrared bands of Landsat
TM proved marginally more successful than the near-infrared
band. The narrow nature of the water bodies in question
creates many fringing mixed pixels that often contain significant
responses from vegetation. Also, some of the water bodies
are very turbid, which increases the upper range of values in
near-infrared more than it does in the mid-inhared bands. The
combination of mixed pixels and high turbidity limits the use
of a single near-infrared band to classify water bodies. This
limitation has significant ramifications for the use of optical
remotely sensed data such as those acquired by Landsat MSS
Or SPOT.
Acknowledgments
The authors thank Rodney Rumbachs for preparation of the
aerial photography and the production of the location map and
John Louis, David Roshier, David Lamb, and the anonymous
reviewers from PE&RS for their comments on the manuscript.
References
Baumann, P., 1999. Flood Analysis, http://www.research.umbc.edu/
tbenja~/baumann/mod2.html1,0 February 1999.
Bennett, M.W.A., 1987. Rapid monitoring of wetland water status using
density slicing, Proceedings of the 4th Australasian Remote Sensing
Conference, [14-18 September], Adelaide, pp. 682-691.
Blackman, J.G., S.J. Gardiner, and M.G. Morgan, 1995. Framework for
biogeographic inventory, assessment, planning and management
of wetland systems: the Queensland approach, Workshop Proceedings,

Wetland Research in the WetlDry Tbopics of Australia, (C.M.
Finlayson, editor), Jabiru, NT,Australia, Supervising Scientist
Report No. 101, Department of Environment and Heritage, Canberra,
pp. 114-122.
Brady, A., M. Shaikh, A. King, and P. Sharma, 1999. Remote sensing
and the Great Cumbung Swamp, Wetlands Australia, 7:596-606.
Jensen, J.R., 1996. Introductory Digital Image Processing: A Remote
Sensing Perspective. Prentice Hall, Upper Saddle River, New Jersey,
316 p.
Johnston, R., and M. Barson, 1993. Remote sensing of Australian wetlands:
An evaluation of Landsat TM data for inventory and classification,
Australian Journal of Marine and Freshwater Research,
44235-252.
Kingsford, R.T., R.F. Thomas, P.S. Wong, and E. Knowles, 1997. GIs
Database for Wetlands of the Murray Darling Basin, Final Report
to the Murray-Darling Basin Commission, National Parks and
Wildlife Service, Sydney, Australia, 85 p.
Lee, K.H., andR.S. Lunetta, 1995. Wetland detection methods, Wetland
and Environment Applications of GIS U.G. Lyon and J. McCarthy,
editors), Lewis, Boca Raton, Florida, pp. 249-284.
Read PE&RS on-line!

Excerpts of PEbRS are now
available on-line.. .
Plus, stay tuned for
many more advances to the
ASPRS web site.
Manavalan, P., P. Sathyanath, and G.L. Rajegowda, 1993. Digital image
analysis techniques to estimate waterspread for capacity evaluations
of reservoirs, Photogrammetric Engineering b Remote Sensing,
59(9):1389-1395.
Morse, A., T.J. Zarriello, and W.J. Kramber, 1990. Using remote sensing
and GIs technology to help adjudicate Idaho water rights, Photogrammetric
Engineering b Remote Sensing, 56(3):365-370.
Olive, L.J., J.M. Olley, Murray, and P.J Wallbrink, 1994. Spatial variation
in suspended sediment transport in the Mumbidgee River,
New South Wales, Australia, Variation in Stream Erosion and
Sediment flonsport, (L.J. Olive, R.J. Loughran, and J.A. Kesby,
editors), IAHS Publication 224, International Association of
Hydrological Sciences, Canberra, pp. 241-249.
Olive, L.J. and J.M. Olley, 1997. River regulation and sediment transport
in a semi-arid river: The Murrumbidgee River, New South Wales,
Australia, Proceedings of the Rabat Symposium, Human Impact
on Erosion and Sedimentation, April, Rabat, Morocco, pp.
283-290.
Overton, I., 1997. Satellite Image Analysis of River Murray Floodplain
Inundation, NRMS Project R6045, Murray-Darling Basin Commission,
Adelaide, 12 p.
Page, K.J., 1988. Bankfull discharge frequency for the Murrumbidgee
River, New South Wales, Fluvial Geomorphology of Australia (R.F.
Warner, editor), Academic Press, Sydney, pp. 267-281.
Phinn, S.R., L. Hess, and C.M. Finlayson, 1999. An Assessment of
Remote Sensing for Monitoring and Inventorying Wetland Environments
in Australia, report prepared for the National Wetlands
Program, Environmental Research Institute of the Supervising
Scientist, Environment Australia, Jabiru, Australia, 52, 45 p.

Pinneena, 1998. New South Wales Suqface Water Data Archive. Version
6, Department of Land and Water Conservation, NSW, Australia
(on a CD).
Shaikh, M., A. Brady, and P. Sharma, 1997. Applications of remote
sensing to assess wetland inundation and vegetation response in
relation to hydrology in the Great Cumbung Swamp, Lachlan Valley,
NSW, Australia, Wetlands for the Future (A.J. McCumb and
J.A. Davies, editors), Gleneagles Publishing, Glen Osmond, South
Australia, pp. 595-606.
Smith, L.C., 1997. Satellite remote sensing of river inundation area,
stage, and discharge: A review, Hydrological Processes,
11:1427-1439.
Thorley, N., S. Clandillon, and P. De Fraipont, 1997. The contribution
of spaceborne SAR and optical data in monitoring flood events:
Examples in northern and southern France, Hydrological Processes,
11:1409-1413.
(Received 26 March 1999; revised and accepted 09 November 1999)
Grids & Datums

Abstracts
Software ~evieew
Calendar
Classifieds
Book Review
and many more
prs. org
PHOTOGRAMMETRIC ENGINEERING 81 REMOTE SENSING December 2000 1467

Delineation
Abstract

Introduction
around 40 percent.

selection.

additional band.
procedure.
0099-1112I0OI6612-1461$3.00/0
and Remote Sensing


techniques using a quantitative spatial accuracy assess- - ment is needed.

Landsat TMdata,
Y1.E
PI'%
(C)

boundary mapping.
The Study Area

Methods
conditions.
Materials

Small Pool
Dam
(105 features)
features)
Max depth 8 to 10 m
Max depth 1to 4m
Area 6 to 18 ha
Variable shape
Variable shape
Max depth 1 to 8 m
Area 2.8 to 14.5 ha
Marginal trees
< 5% crown cover
Marginal trees
< 5% crown cover
Marginal trees
< 10% crown cover
Nil

resulting images
digital
eight
GCPS were
water bodies
likelihood of a
Aedal photograph lnterpretatlon warped using a linear geometric transformation, nearest-neigh-
The aerial photographs were scanned at 300 DPI using a Micro- borhood intensity interpolation, and a pixel
resampling size of

20WX)
1-
~ ~ z ~ q ~ ~ g ~ ~ ~ ~ s~~s$>~g2>~- ~- -$
s~:a:~~
t-4

J

variation.


Density Slicing

Mean Pixel Value
Lagoon 1
Lagoon 2
Lagoon 3
Lagoon 4
Lagoon 5
Lagoon 6
River 1
River 2
River 3
River 4
Dam 1
Dam 2
Standard Deviation
AREAS.

Band Value Value
1464 December 2000

irrigation.

(Table 4).
Band 1
Band 2 Band 7
Band 3
Water Not water
Ground truth Image
Band 4
I
I
R~ver Dam Not Water
Lagoon Small

Band 2
9 00 5 ; K i
Q5"m
.I 0 h
0 255 0 255
Band 4
g5;m
95q-J
.I P
rn
0 255 0255
Band 6
5 0 255
9yrl
8 0 255
-0
8K

pixel ranges.

17.799.65.0
26.496.74.8
38.096.54.9
493.677.142.6
596.981.764.5
794.474.445.4

I!

pixel mode.
Classification Data
Percentage
Dam 1642919385.0%
River 2626419304586.2%
Lagoon 77717395081.8%
Not Water 2079886639074297.7%
Classification Data
Dam
River
Lagoon
Not Water
Total
Percentage Correct
lrr~gatedP asture
II
mm

r
Small unclassi
polygons

Conclusion

of the water area.
Or SPOT.
Acknowledgments
References
pp. 114-122.
316 p.
44235-252.

Read PE&RS on-line!

ASPRS web site.
59(9):1389-1395.
283-290.
Adelaide, 12 p.
(on a CD).
11:1427-1439.

11:1409-1413.
Grids & Datums

Abstracts
Software ~evieew
Calendar
Classifieds
Book Review
and many more
prs. org
SPECTRAL
REFLECTANCE OF
VEGETATION, SOIL
AND WATER
LWR314
Fifteenth International Water Technology Conference, IWTC-15 2011, Alexandria, Egypt
FLOOD MAPPING INFERRED FROM REMOTE SENSING DATA

Jean-Franois Crtaux 1, Muriel Berg-Nguyen 1, Marc Leblanc 2, Rodrigo Abarca Del
Rio3, Francois Delclaux 4, Nelly Mognard 1, Christine Lion1, Rajesh Kumar Pandey1,
Sarah Tweed 2, Stephane Calmant 5, and Philippe Maisongrande 1
1 CNES/Legos, 14 Av Edouard Belin, 31400, Toulouse, France, jeanfrancois.
cretaux@legos.obs-mip.fr
2 School of Earth and Env. Sc., James Cook Univ. Cairns, QLD,4870, Australia,
marc.leblanc@jcu.edu.au
3 Departamento de Geofsica (DGEO), Facultad de Ciencias Fsicas y Matemticas,
Universidad de Concepcin, Concepcin, Chile, roabarca@udec.cl
4 HSM, Univ. Montpellier 2 Case MSE- 34095 Montpellier Cedex,
delclaux@msem.univ-montp2.fr
5 IRD/Legos 14 Av Edouard Belin, 31400, Toulouse, France, Stephane.calmant@ird.fr
ABSTRACT
In ungauged basin, space-based information is essential for the monitoring of
hydrological water cycle, in particular in regions undergoing large flood events where
satellite data may be used as input to hydrodynamic models. A method for near 3D flood
monitoring has been developed which uses synergies between radar altimetry and high
temporal resolution multi-spectral satellite. Surface Reflectance from the MODIS Terra
instrument are used to map areas of open water as well as aquatic vegetation on a weekly
basis, while water level variations in the inundated areas are provided by the radar
altimetry from the Topex / Poseidon (T/P) and Envisat satellites. We present this
synergistic approach to three different regions: Niger Inner delta and Lake Tchad in
Africa, and Ganga river delta in Asia. Based mainly on visible and Near Infra Red (NIR)
imagery is suitable to the observation of inundation extent. This method is well adapted
for arid and semi arid regions, but less for equatorial or boreal ones due to cloud
coverage.
This work emphasizes the limitations of current remote sensing techniques for full 3Ddescription
of water storage variability in ungauged basins, and provides a good
introduction to the need and the potential use of the future SWOT (Surface Water and
Ocean Topography) satellite mission.
Keywords: radar altimetry, MODIS, Flood mapping, SWOT, Arid climate
1. INTRODUCTION
The main purpose of this study is to provide space-based tool for the monitoring of
inundated areas in large wetlands and floodplains located in arid and semi arid regions.
Space and airborne technologies are increasingly found to be a key and unique source of
spatial information for wetlands conservation and management as many of the Worlds
wetlands have insufficient on-ground data partly due to their size, number and limited
accessibility. Finding complementary and new methods to monitor inundation patterns
for large wetlands and floodplains is consequently important with expectation to
assimilate such space-based hydrological information into models (e.g.climate, land
surface, water-management and eco-hydrological models)
Comprehensive time series of inundation are required from space observations to: i)
investigate links between floods in remote areas and climate variability; ii) model the
hydrodynamic of a floodplain at high spatio-temporal resolution; iii) understand the
interaction among inter-annual and seasonal flood cycle and land use patterns in and
around area of flooding; iv) to examine the vulnerability of an ecosystem to inundation;
v) to develop alert systems for inundation in a given wetland.
In recent years, remote sensing techniques have clearly shown their capability to monitor
components of the water balance of large river basins on time scales ranging from
months to decades. Satellite altimetry, which has been developed and optimized for open
oceans, was also widely used in different fields of continental hydrology based on
coupled satellite altimetry / in-situ gauges measurements (Crtaux and Birkett [1];
Cretaux et al. [2], and Calmant et al. [3]). The global altimetry data set has now an 18
years-long lifetime and is intended to be continuously updated in the coming decade.
Moreover several authors have addressed the issue of water extent mapping over
floodplain from Remote Sensing data. Toyra et al., [4], [5], have used a combination of
Radarsat and Spot scenes to study extent of water in wetlands and produced multi-year
maps time series of flooded area in the Peace-Athabasca delta in Canada with high
spatial resolution. Frappart et al., [6], [7], have combined radar altimetry with SAR
images onboard the Japanese Earth Resources Satellites (JERS-1) or visible images of
the Vegetation instrument onboard the Spot Satellites in order to study floods over the
Rio Negro (Amazon) and Mekong Basin. The ASAR instrument is also very effective
sensor to detect flooded areas in the particular cases of cloud cover regions with high
spatial resolution (Henry et al. [8]). Bartsch et al. [9] have also used ENVISAT ASAR
data for mapping of fresh water ecosystems in Siberia, and their analysis have shown that
numerous areas previously mapped as tundra are in fact covered by water. Peng et al,
[10] have used the MODIS data to develop a method of water extent and level
monitoring, which however depends on the knowledge of topographic map of the surface
study or on a relation between surface and level of water. Over large flood regions,
medium resolution multi-spectral imagery like MODIS is well suitable as demonstrated
by Sakamoto et al. [11]. They have produced from this instrument weekly time series of
inundation maps of the Mekong basin over a multi-year period. Synthetic Aperture Radar
(SAR) Interferometry (Alsdorf and Lettenmaier, [12], Alsdorf et al. [13]) and passive
and active microwave observations (Prigent et al. [14]) also offer important information
on land surface waters, such as changing area extent of large wetlands. A common
problem in remote sensing of wetlands inundation is the detection of water under aquatic
vegetation. Leblanc et al. [15] used Meteosat thermal images to derive monthly flood
maps of Lake Chad that captured water under aquatic vegetation and adequately
reconstructed the drying of Lake Chad since the 1980s.
Current challenges remain especially in providing global mapping of inundation and
estimates of water storage changes and this has driven the decision of CNES and NASA
to propose a new concept satellite mission (Surface Water and Ocean Topography:
SWOT).
2. METHODOLOGY OF FLOOD MAPPING

The Modis instrument is a multispectral imaging system installed onboard the Terra and
Aqua satellites at sun-synchronous near polar orbit at an altitude of 705 km. It observes
the whole Earth every day. The basic measurements used to classify earth surface are
surface reflectance measured over 7 spectral bands from the visible to the middle
Infrared. The surface reflectance product we used (MOD09GHK) is corrected for
atmospheric effects.The Modis images are very useful because they offer high temporal
and spectral resolution, with images free of charge, covering wide areas of few tens of
thousands square kilometres. Spatial resolution is 500 meters for the images used in our
study. Here, we used MODIS images to detect open water and aquatic vegetation in arid
and semi arid regions.
MODIS data also provided a means to select altimetry data along the tracks of T/P and
Envisat satellites in order to estimate water level variations in different part of the
inundation area. Designed to operate over ocean surface, the classical radar altimeters
measurements that mainly consist on waveform or backscatter energy are much more
complex over land surface. It can be affected by many radar echoes due to the presence
of several reflectors under the footprint: water, vegetation, sand, etc or inhibited by
rapidly varying topography (Frappart et al. [16]).The radar altimetry is a profiling
technique, which does not allow global view of the Earth surface, hence limits
worldwide surveying as well spatial resolution in the cross-track satellite direction.
Despite those limitations the altimetry is a technique, which has a proven potential for
hydrology science: the data are freely available on all the whole earth, and for a lot of
remote areas it is the only source of information, as for the flood plain where in-situ data
(when they exist) are limited to some points near main stream of the rivers. It however
remains a significant limitation when using radar altimetry for flood monitoring. Due to
revisit interval of several days (10 for T/P or Jason1 & 2, 35 for Envisat) the chance to
collect altimetry data at the time of a flood peak is reduced, with very few measurements
for quick inundation.
The methodology developed is illustrated by the Fig. 1. It is separated into 3 steps: a
preprocessing
phase of MODIS (georeferencing and mosaicking) and altimetry data, a
processing of the MODIS data for pixel classification (to detect open water and aquatic
vegetation in particular), and a processing of the radar altimetry for the measurements
that have been classified as open water by MODIS. The MODIS data along each track of
altimetry satellite above the area of study are interpolated in space and time, then, this
information gives a criterion of selection for altimetry data. Below we present the main
results obtained from this method over few case studies around the world
Fig. 1: Algorithm developed for the flood mapping from MODIS and water height
determination from radar altimetry. Band Unit of reflectance is internal HDF-EOS data
format specific to the Modis data and do not correspond to usual reflectance unit.
Water + dry
Water pixel (Selected)
Non Water pixel (Rejected)
Yes
MODIS PREPROCESSING
Extraction of HDF
images files from
Georeferencing and
mosaicking
Extraction of Band 1,2
and 5 over the ROI
Extraction of Envisat,
& T/P over the ROI
Pre-selection of valid data
from Modis cassification
Open Water
Band 5 < 1200
No
NDVI <0.4 NDVI <0.4

Band 5 < 2700
MODIS Data
No
No
Yes
Yes No Yes
Acquatic Vegetation Dry Vegetation
Weekly Time series

Interpolation over
Altimetry tracks
Pre-selection of valid data
from editing parameters
Water level calculated from altimetry data selected
ALTIMETRY
Products
3. CASE STUDIES
3.1 Niger Inner Delta
The Inner Niger Delta (IND) is located in Central Mali in the semi-arid Sahelian zone, in
the south of the Sahara Desert. It is a shallow area of about 70,000 km2 constituted by
channels, swamps and lakes, and its water extent varies seasonally from around 1000 to
12000 km2 with large inter-annual changes (a factor of 1 to 5 has been observed
according to Mahe et al. [17]). The IND is at the junction of 2 rivers, the Niger and the
Bani which supply approximately 1490 m3/s of water to the delta (Mahe et al. [17]).The
factor, generating this variability is the precipitation rate change, in time and with
latitude. From October to May, the IND is hot and dry, and water balance is mainly
driven by evaporation. From June to September, direct precipitation on the IND and in
the upstream part of Niger and Bani Basins generate the seasonal flood. Precipitation
over the IND are weak (~300mm/yr over the northern part of the IND) while they can
between 1200 and 1500 mm/yr over the upstream parts of the Niger and Bani Basins,
mainly occurring during the four months of the wet seasons (June to September).
Interannual variability of rainfall is also marked as shown in Fig. 2. Therefore, the extent
of water in the IND is changing year to year with very complex patterns depending on
the topography of the river channels in the delta, or the presence of vegetation, and on
the total amount of water filling the IND. During flood periods, the IND land is also
subject to noticeable vegetation growth, mixed with water and dry soil. In that context,
reliable frequent information on the water extent, spatial distribution and temporal
variation of IND floodplain are vital for: i) water resources management, ii) a better
understanding of the influence of climate change and the feedback of this floodplain on
regional climate, iii) the monitoring of this important ecosystem and socio-economical
resource. Moreover, our ability to measure, monitor, and forecast supplies of fresh water
in the IND using in-situ methods is almost impossible because of limited access and the
physics of water flow across vast lowlands. This reinforce the needs for methodologies
based on space techniques, e.g., visible and radar satellite imagery to measure water
extent over the wetlands and floodplains, and radar altimetry to measure the water level
variations with time. The flood over the IND is a combination of high water in both Bani
and Niger rivers but with changes in their role from one year to another one. This is a
factor of complexity of understanding the flood over the IND as this implies to model
both river basins hydrodynamic. Modis mapping of inundation every week over interannual
period can therefore provide an invaluable source of information for both
understanding the flood regime and to serve as external validation data for everyone who
aim to develop model of the flood over the IND and more generally to model the Niger
basin hydrodynamic.
We applied the synergistic mapping with radar altimetry and MODIS to the IND for the
2000-2008 period. From January to May, the IND and surrounding regions are drying
out, and active vegetation is fully disappearing. In June, the permanent lake in the IND
starts to grow and in July aquatic vegetation initiates its seasonal cycle. Until August the
aquatic vegetation is growing regularly and it covers the entire region. At end of August
flood waters start to reach the IND and continue to rise during two months until a peak
between mid and late of October. From November to January the water over the IND
evaporates or flows downward and the only region which still presents some vegetation
covers are the areas that just dried out, exclusively in the IND.
Precise correlations between flood sequence and precipitation patterns in the upstream
part of the Niger and Bani river watershed and in the IND have been established over the
period 2000-2010. Fig. 2a shows that open water maximum is shifted by one month and
half from the maximum of rainfall over the Bani River (the same has been observed with
Niger upstream watershed rainfall). Fig. 2b shows that a first apparition of open water is
observed in August over the IND due to direct precipitation, which is however much
weaker than the main inundation due to flow from Niger and Bani. Fig. 2c shows that
vegetation starts growing immediately after the first rainfall occurring in the IND, in
July, and also that vegetation covers over the IND presents a second smaller peak in end
of November after the main inundation.
Fig. 2: Modis classification over the IND versus precipitation rate over the Bani River and
the IND. (TRMM data set has been used for the computation of rainfall).
Modis classification / rainfall (TRMM)
0
0,2
0,4
0,6
0,8
1
1,2
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
normalized data
open water
Rainfall on Bani
Modis Classification / rainfall (TRMM)
0
0,2
0,4
0,6
0,8
1
1,2
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
normalized data
open water
rainfall on IND
Modis Classification / Rainfall (TRMM)
0
0,2
0,4
0,6
0,8
1
1,2
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
normalized data
Vegetation
rainfall on IND
The 8-days mapping of flood and vegetation dynamic over the IND has also been done
for the 10 years period, and one sees on figure 3 the high inter-annual variability of water
extent from dry to wet years.
Fig. 3: maps of IND during the maximum of inundation of four different years,
from dry years (2002, 2004) to a wet year (2001, 2005). Dark blue is representing
free water, light blue: mixed water and dry land, light green: aquatic vegetation,
dark green: vegetation on dry land, and yellow: the other type of surface.
To complete the information given by MODIS data, satellite radar altimetry
measurements from T/P (one track) and Envisat (5 tracks) mission have been used. They
have allowed calculating the water level variations over the IND, at the location where
the satellite altimetry tracks are passing over open waters pixel as determined by the
MODIS weekly. Fig. 4 shows some results obtained with Envisat satellite along a track
which crosses the IND from South to North. The low temporal resolution of Envisat
satellite is a clear limitation if one needs to monitor the exact water height variations
from the beginning to the end of each inundation, but relative water height from upper
part to lower part of the IND is measured, and it also can serve as control data of
hydrodynamic model of the flood over the IND.
Fig. 4: level variations (m) over the IND from Envisat track n 474
In summary, the monitoring of the flood inter-annual variability also offers an interesting
feedback to the study of convective storms in the surrounding regions. It has been
demonstrated in Taylor [18] that inundation over the IND influences convection which
provokes rainfall over large regions in West Africa. They conclude that planned new
dams in Guinea (upstream to the delta) could have enormous influence, both in the
inundation in the IND and in rainfall patterns across hundreds of km2.
3.2 Lake Tchad

The Lake Chad Basin (LCB) is located in Central Africa, between latitudes 5N and
25N and longitudes 7E and 27E. It extends over 2. 5 Mkm2 across the territory of
Niger, Nigeria, Chad and Cameroon, and the borders of Algeria, Libya, Sudan and the
Central African Republic, making it the largest endoreic basin in the world. Enclosed
between mountain ranges (Tibesti, Darfour) and high plateaus (Adamoua), this entire
drainage network converges on a central depression, in which Lake Chad is situated. The
climate of this region is related to the movement of the Inter Tropical Convergence Zone
(ITCZ) (Olivry et al. [19]). In the south, the monsoon flow brings heavy precipitation
whereas in the north, the Harmattan produces a dry climate, at the latitude of Lake Chad,
the climate is of Sudano Sahelian type because the ITCZ only reaches this zone between
June and August, the rainy season during which total precipitation barely exceeds 300
mm.
The water balance of the lake is represented in Figure 5c. During the period preceding
the 1970s drought, the Chari and Logone rivers supplied Lake Chad with, in average,
nearly 85% of the total inflows. The other significant tributary into the lake is the
intermittent Nigerian river the Komadougou Yob which contribution to the lake is about
1 km3/yr. Precipitation over the lake contributes 6 km3/yr. Concerning losses,
evaporation is the dominant process, accounting for an average annual volume of 43

km3/yr. Losses due to infiltration are estimated at 3 km3/yr. It should be noticed that due
the recent droughts at the beginning of the 1970s and 1980s, these volumes have been
divided by two, and that the total surface of the lake decreased from about 25 000 km2 to
about 5 000 km2.
Lake Chad is of great ecological and socio-economical value for the region. This high
hydrological variability and regional importance confirm the need for an effective
monitoring of the lake. Unfortunately, in-situ observations of the lake are scarce. Only
one staff gauge is currently operational (Bol station) and does not allow for the
complexity of the inundation patterns to be captured. Specifically, as Lake Chad dries it
can split into several isolated pools (Lemoalle et al. [20]). For these reasons, remote
sensing data are the only means of gaining consistent hydrologic observations across the
entire lake. These space observations are used to simulate the hydrology of Lake Chad
(Delclaux et al. [21]; Lemoalle et al. [22]; Bader et al. [23]). Radar altimetry (Fig. 5a)
provides water height time series over the Lake Chad since 1993, from multi satellite
tracking (T/P from 1993 to 2002, then Envisat up to 2010 and Jason 2 from 2008), and
Modis imagery allow the monitoring of flood water dynamic in the south and north
basins since 2000 (Fig. 5b). Since 2001, a small decline in the south pool of Lake Chad
is observed from both radar altimetry and Modis images analysis.
Fig. 5: (from up to low) Level variation of Lake Tchad from altimetry multi-mission (T/P
(up to 2002), Envisat (up to 2008) and Jason-2(up to 2010)) (a), surface variations from
Modis images (b) and water balance of the lake (c).
Multi-satellite altimetry water level variation (m)
277
277,5
278
278,5
279
279,5
280
280,5
281
1993 1995 1997 1999 2001 2003 2005 2007 2009
0
500
1000
1500
2000
2500
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Surface of open water (km2)
3.3 Ganga river basin

The Ganga River is continuously fed by the melting of ice on glaciers and other rivers
originating in Himalaya. Rain water, during the period mid June to Sept, over the
Gangetic plain is also a major contributor in the river water which is causing the flood in
northern India and Bangladesh. Though the various rivers in India feed the water from
glaciers it is also affected by the Indian Monsoon. The climate of India is classified in
four categories: winter, pre-monsoon, monsoon and post- monsoon. Monsoon period,
also known as the rainy season, lasting from June to September, causes the flood in most
of the region if India. In pre-monsoon period, April to June is the driest season and this
time river sink to its minimum.
Like the other rivers in India, the inundation along the river Ganga is also mostly
depending upon the rain in catchments of the Ganga basin and its tributaries. The water
drained to river from catchments is at his maximum in the middle of the rainy season.
Although the region is rather cloudy in rainy season, the Modis images provide an
interesting view of phase of inundation over the Ganga delta with however a proportion
of rejected images higher than what was observed in the other regions described in this
study. One hence could estimate average water extent on the delta at inter-annual time
scale.
Modis classification over the Ganga river
0
5000
10000
15000
20000
25000
30000
35000
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
surface of open water (km2)
Fig. 6: Surface of open water detected from Modis classification for the down Ganga
River.
For example the variation of water logging, Fig. 6, in lower Ganga basin shows a good
annual repetition in rainy season (maximum in August-September). There is a high peak
in 2007 (also 2008) which is because of the huge amount of water discharge in Kosi
River, due to high rain, from Nepal. Large flood area south to the Ganga River is also
observed from MODIS in Bihar in 2000 during which, the width of the Ganga River
increases significantly, causing lateral overflow from the bank of the river. Analysis of
images from 2000 to 2010 has also showed high inter-annual amplitudes of the
inundation over the Ganga River, therefore over the delta in Bangladesh. Between the 2
dry years of 2005 and 2006, and the wet years of 2007 and 2008 there is a factor two of
the inundated surface (Fig. 6).
4. DISCUSSION ON FUTURE MISSION AND CONCLUSIONS

What are the perspectives for the next decade on flood monitoring from space?
Multispectral imagery from Modis has proven its ability to monitor water extent over
large areas with continuous data at relatively high temporal resolution. Other sensors
exist or are programmed for the coming years. The Meris instrument onboard the Envisat
platform is based on the same measurements principle. However the Envisat mission is
at the end of its life and one would not be able to rely on this satellite in the near future
for multi-spectral and radar altimetry data. In the frame of the establishment of new
Global Monitoring for Environment and Security GMES capacities by the European
Union, a panel of new satellites have been planned for the next decade, with dedicated
missions in land monitoring from multispectral sensor (Sentinel-2), and radar altimetry
in dual Ku-C bands (Sentinel-3). Sentinel-2 will provide multispectral imagery at high
resolution (4 spectral band at 10m, 6 at 20m and 3 at 60m), with a full coverage of the
Earth every 5 days. It will consist in pair of satellites, with initial launch in 2013.
Sentinel-3, mission is dedicated to the measurements of sea surface topography but could
also be used for water level estimation on continental water bodies, in the same way that
Envisat is currently used. Sentinel-3 will also consist in a pair of satellites with expected
first launch in 2013. In 2011, the Centre National dEtudes Spatiales (CNES) and ISRO
Indian Space Research Organisation (ISRO) will launch the Saral/Altika mission, which
will be the first altimeter operating in Ka band which will have the main advantage of a
better spatial resolution due to smaller footprint of the radar signal (150 m instead of few
km), allowing a better discrimination of water in floodplains and anastomosed rivers
with a large number of small channels. This mission will be placed on the same orbit
than Envisat and should cover the needs for continental water level monitoring, for lakes,
rivers, and floodplains. In 2013, the CNES, EUMETSAT, and NASA will continue the
Jason program, with the launch of Jason-3 radar altimeter in Ku and C bands.
However, none of those missions is dedicated exclusively to continental hydrology. The
future SWOT mission is the first satellite mission dedicated to the measurement of
continental surface water. SWOT will provide a global inventory of all terrestrial water
bodies whose surface area exceeds (250 m2) and river whose width exceeds 100 m, at
sub-monthly, seasonal and annual time scales (Biancamaria et al., [24]). The principal
instrument of SWOT will be a Ka-band Radar Interferometer (KaRIN), which will
provide heights and co-registered all weather imagery of water over 2 swaths, each 60
km wide, with an expected precision of 1cm/km for water slopes, and absolute height
level precision of 10 cm / km2 (Fig. 7). SWOT will provide an estimate of river
discharge, and map floodplain topography and channel reaches. SWOT will allow us to
better quantify the exchange between rivers and floodplains for improved prediction of
inundations (Biancamaria et al., [25]). For floodplain mapping the improvement should
be very significant since SWOT will have a global coverage with a resolution of 250 by
250m. As shown above, the high spatio-temporal heterogeneity of inundations is a key

limitation for actual remote sensing system. Combination of altimetry and imagery
provide interesting control data to validate and assess quality of hydro dynamical model.
They allow better understanding of the linkage with climate variability, changes due to
monsoon or drought in Africa or India, impact of PDO or El Nio in South America and
Central Australia, but the spatial and temporal sampling of data actually does not allow
having a global view of surface water elevation and storage variation at high spatiotemporal
resolution. Additional information like precise topography of the floodplain
inferred from digital elevation model for example, is still needed if one expects to
determine water volume variation. Other very important information that SWOT will
provide, is the water discharge at each river channel entering and leaving a floodplain
like the IND for example. This information would be invaluable for assimilation in
hydrodynamic model of inundation (Biancamaria et al. [26]) and to understand the role
played by such floodplain in the regional water availability with high societal interest.
The potential of SWOT measurements will be enhanced if coupling with other Remote
Sensing data (SMOS-like, radar altimetry, imagery, gravimetry and meteorological
satellite data sets) and it will considerably improve our understanding of flood process.
Fig. 7: SWOT mission. Radar Interferometry principle for the determination of water
height over wide swath of 120 km along the orbital track of the satellite.
The combined radar and multispectral approach developed in this paper demonstrates our
current capabilities for operational space monitoring of inundated extents and levels for
three case studies across the World. It also raises serious limitations which could be
overcome by future missions. By complementing in situ observations and hydrological
modelling, space observations have the potential to improve significantly our
understanding of hydrological processes at work in large river basins and their influence
on climate variability, geodynamics and socio-economic life. It offers global

geographical coverage, good spatio-temporal sampling, continuous monitoring with
time, and capability of measuring water mass change occurring at or below the surface.
ACKNOLEDGMENT
This work has been supported by the Centre National dEtudes Spatiales (CNES) in
the frame of the TOSCA program. It has also been supporter by the CEFIPRA
program. The altimetry data are downloaded from the Centre de Topographie des
Ocans et de lHydrosphre (CTOH) of Legos. The Modis data are downloaded from
the WIST data centre: Earth Observing System Data and Information System
(EOSDIS). 2009. Earth Observing System ClearingHOuse (ECHO) / Warehouse
Inventory Search Tool (WIST) Version 10.X [online application]. Greenbelt, MD:
EOSDIS, Goddard Space Flight Center (GSFC) National Aeronautics and Space
Administration (NASA). URL: https://wist.echo.nasa.gov/api/
REFERENCES
[1] Crtaux J-F and Birkett, C., Lake studies from satellite altimetry, C R Geoscience,
doi: 10.1016/J.cre.2006.08.002, 2006
[2] Crtaux J-F, Calmant, S., Abarca Del Rio, R., Kouraev, A., and Berg-Nguyen, M.,
Lakes studies from satellite altimetry, Handbook on Coastal altimetry, Springer ed.,
chap. 19, 2010
[3] Calmant, S., Seyler, F. and Cretaux, J.F., Monitoring Continental Surface Waters by
Satellite Altimetry, Survey in Geophysics, special issue on Hydrology from Space, Vol
29, Issue 4-5, pp 247-269, 2008.
[4] Tyr, J., Pietroniro, A. and Martz, L.W. Multisensor Hydrologic Assessment of a
Freshwater Wetland, Remote Sensing of Environment, 75(2), 162-173, 2001
[5] Tyr, J., and Pietroniro A., Towards operational monitoring of a northern wetland
using geomatics-based techniques, Remote Sensing of Environment, 97, 174-191, 2005
[6] Frappart F., Seyler, F., Martinez, J-M., Leon J., and Cazenave, A., Determination of
the water volume in the Negro River sub basin by combination of satellite and in situ
data, Remote Sensing of Environment, 99, 387-399, 2005.
[7] Frappart, F., Dominh, K., Lhermitte, J., Ramilllien, G., Cazenave, A. and LeToan, T.,
Water volume change in the lower Mekong Basin from satellite altimetry and imagery
data, Geophys. J. Int., 167, 570-584, 2006a.
[8] Henry, J-B., Chastanet, P., Fellah, K., and Desnos, Y-L., ENVISAT Multi-Polarised
ASAR data for flood mapping, International Journal of Remote Sensing, Vol 27, no 10,
1921-1929, 2006
[9] Bartsch A., Pathe, C., K. Scipal, K., and Wagner, W., Detection of permanent open
water surfaces in central Siberia with ENVISAT ASAR wide swath data with special
emphasis on the estimation of methane fluxes from tundra wetlands, Hydrology
Research, 39.2, 89-100, 2008
[10] Peng, D.Z., L. Xiong, S. Guo, and N. Shu, Study of Dongting Lake area variation
and its influence on water level using MODIS data, Hydrological Sciences, 50(1), pp
31-44, 2005
[11] Sakamoto T., Nguyen, N.V., Kotera, A., Ohno, H., Ishitsuka, N., and Yokozawa,
M., Detecting temporal changes in the extent of annual flooding within the Cambodia
and the Vietnamese Mekong Delta from MODIS time series imagery, Remote Sensing of
Environment, 109, Issue 3, 295-313, doi:10.1016/j.rse.2007.01.011, 2007

[12] Alsdorf, D.E., and Lettenmaier D.P., Tracking fresh water from space, Science, 301,
1485-1488, 2003.
[13] Alsdorf D.E., Rodriguez E., and Lettenmaier D., Measuring surface water from
Space, Review of Geophysics, 45, RG2002, 2007
[14] Prigent, C., Matthews, E., Aires, F., and Rossow, W.B., Remote sensing of global
wetland dynamics with multiple satellite data sets, Geophys. Res. Let., 28 , 4631-4634,
2001.
[15] Leblanc, M., Lemoalle, J., Bader, J-C., Tweed, S., and Mofor, L., 15 years of
inundation patterns for Lake Chad: new observations during a drought period from
satellite thermal data. Journal of Hydrology, In Press. 2011
[16] Frappart F., Calmant, S., Cauhop, M., Seyler, F., and Cazenave, A., Preliminary
results of ENVISAT RA-2-derived water levels validation over the Amazon basin,
Remote sensing of Environment, 100, 252-264, 2006b
[17] Mah, G., Bamba, F., Soumaguel, A., Orange D., and Olivry, J-C., Water losses in
the inner delta of the River Niger: water balance and flooded area, Hydrol. Process. 23,
3157-3160, 2009
[18] Taylor, C.M, Feddbacks on convection from African wetland, Geophys. Res. Lett.
Vol 37, L05406, doi: 10.1029/2009GL041652, 2010
[19] Olivry, J.C., Chouret, A., Vuillaume, G., Lemoalle, J., and Bricquet, J.P. Hydrologie
du lac Tchad. Monographie hydrologique 12. ORSTOM ditions. 266 p. Paris :
ORSTOM. 398 p, 1996
[20] Lemoalle, J., Bader, J-C., and Leblanc, M.,. Lake Chad. Encyclopedia of Lakes
and Reservoirs. Springer, Dordrecht, 2011
[21] Delclaux F., Le Coz M., Coe M., Favreau G., and Ngounou Gatcha B. Confronting
Models with Observations for evaluating Hydrological Change in the Lake Chad Basin,
Africa. XIIIth World Water Congress, Aout 2008, Montpellier (France), 2008
[22] Lemoalle, J., Bader, J-C., and Leblanc, M., The variability of Lake Chad:
hydrological modelling and ecosystem services. Proceedings of the 13th IWRA World
Water Congress, 01-04 September 2008, Montpellier, France, 2008
[23] Bader, J-C., Lemoalle, J., and Leblanc, M., Modle hydrologique du lac Tchad.
Journal des Sciences Hydrologiques Hydrological Sciences Journal, 2011
[24] Biancamaria, S., Andreadis, K.M., Durand, M., Clark, E.A., Rodriguez, E.,.
Mognard, N.M., Alsdorf, D.E., Lettenmaier, D.P., and Oudin, Y., Preliminary
characterization of SWOT hydrology error budget and global capabilities. IEEE
JSTARS, Special Issue on Microwave Remote Sensing for Land Hydrology Research
and Applications, 3 (1), 6-19, doi:10.1109/JSTARS.2009.2034614, 2010a.
[25] Biancamaria S., Durand M., Andreadis, K.M., Bates, P.D., A. Boone, A., Mognard,
N.M., Rodriguez, E., Alsdorf, D.E., Lettenmaier, D.P., and Clark, E.A., Assimilation of
virtual wide swath altimetry to improve Arctic river modeling. Remote Sensing of
Environment, doi:10.1016/j.rse.2010.09.008, in press, 2010b.
[26] Biancamaria, S., Boone, A., Bates, P., and Mognard, N.M., Large-scale coupled
hydrologic and hydraulic modelling of the Ob river in Siberia, J. of Hydrology, 379,
136150, 2009.
See discussions, stats, and author profiles for this publication at:
http://www.researchgate.net/publication/222427328
Evaluation of on-line DEMs for flood

inundation modeling. Adv Water Res
ARTICLE in ADVANCES IN WATER RESOURCES AUGUST 2007
Impact Factor: 3.42 DOI: 10.1016/j.advwatres.2007.02.005
CITATIONS
119
READS
284
1 AUTHOR:
Brett Sanders
University of California, Irvine
88 PUBLICATIONS 1,893 CITATIONS
SEE PROFILE
All in-text references underlined in blue are linked to publications on ResearchGate,
letting you access and read them immediately.
Available from: Brett Sanders
Retrieved on: 30 November 2015
Evaluation of on-line DEMs for flood inundation

modeling
Brett F. Sanders
*,1
Department of Civil and Environmental Engineering, University of California, Irvine, CA, United States
Received 19 October 2006; received in revised form 6 February 2007; accepted 8 February 2007
Available online 21 February 2007
Abstract
Recent and highly accurate topographic data should be used for flood inundation modeling, but this is
not always feasible given time
and budget constraints so the utility of several on-line digital elevation models (DEMs) is examined
with a set of steady and unsteady test
problems. DEMs are used to parameterize a 2D hydrodynamic flood simulation algorithm and
predictions are compared with published
flood maps and observed flood conditions. DEMs based on airborne light detection and ranging (LiDAR)
are preferred because of horizontal
resolution, vertical accuracy (_0.1 m) and the ability to separate bare-earth from built structures and
vegetation. DEMs based on
airborne interferometric synthetic aperture radar (IfSAR) have good horizontal resolution but gridded
elevations reflect built structures
and vegetation and therefore further processing may be required to permit flood modeling. IfSAR and
shuttle radar topography mission
(SRTM) DEMs suffer from radar speckle, or noise, so flood plains may appear with non-physical relief
and predicted flood zones may
include non-physical pools. DEMs based on national elevation data (NED) are remarkably smooth in
comparison to IfSAR and SRTM
but using NED, flood predictions overestimate flood extent in comparison to all other DEMs including
LiDAR, the most accurate. This
study highlights utility in SRTM as a global source of terrain data for flood modeling.
_ 2007 Elsevier Ltd. All rights reserved.
Keywords: Flood inundation; Flood modeling; DEM; DTM; LiDAR; IfSAR; SRTM
1. Introduction
Flooding is the most common and damaging natural
hazard faced by civilization, and flooding threats are likely
to increase given current climate change predictions that
suggest higher sea levels and more intense cyclonic weather
systems and precipitation. The frequency, distribution and
causes of floods over the last thirty years has been analyzed
and reported by the Dartmouth flood observatory,

and a five-fold increase in the number of floods per year
has been observed since the 1980s [11]. In the UK, a 20fold increase in the economic risk of flooding has been
projected [12].
Strategies to cope with flooding such as flood-plain
building regulations, insurance, structural measures such
as reservoirs and levee projects and emergency preparedness
all rely on flood predictions to some extent, and hence
the effectiveness of these measures is linked to the quality of
flood predictions. Topographic data are crucial for flood
inundation modeling and it is best to use recent and highly
accurate topographic data. However, this is not always feasible
given time and budget constraints and therefore it is
of interest to learn what can be accomplished using digital
elevation models (DEMs) that can be accessed on-line and
downloaded without charge.
This paper begins with an overview of digital elevation
models (DEMs), describing where each can be accessed
on-line and reviewing aspects relevant to flood inundation
modeling. Next, a flood inundation model is presented and
applied to a set of test cases. DEMs are used to assign elevations
to nodes of the computational grid, and the sensitivity
of flood model predictions to DEM type, resolution
and accuracy is examined.
0309-1708/$ - see front matter _ 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.advwatres.2007.02.005
* Tel.: +1 949 824 4327; fax: +1 949 824 3672.
E-mail address: bsanders@uci.edu.
URL: http://gram.eng.uci.edu/~bfs/sanders2.html.
1 On sabbatical at School of Civil Engineering, University of Nottingham,
United Kingdom.
www.elsevier.com/locate/advwatres
Advances in Water Resources 30 (2007) 18311843
2. Digital elevation models

DEMs considered in this study include airborne light
ranging and detection (LiDAR) data at 1/9 s resolution,
airborne interferometric synthetic aperture radar (IfSAR)
data at 1/9 s resolution, United States Geologic Survey
(USGS) national elevation data (NED) at 1/9, 1/3 and
1 s resolution, and shuttle radar topography mission
(SRTM) data at 1 and 3 s resolution. Note that 1 s corresponds
to about 30 m. Table 1 lists these DEMs along with
internet addresses where data can be accessed.
DEMs listed in Table 1 are distributed via portals that
enable the user to select a region of interest and dispatch
several files that include elevation data and meta data.
Note, in particular, the horizontal and vertical coordinate
system reported in the meta data because these are not
standardized. Sometimes data are distributed with a projected
coordinate system (e.g. universal transverse mercator
or UTM), but more often data are in a geodetic coordinate
system (longitude and latitude).
In this study, DEM is used to mean a grid of elevation
data, digital terrain model (DTM) is adopted when elevation
data corresponds to bare earth, and digital surface
model (DSM) is adopted when elevation data correspond
to a mixture of bare earth, tree tops, and buildings. Some

DTMs capture bare earth elevation better than others,
and this has implications relative to flood modeling, so
the process by which elevation measurements are acquired,
processed, and gridded is discussed below with an eye on
DTM quality.
LiDAR DEMs resolve terrain with a high degree of vertical
accuracy, 0.050.2 m (RMSE) [1,16,35,39]. An aircraft
fitted with a laser measures its distance to the ground, and
by tracking the position, pitch, and roll of the aircraft the
coordinates of the ground surface can be calculated
[1,16,35,39]. The laser works with pulses of light that reflect
off surfaces back to the aircraft. The pulse may reflect off
multiple surfaces, such as foliage and the ground, leading
to multiple returns. It is common to use the first and last
return to model vegetation and bare earth heights, respectively,
although additional features can be extracted by
studying the waveform of the return [35]. The high accuracy
of LiDAR DEMs, and the success with which bare-earth is
discerned from vegetation and buildings, has resulted in
many flood modeling applications of LiDAR [9,10,22,
23,2]. LiDAR DEMs for USA can be accessed from several
on-line sources (Table 1), and internationally there are offline
sources as well (e.g., UK Environment Agency).
IfSAR DEMs and SRTM DEMs are both based on
interferometry, which involves the reception of a backscattered
radar signal by two antennae. Terrain heights are
deduced by trigonometry using the phase difference
between signals, the distance between antennae, and the
location of the aircraft [28,29,35,38]. Vertical accuracy
depends on altitude but commercial IfSAR DEMs typically
achieve _1 m (RMSE) with an airborne platform, while the
accuracy of SRTM is _10 m (RMSE) due to its space
borne altitude [2830,35,38]. Vegetative cover will degrade
vertical precision, i.e., radar will penetrate vegetation some
distance before reflecting. Another drawback to radar
based DEMs is speckle, or random noise, which degrades
relative vertical accuracy particularly on floodplains
[13,19]. That is, speckle adds waviness to DTMs with
amplitudes of _0.1 and 1.0 m for IfSAR and SRTM,
respectively [13,19]. Hence, IfSAR and SRTM DEMs are
good examples of DSMs and may require further processing
to support flood modeling. For example, wavelet filters
have been applied to de-noise SRTM topography [13]. Several
versions of SRTM data available, and finished data
is accessible from the USGS (Table 1). The global SRTMis
3 s horizontal resolution and covers latitudes between 60 N
and 56 S. In the USA, 1 s SRTM is available but its true
resolution is closer to 2 s [34]. IfSAR DSMs cover many
locations internationally and can be obtained for a fee,
while data covering parts of USA are available at no cost
(Table 1).
Table 1
On-line accessible topographic datasets potentially useful for flood inundation modeling
Dataset Link Coverage Horiz. Res. (m) Vert. Accur. RMSE
(m)
NED-1/9 s seamless.usgs.gov USA: WV and parts of NB, TX, and

WA
_3 }0.30.5
NED-1/3 s seamless.usgs.gov Contiguous USA and HI _10 }7
NED-1 s seamless.usgs.gov Contiguous USA and HI, PR and VI _30 }715
NED-2 s seamless.usgs.gov Alaska _60 }715
SRTM-1 s seamless.usgs.gov USA and territories _30 }10
SRTM-3 s seamless.usgs.gov Global: 56 S 60 N _90 }10
Idaho LiDAR inside.uidaho.edu/geodata/
LiDAR/
ID _2 }0.05
Louisiana LiDAR atlas.lsu.edu LA _3 }0.200.25
North Carolina
LiDAR
www.ncfloodmaps.com NC _3 }0.200.25
NOAA CSC IfSAR www.csc.noaa.gov USA: CA 3 }2.5
NOAA CSC LiDAR www.csc.noaa.gov Coastal USA Variable }0.15
Intermap Corp. www.intermap.com International Variable Variable
1832 B.F. Sanders / Advances in Water Resources 30 (2007) 18311843
NED covers USA and is distributed at 2, 1, 1/3 and

1/9 s resolution by USGS (Table 1). NED is a patchwork
of topographic data from several sources ranging from
USGS quadrangle maps to LiDAR surveys. When data
are downloaded, a source description is included as meta
data. In addition to adopting uniform horizontal and vertical
control datums, NED are processed to smoothly
marry data from different sources (e.g., neighboring
7.5 min quadrangle maps) and smooth out artifacts such
as local minima that incumber some hydrologic analyses.
Alaska is covered by 2 s NED, and the remainder of
USA is covered by 1 and 1/3 s NED. 1/9 s NED is based
on LiDAR and presently has spotty coverage. The vertical
accuracy of NED depends on the data source but
USGS reports 7 m (RMSE) for 1/3 s NED and 715 m
for 1 s NED. This would appear horribly crude for the
purposes of flood modeling, where heights of decimeters
can mean the difference between flooded and dry land.
However, in the experience of the author the data are
far more accurate and in fact quite useful for studies of
flood plain inundation and overland flow as will be illustrated
later.
3. Flood simulation method
Flood inundation modeling can be accomplished with
one of several approaches. These could be organized by
dimensionality, ranging from zero-dimension models
consisting of a water level versus flow rate rating curve,
one-dimensional (1D) models such as the well known
HEC-RAS code, two-dimensional (2D) models that solve
the shallow-water equations, three-dimensional (3D) models
that solve NavierStokes equations and hybrids that
combine, for example, 1D and 2D approaches [18,21].
For this study a 2D approach is adopted for a number of
reasons. First, it is compatible with the native structure
of geospatial elevation data which is mapped on a horizontal
grid, compared with 1D approaches which require
derivative products such as storage and conveyance tables.
These in fact require a-priori knowledge of the flow path
and direction which can be difficult to ascertain in cases
of severe flooding. Second, flood hydrodynamics can be
truly multi-dimensional and in such cases its description

warrants a multi-dimensional model formulation. Third,
the 2D approach is justified based on the relative depth
and breadth of floods and the appropriateness of the
hydrostatic pressure approximation. Fourth, 2D flood
inundation modeling has been an active area of research
the past decade, particularly on the topic of Godunov-type
finite volume schemes, and today there are many robust
simulation codes [22,24].
The primary hypothesis of the 2D approach is that the
vertical pressure distribution is hydrostatic [20], and furthermore
that turbulent momentum dissipation can be
modeled with bottom stresses according to a quadratic
drag law [17]. For this study, the drag coefficient was scaled
by a Manning coefficient [4].
Solution of the shallow-water equations has long been
of interest to hydraulic engineers and over the last decade
or so significant advances have come with the advent of
Godunov-type finite volume schemes. Toro [36] provides
a review of many developments, and the accuracy of several
dam-break algorithms is reviewed by Morris [24].
The algorithm adopted in this study, BreZo (FlowSimulation,
LLC, Irvine, CA) is an operational version of the triangular
grid algorithm reported by Begnudelli and Sanders
[4] and is similar to the quadrilateral grid algorithm
reported by Bradford and Sanders [7,8] and Begnudelli
and Sanders [5]. BreZo does not account for infiltration
of water into the soil or changes in topography resulting
from flow induced erosion or sediment deposition. Grid
generation for this study was accomplished using a Delaunay
mesh generation tool [33]. Further, elevation at the
vertices of the triangular grid was estimated from the
DEM by linear interpolation. There are advantages to
meshes that adapt based on topographic features [2], but
to compare DEM effects a common mesh should be used
[22].
4. Application to Santa Clara River
Santa Clara River (SCR) near Castaic Junction in
southern California serves as the first of two test sites to
examine hydraulically relevant features of DEMs. This site,
shown in Fig. 1, was chosen because airborne IfSAR data
are available in addition to NED and SRTM data. There
are also many land uses (and hence, different land surface
textures) along the river including residential, commercial,
agricultural and open space. Thirdly, there are several river
junctions along this reach. Castaic Creek joins SCR from
the north of Castaic Junction, while San Francisquito
Creek (SFC) joins SCR from the northwest. Santa Clara
Valley was inundated in 1928 following collapse of St.
Francis Dam which was located along SFC. Data from this
flood provide a basis for model validation [26]. From a
transportation perspective, Castaic Junction is important
because the primary NorthSouth highway on the west
coast of USA, Interstate-5 (I-5), crosses SCR and serves
as the main link between the greater Los Angeles basin,
to the south, and the great Central Valley and the San
Francisco Bay Area to the north.
Fig. 2 shows how topography is represented by IfSAR,
NED and SRTM DEMs. NED appear smooth in comparison
to SRTM and IfSAR data, and depressions associated
with river channels and overbank flood plains can be seen
clearly. Comparing 1 s NED and SRTM, NED appear
smooth while SRTM show a subtle waviness that is a signature
of radar speckle [13]. IfSAR data reveal the most
detail, and upon close inspection it becomes clear that
the apparent roughness is linked to topographic attributes
such as vegetation and buildings. For example, large rectangular
structures of a commercial business district are visible
east of I-5, and tree tops appear as small mounds or
humps along the SCR corridor. The magnified views in
B.F. Sanders / Advances in Water Resources 30 (2007) 18311843 1833
Fig. 2 highlight the crossing of SCR by I-5, a divided highway

that is clearly visible in Fig. 2A. Fig. 2B shows that the
bridge is partially stripped from 1/3 s NED, i.e., the
divided road bed is visible on the northern half of the channel
bed. A faint trace of this feature is resolved by 1 s NED
but no trace is visible in SRTM data (Fig. 2D and E). In
fact, it is difficult to see whether SRTM resolves SCR.
To quantitatively compare IfSAR, NED and SRTM
DEMs, six square swaths of terrain measuring 255
255 m were gridded at 1 m resolution and bed elevation
at each grid point was estimated by linear interpolation
using five different topographic data sources: IfSAR,
1/3 s NED, 1 s NED, 1 s SRTM and 3 s SRTM. The six
swaths appear in Fig. 2 and are labeled in Fig. 2B as follows:
Pasture, Riparian, Commercial, Mountain, Old Residential
and New Residential. The first three are located
within the St. Francis dam-break flood zone or valley bottom,
while the last three are above it. Two residential
swaths were used because one corresponds to an older
housing development with mature vegetation, while the
other corresponds to a new housing development without
significant tree growth but with evidence of grading or
earth work which is a common development practice in
California and elsewhere. For each swath, the relative elevation
of topography according to these DEMs was compared
based on swath-mean values. Table 2 summarizes
these results. For reference purposes, the NED can be
traced to Val Verde and Newhall 1:24,000 scale quadrangle
maps last revised in 1995, the SRTM flew in 2000, and the
IfSAR data were acquired in 20022003.
The swath comparison (Table 2) shows that DEMs
compare better within the valley bottom than above it.
SRTM has been found to underestimate elevation of
high-relief and high-altitude terrain [3,13,14], and all
swaths above the valley bottom (Mountain, Old Residential
and New Residential) reflect this trend. SRTM underestimates
elevation 3.610.3 m at these sites compared to
NED. Along the valley bottom, the agreement between
DEMs varies by land use. DEMs compare best for the
Pasture swath where mean elevations differ by at most

0.91 m across all five data sources. This site lacks vegetative
cover and surface features that would cause radar
shading errors so it ideal for radar-based mapping. In
addition, its long term agricultural land use would suggest
that older USGS topographic maps (the basis of the
NED) and recent aerial or space-based surveys would
have mapped virtually the same surface, not a surface
reworked for development purposes. DEMs differ by up
to 3.0 m for the Riparian swath and 4.0 m for the Commercial
swath. The higher elevations reported by IfSAR
and SRTM for the Commercial swath are consistent with
buildings that would reflect radar, while lower elevations
reported by IfSAR and SRTM for the Riparian swath
are difficult to explain particularly considering that vegetation
would tend to push the radar-based DEMs even
higher. It is been reported that basin elevations can be
Fig. 1. Map of Santa Clara River and Buffalo Bayou study sites selected for DEM and flood model analyses. Santa
Clara River flows west from the
Castaic Junction study site (rectangle) to the Pacific Ocean at Ventura. Buffalo Bayou flows east from the Houston
study site (rectangle) towards
Galveston Bay and the Gulf of Mexico.
overestimated by NED [13], but the Pasture and Commercial

swaths are inconsistent with this possibility. Overall,
this comparison shows that elevations along the basin of
the Santa Clara Valley differ by up to 4 m based on the
DEM. This should serve to caution flood modelers
because much smaller flood depths are common.
Now, to examine the sensitivity of flood predictions to
these DEMs, BreZo was configured to simulate SCR flooding
from the 1928 St. Francis Dam failure [6,26]. St. Francis
Dam was a curved concrete dam 57 m high, 213 m long
which stored 47 million cubic meters of water. The dam
failed on March 12, 1928, when the reservoir was first filled
Fig. 2. Representation of Santa Clara River terrain near Castaic Junction by various DEMs. The x and y directions
correspond to UTM Easting and
Northing, respectively. (A) 1/9 s IfSAR; (B) 1/3 s NED; (C) 1 s NED; (D) 1 s SRTM and (E) 3 s SRTM.
to its crest. A wave of water then traveled 87 km to the

Pacific Ocean, first south along San Francisquito Canyon,
then west along Santa Clara Valley towards Ventura.
There were at least 500 casualties and economic losses at
the time were USD 1025 million.
A computational grid was prepared to span the path of
the flood between the reservoir and the ocean, and a grid
resolution of 1520 m was used between the reservoir and
Castaic Junction. Downstream of Castaic Junction, a relatively
coarse resolution was used for computational efficiency.
The grid consisted of 84,332 nodes and 167,943
cells. Five simulations were completed corresponding to
the following DEMs: (A) 1/9 s IfSAR, (B) 1/3 s NED,
(C) 1 s NED, (D) 1 s SRTM and (E) 3 s SRTM. In each
case, elevation was assigned to the vertices of the mesh
by linear interpolation and no other processing or filtering
of the DEM was completed.
Each simulation used a time step of 0.1 s and a total of
54,000 steps. The 90 min simulation period was long

enough for peak flood conditions to pass Castaic Junction,
but not reach the ocean. The flood reached the ocean 5 h
and 20 min after failure of the dam [26]. Using 1/3 s
NED and a spatially uniform Manning coefficient n, Begnudelli
and Sanders [6] report that model predictions of
St. Francis dam-break flooding are most accurate with
n = 0.02. This value was used in all cases considered here,
though more accurate predictions might be achieved with
a spatially distributed parameter. The maximum depth
and the flood travel time, defined by the elapsed time
between dam failure and initial wetting, were recorded during
each simulation and used to evaluate the sensitivity of
flood predictions to DEMs. Further, model predictions of
the flood zone were compared to a dam-break flood map
developed from aerial photographs of the flood in its aftermath
[31].
Fig. 3 shows contours of the maximum depth and travel
time for cases (AE), along with curves corresponding to
previously reported flood limits [31]. There are clearly local
differences in depth predictions, but flood extent is remarkably
similar. Table 3 presents flood extent and average
flood depth (i.e., spatial average of the maximum depth
recorded during the flood) for two sub-domains: San Francisquito
Canyon and Santa Clara Valley. The flood was
bounded by the steep canyon walls within San Francisquito
Canyon but Santa Clara Valley is broad by comparison.
The results (Table 3) show that all DEMs give nearly the
same flooded area for San Francisquito Canyon, with the
exception of 3 s SRTM which yielded a 4% smaller area.
According to the flood map [31], the DEMs under predict
flooded area in San Francisquito Canyon by 1518%. A
larger spread in model predictions is seen in Santa Clara
Valley (Table 3), where flooded area differs by at most
12%. Predictions all fall within 5% of the reported flood
extent, but it is clear from Fig. 3 that the boundaries of
the predicted and observed flood zone are not in alignment.
It is difficult to argue that one particular prediction compares
best with the flood map.
Average flood heights in San Francisquito Canyon (Table
3) are predicted to be roughly 18 m with all DEMs except
1/9 s IfSAR, which predicted a 17% higher flood height.
Within Santa Clara Valley, flood height was more sensitive
to DEM and again the largest resulted from 1/9 s IfSAR,
30% greater than the smallest flood height resulting from
1 s NED. Flood travel time predicted by 1/9 s IfSAR was
also longest, nearly 50% longer than the shortest travel time
resulting from 1 sNED. Travel times reported in Table 3 correspond
to the point labeled P in Fig. 3.
For a second application, BreZo was configured to predict
steady state conditions resulting from an extreme flood
with discharge of 1,950 m3/s, the largest flood on record for
USGS gage 11108500 located along SCR just west of the
study site. The 8.7 km flood path was gridded with a resolution
of roughly 6 m, resulting in 92,354 nodes and
184,069 cells. Further, resistance was modeled by a spatially

uniform Manning coefficient of 0.04. The downstream
cross-section was modeled as an open, nonreflecting
boundary [32]. A free-slip wall was placed across
the upstream boundary of the domain, and inflow to channel
was accomplished with a point source placed just inside.
As before, five different simulations were performed corresponding
to the DEMs previously labeled (AE) for the
dam-break test case. For each simulation, a time step of
0.05 s and a total of 100,000 time steps were used.
Fig. 4 shows contours of depth predicted for cases (B
E). Case (A) did not yield results because the IfSAR
DSM resolves the top of the I-5 crossing of SCR, effectively
creating a dam where none exists in reality. Additional processing
of the IfSAR DEM is clearly needed to remove
these features and enable a more realistic simulation. A
few landmarks have been added to each contour plot for
referencing purposes including I-5 and the junctions of
San Francisquito Creek and Castaic Creek. The inundation
Table 2
Properties of elevation data in selected swaths
Mean Elev. (m)a 1/9 s IfSAR 1/3 s NED 1 s NED 1 s SRTM 3 s SRTM
Pasture _0.91 0.00 0.15 _0.02 0.22
Riparian _2.26 0.00 0.30 _4.08 _1.61
Commercial 3.05 0.00 0.05 2.89 2.03
Mountain _1.68 0.00 _0.85 _4.69 _8.79
Old Residential _2.06 0.00 _0.81 _3.59 _5.65
New Residential _2.89 0.00 _0.76 _7.49 _10.31
a Relative to 1/3 s NED.
maps are surprisingly similar given DEM resolution varies

from 1/3 to 3 s (1090 m), and it is particularly surprising
how well 3 s SRTM based predictions compare to predictions
based on higher resolution data. Table 4 shows
flooded area, average flood depth, and average flood width
for cases (BE). Flooded area differs by at most 25%
between simulations, with 1/3 s NED predicting the smallest
area and 3 s SRTM predicting the largest area. Further,
flood depths differ by at most 12% with 1/3 s NED predicting
the smallest and 1 s SRTM predicting the largest. In
places the flood path is predicted to be less than 200 m wide
and while there are at least 20 computational cells spanning
the cross-section, its shape is crudely defined by only three
or four SRTM data points.
Upstream of I-5, there is a channel constriction labeled
P in Fig. 4 corresponding to a surface road that crosses
Fig. 3. Contours of maximum flood depth (left) and flood travel times (right) predicted by BreZo for St. Francis dambreak test case. The flood arrived
from the north and then progressed west. Solid curves denote observed flood extent [31]. (A) 1/9 s IfSAR; (B) 1/3 s
NED; (C) 1 s NED; (D) 1 s SRTM and
(E) 3 s SRTM.
SCR. This and the I-5 crossing are well resolved by 1/3 s
NED and visible in 1 s NED, but poorly resolved by 1
and 3 s SRTM. Fig. 4 shows what happens when such
hydraulically important features are not included in the
geometric representation of the channel. The model predicts
a relatively uniform-width flood path using SRTM,
while NED shows more realistically the regions where a

narrow and broad flood path are to be expected. The consequences
of radar speckle relative to flood modeling is also
evident in Fig. 4. The flood zone appears as an assemblage
of lobes or pools in contrast to NED predictions.
5. Application to Buffalo Bayou
To further examine the utility of on-line topographic data
for flood inundation modeling, Houston, Texas was selected
as a second test site. Houston is relatively wet, flat and highly
vegetated compared to the SCR site in southern California
which is relatively dry, mountainous and sparsely vegetated.
Houston was selected because 1/9 s NED (LiDAR data) are
accessible, along with 1 and 1/3 s NED as well as 1 and 3 s
SRTM data. Houston is also an attractive test site because
comprehensive flood analyses have recently been completed
as part of the Tropical Storm Allison Recovery Project
(TSARP) and results of these analyses, including flood flows
and Federal Emergency Management Agency (FEMA)
Flood Information Rate Maps (FIRMs) can be readily
obtained [37]. Used in the USA, FIRMs depict terrain susceptible
to inundation by events such as 50, 100, and 500 year
floods where a T-year flood corresponds to a 1/T annual
exceedance probability. Tropical Storm Allison struck the
Table 3
Comparison of St. Francis dam-break flood predictions
Flooded area (106 m2) Avg. depth (m) Travel time to P (min)
SFCa SCVb SFC SCV
(A) 1/9 s IfSAR 5.0 13.7 21.2 4.0 70.1
(B) 1/3 s NED 5.0 13.1 17.9 3.1 48.2
(C) 1 s NED 5.0 12.6 17.4 3.1 47.4
(D) 1 s SRTM 5.0 12.2 18.3 3.6 57.0
(E) 3 s SRTM 4.8 12.3 18.1 3.4 49.9
a San Francisquito Canyon flooded area estimated from aerial imagery as 5.9 103 m2 [31].
b Santa Clara Valley flooded area estimated as 13.0 103 m2 [31]. The location of P is shown in Fig. 3.
Fig. 4. Contours of Santa Clara River flood depth predicted by BreZo using NED and SRTM datasets. Q = 1950 m3/s
specified at upstream (East)
boundary. Shown also are Castaic Creek (CC) and San Francisquito Creek (SFC) junctions, Interstate-5 (I-5) and point
P which marks a constriction in
the river.
Table 4
Comparison of Santa Clara River extreme flood predictions
Flooded area
(106 m2)
Avg. depth
(m)
Avg. widtha
(m)
(A) 1/9 s
IfSAR
(B) 1/3 s NED 2.51 2.44 288

(C) 1 s NED 2.52 2.46 290
(D) 1 s SRTM 2.94 2.73 338
(E) 3 s SRTM 3.12 2.48 359
a Defined as flooded area divided by the reach length of 8711 m.
Houston area in June, 2001 causing over USD 5 billion in

damage and 22 fatalities.
An 8 km reach of Buffalo Bayou near downtown Houston,
shown in Fig. 1, was selected for analysis and modeling.
Fig. 5 shows topography depicted by (A) 1/9 sNED(derived
from LiDAR), (B) 1/3 s NED, (C) 1 s NED, (D) 1 s SRTM
and (E) 3 s SRTM. The channel is clearly visible in (AC),

but poorly resolved by (D) and undetectable in (E). The discrepancy
between (C) and (D) is interesting because both are
1 s resolution, yet one clearly achieves a superior depiction of
terrain. This is probably a consequence of radar shadow
within the deep and narrow bayou [29,28].
BreZo was configured to predict the 100 and 500-year
flood plains using (A) 1/9 s NED (derived from LiDAR),
(B) 1/3 s NED and (C) 1 s NED. SRTM was found to be
inadequate for flood modeling at this site. The flood path
Fig. 5. Representation of Buffalo Bayou near Houston using various DEMs. The x and y directions correspond to UTM
Easting and Northing,
respectively. (A) 1/9 s NED (LiDAR); (B) 1/3 s NED; (C) 1 s NED; (D) 1 s SRTM and (E) 3 s SRTM.
was resolved by a roughly 10 m resolution mesh consisting

of 183,439 cells and 92,086 nodes. The volumetric discharge
was obtained from flood frequency analyses of
USGS Gage 08074000 [27] and specified at the upstream
boundary. The water level at the downstream boundary
of the reach was obtained from the FEMA FIRM and used
as a boundary condition [37]. The values of the discharge
and water level were 437 m3/s and 11.28 m, respectively,
for the 100 year event and 520 m3/s and 13.11 m, respectively,
for the 500 year event. A uniform Manning coefficient
of 0.06 was arbitrarily selected, without knowledge
of the value used to prepare the FIRM. Starting from a still
water initial condition defined by the downstream boundary
condition, BreZo was integrated 50,000 time steps
using a time step of 0.1 s. This duration was found to be
long enough to converge within 0.1% of the steady state
water level. A description of the procedure used to create
the FEMA FIRM could not be identified from [37], but
in the USA it is common to use 1D flood models such as
HEC-RAS [15] to predict flood stage, and to map inundation
by extrapolating channel centerline flood stage predictions
to overbank areas.
Fig. 6a presents BreZo 500 year flood zone prediction
(in blue2) based on 1/9 s NED superimposed upon the
FEMA FIRM (shaded gray and outlined in black). The
two compare well. The agreement on the eastern end of
the domain indicates that the boundary condition used
by the model is consistent with the FIRM. On the other
hand, the agreement on the western end shows that the
2D modeling approach used here and the 1D approach
most likely used to create the FIRM predict nearly equal
flood heights. The 100 year BreZo prediction based on
1/9 s NED compares equally well with the FIRM, and
for this reason is not shown. Fig. 6a also shows a polyline
that traces the centerline of the channel, and flood stage
along this path was extracted from the 2D prediction and
Fig. 6. (A) Buffalo Bayou 500-year flood extent predicted by BreZo using 1/9 s NED, superimposed on 100- and 500year FEMA FIRMs. Distances in
meters are marked at twelve points along the channel centerline. (B) Predictions of flood stage for 500 year flood
along channel centerline. (C) Predictions
of flood stage for 100 year flood.
2 For interpretation of color in Fig. 6, the reader is referred to the web
version of this article.
is shown in Fig. 6b and c. Note that the 1/3 and 1 s NED

predictions closely track each other, while 1/9 s NED predictions
deviate from these by as much as 0.06 m, or about
12% of the reach-scale stage difference of 0.5 m based on
the 500 year event. Table 5 shows flooded area, average
depth, and average width of the flood zone for all predictions.
Based on the 500 year event, flooded area varies by
up to 18% depending on the DEM and flood depth varies
by up to 13%. Smaller differences are associated with the
100 year event, where flooded area varies by 5% and flood
depth varies by 4%. It is interesting that predictions based
on 1/9 s NED imply the largest flood zone, even though
1/9 s NED predictions lower flood heights along most of
the reach (Fig. 6b and c). This is due to relatively lower
overbank heights being reported by 1/9 s NED. Fig. 7
shows the flood zone based on 1/9, 1/3 and 1 s NED. Note
how much better the prediction based on 1/9 s NED tracks
the FEMA FIRM, compared to the 1/3 and 1 s NED based
predictions. Contours of velocity are also shown to highlight
the multidimensional aspects of channel flow that
are resolved by the model.
6. Discussion
The preceding results show remarkable value in relatively
coarse DEMs that can be accessed on-line, one of which has
global coverage (3 s SRTM). In the unsteady-flow St. Francis
application, 3 s SRTM yielded a flood zone just 12% larTable 5
Comparison of Buffalo Bayou extreme flood predictions
Flooded area (106 m2) Avg. depth (m) Avg. widtha (m)
100 yearb 500 year 100 year 500 year 100 year 500 year
(A) 1/9 s LiDAR 2.12 3.04 4.98 4.99 275 394
(B) 1/3 s NED 2.01 2.58 5.18 5.62 260 334
(C) 1 s NED 2.03 2.64 4.98 5.39 264 343
a Defined as the flooded area divided by the reach length of 7705 m.
b Return period of flood.
Fig. 7. BreZo prediction of 500 year flood extent at western end of reach using (A) 1/9 s NED, (B) 1/3 s NED, and (C)
1 s NED. Contours show velocity
magnitude (m/s) to highlight differences in flood predictions.
ger than NED predictions and flood travel time predictions

less than 6% longer. In the steady-flow Santa Clara River
application, 3 s SRTM yielded a 25% larger flood zone.
Whether or not this is a tolerable level of uncertainty will
surely depend on the application. In developed countries,
the added precision of high resolution surveys such as
LiDAR will likely justify the cost. On the other hand, 3 s
SRTM may appear quite attractive in cases where LiDAR
or IfSAR surveys are cost-prohibitive and there are other,
larger factors of uncertainty. It is advantageous for flood
modeling that SRTM vertical accuracy is better on relatively
flat terrain, such as flood plains, compared with high relief
area [3,13,14]. Filters to remove radar speckle could further
enhance the utility of SRTM for flood modeling purposes,
for these DEM errors give rise to non-physical pools within
flood zones as shown in Figs. 3 and 4.
Another important question is whether 1 s SRTM is any
better for flood modeling than 3 s SRTM. Coverage of the
former is limited to USA, so any flood simulation benefit

that it offers has limited impact particularly considering
that NED covers the same terrain. Results of the unsteady
St. Francis and steady-state Santa Clara River applications
show that 1 s SRTM predictions compare to NED predictions
as well or better than 3 s SRTM predictions. Hence,
there appears to be an advantage even though the true resolution
of 1 s SRTM is about 2 s [34]. One caveat is that
travel time predictions based on 1 s SRTM did not agree
with NED as well as 3 s SRTM.
Longer travel times are an indication of greater flow
resistance. The longest travel time resulted from using the
IfSAR DEM, which is sensitive to small length-scale terrain
features such as built structures and trees. These features
have been called mega-roughness (A. Valiani,
private communication) based on the well-known linkage
between surface roughness and flow resistance. Flood inundation
models generally rely on two sets of parameters, one
which defines geometry (DEM) and another that scales
flow resistance [22]. However, these results and others have
shown that the DEM also affects resistance [25]. Therefore,
to demonstrate numerical convergence modelers should
adopt a refinement process that does not add additional
length scales to the terrain description.
When using on-line DEMs for flood modeling, there are
several limitations to consider. No DEM can resolve channels
smaller in width than twice the DEM resolution. Three
survey points, one in the middle of the channel and one on
each bank, is the bare minimum for a crude definition of
the cross-section. When channels are not resolved, models
are likely to overestimate flood extent and underestimate
flood speeds. It is good practice to perform ground surveys
of study sites when DEMs of questionable accuracy are utilized.
Ground survey data can be used to estimate DEM
accuracy or can even support additional data processing
by providing a basis for ground-truthing the DEM. Secondly,
the suitability of a DEM for flood modeling may
depend on flood frequency. Relatively frequent floods (2
10 year events) could require greater DEM resolution than
infrequent floods (100500 year events) or dam-break
floods. That is, the highest resolution may be required to
resolve small channels that bound frequent floods, while
such features are less important when completely submerged
by infrequent floods. Third, success in predicting
flood extent should not be mistaken for locally accurate
predictions of water depth and velocity which can only
be captured in a spatially averaged sense at scales larger
than the DEM resolution.
7. Conclusions
LiDAR-based DEMs represent the best source of terrain
data for flood modeling due to their horizontal resolution
and vertical accuracy, which is _0.1 m. The ability to
detect bare earth as well as vegetation and building heights
is also appealing. 1/9 s NED available from the USGS
achieve a good depiction of ground heights, i.e., buildings
and vegetation are effectively stripped away, so this DEM

is a good DTM.
Airborne IfSAR based DEMs available from NOAA/
CSC or commercially (e.g., from Intermap Corp.) have
excellent horizontal resolution (1/9 s) so features such as
buildings and bridges are resolved, but these are truly
DSMs and therefore further processing is required to
depict bare-earth elevation in urban and vegetated areas.
In the steady-state Santa Clara River application, flood
simulation with airborne IfSAR was not possible because
Interstate-5 was resolved by the DSM and acted like a
dam across the river. IfSAR DEMs including SRTM
may also be contaminated by radar speckle which adds
noise or undulations to otherwise flat surfaces. In the context
of a flood simulation, this noise may make the flood
zone appear as a network of pools.
In addition to errors from radar speckle, DEMs based
on 1 and 3 s STRM may have data gaps caused, for example,
by radar shadow which make flood modeling impossible.
The Buffalo Bayou application exemplifies SRTM
failure to capture a channel 300 m wide. On the other hand,
the Santa Clara River application highlighted remarkable
value in SRTM data. Santa Clara River was only 200 m
wide in places (_7 s) and SRTM data yielded flood zone
predictions only 25% different than NED. Further, predictions
based on 1 s SRTM yielded only slightly more accurate
flood zones than predictions based on 3 s SRTM.
DEMs based on 1/3 and 1 s NED are truly DTMs and
map terrain with remarkable smoothness but vertical accuracy
can be poor. NED were processed with hydrologic
modeling in mind, e.g. local minima are removed, and this
leads to a good representation of channel and overland
flow with numerical models. But using 1/3 and 1 s NED,
flood zones are predicted to be as much as 25% smaller
than flood zones predicted with other DEMs such as
LiDAR, IfSAR, and SRTM. Therefore, these DEMs
should be used cautiously in the context of flood zone mapping
for they may cause a systematic underestimation of
flood risks.
Acknowledgments
This study was made possible by a Grant from the National
Science Foundation Multi-Disciplinary Center for
Earthquake Engineering Research (MCEER) and by research
support provided by the University of Nottingham
School of Civil Engineering, where the first author spent
his sabbatical during the 20062007 academic year. The
authors thank H. Morvan, M. Smith and N. Wright (U.
Nottingham), P. Bates (U. Bristol), J. Carter (NOAA/
CSC) and D. Finnegan (USACE/CRRE) for discussions
and assistance that contributed to this study.
References
[1] Baltavias EP. Airborne laser scanning: basic relations and formulas.
ISPRS J Photogramm 1999;54:199214.
[2] Bates PD, Marks KJ, Horritt MS. Optimal use of high-resolution
topographic data in flood inundation models. Hydrol Process
2003;17:53757.
[3] Berthier E, Arnaud Y, Vincent C, Remy F. Biases of SRTM in highmountain
areas: implications for the monitoring of glacier volume
changes. Geophys Res Lett 2006;33:L08502. doi:10.1029/
2006GL025862.
[4] Begnudelli L, Sanders BF. Unstructured grid finite volume algorithm
for shallow-water flow and transport with wetting and drying. J
Hydraul Eng 2006;132(4):37184.
[5] Begnudelli L, Sanders BF. Conservative wetting and drying methodology
for quadrilateral grid finite volume models. J Hydraul Eng
2006;133(3).
[6] Begnudelli L, Sanders BF. Simulation of the St. Francis dam-break
flood. J Eng Mech, in review.
[7] Bradford SF, Sanders BF. Finite-volume model for shallow-water
flooding of arbitrary topography. J Hydraul Eng 2002;128(3):28998.
[8] Bradford SF, Sanders BF. Performance of high-resolution, non-level
bed, shallow-water models. J Eng Mech 2005;131(10):107381.
[9] Cobby DM, Mason DC, Davenport IJ. Image processing of airborne
scanning laser altimetry data for improved river flood modeling.
ISPRS J Photogramm 2001;56:12138.
[10] Cobby DM, Mason DC, Horritt MS, Bates PD. Two-dimensional
hydraulic flood modelling using a finite-element mesh decomposed
according to vegetation and topographic features derived from
airborne scanning laser altimetry. Hydrol Process 2003;17:19792000.
[11] Dartmouth Flood Observatory http://www.dartmouth.edu/
~floods.
[12] Evans E, Hall J, Penning-Roswell E, Sayers P, Thorne C, Watkinson
A. Future flood risk management in the UK. Proc Inst Civ Eng
Water Management 2006;159(WM1):5361.
[13] Falorni G, Teles V, Vivoni ER, Bras RL, Amaratunga KS. Analysis
and characterization of the vertical accuracy of digital elevation odels
from Shuttle Radar Topography Mission. J Geophys Res
2005;110:F02005. doi:10.1029/2003JF000113.
[14] Guth PL. Geomorphometry from SRTM: comparison to NED.
Photogramm Eng Rem S 2006;72(3):26977.
[15] HEC-RAS software is described and accessed at www.hec.usace.
army.mil.
[16] Hodgson ME, Jensen JR, Schmidt L, Schill S, Davis B. An evaluation
of LIDAR- and IFSAR-derived digital elevation models in leaf-on
conditions with USGS Level 1 and Level 2 DEMs. Remote Sens
Environ 2003;84:295308.
[17] Hogg AJ, Pritchard D. The effects of hydraulic resistance on dambreak
and other shallow inertial flows. J FluidMech 2004;501:179212.
[18] Horritt MS, Bates PD. Evaluation of 1D and 2D numerical
models for predicting river flood inundation. J Hydrol 2002;268:
8799.
[19] Intermap Corp. Product handbook and quick start guide. Version
3.3. June 15; 2004.
[20] Katopodes ND, Strelkoff T. Computing two-dimensional dam-break
flood waves. J Hydraul Div ASCE 1978;104(HY9):126988.
[21] Lin B, Wicks JM, Falconer RA, Adams K. Integration of 1D and 2D
hydrodynamic models for flood simulation. Proc Inst Civ Eng
Water Management 2006;159(WM1):1926.
[22] Marks K, Bates P. Integration of high-resolution topographic
data with floodplain flow models. Hydrol Process 2000;14:
210922.
[23] Mason DC, Cobby DM, Horritt MS, Bates PD. Floodplain friction
parameterization in two-dimensional river flood models using vegetation
heights derived from airborne scanning laser altimetry. Hydrol
Process 2003;17:171132.
[24] Morris MW. CADAM concerted action on dambreak modelling.
Report SR 571, HR Wallingford; 2000.
[25] Mrse RD. Impact of resolved bed forms on model predictions of flow
resistance, M.S. Thesis, Department of Civil and Environmental
Engineering, University of California, Irvine; 2006.
[26] Outland CF. Man-made disaster: the story of St. Francis Dam. The
Arthur H. Clark Company; 1963. p. 249.
[27] PBS&J, LJA Engineering and Land Surveying. Flood flow frequency
analysis for tropical storm allison recovery project, Final Report,
Prepared for Harris County Flood Control District and Federal

Emergency Management Agency, October; 2002. Available on-line at
www.tsarp.org.
[28] Rabus B, Eineder M, Roth A, Bamler R. The shuttle radar
topography mission a new class of digital elevation models
acquired by spaceborne radar. ISPRS J Photogramm 2003;57:
24162.
[29] Rodriguez E, Martin JM. Theory and design of interferometric
synthetic aperature radars. IEE Proc-F 1992;139(2).
[30] Rodriguez E. A global assessment of the SRTM performance.
Photogramm Eng Rem S 2006;72(3):2618.
[31] Rogers JD, James K. Mapping the St. Francis dam outburst flood
with geographical information systems. Presentation given at the 46th
annual meeting of the association of engineering geologists, Vail CO
2003, Program with Abstracts; 2003. p. 72 http://web.umr.edu/
~rogersda/st_francis_dam.
[32] Sanders BF. Non-reflecting boundary flux function for finite volume
shallow-water models. Adv Water Resour 2002;25:195202.
[33] Shewchuk JR. Triangle: engineering a 2D quality mesh generator and
Delaunay triangulator. In: Lin MC, Manocha D, editors. Applied
computational geometry: towards geometric engineering. Lecture
Notes in Computer Science, vol. 1148. Springer; 1996. p. 20322.
Software may be obtained at http://www-2.cs.cmu.edu/
~quake/triangle.html.
[34] Smith B, Sandwell D. Accuracy and resolution of shuttle radar
topography mission data. Geophys Res Lett 2003;30(9):1467.
doi:10.1029/2002GL016643.
[35] Smith MJ, Edwards EP, Priestnall G, Bates PD. Exploitation of new
data types to create digital surface models for flood inundation
modeling, FRMRC Research Report UR3, FRMRC, UK; 2006.
Report available on-line at www.floodrisk.org.uk.
[36] Toro EF. Shock-capturing methods for free-surface shallow
flows. Chichester, UK: Wiley; 2001.
[37] Tropical Storm Allison Recovery Project. http://www.tsarp.org.
[38] van Zyl JJ. The Shuttle Radar Topography Mission (SRTM): a
breakthrough in remote sensing of topography. Acta Astronaut
2001;48(5):512.
[39] Wehr A, Lohr W. Airborne laser scanning-an introduction and
overview. ISPRS J Photogram 1999;54:6882.
Flood mapping in ungauged basins using fully

continuous hydrologichydraulic modeling
ARTICLE in JOURNAL OF HYDROLOGY APRIL 2013
Impact Factor: 3.05 DOI: 10.1016/j.jhydrol.2013.02.023
CITATIONS
16
READS
145
4 AUTHORS, INCLUDING:
Salvatore Grimaldi
Tuscia University

SEE PROFILE
A. Petroselli
Tuscia University
49 PUBLICATIONS 359 CITATIONS

SEE PROFILE
Fernando Nardi
Universit per Stranieri di Perugia

SEE PROFILE
Available from: Salvatore Grimaldi
Retrieved on: 30 November 2015
Flood mapping in ungauged basins using fully continuous

hydrologichydraulic modeling
Salvatore Grimaldi
a,b,c,,
Andrea Petroselli d, Ettore Arcangeletti d, Fernando Nardi
Dipartimento per la innovazione nei sistemi biologici, agroalimentari e forestali (DIBAF Department), University of Tuscia, Via San Camillo De
Lellis snc, 01100 Viterbo, Italy
b Honors Center of Italian Universities (H2CU), Sapienza University of Rome, Via Eudossiana 18, 00184 Roma, Italy
c Department of Mechanical and Aerospace Engineering, Polytechnic Institute of New York University, Six MetroTech Center, Brooklyn, NY
11201, United States
d Dipartimento di scienze e tecnologie per lagricoltura, le foreste, la natura e lenergia (DAFNE Department), University of Tuscia, Via San
Camillo De Lellis snc, 01100 Viterbo, Italy
e WARREDOC, Universit per Stranieri di Perugia, Piazza Fortebraccio 4, 06123 Perugia, Italy
a
articleinfo
Article history:
Received 15 August 2012
Received in revised form 8 February 2013
Accepted 14 February 2013
Available online 26 February 2013
This manuscript was handled by Andras
Bardossy, Editor-in-Chief, with the
assistance of Niko Verhoest, Associate Editor
Keywords:
Flood hazard
Flood mapping
Continuous hydrologic model
2D hydraulic model
WFIUH
summary
In this work, a fully-continuous hydrologichydraulic modeling framework for flood mapping is introduced
and tested. It is characterized by a simulation of a long rainfall time series at sub-daily resolution
that feeds a continuous rainfallrunoff model producing a discharge time series that is directly given as
an input to a bi-dimensional hydraulic model. The main advantage of the proposed approach is to avoid
the use of the design hyetograph and the design hydrograph that constitute the main source of subjective
analysis and uncertainty for standard methods. The proposed procedure is optimized for small and ungauged
watersheds where empirical models are commonly applied. Results of a simple real case study confirm
that this experimental fully-continuous framework may pave the way for the implementation of a
less subjective and potentially automated procedure for flood hazard mapping.
_ 2013 Elsevier B.V. All rights reserved.
1. Introduction
Flood hazard mapping is a fundamental non-structural measure
implemented by governments for sustainable urban and land planning,
protecting human properties, activities, lives, and preserving
the ecohydrology of river corridors (European Parliament, 2007).
While hydrogeomorphic methods, that make extensive use of Digital
Elevation Models (DEMs) and terrain analysis have been recently
adopted for preliminary floodplain characterization at the
basin scale (e.g. Dodov and Foufoula-Georgiou, 2006; Nardi et al.,
2006), detailed inundation maps are generally based on advanced
hydrologic and hydraulic modeling. In particular, the hydrologic
forcing is estimated through a rainfallrunoff model that quantifies
flood peak discharges or a flow hydrograph of given return period
(T). The hydraulic analysis is, developed, by means of a mono
dimensional (1D) surface water model in a steady state condition
and/or by means of an unsteady 1D or bidimensional (2D) flood
routing algorithm for the accurate spatially distributed evaluation
of the flow and velocity dynamics.
The more complex and detailed 2D models suffered in the past
of limited amounts of data and of the computational issue. However,
latest 2D models, overcoming these limitations, are now
increasingly used as the new standard approach for accurately simulating
spatial and temporal dynamics of the flooding process (e.g.
Horritt et al., 2007; Di Baldassarre et al., 2009). As a result, 2D simulations
represent a mandatory choice for detailed small scale urban
development projects as well as for large scale ones, especially
for complex urban settings where the 1D hypothesis is often not
applicable and the socio-economic impact of floods must also be
assessed (European Parliament, 2007).
Nevertheless, current hydrologichydraulic modeling procedure
for flood mapping is characterized by three main issues: (1)

the availability of detailed topographic information; (2) choosing
the flood propagation hydraulic model (3) the impact that the
hydrologic forcing has on flood mapping results (i.e. design hydrograph
estimation). The first two issues benefit of the latest technological
advancements, in particular of the increasing accuracy,
availability, and usability of high resolution Lidar DEMs and 2D
hydraulic models (e.g. Gilles et al., 2012; Sampson et al., 2012).
This work investigates the third issue for small and ungauged ba0022-1694/$ - see front matter _ 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jhydrol.2013.02.023
Corresponding author at: Dipartimento per la innovazione nei sistemi biologici,
agroalimentari e forestali (DIBAF Department), University of Tuscia, Via San Camillo
De Lellis snc, 01100 Viterbo, Italy. Tel.: +39 0761 357326; fax: +39 0761 357453.
E-mail address: salvatore.grimaldi@unitus.it (S. Grimaldi).
Journal of Hydrology 487 (2013) 3947
Contents lists available at SciVerse ScienceDirect
Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol

sin, comparing event-based approaches with respect to new experimental
procedures based on the implementation of continuous
rainfallrunoff methods.
For event-based approach (EBA) we refer to those schemes that
implement a rainfallrunoff transformation for estimating the design
hydrograph once given the design hyetograph of assigned return
period, (Alfieri et al., 2008; Hsieh et al., 2006; Laforce et al.,
2011; Soczyska et al., 1997). EBA procedures are widely adopted
in engineering practice for their simplicity, even if some significant
drawbacks must be accounted for, and in particular: the predefined
hyetograph shape and/or estimation method, the critical rainfall
duration parameter, the difficulty in assigning the pre-event soil
moisture conditions, and the pivotal hypothesis of assigning the
same T value for design rainfall to the corresponding hydrograph
(Rahman et al., 2002; Verhoest et al., 2010).
As a result, in recent times, the use of various continuous approaches
has been investigated (Blazkova and Beven, 2009; Calver
et al., 2009; Grimaldi et al., 2012b, in press-a; Moretti and Montanari,
2008). In those schemes, a runoff time series is obtained
through the implementation of a continuous rainfallrunoff transformation.
The hydrologic input is a long synthetic or observed
rainfall time series. The corresponding simulated runoff signal is
then statistically analyzed for extrapolating a Synthetic Design
Hydrograph (SDH) related to a return period (Pramanik et al.,
2010; Serinaldi and Grimaldi, 2011). In the current work, this approach
is referred to as the Semi-Continuous Approach (SCA).
SCA is characterized by the fact that statistical analysis is not performed
during the rainfall estimation step (as for the EBA), but on
the derived simulated hydrographs.
Several works evaluated the EBA and SCA performances in
terms of hydrograph properties and uncertainty of results. Statistical
and comparative analyses of the hydrograph peak discharge,
volume, and duration suggest that EBA tend to underestimate the
flood volume (Grimaldi et al., 2005, 2012b; Nishat et al., 2010;
Nnadi et al., 1999; Viglione and Bloschl, 2009; Viglione et al.,
2009). In particular, several researchers demonstrated the uncertainty
that is induced through a selection of subjective parameters
and the simplification of a physical hydrologic and hydraulic process
that indirectly influences the flood inundation mapping procedure
(e.g. Aronica et al., 2002; Bates et al., 2004; Di Baldassarre
et al., 2010; Hall et al., 2005; Kalyanapu et al., 2012; McMillan
and Brasington, 2008; Merwade et al., 2008; Pappenberger et al.,
2006).
In the following, we introduce a Fully Continuous Approach
(FCA) and we test its potential to diminish the impact of subjective
preliminary steps in hydrologichydraulic modeling. This is
achieved mainly by skipping the SDH estimation, still present in
the SCA approach, and routing the entire simulated flow time series
along the floodway. FCA performances are then compared to
EBA and SCA by qualitatively analyzing inundation results.
The paper is organized as follows: after this brief introduction,

Section 2 provides a more detailed description of EBA, SCA and
FCA approaches. In Sections 3 and 4, the case study and the proposed
procedure application are presented, while in Section 5, results
are discussed followed by conclusive remarks of Section 6.
2. Event-based, semi-continuous and fully-continuous
approaches
The three methodologies, EBA, SCA and FCA, are described and
summarized in the flow diagram inserted in Fig. 1.
The described procedures are valid for small and ungauged basins.
However, the general concept and related application may be
extended to any watershed. The term small refers to basin for
which rainfall is assumed spatially uniform, while ungauged
means that flow measurements are not available even for regionalization.
For these small and ungauged basins, where only empirical
analytical procedures are applicable, we assume that the following
data are available: rainfall registration (at daily and sub-daily resolution),
a DEM of 2030 m resolution, and soils information.
The three approaches, depicted in Fig. 1, include five steps: (1)
rainfall input, (2) rainfall excess, (3) design hyetograph estimations,
(4) rainfallrunoff transformation, and (5) hydraulic flood
routing. A brief summary of steps 2, 4, and 5, shared by the three
procedures, is hereafter reported, while a more detailed description
of the different methodological steps 1 and 3 for EBA, SCA
and FCA is included in the next corresponding subsections.
The rainfall excess (step 2) is estimated using the Soil Conservation
Service Curve Number (SCS-CN) method (USDA, 1986).
While we are aware that this approach is not appropriate for a
sub-daily rainfall series (Grimaldi et al., in press-b, in press-c), it
is here applied because it does not influence the results of the comparison
tests.
The rainfallrunoff transformation (step 3) is implemented
using a parsimonious version of the Width Function Instantaneous
Unit Hydrograph (WFIUH), namely WFIUH-1par (Grimaldi et al.,
2012a) that optimizes the available DEM by means of advanced
and automated terrain analysis techniques. The procedure for performing
this step is summarized in the following three sub-steps:
(a) DEM pits and flat areas are removed using the Physicallybased
Erosion Model for PIT removal PEM4PIT (Grimaldi
et al., 2004, 2007; Santini et al., 2009).
(b) Flow directions are defined using a mixed flow direction
approach, combining an advanced single-flow approach in
the channel and a multi-flow approach on the hillslopes
(Nardi et al., 2008). The river network is then extracted using
the drop-analysis approach (Tarboton et al., 1991; Tarboton
and Ames, 2001).
(c) Given the drainage network, the WFIUH-1par is obtained by
estimating a residency time distribution for rainfall drops
flowing throughout the domain. Surface hillslope flow velocities
are estimated using the NRCS formula (Grimaldi et al.,
2010; NRCS, 1997), while channel flow velocity is calibrated
so that the WFIUH base time is equal to the basin concentration
time (Tc). Tc is estimated using the Giandotti empirical
formula widely used in Italy (Giandotti, 1934). The adopted
WFIUH-1par approach does not need observed data to be
calibrated (Grimaldi et al., 2012a).
The last step (5) of the flood routing is developed by implementing
the 2D commercial hydraulic model FLO-2D (http://
www.flo-2d.com/). The FLO-2D scheme is based on the dynamic
wave momentum equation solved on a numerical grid of square
cells through which the hydrograph is routed propagating the surface
flow along the eight cardinal directions (OBrien et al., 1993;
FLO-2D Reference Manual, 2012).
2.1. Event Based Approach (EBA)
Standard EBA is characterized by an empirical rainfall input
estimation procedure that is developed as follows. A critical rainfall
depth for the predefined return period is quantified using the
IntensityDurationFrequency (IDF) curve assuming the watershed
concentration time as duration. This critical rainfall depth
is distributed over time through a synthetic hyetograph. As previously

underlined, this design hyetograph characterization method
(Alfieri et al., 2008) is affected by four main issues: the concept of
critical rainfall for a given duration that tends to underestimate the
runoff volume, the uncertainty in the estimation of the time of concentration
(Grimaldi et al., 2012c), the design hyetograph itself
40 S. Grimaldi et al. / Journal of Hydrology 487 (2013) 3947
that is arbitrary and far from representing real rainfall storms, and
the hypothesis that the design hyetograph and corresponding hydrograph
have the same return period.
For estimating the IDF, we selected the three-parameter formula
(Chow et al., 1988):
id; T
aT
bT d_cT 1
where i(d,T) is the gross rainfall intensity, and d the rainfall duration.
a(T), b(T) and c(T) are determined through a nonlinear leastsquares
method. The i(d,T) is estimated on the annual maxima for
the corresponding durations of 1, 3, 6, 12 and 24 h according to
the empirical cumulative distribution of the Weibull formula:
j
n11_
1
T 2
where j = 1,. . . ,n represents the jth observation of the samples arranged
in ascending order.
The selected hyetograph estimation method is the symmetric
Chicago hyetograph (Keifer and Chu, 1957), that, given as an input
to the rainfallrunoff procedure (step 4), provides the corresponding
design hydrograph.
2.2. Semi Continuous Approach (SCA)
The above mentioned four issues of the EBA correspond to four
significant theoretical and conceptual hypotheses representing
sources of uncertainty and/or user-derived subjective choices
affecting the design hydrograph estimation procedure. The SCA
aims to reduce the impact of these conceptual hypotheses by
substituting the IDF-derived critical rainfall with a simulated rainfall
time series. This calibrates a statistical model on the observed
rainfall data and provides it to the rainfallrunoff model without
any prior estimation of a single design event. In this way, the impact
of the uncertainty of the IDF, critical rainfall, and design hyetograph
hypotheses is substituted by the need for the proper
selection of a rainfall simulation model. The return period is also
estimated directly on the simulated runoff series allowing it to
omit another source of uncertainty given by the assumption that
the design discharge preserves the rainfall time period.
Fig. 1. Flow-chart of the three flood modeling approaches: event-based, semi-continuous, and fully-continuous.
S. Grimaldi et al. / Journal of Hydrology 487 (2013) 3947 41
In this paper, steps 1 and 3 are developed applying the COSMO4SUB

framework (Grimaldi et al., 2012b) where the rainfall
simulation is performed using a copula-based daily rainfall generator
(Serinaldi, 2009) and a continuous-in-scale universal multifractal
model (Schertzer and Lovejoy, 1987) to disaggregate the
daily rainfall. The simulated rainfall time series is given as an input
to the runoff model applied in continuous using a predefined event
separation time parameter (Ts) that allows to separate the rainfall
events for the appropriate implementation of the SCS-CN scheme.
Selected the annual maximum hydrographs from the obtained runoff
time series, the SDHs are estimated using the peak discharge as
the driving variable for each return period following the procedure
described in Serinaldi and Grimaldi (2011) where the hydrograph
shape is assumed to be similar to the Generalized Standard TwoSided Power distribution.
2.3. Fully Continuous Approach (FCA)
The significant issue of the SDH definition for the SCA scheme is
that, even if the peak and volume information are generally preserved,
it tends to provide unrealistic hydrographs characterized
by symmetric or averaged shapes. This issue is here solved by
introducing a fully continuous rainfallrunoff algorithm that is
the proposed FCA that provides as input to the 2D hydraulic propagation

model a complete runoff time series (i.e. multi-event hydrograph)
rather than a single event SDH. In this way, the entire
physical process is represented in continuous from the hydrologic
forcing (i.e. rainfall) to the river flow routing reducing the subjectivity
and uncertainty of user-derived hypotheses. Moreover, a
more complete dataset is available for the inundation mapping
interpretation and in particular, a water depth time series (and
other hydraulic parameters like velocity, flow directions, and the
energy impact of the flood wave among others) for each node of
the floodplain domain.
The simulation of the entire runoff series is limited, today, by
computer hardware and memory for handling this enormous
amount of data. As a result, given the entire synthetic runoff time
series, the 2D hydraulic model is applied only on the n maximum
annual hydrographs that are selected here using the peak discharge
as driving variable and the duration as a continuous sequence
of no-zero runoff values. However, this does not affect
the validity and generalization of the procedure. The nth flood
map is derived by selecting the highest simulated water depths
for each node and the synthetic inundation surface is characterized
by water surface peaks related to different simulation time steps.
The final inundation map for each return period is the result of
the cell-by-cell statistical analysis of the n flood maps. In order
to estimate the corresponding return period, the n water depth values
for each cell are analyzed using the mixed Gumbel distribution
formula (Kedem et al., 1990; Yoo et al., 2005):
Gx 1 _ pHx pFx 3
where x is the random variable for the flow depth for each cell, p is
the probability of wet cell (x > 0), 1 _ p is the probability of a dry
event (x = 0), G(x) is the mixed distribution function considering
both wet and dry events, H(x) is the step function: H(x) = 0, if
x = 0; H(x) = 1, if x > 0, and F(x) is the probability density function
estimated for non-zero values.
3. The Rio Torbido case study
The Rio Torbido, a 61.67 Km2 river basin located in central Italy
draining directly into the main channel of the Tiber River, is selected
for this case study. Elevations range from 85 mto 625 m, the average
slope is 21.9% and the maximum hydrologic distance of the outlet
from the watershed divide is 25.8 Km. The selected hydraulic modeling
domain is located in the proximity of the confluence of the Tiber
where the Torbido meanders within a large floodplain. This area,
shown in Fig. 2, is 1.5 Km2 large and 2 Km long wide.
The DEM, with integer precision and 20 mof spatial resolution, is
extracted from the Italian Geographic Military Institute (IGMI, 2003)
MATRIX raster dataset. The IGMI Matrix DEM, derived from the
interpolation of 1:25.000 contour lines, is characterized by a spatial
and vertical accuracy of respectively 510 mand 15 m. Land cover
is derived from the CORINE project (European Commission, 2000).
Rainfall measurements include 49 years of observed data at a daily
time scale and 10 years at 5 min of resolution (Serinaldi, 2010).
Fig. 3 shows the Rio Torbido basin DEM with the simulated river
network implementing the terrain analysis procedure described in
Section 2. The detailed topography, bathymetry and roughness conditions
of the 2D hydraulic modeling project also including the surveyed
river cross sections (Fig. 2) were derived by integrating the
available digital data with the results of an extensive field campaign.
The final 2D flood model resolution is 20 m that is the size
of the grid cell representing the domain topography. Upstream
and downstream boundary conditions for the hydraulic model are
respectively the inflow hydrograph for the first channel grid element
and the uniform flow assigned to the boundary grid elements.
4. Comparison test
EBA, SCA and FCA are compared by means of flood maps that are
derived using the above described procedure. Although a case
study with available observations would be better representative
and would allow a quantitative FCA validation, the aim here is to
provide a qualitative evaluation of the differences among the three
procedures. The comparison is here proposed to visualize the impact
of the fully continuous approach that, while neglecting the

subjectivity affecting standard EBA procedure, show its potential
practical advantage for inundation mapping.
The conditions and parameters used in the method implementation
are the following:
The rainfall scenario is obtained simulating 500 years of a synthetic
rainfall time series at 5 min of resolution. The IDF presented
in the EBA is estimated directly on this rainfall
scenario using formula (1).
Properties and conditions of the simulated 500 years of 5 min
rainfall are reported in Grimaldi et al. (in press-a).
For the SCS-CN method, adopted in all three procedures for the
net rainfall estimation, the parameters are fixed to: k = 0.2 as
suggested in USDA-NRCS (2010), and CN = 74.6. This value is
estimated averaging the CN assigned to homogeneous subareas
of the basin referring to the CORINE soil classification.
The time of concentration, useful either for EBA (for design hyetograph
estimation and rainfallrunoff model application) and
for SCA and FCA (for only rainfallrunoff model application), is
estimated using the Giandotti formula and fixed to 4.5 h.
The separation time Ts, useful in SCA and FCA for continuous
application of rainfallrunoff transformation, is set to 24 h as
suggested in Grimaldi et al. (2012b).
Roughness conditions that is Manning values, for the hydraulic
model, are estimated using literature values (Chow et al., 1988)
of respectively 0.04 for the main channel and in the range of
0.050.12 for the floodplain.
5. Results and discussion
The results of the application of the EBA, SCA and FCA procedures
are included here and discussed with specific regard to the
design hydrograph and hydraulic modeling results of steps 4 and 5.
Table 1 represents the main properties of the design hydrographs

obtained from the EBA and SCA with information on peak
discharge (Qp), volume (V) and duration (D), and the main statistical
properties (minimum, maximum, average, standard deviation
and quantiles at 25% (q25) and at 75% (q75)) of the 500 annual
maximum FCA hydrographs. For the FCA, the Qp, V and D values
are not summarized using the return time since V and D are not
necessarily maximum annual values .
The EBA is characterized by peak discharges that range from
16.6 m3/s to 186 m3/s, while the SCA ranges from 23.2 m3/s to
210.5 m3/s with an average difference of 34.1%. The FCA peak discharges
go from 0 m3/s to 434.8 m3/s. While it seems that the variability
associated with the FCA peak discharges is higher than the
EBA and SCA, the differences in the quantiles of the peak distribution
are limited.
The comparison of volume and duration of the simulated hydrographs
shows more significant differences: EBA and SCA range
from 163.017 m3 to 1.780.300 m3 respectively and from
Fig. 2. Rio Torbido watershed. Case study 2D hydraulic modeling domain.
Fig. 3. Rio Torbido watershed. DEM and drainage network.
Table 1
Summary of design hydrograph characteristics obtained with EBA, SCA, and FCA
approaches.
Event Based Approach (EBA)
T (years) 2 5 10 20 50 100
Qp (m3/s) 16.6 46.4 68.6 95.9 146.5 186.0
V (m3 _ 104) 16.3 45.6 66.8 93.0 140.9 178.0
D (h) 6.13 6.7 6.7 6.8 7.0 7.1
Semi Continuous Approach (SCA)
T (years) 2 5 10 20 50 100
Qp (m3/s) 23.2 65.2 99.1 140.8 175.7 210.5
V (m3 _ 104) 37.5 105.3 159.9 227.2 283.5 339.6
D (h) 15.2 17.3 19.0 21.0 22.8 24.5
Fully Continuous Approach (FCA)
min q25 average q75 max St. dev.
Qp (m3/s) 0 5.7 38.9 56.7 434.8 47.7
V (m3 _ 104) 0.0 7.2 63.4 81.3 526.9 86.0
D (h) 0 7.2 13.8 18.8 49.1 9.5
374.929 m3 to 3.395.933 m3 with an average percentage difference

of 122.8%. This is consistent with previous studies (Grimaldi et al.,
2012b) demonstrating that the SDH duration is approximately

twice as much as the design hydrograph obtained with EBA. The
FCA results also show the volume and duration differences for
EBA and SCA.
Figs. 4 and 5 illustrate the design hyetographs and the design
hydrographs related to the EBA for the corresponding return time
T, while in Fig. 6, the SDHs for the SCA are reported. Fig. 7 represents
the maximum annual hydrograph related to the 0.98 peak
quantile obtained with FCA.
The interpretation of results for the design hydrographs clearly
shows a significant difference between the EBA and the SCA/FCA
schemes in terms of peak and, even more predominantly, event
volume, while the proposed FCA is able to preserve the realistic
shape of the hydrologic forcing.
An example of flood inundation results is inserted in Fig. 8 that
include the maximum floodplain flow depth that is the maximum
height of the flow above the top of the channel banks for the design
event (EBA and SCA) or for the entire simulation time (FCA). Fig. 8
includes, in fact, the three flood maps corresponding to the routing
of the EBA, SCA and FCA design hydrograph along with the selected
hydraulic modeling domain. Differences of the three flooding scenarios
are quite significant. Note that some areas are not associated
to any flow depth meaning that the channel is sufficient to convey
the routed flow in that area. In particular, the EBA underestimates
the inundated area with respect to the SCA and FCA that provide
similar results. Table 2 summarizes results for the three applications
in quantitative terms and for all the return periods with the
corresponding flood area and volume. In Table 3, the percentage
differences among the three approaches are shown confirming
the major differences for the EBA with respect to the SCA and
FCA. These flood areas and volumes are sensibly greater than the
standard EBA case. In particular, in Table 3, it is noted that the differences
between the SCA and FCA are limited to a maximum of
16.9% for flood volumes and 5.6% for flood areas. This was estimated
by analyzing the floodplain and not the channel. While differences
between the SCA and FCA with respect to the EBA go up to
a maximum of 201.9% for flooded volumes and 127.8% for flooded
areas.
Results show that inundation mapping based on a design hydrograph
using EBA should be used with caution since the flood volume
may be significantly underestimated. This is particularly
important when the extra-fluvial process forced by extreme hydrologic
conditions requires 2D unsteady hydraulic model applications
whose results are significantly impacted by the hydrograph
shape and volume rather than just the flood peak. While the present
work shows differences between the design hydrograph derived
from the SCA and the proposed FCA, the corresponding
flood maps seem to be characterized by similar results. However,
this could be due to either the simple flood modeling domain that
is characterized by a large flat area limited by a well defined break
in slope and by the fact that the hydrograph shape does not meaningfully
influence the inundation dynamics.
In any case, results show the accomplishment of the two goals
of the present work: (1) to underline differences in the methodologies
for design hydrograph estimation using the standard EBA
with respect to the SCA and FCA, keeping in mind that EBA is still
used as the standard method by professionals. In fact, EBA is optimized
to estimate the flood peak discharge as the main input for
steady 1D hydraulic models without paying much attention to
the other flood hydrograph components. In particular, it indicates
that the volume, shape and duration govern the modeling results
of 2D hydraulic modeling applications. (2) to present a simple
Fig. 4. Design hyetographs estimated with the EBA approach. and parsimonious procedure for hydrologic forcing characterizaFig. 5. Design hydrographs estimated with the EBA approach.
Fig. 6. Design hydrographs estimated with the SCA approach.
Fig. 7. Example of hydrograph estimated with the FCA approach.
tion reducing at best the conceptual hypotheses and the corresponding

uncertainties that significantly impact the EBA, and also
in part, the SCA methods.
6. Conclusions
In this paper three hydrograph estimation procedures for defining
the design hydrologic input of flood modeling and mapping are
investigated. By taking into consideration small and ungauged river
basins that usually requires subjective and empirical approaches
for rainfallrunoff modeling, the following three methodologies
have been selected and compared with specific regard to the eventual
advantages of the FCA.
The so-called event-based approach, or EBA, is based on a synthetic
representation of both the rainfall input (i.e. IDF method)
and runoff output for hydrograph characterization. The EBA has
Fig. 8. Water depth distribution in selected inundation domain related to 20-years return period: (a) EBA approach; (b) SCA approach; (c) FCA
approach.
been the most simple and commonly applied procedure for decades,
but it is severely impacted by conceptual hypotheses,
assumptions, related errors and uncertainties in hydrologic terms.
The more recent Semi Continuous Approach, or SCA, overcomes
most of the conceptual hypotheses of the EBA and introduces the
use of a simulated rainfall time series by which the rainfallrunoff
model is applied using a continuous implementation. This allows
the extrapolation of a design hydrograph from the runoff synthetic
time series. This SCA also removes the conceptual steps of the EBA
method in the rainfall input estimation, but it is still predominantly
synthetic in the extrapolation of the corresponding
hydrograph.
The proposed Fully Continuous Approach, or FCA, neglects any
preliminary conceptualization of the hydrologic modeling procedure
for design hydrograph estimation by implementing a method
that provides, as input to the hydraulic model, the entire simulated
runoff time series. In this way, there is no conceptualization until
the last step of flood modeling and mapping analysis. Furthermore,
it performs the last necessary analytical and statistical evaluation,
without prior simplification, of the hydraulic modeling results to
identify the flood inundation map and the corresponding design
flow depths and velocities for each cell of any given return period.
Even if the selected coarse resolution DEM is not appropriate for
a detailed quantitative evaluation, it is appositely used for demonstrating
the expected different behavior on the three approaches
while using a topographic resolution of common use for practical
flood mapping in ungauged basins. More specifically, results show
that the three methodologies provide significantly different outputs
with major differences characterizing the EBA with respect
to the SCA and FCA. Those differences confirm that the EBA should
not be used for flood modeling applications that are mainly governed
by the extra-channel physical processes and dynamics. For
these cases the hydrograph shape, duration and volume play the
major role rather than just the flood peak. This is always the case
of high return time flood mapping studies that require an accurate
and detailed characterization of the input hydrograph and the
application of a 2D hydraulic model. It is also shown that the differences
in design hydrograph estimation of the SCA with respect
to the FCA are not transferred into the corresponding flood mapping,
but this is reasonable considering that the two approaches
differ only for the hydrograph shape.
The described proof of concept suggests to prefer the FCA procedure
since it focuses on the implementation of a more realistic
representation of the rainfallrunoff process by means of a simple,
parsimonious model that could be easily implemented by professionals
for practical flood modeling and mapping applications in
small and ungauged river basins.
Future research shall validate the FCA on a small watershed
where flood maps as well as runoff observations shall be available
with the aim of investigating the potential use of such method for
gauged basins. As a result several DEMs at different resolution
should be selected to evaluate the topographic data impact on
model performances. Moreover, the validation, using a DEM with
proper resolution and precision, (<5 m of resolution and <1 m of
precision), shall be also developed. Finally, since the proposed
FCA method provides flow height and velocity time series for each
cell of the inundated domain, a specific statistically-based post
processing analysis shall be implemented for the estimating the
uncertainty of flood inundation maps with the final aim of providing
to floodplain managers and urban planners a detailed decision
making framework.
Acknowledgments
This research was partially supported by the Honors Center of
Italian Universities.
Authors are particularly grateful to Dr. Francesco Serinaldi for
his contribution in the statistical analysis of the inundation maps,
to the eponymous reviewer, Dr. Giuliano Di Baldassarre, and to the
anonymous reviewer for their useful comments.
Table 2
Summary of flood volumes and areas for six return period values estimated for the EBA, SCA and FCA.
T (years) 2 5 10 20 50 100
Event Based Approach (EBA)
V (floodplain and channel) (m3 _ 103) 69.8 115.6 142.9 182.2 269.1 341.5
V (floodplain)(m3 _ 103) 0.0 2.0 7.6 23.6 80.0 134.7
A (floodplain and channel) (m2 _ 103) 34.8 42.0 58.8 80.8 130.4 158.0
A (floodplain) (m2 _ 103) 0.0 7.2 24.0 46.0 95.6 123.2
Semi Continuous Approach (SCA)
V (floodplain) (m3 _ 103) 0.0 5.2 20.4 56.6 101.2 153.5
A (floodplain) (m2 _ 103) 0.0 16.4 42.8 88.0 108.8 126.0
Fully Continuous Approach (FCA)
V (floodplain) (m3 _ 103) 0.0 5.6 23.0 66.1 110.7 166.9
A (floodplain) (m2 _ 103) 0 16.4 45.2 91.6 112.4 130.4
Table 3
Summary of flood volume and area percentage differences for the EBA, SCA and FCA.
T (years) 2 5 10 20 50 100
Semi continuous event based
DV (%) (floodplain and
channel)
8.9 16.4 22.8 29.6 10.6 7.1
DV (%) (floodplain) 162.9 167.8 139.3 26.4 14.0
DA (%) (floodplain and
channel)
0.0 21.9 32.0 52.0 10.1 1.8
DA (%) (floodplain) 127.8 78.3 91.3 13.8 2.3
My continuous event based
channel)
10.9 17.7 26.5 37.1 15.3 12.1
DV (%) (floodplain) 183.3 201.9 179.7 38.3 23.9
channel)
0.0 21.9 36.1 56.4 12.9 4.6
DA (%) (floodplain) 127.8 88.3 99.1 17.6 5.8
Fully continuous semi continuous
channel)
1.8 1.1 3.0 5.8 4.2 4.7
DV (%) (floodplain) 7.7 12.7 16.9 9.4 8.7
channel)
0.0 0.0 3.1 2.9 2.5 2.7
DA (%) (floodplain) 0.0 5.6 4.1 3.3 3.5
References
Alfieri, L., Laio, F., Claps, P., 2008. A simulation experiment for optimal design
hyetograph selection. Hydrol. Process. 22, 813820.
Aronica, G., Bates, P.D., Horritt, M.S., 2002. Assessing the uncertainty in distributed
model predictions using observed binary pattern information within GLUE.
Hydrol. Process. 16, 20012016.
Bates, P.D., Horritt, M.S., Aronica, G., Beven, K., 2004. Bayesian updating of flood
inundation likelihoods conditioned on flood extent data. Hydrol. Process. 18,
33473370.
Blazkova, S., Beven, K., 2009. A limits of acceptability approach to model evaluation
and uncertainty estimation in flood frequency estimation by continuous
simulation: Skalka catchment, Czech Republic. Water Resour. Res. 45, W00B16.
Calver, A., Stewart, E., Goodsell, G., 2009. Comparative analysis of statistical and
catchment modelling approaches to river flood frequency estimation. J. Flood
Risk Manage. 2, 2431.
Chow, V.T., Maidment, D.R., Mays, L.W., 1988. Applied Hydrology. Mc-Graw-Hill,
New York.
European Commission. 2000. CORINE (Coordination of Information on
Environment) Database, a Key Database for European Integrated
Environmental Assessment. Programme of the European Commission,
European Environmental Agency (EEA).

Di Baldassarre, G., Castellarin, A., Montanari, A., Brath, A., 2009. Probability
weighted hazard maps for comparing different flood risk management
strategies: a case study. Nat. Hazards 50 (3), 479496.
Di Baldassarre, G., Schumann, G., Bates, P.D., Freer, J.E., Beven, K.J., 2010. Flood-plain
mapping: a critical discussion of deterministic and probabilistic approaches.
Hydrol. Sci. J. 55 (3), 364376.
Dodov, B., Foufoula-Georgiou, E., 2006. Floodplain morphometry extraction from a
high resolution digital elevation model: a simple algorithm for regional analysis
studies. IEEE Geosci. Remote Sens. Lett. 3 (3), 410413.
European Parliament. 2007. Directive 2007/60/Ec of the European Parliament
and of the Council of 23 October 2007 on the Assessment and Management
of Flood Risks. <http://ec.europa.eu/environment/water/flood_risk/key_docs.
htm#Directive> (accessed 15.03.2011).
FLO-2D. 2012. FLO-2D Reference Manual, <https://www.flo-2d.com/downloa/>.
Giandotti, M., 1934. Previsione delle piene e delle magre dei corsi dacqua. vol. 8,
Istituto Poligrafico dello Stato. pp. 107117.
Gilles, D., Young, N., Schroeder, H., Piotrowski, J., Chang, Y.-J., 2012. Inundation
mapping initiatives of the iowa flood center: statewide coverage and detailed
urban flooding analysis. Water 4, 85106.
Grimaldi, S., Teles, V., Bras, R.L., 2004. Sensitivity of a physically-based method of
terrain interpolation to variance of input data and to the constraint of an
enforced surface drainage direction. Earth Surf. Process. Land. 29 (5), 587597.
Grimaldi, S., Serinaldi, F., Napolitano, F., Ubertini, L., 2005. A 3-copula function
application for design hyetograph analysis. IAHS-AISH Publ. 293, 203211.
Grimaldi, S., Nardi, F., Di Benedetto, F., Istanbulluoglu, E., Bras, R.L., 2007. A
physically based method for removing pits in digital elevation models. Adv.
Water Resour. 30, 21512158.
Grimaldi, S., Petroselli, A., Nardi, F., Alonso, G., 2010. Flow Time estimation with
variable hillslope velocity in ungauged basins. Adv. Water Resour. 33 (10),
12161223.
Grimaldi, S., Petroselli, A., Nardi, F., 2012a. A parsimonious geomorphological unit
hydrograph for rainfallrunoff modelling in small ungauged basins. Hydrol. Sci.
J. 57 (1), 7383. http://dx.doi.org/10.1080/02626667.2011.636045.
Grimaldi, S., Petroselli, A., Serinaldi, F., 2012b. A continuous simulation model for
design-hydrograph estimation in small and ungauged watersheds. Hydrol. Sci. J.
57 (6), 117.
Grimaldi, S., Petroselli, A., Tauro, F., Porfiri, M., 2012c. Time of concentration: a
paradox in modern hydrology. Hydrol. Sci. J. 57 (2), 217228. http://dx.doi.org/
10(1080/02626667), 2011, 644244.
Grimaldi, S., Petroselli, A., Serinaldi, F., in press-a. Design hydrograph estimation in
small and ungauged watersheds: continuous simulation method versus eventbased
approach. Hydrological Processes. http://dx.doi.org/10.1002/hyp.8384.
Grimaldi, S., Petroselli, A., Romano, N., in press-b. Green-ampt curve number mixed
procedure as an empirical tool for rainfallrunoff modelling in small and
ungauged basins. Hydrological Processes. http://dx.doi.org/10.1002/hyp.9303.
Grimaldi, S., Petroselli, A., Romano, N., in press-c. Curve-Number/Green-Ampt
mixed procedure for streamflow predictions in ungauged basins: parameter
sensitivity analysis. Hydrological Processes. http://dx.doi.org/10.1002/
hyp.9749.
Hall, J.W., Tarantolo, S., Bates, P.D., Horritt, M., 2005. Distributed sensitivity analysis
of flood inundation model calibration. J. Hydraulic Eng. 131 (2), 117126.
Horritt, M.S., Di Baldassarre, G., Bates, P.D., Brath, A., 2007. Comparing the
performance of a 2-D finite element and a 2-D finite volume model of
floodplain inundation using airborne SAR imagery. Hydrol. Process. 21, 2745
2759. http://dx.doi.org/10.1002/hyp. 6486.
Hsieh, L.-S., Hsu, M.-H., Li, M.-H., 2006. An assessment of structural measures for
flood-prone lowlands with high population density along the Keelung River in
Taiwan. Nat. Hazards 37 (12), 133152.
IGMI (Italian Geographic Military Institute). 2003. Raster (Matrix) Numerical DEM
of Italy (Internal Factsheet, in Italian). <http://www.igmi.org/pdf/
info_matrix2003.pdf>.
Kalyanapu, A., Judi, D., McPherson, T., Burian, S., 2012. Monte Carlo-based flood
modelling framework for estimating probability weighted flood risk. J. Flood
Risk Manage. 5, 3748. http://dx.doi.org/10.1111/j.1753-318X.2011.01123.x.
Kedem, B., Chiu, L.S., Karni, Z., 1990. An analysis of the threshold method for
measuring area-average rainfall. J. Appl. Meteorol. 29 (1), 320.
Keifer, C.J., Chu, H.H., 1957. Synthetic storm pattern for drainage design. ASCE, J.
Hydraulics Div. 83(HY4), 125.
Laforce, S., Simard, M.-C., Leconte, R., Brissette, F., 2011. Climate change and
floodplain delineation in two southern quebec river basins. J. Am. Water Resour.
Assoc. 47 (4), 785799.
McMillan, H.K., Brasington, J., 2008. End-to-end flood risk assessment: a coupled
model cascade with uncertainty estimation. Water Resour. Res. 44, W03419.
http://dx.doi.org/10.1029/2007WR005995.
Merwade, V., Olivera, F., Arabi, M., Edleman, S., 2008. Uncertainty in flood
inundation mapping: current issues and future directions. J. Hydrol. Eng. 13
(7), 608620.
Moretti, G., Montanari, M., 2008. Inferring the flood frequency distribution for an
ungauged basin using a spatially distributed rainfallrunoff model. Hydrol.
Earth Syst. Sci. 12, 11411152.
Nardi, F., Vivoni, E.R., Grimaldi, S., 2006. Investigating a floodplain scaling relation
using a hydrogeomorphic delineation method. Water Resour. Res. 42, W09409.
http://dx.doi.org/10.1029/2005WR004155.
Nardi, F., Grimaldi, S., Santini, M., Petroselli, A., Ubertini, L., 2008. Hydrogeomorphic
properties of simulated drainage patterns using DEMs: the flat area issue.
Hydrol. Sci. J. 53 (6), 11761193.
Nishat, S., Guo, Y., Baetz, B.W., 2010. Antecedent soil moisture conditions of
different soil types in South-western Ontario, Canada. Hydrol. Process. 24,
24172424.
Nnadi, F.N., Kline, F.X., Wary, H.L., Wanielista, M.P., 1999. Comparison of design
storm concepts using continuous simulation with short duration storms. J. Am.
Water Resour. Assoc. 31 (1), 6185.
NRCS. 1997. Ponds-Planning, Design, Construction. Agriculture Handbook N. 590.
US Natural Resources Conservation Service, Washington DC.
OBrien, J.S., Julien, P.Y., Fullerton, W.T., 1993. Two-dimensional water flood and
mudflow simulation. J. Hydraulic Eng. 119 (2), 244261.
Pappenberger, F., Matgen, P., Beven, K.J., Henver, J.B., de Pfister, L., Fraipont, P., 2006.
Influence of uncertainty boundary conditions and model structure on flood
inundation predictions. Adv. Water Resour. 29, 14301449.
Pramanik, N., Panda, R.K., Sen, D., 2010. Development of design flood hydrographs
using probability density functions. Hydrol. Process. 24 (4), 415428.
Rahman, A., Weinmann, P.E., Hoang, T.M.T., Laurenson, E.M., 2002. Monte Carlo
simulation of flood frequency curves from rainfall. J. Hydrol. 256, 196210.
Sampson, C.C., Fewtrell, T.J., Duncan, A., Shaad, K., Horritt, M.S., Bates, P.D., 2012.
Use of terrestrial laser scanning data to drive decimetric resolution urban
inundation models. Advances in Water Resources, 41, 117, ISSN 0309-1708.
10.1016/j.advwatres.2012.02.010.
Santini, M., Grimaldi, S., Nardi, F., Petroselli, A., Rulli, M.C., 2009. Pre-processing
algorithms and landslide modelling on remotely sensed DEMs. Geomorphology
113 (12), 110125.
Schertzer, D., Lovejoy, S., 1987. Physical modeling and analysis of rain and clouds by
anisotropic scaling of multiplicative processes. J. Geophys. Res. 92, 96939714.
Serinaldi, F., 2009. A multisite daily rainfall generator driven by bivariate copulabased
mixed distributions. J. Geophys. Res. 114, D10103. http://dx.doi.org/
10.1029/2008JD011258.
Serinaldi, F., 2010. Multifractality, imperfect scaling and hydrological properties of
rainfall time series simulated by continuous universal multifractal and discrete
random cascade models. Nonlin. Processes Geophys. 17, 697714. http://
dx.doi.org/10.5194/npg-17-697-2010.
Serinaldi, F., Grimaldi, S., 2011. Synthetic design hydrographs based on distribution
functions with finite support. Journal of Hydrologic Engineering, 16, (5) 434
446; doi:10.1061/(ASCE)HE.1943-5584.0000339.
Soczyska, U., Nowicka, B., Somorowska, U., Kupczyk, E., Suligowski, R., 1997.
Prediction of design storms and floods. IAHS-AISH Publ. 246, 297303.
Tarboton, D.G., Ames, D.P., 2001. Advances in the Mapping of Flow Networks from
Digital Elevation Data. World Water and Environmental Resources Congress,
Orlando, Florida, May 2024, ASCE.
Tarboton, D.G., Bras, R.L., Rodriguez-Iturbe, I., 1991. On the extraction of channel
networks from digital elevation data. Hydrol. Process. 5 (1), 81100.
USDA-SCS. 1986. Urban Hydrology for Small Watersheds. Technical Release 55.
Washington, DC.
USDA-NRCS. 2010. National Engineering Handbook, Part 630 Hydrology. <http://
www.info.usda.gov/viewerFS.aspx?hid=21422>.
Verhoest, N.E.C., Vandenberghe, S., Cabus, P., Onof, C., Meca-Figueras, T.,
Jameleddine, S., 2010. Are stochastic point rainfall models able to preserve
extreme flood statistics? Hydrol. Process. 24, 34393445.
Viglione, A., Bloschl, G., 2009. On the role of storm duration in the mapping of
rainfall to flood return periods. Hydrol. Earth Syst. Sci. 13, 205216.
Viglione, A., Merz, R., Bloschl, G., 2009. On the role of the runoff coefficient in the
mapping of rainfall to flood return periods. Hydrol. Earth Syst. Sci. 13, 577593.
Yoo, C., Jung, K.S., Kim, T.W., 2005. Rainfall frequency analysis using a mixed
Gamma distribution: Evaluation of the global warming effect on daily rainfall.
Hydrol. Process. 19 (19), 38513861.
SpectralReflectancePropertiesof
Hydrocarbons:
RemoteSensingImplications
EDWARDA.CLOUTIS
Thespectralreflectancepropertiesofbituminoustarsandswere
examinedinthe
wavelengthrangefrom0.35to2.6micrometers.Uniqueabsorption
featuresduetoall
themajorphases,exceptquartz,appearinthespectra.The
intensitiesoftheabsorption
featurescorrelatewiththeabundancesofthevariousphases.The
resultsofthisstudy
haveapplicationstotheremotesensingofmanyterrestrialand
extraterrestrialtargets
becauseofpotentialsimilaritiesbetweentarsandandother
hydrocarbonoccurrences.
Forexample,itisfoundthathighlypolymerizedhydrocarbonsare
aplausible
constituentofthedarkmaterialononeofSaturn'ssatellites,
Iapetus.
ASYSTEMATICSTUDYOFTHESPECtral
reflectancepropertiesofbituminous
tarsandswasundertakenin
ordertodeterminewhetherthisclassof
materialshasdistinctspectralproperties(1).
Tarsandsarecomposedofamixtureof
clays,bitumen(acomplexarrayofvariously
polymerizedhydrocarbons),quartzgrains,
water,andminoraccessoryminerals(24).
Understandingthespectralpropertiesof
thismaterialisessentialforgeologicalremote
sensingofterrestrialandextraterrestrial
targets,becauseverylittleisknownabout
thespectralreflectancepropertiesofhydrocarbon
bearingmaterials.
Amongtheleadingextraterrestrialcandidates
forpossiblehydrocarbonoccurrences
areTrojanasteroids,cometarynuclei,and
thedarksideofoneofSaturn'ssatellites,
Iapetus(5,6).Tarsandsmayserveasreasonable
spectralanalogsoftheseobjectsand
otherbodieswhosesurfacesarebelievedto
becomposedofvariouscombinationsof
organicmatter,clays,andH20.Spectral
analysisofthetarsandsmayalsomake
possibletheremotesensingdetectionofterrestrial,
surficialhydrocarbonseeps,because
highlypolymerizedhydrocarbonsarecharacteristic
ofbothtarsandsandsurfaceexposures
(7).Thetermkerogenisoftenusedto
refertoextraterrestrialoccurrencesofhydrocarbons.
Kerogenandbitumendiffer
primarilyinthehydrogen/carbonratio;in
bitumentheratioishigherbecauseofgreater
maturation(7).Rapidcharacterizationof
tarsandsandheavyoilsampleswouldbe
possiblewiththedevelopmentofquantitative
spectralanalysistechniques.Thiscould
improvecommercialbitumenextractionefficiencies
andsimplifytheonerousprocedures
nowrequiredfortarsandcharacterization.
Spectralreflectancetechniquesmay
eventuallybeextendedtotheanalysisof
othernonrenewableenergyresourcessuch
asoilshalesandcoals.
AllthephasespresentintheAthabascatar
sands,exceptquartz,displaydistinctabsorption
bandsthatarepotentiallyresolvable.
Theshapes,intensities,andwavelengthpositions
ofthesebandsmayprovideinformation
onthephysicalandchemicalproperties
ofthevariousphases(811).
Manyorganiccompoundsdisplayelectronic
transitionsarisingfromexcitationsof
bondingelectronsinthewavelengthregion
from0.1to0.35,um.Asthecomplexityof
theorganicmoleculesincreases,themaximum
absorptionshiftstowardlongerwavelengths
andindividualbandsbecomeless
distinctbecauseofincreasingoverlap(12
14).Amaterialascomplexasbitumenisnot
expectedtoexhibitindividual,resolvable
absorptionbandsintheultravioletandvisible
spectralregion.Abroadoverallreflectance
decreasetowardshorterwavelengthsis
seenincoals,chars,oilshales,andtarsands
(12,1417).
Thewavelengthpositionsofthemost
intense,majororganicfundamentalbands
arelistedinTable1.Becausetheenergiesof
thefimdamentalabsorptionbandsdepend
onanumberoffactorssuchaslocalatomic
configurations(18),thewavelengthpositions
oftheovertoneandcombination
bandsaresomewhatuncertain(Table2).
Thelowoverallreflectanceoftarsandsis
expectedtosuppressallbutthemostprominent
ofthesebands.Themostpromising
regionsinwhichtosearchfororganicabsorption
bandsarenear1.7,umandbetween
2.2and2.6,um(16,19).The1.7,umregion
isdominatedbyvariousCHstretching
overtonesandcombinationbands.Thefundamental
bandsthatcontributetothisfeature
arethemostintenseabsorbersinthe
infraredspectraofbitumen(20).Theregion
from2.2to2.6,umisaffectedbynumerous
overlappingcombinationandovertone
bands.Bitumentransmissionspectrashow
moderatelyintenseabsorptionbetween
1750and900cmlwhichcancombine
inanumberofwaysintheregionfrom2.2
to2.6,um(2,12,20,21).Becauseofthe
sheernumberofpossiblebands,theregion
hasalowoverallreflectancewithonlythe
mostintenseabsorptionbandsbeingpartially
resolvable(14,15,22).
Transitionserieselements,particularlyvanadium
andnickel,complexedwithporphyrin
typemoleculesarequitecommoninthe
Table1.Positionsofthemajororganicfundamental
absorptionbandsinenergyspace.
Fundamental
BandfrequencyAssignment
(cm')
a3030Alkene,aromaticCH
stretch
b2950AsymmetricCH3
stretch
c2920AsymmetricCH2
stretch
d2875SymmetricCH3stretch
e2850SymmetricCH2stretch
f1700CarbonylcarboxylCO
stretch
g1600Aromaticcarbon
stretch
h1450AsymmetricCH2,CH3
bend
i1375SymmetricCH3bend
Table2.Predictedwavelengthpositionsofthe
mostintenseovertonesandcombinationsofthe
majorfundamentalorganicabsorptionbandslisted
inTable1.
Combinationovertone
band
2a
a+c
2b
a+e
2c
b+d
c+e
2d
2e
e+f
c+g
e+g
b+h
c+h
b+i,d+h
e+h
d+i
c+j
Wavelength
(Lm)
1.65
1.68
1.69
1.70
1.71
1.72
1.73
1.74
1.75
2.20
2.21
2.25
2.27
2.29
2.31
2.33
2.35
2.53
REPORTSI65
DepartmentofGeology,UniversityofAlberta,Edmonton,
Alberta,CanadaT6G2E3.
I4jULYI989
Downloaded from www.sciencemag.org on December 5, 2013
0.51.01.52.02.5
Wavelength(gm)
Fig.1.Reflectancespectrum(0.35to2.6,um)of
alowbitumentarsand(sample8205)(1).
asphalticfractionoftarsands.Theydisplaya
numberofabsorptionbandsintheultraviolet
andvisiblespectralregions.Themost
intenseoftheseoccursat0.4p.mandhas
beendetectedinthespectraofoilshaleand
carbonaceouschondrites(23).
ThebitumencontentintheAthabascatar
sandstypicallyrangesfrom0to15%by
weight(3).Thisrangewasarbitrarilydivided
intothree5%intervalsinordertodisplay
spectralvariationsasafunctionofchanging
phaseabundances.Thereflectancespectraof
low,medium,andhighbitumensamples
weremeasured,andarepresentativesample
ofeachgroupwasselectedformoredetailed
analysis.
Thelowbitumengroupischaracterized
byprominentabsorptionbandsduetothe
clays(Fig.1).Theabsorptionbandsat
1.4,1.9,2.2,and2.3to2.6p.mare
assignedtostructuralandadsorbedwaterin
theclaysandtocationOHvibrational
bands(8).Thewavelengthpositionofthe
1.9p.mband(1.92p.m)doesnotcoincide
withthatexpectedforfreewater[1.94p.m
(11)].Thebroadnessofthisbandandthe
1.4p.mbandisconsistentwiththepresence
ofwaterinavarietyofsitesorwithstructurally
disorderedclays(2).Thedecreasein
reflectancefrom2.3to2.6p.misalsocharacteristic
ofaclaydominatedspectrum(8).
Discretebitumenabsorptionbandsareweak
orabsentinspiteoftheamountofbitumen
present(3.2%)andthedarkappearanceof
thesample.Indirectevidenceforbitumen
comesfromthelowoverallreflectanceofthe
sample,whichisnotcharacteristicofmost
clays(8,9).
Spectraofmediumbitumensamplesare
richerindetailthanthoseofthelowbitumen
samplesanddifferinotherways.The
depthofthe1.9p.mbandisreduced,anew
bandappearsat1.7,um,thedepthofthe
bandintheregionfromthe2.3to2.6p.mis
greater(Table3),andtheoverallreflectance
slopeinthisrangeisslightlypositive(Fig.
2).Theabsorptionbandthatappearsat1.7
p.mcanbeattributedtofirstorderovertones
andcombinationsofthevariousCH
stretchingfundamentals(Tables1and2).
Theincreaseinabsorptionintheregion
I66
Table3.Wavelengthpositionsoftheobserved
absorptionbandsandtheirdepths,Db.Some
bandpositionsaregivenasrangesbecauseof
variationsbetweendifferentsamplespectra.
BandSample
minmum
(ILm)820587048619
0.407.0
1.391.414.56.03.0
1.721.751.89.918.6
1.921.9519.010.43.8
2.105.0
2.152.184.19.611.4
2.216.515.514.3
2.3114.840.254.3
2.3513.938.955.0
2.4516.935.647.9
2.5522.133.938.6
from2.3to2.6p.misalsoamanifestationof
theincreasingbitumencontent.Theslightly
positiveslopeinthisregionisnotcharacteristic
ofclays,whichshowanegativeslope.
Thissuggeststhatbitumenisoverridingthe
spectralsignatureoftheclaysinthisregion.
Spectraofsampleswithhighbitumen
contentaredominatedbybitumen(Fig.3),
thesecondmostabundantphaseafterquartz
inthesesamples(Table3).Allthemajor
absorptionbandscanbeattributedtohydrocarbons.
Theexpectedclaywaterbands
at1.4and1.9p.marevirtuallyabsent.The
1.7p.mbitumenabsorptionregionisvery
prominent,anditsnonsymmetricalshapeis
consistentwithmultiple,overlappingCH
absorptionbands.Theabsorptionisstrong
intheregionfrom2.3to2.6p.mandhasa
positiveslope,neitherofwhichischaracteristic
ofclays.Theabsorptionbandat2.3p.m
isveryintense,andthepositiveslopeinthe
regionfrom2.3to2.6p.mistheresultof
thedominanceofthisfeature.
Eachspectralgroup,dividedonthebasis
ofbitumencontent,exhibitsuniquespectral
properties.Thesevariationscanberelatedto
thephysicalandchemicalpropertiesofthe
samplesthathavebeenindependentlydetermined.
Bitumenandclayabundancesare
generallyinverselycorrelated.Thereisno
simplecorrelationbetweenspectralalbedo
andanyofthemajorphaseabundances.The
sameholdstruefortheabundancesof
quartzversusclay.Bitumenandquartz
abundancesarepositivelycorrelated.Spectra
correspondingtosampleswithhighand
lowbitumencontentsaredominatedby
hydrocarbonandclayabsorptionfeatures,
respectively.
Themostprominentabsorptionbandattributable
tothevariousformsofwateris
presentnear1.9p.mandismostintensein
themostwaterandclayrichsamples(Table
4).Thiswavelengthregionisusefulforclaywater
determinationbecause,unlikethere
0.20
=
00..115_/
;0.10
0.05
0.51.01.52.02.5
Wavelength(gm)
amediumbitumentarsand(sample8704).
gionfrom2.3to2.6,um,itisnotsignificantly
overlappedbyhydrocarbonabsorptions.
Thebroadnessandcomplexshapeof
the1.9,umbandisconsistentwiththe
presenceofwaterinanumberofsitesinthe
clay,structurallydisorderedclays,ormore
thanonespeciesofclay.Allthreeinterpretations
areborneoutbyindependentstudies
oftheAthabascaclays(2,3).Theremaybea
moderatecontributionbyfreewaterat1.94
,um,butitspresencecannotbeunambiguously
resolved.
Theabsorptionbandnear1.4,umismost
prevalentinthelowbitumencontentspectra
andisassignedtostructuralandbound
waterintheclays.Theavailablespectral
informationonvariousformsofwater
showsthatfreewaterandpolymerizedwater
absorbatdifferentwavelengths(>1.45p.m)
thanwaterinclays(1.38to1.42,um).The
broadnessofthisbandisinterpretedinthe
samewayasthatofthe1.9p.mband(24).
ClaylatticeOHabsorptionbandsare
expectedintheregionfrom2.2to2.6p.m.
Thisbandisunexpectedlyweakestinthe
spectrumofthemostclayrichsample(Fig.
1).Thehighlyabsorbingnatureofbitumen
inthisregion,anditslowabundanceinthe
clayrichsamples,explainsthisdiscrepant
behavior.Clayspectranormallyexhibithigh
overallreflectance(8,9).Thelowbitumen
spectrum(Fig.1)contains28%claysized
particlesandshowsonlyclayabsorption
bandsbuthasaverylowoverallreflectance.
Thelowreflectancemaybecausedbythe
presenceofthebitumen,ironoxidecontaminants,
ortransitionserieselementsubstitutions
(3).Areflectancespectrumofatransition
elementbearingclay,glauconite(Fig.
4),showsanumberofsimilaritiestothe
clayrichspectrumofFig.1:aconcave
reflectancerisebetween0.9and1.5p.m,a
weakorabsent1.4p.mabsorptionband,
andareflectancedecreasebeyond2.3p.m.
Theoverallreflectanceoftheglauconiteis
stillsignificantlyhigherthanthatofthe
mostclayrichsample(Fig.1).Eventhough
transitionseriesbearingclaysmaybepresent,
glauconiteitselfhasnotbeenidentified
inAthabascatarsands(2,3).Thespectrum
ofthelowbitumensampleisconsistent
SCIENCE,VOL.245
0.10h
co
40.05
0.51.01.52.02.5
Wavelength(pm)
ahighbitumentarsand(sample8619).
withacationsubstitutedclaybutrequires
thepresenceofaphasewithlowoverall
reflectance.Bitumenseemsthelikeliestcandidate
forthedarkeningmaterial.
Thehighbitumenspectrumcorresponds
toasamplewith5%claysizedparticlesand
hasavirtuallyundetectable1.9p.mclaywater
band.Thusalowerlimitof5%can
beplacedonspectraldeterminationofclays
inorganicrichmaterialsofthistype.Bitumen
servesasaneffectivesuppressorofwhat
arenormally,moderatelyintenseabsorption
bands.
Theregionsat1.7and2.3to2.6p.mare
thebestareasforthedetectionofvariousCH
absorptionbands.The1.7p.mregionis
preferablebecauseitisnotoverlappedby
clayandwaterabsorptionbandsandisthe
mostrecognizablefeatureshortwardof2.1
p.m.Thecomplexshapeofthisbandis
consistentwiththepresenceofmultiple,
partiallyoverlappingabsorptionbands,although
noattempthasyetbeenmadeto
deconvolvethisfeatureintoitsconstituent
bands.Theuseofthisabsorptionfeature
placesalowerdetectionlimitof4%on
bitumenabundancesintarsands.
Thenearabsenceofa1.9p.mclaywater
bandinFig.3stronglysuggeststhatthis
spectrumisagoodrepresentativeofpolymerized
hydrocarbonsplusspectrallyneutral
material.Themoderatestrengthabsorption
bandsthatappearinallthesample
spectrabettween2.17and2.25p.mcanbest
beassignedtobothclaylatticeOHand
organicbands,becausethereisnosimple
correlationbetweentheirbanddepthsand
anyonephaseabundance.Theoverallslope
oftheregionfrom2.3to2.6p.missensitive
tobothbitumenandclaycontents.Clays
consistentlyexhibitanegativeslopeinthis
region[see(8,9)andFig.4],whereas
organicmaterialsshowapositiveslope(15,
16,22).Thechangeoverfromnegativeto
positiveslopeoccursinthemediumbitumen
group.
Changesinbitumencontentaffectthe
ultravioletandvisiblespectralregions.The
pointofminimumreflectanceshiftstolonger
wavelengthsasthebitumencontentincreases.
Thewavelengthpositionofthere
Table4.Phaseabundancesofthesamples(in
percentagesbyweight).
Abundances(%)<400>400
pieBituWaSolmeshmesh
menterids(%(%
82053.28.188.732.167.9
87048.16.685.329.370.7
861913.22.583.96.094.0
0.60
0.40
O
aC D
x0.20
0.oo0|I, ,
. ,I..I.I.....
0.300.600.901.201.501.802.102.40
Wavelength(pm)
glauconite,whichhasanumberofsimilaritiesto
theIapetusdarkmaterial.
flectanceminimuminFig.3is0.47p.m,
whichcorrespondstoanaverageofsix
condensedaromaticringspermoleculeifwe
usethecalibrationofBadger(25).Thisisin
goodagreementwiththeconventionalanalytical
resultsoffiveringspermoleculefor
theresinfraction,themostabundantorganic
group(3,20).Theshapeofthereflectance
spectrumbetween0.5and1.5p.m
changesfromtwolinearsegmentswitha
slopebreakat0.9p.mtoasmoother
concaveslopewithincreasingbitumenand
decreasingclaycontents.Theformershape
issimilartothatoftheglauconitespectrum
(Fig.4);thelattershapeisconsistentwitha
continuumofchargetransferenergylevels
expectedincomplexhydrocarbons.
TheintenseSoretband,whichischaracteristic
ofporphyrins,shouldbepresentat
0.4p.m(23).Thelowoverallreflectance
ofthesamplespectramakesidentificationof
thisfeatureverydifficult,anditsunambiguous
presencehasnotyetbeenestablished.
Hydrocarbonsaresuspectedtobepresent
onanumberofextraterrestrialbodies(5,6)
andhavebeenfoundincarbonaceouschondrites
(26).Iapetuswasselectedformore
detailedspectralanalysis.Onehemisphereis
almostentirelycoveredwithaverydark
substanceofunknowncomposition.Earlier
investigatorsusedamixingmodeltoremove
thespectralsignatureofwatericeandisolate
thedarkmaterial(6),whichshowsagradual
riseinreflectancetowardlongerwavelengths
withonlyminorabsorptionbands.
Iftheremovalprocedureforthewaterice
signaturedidnotinadvertentlydeleteadditional
features,thereisnoevidenceforany
oftheexpectedorganicabsorptionbands.In
3
2
0
0.51.01.52.02.5
Wavelength(pm)
Fig.5.Normalizedreflectancespectrumofthe
lowbitumentarsandsamplewiththetelescopic
spectraldatafortheIapetusdarkmaterialfrom
(6)overlain(filledcircles).
particular,theregionsat1.7and2.3to2.6
p,mshownoneofthecharacteristicsof
highlypolymerizedhydrocarbons.
Thebestmatchtothedarkmaterialwas
foundtobeamixtureof90%clayand10%
coaltarrepresentingorganicmatter(6).The
lowbitumensample(Fig.1)consistsofthe
sameratioofclaytoorganicsand,likeits
coaltarcounterpart,isfreeofdetectable
organicabsorptionbands.Theclaycoaltar
spectrumprovidesagoodmatchtothe
overallchangeinreflectancewithwavelength
oftheIapetusspectrum,althoughthe
fittotheregionfrom0.8to1.4p.misnot
verygood.Thelowbitumenspectrum(normalized
tomatchtheIapetusdarkmaterial)
providesamuchbetterfittotheregionfrom
0.8to1.4,umandalsoshowsachangein
slopeat0.9p.m(Fig.5).Beyond1.4,um
thereflectanceofthetarsandisconsistently
higherthanthatoftheIapetusdarkmaterial;
theclayabsorptionbandsat1.9p.mand
2.2to2.5p.marealsomuchmoreintensein
thetarsandspectrum.
Neithertheclaycoaltarmixturenorthe
tarsandareperfectspectralmatchestothe
Iapetusdarkmaterial.Thereflectancespectrum
ofglauconite(Fig.4)exhibitsanumber
ofdesirablespectralfeaturesformatching
withtheIapetusdarkmaterial.The
differencebetweentheminimumandmaximum
reflectanceiscomparabletothatfound
forthedarkmaterial,theslopebreakat
0.9p.mispresent,andaweak1.9p.m
claywaterbandispresent.Mostotherclays
donotshowthesamereflectancedifference
betweenminimumandmaximumreflectance
(8),becausetheycontainonlyminor
amountsoftransitionserieselements.A
darkmaterialmustbeaddedtothisspectrum
tofurthersuppressthe1.9p.mband
andneutralizethereflectancedropoffbeyond
2.2p.m.
Nooneofthesamplespectraisaperfect
matchtotheIapetusdarkmaterial.Because
eachmaterialexaminedhascertaindesirable
spectralcharacteristics,acomplex,intimate
mixtureofdifferentmaterialsmayberequired.
Anironsubstitutedclayseemstobe
YREPORTSI67
.1.
14.JULY1989
anecessaryconstituent(5).Thepresenceof
someamountofahighlypolymerizedhydrocarbon,
likethebitumenpresentintar
sands,maybeneededtomodifytheclay
spectrumintheregionfrom0.8to1.4,um.
Afewpercentofbitumencanbepresentand
notexhibitdistinctabsorptionbands.Finally,
amaterialspectrallysimilartothecoaltar
extractisneeded,whichcaneffectivelymask
thenegativeslopeoftheclaysbetween2.2
and2.6,um,reducethereflectanceriseat
longerwavelengths,andsuppressclayorganic
bandsat2.2to2.6,um.
REFERENCESANDNOTES
1.Thetarsandsamplesusedinthisstudyarefromthe
AthabascadepositinnortheasternAlbertaandconsist
ofviscousorganicmatterembeddedinclastic
sedimentsoftheCretaceousMcMurrayandClearwater
formations[B.NagyandG.C.Gagnon,
Geochim.Cosmochim.Acta23,155(1961)].They
wereobtainedfromtheAlbertaOilSandsTechnology
andResearchAuthority[(AOSTRA),Edmonton,
Alberta]OilSandsSampleBank,alongwith
someanalyticaldata.Thereflectancespectrawere
acquiredattheU.S.GeologicalSurveyspectrometer
facilityinDenver,CO,andtheNationalAeronautics
andSpaceAdministrationReflectanceExperiment
Laboratory(RELAB)spectrometerfacilityat
BrownUniversityinProvidence,RI.Comprehensive
descriptionsoftheinstrumentsanddataprocessing
proceduresareavailableinthefollowing
references:R.N.Clark,Publ.Astron.Soc.Pac.92,
221(1980);C.M.Pieters,J.Geophys.Res.88,
9534(1983);T.V.V.KingandW.I.Ridley,ibid.
92,11457(1987).Thehomogeneityofthetarsand
samplescouldonlybeassessedvisually.Thesamples
weregentlychoppedandmixeduntilavisually
uniformmixtureresulted.Theywerethenplacedin
sampleholdersandlightlycompressedtoprovidea
flatsurfaceforspectralmeasurement.Allsamples
weremeasuredrelativetoHalon.Thetarsand
spectrawereacquiredwiththeuseofanintegrating
spherearrangementattheU.S.GeologicalSurvey
facility.Theglauconitewasspectrallycharacterized
inabidirectionalreflectancemodeattheNASA
RELABfacility,usinganincidenceangleof0and
anemissionangleof150.Aduplicatetarsand
spectrum,acquiredtocheckforreproducibility,was
withintheerrorlimitsofthedata.Theerrorbarsin
thespectraldatarepresent1SDofthemean.Band
depth(Db),usedextensivelyinspectralanalysis,is
heredefinedas
Db=1Rb/Rc
whereRcisthereflectanceofastraightlinecontinuum
atthewavelengthpositionofareflectance
minimumandRbistheminimumreflectanceatthe
samewavelength.Thecontinuumwasconstructed
asastraightlinetangenttothespectrumoneither
sideoftheabsorptionfeatureofinterest.Thecontinuum
fortheregionfrom2.3to2.6,umwastaken
asahorizontallinetangenttothereflectancemaximum
near2.2,um.
2.E.CzarneckaandJ.E.Gillot,ClaysClayMiner.28,
197(1980);T.M.Ignasiaketal.,Fuel62,353
(1983).
3.J.A.Bichard,OilSandCompositionandBehaviour
Research(AOSTRATechnicalPublicationSeries4,
Edmonton,Alberta,Canada,1987).
4.M.L.Seluckyetal.,Fuel56,369(1977);J.W.
Bungeretal.,ibid.58,183(1979).
5.J.GradieandJ.Veverka,Nature283,840(1980);
C.ChybaandC.Sagan,ibid.330,350(1987);E.
K.Jessberger,A.Christoforidis,J.Kissel,ibid.332,
691(1988);D.P.CruikshankandR.H.Brown,
Science238,183(1987).
6.D.P.Cruikshanketal.,Icarus53,90(1983);J.F.
Belletal.,ibid.61,192(1985).
7.J.M.Hunt,PetroleumGeochemistryandGeology
(Freeman,SanFrancisco,1979).
i68
8.G.R.HuntandJ.W.Salisbury,Mod.Geol.1,283
(1970);,C.J.Lenhoff,ibid.4,85(1973);G.
R.Hunt,Geophysics42,501(1977);R.N.Clark,
J.Geophys.Res.86,3074(1981);T.V.V.King,
thesis,UniversityofHawaii(1986).
9.J.D.LindbergandD.G.Snyder,Am.Mineral.57,
485(1972);S.C.Feldman,J.V.Taranik,D.A.
Mouat,inProceedingsoftheAirborneImagingSpectrometer
DataAnalysisWorkshop,G.VaneandA.F.H.
Goetz,Eds.(JetPropulsionLaboratory,Pasadena,
CA,1985),p.56;G.VaneandA.F.H.Goetz,
RemoteSensingEnviron.24,1(1988);S.A.Drury
andG.A.Hunt,Photogramm.Eng.RemoteSensing
54,1717(1988);C.M.PietersandJ.F.Mustard,
RemoteSensingEnviron.24,151(1988).
10.J.E.Zajic,D.G.Cooper,J.A.Marshall,D.F.
Gerson,Fuel60,619(1981);K.BuijsandG.R.
Choppin,J.Chem.Phys.39,2035(1963);H.
Yamatera,B.Fitzpatrick,G.Gordon,J.Mol.Spectrosc.
14,268(1964);M.R.Thomasetal.,J.Phys.
Chem.69,3722(1965);K.B.Whetsel,Appl.
Spectrosc.Rev.2,1(1968);J.D.WorleyandI.M.
Klotz,J.Chem.Phys.45,2868(1966).
11.J.G.Bayly,V.B.Kartha,W.H.Stevens,Infrared
Phys.3,211(1963);R.GoldsteinandS.S.Penner,
J.Quant.Spectrosc.Radiant.Transfer4,441(1964).
12.M.L.BoydandD.S.Montgomery,J.Inst.Pet.
London49,345(1963).
13.B.E.Hudson,Jr.,R.F.Robey,J.F.Nelson,in
ProceedingsoftheFifthWorldPetroleumCongress(Fifth
WorldPetroleumCongress,Inc.,NewYork,1959),
vol.V,p.1;R.M.SilversteinandG.C.Bassler,
SpectrometricIdentificationofOrganicCompounds(Wiley,
NewYork,1967).
14.L.A.Gilbert,Fuel39,393(1960).
15.0.Ito,H.Seki,M.lino,ibid.67,573(1988).
16.W.Kaye,Spectrochim.Acta6,257(1954);R.F.
Goddu,Anal.Chem.29,1790(1957);S.A.Fyshet
al.,Appl.Spectrosc.39,354(1985);N.A.McAskill,
ibid.41,313(1987).
17.T.Yokotaetal.,Fuel65,1142(1986);E.A.
Kmetko,Phys.Rev.82,456(1951).
18.G.Svehla,ComprehensiveAnalyticalChemistry,vol.
VI,AnalyticalInfraredSpectroscopy(Elsevier,Amsterdam,
1976);P.C.Painteretal.,Appl.Spectrosc.35,
475(1981);M.P.Fulleretal.,Fuel61,529(1982);
P.C.Painteretal.,ibid.66,973(1987);P.W.Yang
etal.,EnergyFuels2,26(1988).
19.R.J.Moore,R.J.Gordon,R.C.Eiffert,in
ProceedingsoftheFifthWorldPetroleumCongress(Fifth
WorldPetroleumCongress,Inc.,NewYork,1959),
vol.V,p.13.
20.M.L.BoydandD.S.Montgomery,CanadianDep.
MinesTech.Surv.R78(1961);CanadianDep.Mines
Tech.Surv.R88(1961);CanadianDep.MinesTech.
Surv.104(1962).
21.B.NagyandG.C.Gagnon,Geochim.Cosmochim.
Acta23,155(1961).
22.J.C.DoniniandK.H.Michaelian,InfraredPhys.26,
135(1986);E.K.PlylerandW.S.Benedict,J.Res.
Natl.Bur.Stand.47,202(1951).
23.J.B.F.ChamplinandH.N.Dunning,Econ.Geol.
55,797(1960);G.W.HodgsonandB.L.Baker,
Chem.Geol.2,187(1967);E.W.FunkandE.
Gomez,Anal.Chem.49,972(1977);P.N.Holden
andM.J.Gaffey,Meteoritics22,412(1987).
24.R.B.SingerandT.L.Roush,LunarPlanet.Sci.
Conf14,708(1983).
25.G.M.Badger,StructuresandReactionsoftheAromatic
Compounds(CambridgeUniv.Press,London,
1957).
26.B.Nagy,CarbonaceousMeteorites(Elsevier,Amsterdam,
1975).
27.Supportedbyascholarshipandresearchgrantfrom
AOSTRAandanAmericanAssociationofPetroleum
Geologistsgrantinaid(5821201).Ithank
D.WallaceandD.HenryoftheAOSTRAOil
SandsSampleBankforprovidingthetarsand
samples.IthankR.ClarkandG.SwayzeoftheU.S.
GeologicalSurveyinDenver,C.PietersandS.Pratt
atBrownUniversityforprovidingaccesstotheir
respectivespectrometerfacilities,andM.Gaffeyfor
providingthespectrumanalysisprogramsandencouragement
throughout.Iamalsogratefultotwo
anonymousreviewersformanyusefulsuggestions
andcorrections.
30December1988;accepted19April1989
TemperatureMeasurementsinCarbonatiteLava
LakesandFlowsfromOldoinyoLengai,
Tanzania
MAURICEKRAFFTANDJORGKELLER
Thepetrogenesisofcarbonatiteshasimportantimplicationsfor
mantleprocessesand
forthemagmaticevolutionofmantlemeltsrichincarbon
dioxide.OldoinyoLengai,
Tanzania,istheonlyactivecarbonatitevolcanoonEarth.Its
highlyalkalic,sodiumrich
lava,althoughdifferentincompositionfromthemorecommon
calciumrich
carbonatites,providestheopportunityforobservationsofthe
physicalcharacteristics
ofcarbonatitemelts.Temperaturemeasurementsonactive
carbonatiticlavaflowsand
fromcarbonatiticlavalakeswerecarriedoutduringaperiod
ofeffusiveactivityinJune
1988.Temperaturesrangedfrom491to519C.Thehighest
temperature,measured
fromacarbonatiticlavalake,was544C.Thesetemperaturesare
severalhundred
degreeslowerthanmeasurementsfromanysilicatelava.Atthe
observedtemperatures,
thecarbonatitemelthadlowerviscositiesthanthemostfluid
basalticlavas.The
unusuallylowmagmatictemperatureswereconfirmedwith1
atmospheremelting
experimentsonnaturalsamples.
ARBONATITESAREIGNEOUSROCKSOldoinyoLengai,Tanzania,theonly
richinprimarycarbonates,domiactivecarbonatitevolcanoon
Earth,isfanantly
calciteanddolomite.Almousforitsunusualalkalirichmagma,
thoughrarecomparedwithsilicaterocks,termednatrocarbonatite.
Thevolcano,a
carbonatitesprovideimportantconstraints
onpartialmeltingprocesses,volatileconM.Krafft,Centre
Vulcain,F6700Cernay,France.
tent,andchemicalcompositionintheJU.nivKeerllseirt,at,MiDn
e7ra8l0o0giFsrcehi
bPuertgr,ogFreadpehrailscRheepsubIlnistcitouftGedrer
earth'smantle.many.
SCIENCE,VOL.245
A change detection approach to flood

mapping in urban areas using
TerraSAR-X
Article
Accepted Version
Giustarini, L., Hostache, R., Matgen , P., Schumann, G.,

Bates, P. D. and Mason, D. C. (2013) A change detection
approach to flood mapping in urban areas using TerraSARX.
IEEE Transactions on Geoscience and Remote Sensing, 51
(4). 2417-2430 . ISSN 0196-2892 doi:
10.1109/TGRS.2012.2210901 Available at
http://centaur.reading.ac.uk/29866/
It is advisable to refer to the publishers version if you intend to cite from the
work.
To link to this article DOI: http://dx.doi.org/10.1109/TGRS.2012.2210901
Publisher: IEEE Geoscience and Remote Sensing Society
All outputs in CentAUR are protected by Intellectual Property Rights law,

including copyright law. Copyright and IPR is retained by the creators or
other
copyright holders. Terms and conditions for use of this material are defined in
the End User Agreement .
www.reading.ac.uk/centaur
CentAUR
Central Archive at the University of Reading
Readings research outputs online
1
A change detection approach to flood mapping in urban areas using TerraSARX
Laura Giustarini, Renaud Hostache, Patrick Matgen, Guy Schumann, Paul D. Bates,
and David C. Mason
L. Giustarini, R. Hostache and P. Matgen are with the Centre de Recherche Public Gabriel Lippmann, Dpartement Environnement et Agro-biotechnologies, L-4422
Belvaux, Grand-Duchy of Luxembourg (email: giustari@lippmann.lu;
hostache@lippmann.lu; matgen@lippmann.lu)
G. Schumann and P. D. Bates are with the School of Geographical Sciences,
University of Bristol, BS8 1SS Bristol, U.K. (email: guy.schumann@bristol.ac.uk;
paul.bates@bristol.ac.uk)
D. C. Mason is with the Environmental System Science Centre, University of
Reading, RG6 6AL Reading, U.K. (email: dcm@mail.nerc-essc.ac.uk)
ABSTRACT
Very high-resolution Synthetic Aperture Radar sensors represent an alternative to
aerial photography for delineating floods in built-up environments where flood risk is
highest. However, even with currently available SAR image resolutions of 3 m and
higher, signal returns from man-made structures hamper the accurate mapping of
flooded areas. Enhanced image processing algorithms and a better exploitation of
image archives are required to facilitate the use of microwave remote sensing data for
monitoring flood dynamics in urban areas. In this study a hybrid methodology
combining radiometric thresholding, region growing and change detection is
2
introduced as an approach enabling the automated, objective and reliable flood extent
extraction from very high-resolution urban SAR images. The method is based on the
calibration of a statistical distribution of open water backscatter values
inferred
from SAR images of floods. SAR images acquired during dry conditions enable the
identification of areas i) that are not visible to the sensor (i.e. regions
affected by
layover and shadow) and ii) that systematically behave as specular
reflectors (e.g.
smooth tarmac, permanent water bodies). Change detection with respect to a pre- or
post flood reference image thereby reduces over-detection of inundated areas. A case
study of the July 2007 Severn River flood (UK) observed by the very high-resolution
SAR sensor on board TerraSAR-X as well as airborne photography highlights
advantages and limitations of the proposed method. We conclude that even though the
fully automated SAR-based flood mapping technique overcomes some limitations of
previous methods, further technological and methodological improvements are
necessary for SAR-based flood detection in urban areas to match the flood mapping
capability of high quality aerial photography.
I. INTRODUCTION
The support of remote sensing for mapping changes in water surface extents and
elevations has been demonstrated widely (for detailed reviews see [1]-[3]). The
success of these research studies together with recent public and political awareness
for quantifying global environmental change has led to a significant increase in the
number of satellites dedicated to flood monitoring and hydrology in the wider sense.
Importantly, flood monitoring from space has the advantage of large area coverage
and relatively fast response services (see for example the International
Charter Space
3
and Major Disasters initiated by major space agencies:
http://www.disasterscharter.org/).
The vast majority of a flooded area is rural rather than urban and accordingly most
literature on remote sensing based flood detection to date has focussed on the rural
case. However, it is perhaps more important to detect the urban flooding because of
the increased risks and costs associated with it. Flood extent can be detected in rural
floods using SARs such as ERS and ASAR, but these have too low a resolution (25
m) to detect flooded streets in urban areas. However, a number of SARs with spatial
resolutions as fine as 3 m or better have recently been launched and are potentially
capable of detecting urban flooding. They include TerraSAR-X, RADARSAT-2, and
the four COSMO-SkyMed satellites.
In an operational context, reference [4] have proposed a hybrid methodology for
SAR imagery, which combines radiometric thresholding and region growing as an
approach enabling the automatic, objective and reliable flood extent extraction from
SAR images. First results on moderate and low resolution image data indicate that the
proposed method may outperform manual approaches if no training data are available
even if the parameters associated with these methods are determined in a non-optimal
way. The results demonstrate the algorithms potential for accurately
processing data
from different SAR sensors.
Notable examples of research into automatic near real-time flood detection
algorithms using single-polarisation high resolution (greater than a few metres) SAR
imagery have been shown by references [5]-[6] on TerraSAR-X data and reference [7]
on COSMO-SkyMed data. The algorithms by references [5]-[6] search for water as
regions of low SAR backscatter using a region-growing iterated segmentation /
classification approach whereas the technique by reference [7], is based on a fuzzy
4
logic approach which integrates theoretical knowledge about the radar return from
inundated areas based on backscattering models, with simple hydraulic considerations
and contextual information. Both algorithms are very effective at detecting rural
floods, but would require substantial modification to work in urban areas containing
radar shadow and layover.

A semi-automatic algorithm for the detection of floodwater in urban areas using
TerraSAR-X has also been developed by reference [8]. It uses a SAR simulator [9] in
conjunction with LiDAR terrain data to estimate regions of the image in which water
would not be visible due to shadow or layover caused by buildings and taller
vegetation. Ground will be in radar shadow if it is hidden from the radar by an
adjacent intervening building. The shadowed area will appear dark, and may be
misclassified as water even if it is dry. In contrast, an area of flooded ground in front
of the wall of a building viewed in the range direction may be allocated to the same
range bin as the wall, causing layover which generally results in a strong return, and a
possible misclassification of flooded ground as un-flooded. The algorithm is aimed at
detecting flood extents for validating an urban flood inundation model in an offline
situation, and requires user interaction at a number of stages.
Follow-up work from this was carried out by reference [10]. Here the objective
was to build on a number of aspects of the existing algorithms to develop an
automatic near real-time method for flood detection in urban and rural areas. In the
urban area, 75% of the urban water pixels visible to TerraSAR-X were correctly
detected, though this percentage reduced somewhat if the urban flood extent visible in
the aerial photos and detected by TerraSAR-X was considered, because flooded pixels
in the shadow/layover areas not visible to TerraSAR-X then had to be taken into
account. Better flood detection accuracy was achieved in rural areas, with almost 90%
5
of water pixels being correctly detected by TerraSAR-X. The algorithm assumes that
high resolution LiDAR data are available for at least the urban regions in the scene, so
that a SAR simulator may be run in conjunction with the LiDAR data to generate
maps of radar shadow and layover in urban areas. It is therefore limited to urban
regions of the globe that have been mapped using LiDAR.
To contribute to these recent developments in high performance flood detection
algorithms for high resolution SAR data, we propose a simple but effective technique
based on image differencing as proposed by reference [4], which may compete with
existing algorithms in terms of accuracy and level of automation. For this we also
focus on flood detection inside urban areas and use the TerraSAR-X image of the
England summer 2007 floods as demonstration. Although this is only a single test,
and different results may be obtained for other urban areas where the built
environment is different to the UK case studied here, it does provide a first
demonstration of the potential of the method.
II. METHODOLOGY
Reference [5] recently highlighted an apparent lack of traceability and
standardisation in many SAR-based flood-mapping methodologies. This concern has
led to the introduction of two variants of an automated and physically-based SARbased
flood-mapping algorithm [4]. Both variants, which are termed M1 and M2a
respectively, exploit the statistics of backscattering coefficients retrieved from SAR to
segment an image into its flooded and non-flooded parts. While M1 only considers a
single SAR flood image to extract pixels corresponding to open water via
thresholding and region growing, M2a adds change detection with respect to a nonflood
reference image to improve the algorithms performance. In this study we
6
introduce an enhanced version of M2a, which we term M2b. This method addresses
some of the shortcomings of M2a that reference [4] identified in two representative
case studies. This section provides a detailed overview of all processing steps of the
flood extraction algorithm M2b, together with the associated parameters defining each
process and a list of differences with respect to the M1 and M2a algorithms
previously introduced. Besides standard pre-processing steps commonly involved
with Level 1 SAR data, the M2b algorithm consists of four processing steps (Fig. 1).
A. Statistical distribution of the open water backscatter
The flood extraction algorithm uses as input Level 1 SAR data that are geo-coded,
co-registered and calibrated. The first step is the estimation of the probability density
function (PDF) of backscattering values associated with open water. The
aim of this
processing step is the calibration of a theoretical PDF that optimally fits the empirical
distribution of backscatter values from open water inferred from the SAR
image.
According to reference [13] the backscatter variability on a homogeneous surface is
mainly due to speckle and the theoretical PDF that best describes the distribution of
backscatter originating from a homogeneous surface is of gamma type. Here, we
hypothesize open water to be a homogeneous surface, which means that a
potential
limitation of the approach and SAR mapping of inundated surfaces in general relates
to the possible roughening of open water caused by emerging vegetation, wind or
rainfall. We therefore assume that the PDF of type gamma can be used to estimate the
distribution function of open water backscatter:
7
1
0
0
1
0
0
1
0
01
1
0
0
0
1
1
/
k
e
k
fk=m
k
m
k
m
(1)
where k is the shape parameter of the gamma distribution and 0
m is the gamma
distribution mode. The parameter 0
1 is the minimum backscatter value in the SAR
image, which needs to be applied so that the gamma distribution is calculated only for
positive values.
Two parameters thus need to be optimized to identify the theoretical gamma
function f that best fits the empirical distribution of backscatter values from
open
water h (i.e. image histogram). The optimization of the two parameters k and 0
m
consists in minimizing the root mean squared error (RMSE) between the image
histogram and the gamma distribution for backscatter values lower than 0
thr , with the
parameter 0 0
thr m representing the point where the distributions f and h start
deviating. The optimization is performed with sequentially increasing values of 0
m
and 0
thr values. For each set of 0
m and 0
thr values, the parameter k is optimized
using the nonlinear fitting process of reference [14]. The RMSE between the
theoretical density function f and the empirical density distribution h is calculated for
each parameter set and over all backscatter values lower than 0
thr . Finally, the
parameter set ( 0
m , k, 0
thr ) providing the lowest RMSE is set as optimal.
B. Radiometric thresholding
The aim of the following step of the algorithm is to extract seeds of open
water
areas from the flood image, being either individual pixels or regions. The parameter
8
0
thr
represents the maximum backscatter value for which the fit between the
theoretical and empirical PDF is satisfactory. For backscattering values higher than
0
thr
the distribution functions f and h start deviating, which leads to an immediate and
substantial increase of over-detected water pixels. As a matter of fact, 0

thr is
considered the maximum backscatter value for which there is no significant overlap
between radiometric distributions corresponding to water bodies and other land use
types. Since the backscatter values from water surfaces are comparatively low, this
value is used to extract the seeds of water bodies by selecting the pixels having
backscatter values lower than 0
thr . The thresholding yields a preliminary flood
inundation map that represents the seed region for a subsequent region growing
process.
Moreover, in order to be able to map permanent water bodies, the threshold
computed on the flood image, 0
thr , is also applied on the reference SAR image to
classify seeds of permanent water bodies. It is worth noting that these seeds include in
addition to permanent water bodies other smooth surfaces with a water surface-like
radar response as well as all shadow-affected areas.
C. Region growing
Next, the extracted water bodies, representing the seeds, are dilated using the
region growing approach of reference [15]. The procedure iteratively grows the seeds
until a given tolerance level is reached. The sequence of thresholding and region
growing only adds pixels to the seeds that are located in the vicinity of the
preliminary flood extent, thereby limiting the risk of over-detection on areas distant
from the flooded area (i.e. misclassification of dry pixels as wet). The
tolerance
9
parameter characterises the regional homogeneity of the backscattering behaviour.
The tolerance criterion adopted here is based on the percentiles of the theoretical
gamma distribution of open water pixels.
The iterative procedure incorporates pixels with backscatter values lower than
0
, corresponding to a given percentile of the theoretical gamma distribution of

water pixels in the image. In this study we propose a simultaneous calibration
recently advocated by reference [4]. The approach optimizes the tolerance criterion
rg
together with the change detection parameter introduced in the next section.
Region growing, with the same threshold value 0
rg is also applied to dilate the
seeds of permanent smooth surfaces obtained from the reference image. The approach
provides a mask of water surface-like radar response areas that is used to limit the
rg
region growing applied on the flood image, thereby preventing the spreading of
flooded areas into permanent smooth areas.
D. Change detection
Reference [4] argued that flood maps resulting from region growing should
include all open water pixels connected to the seeds. The region growing should
thus extend into the high percentiles of the gamma distribution. However, the
resulting over-detection needs to be removed by the subsequently applied change
detection step. Change detection thus aims at removing pixels from the flood extent
map that do not correspond to flood water. To do so, only pixels that significantly
change their backscatter values with respect to their baseline backscatter values are
kept in the flood extent map, while pixels that did not decrease their backscatter
values by a minimum amount are removed. This means that the main river channel,
10
which is a permanent water body, is not any longer an integral part of the flooded
area.
The specific parameter of the change detection is 0, defined as the required
minimum change in backscatter between the reference and the flood image for a pixel
being considered as flooded. In order to determine the optimal criterion for the
required minimum change in backscatter, an iterative procedure is adopted.
As mentioned earlier, the two parameters 0
rg and 0 are optimized through a
simultaneous calibration, minimizing the RMSE computed over the whole range of
backscatter values in the flood image between the theoretical gamma distribution and
the empirical distribution of open water pixels. This means that different
threshold
values, 0
rg , which correspond to different percentiles of the theoretical gamma
distribution, are sequentially selected from an interval of plausible values, and a
corresponding minimum change detection parameter 0 is optimized for each tested
0
threshold value. For every parameter set ( 0

rg , 0), the sequence of region
growing and change detection processes is applied on the area conditioned by the
permanent smooth area mask. At the end of each iteration, the histogram of flood
water pixels is computed. The corresponding empirical PDF is compared
against the
initially calibrated theoretical gamma distribution (1). The parameter set ( 0
rg , 0)
providing the lowest RMSE value is set as optimal.
To summarize, M2b essentially represents an improved version of the M2a
method introduced by reference [4]. The two algorithms both take into account a
reference SAR image and include four inter-related processing steps (i.e. calibration
of gamma distribution function, radiometric thresholding, region growing and change
detection). However, while M2a pre-defines the region growing parameter 0
rg as the
11
rg
99% percentile of the water backscatter gamma distribution, M2b adds

flexibility to
the optimization process by calibrating the tolerance criterion 0
rg that, together with
the associated change detection parameter 0, minimizes the RMSE between
empirical and theoretical distribution functions.
This modification implemented in M2b constitutes an important change as it
renders the algorithm fully automated, without any requirement of manual user inputs.
Therefore, the mapping process is believed to be entirely objective. Another important
improvement is that M2b, unlike M2a, makes use of the reference image to build a
mask of permanent water surface-like radar response areas. Indeed, to render the
algorithm suitable for urban flood mapping, it is necessary to mask out not only
smooth surfaces like tarmac, paved roads and parking lots, but also all regions in
shadow-affected areas unseen by the satellite. In urban areas the latter are particularly
important as they potentially lead to a significant part of over-detected flooded areas.
It should also be noted that method M2a and its enhanced version M2b both rely
on the availability of reference images acquired from the same orbital track, with the
same incidence angle, polarization and resolution, and prior to the onset of flooding.
Moreover, the adequate choice of a season-dependent reference image might help
reducing the effects of changes in vegetation and in soil moisture conditions, as
argued recently by reference [16]. With the advent of relatively new sensors such as
TerraSAR-X, it can be difficult to find an image that satisfies these selection criteria.
However, as image archives are gradually being built up, this should be less of a
problem in the near future. If no reference image is available, method M1 can be
applied.
12
III. STUDY AREA AND AVAILABLE DATASET
This section describes the study area, the flooding event and the available remote
sensing images for testing and evaluating the proposed automated flood delineation
algorithm.
The image data used for this study were acquired for the 1-in-150 year flood that
took place around Tewkesbury, U.K., in July 2007. Extreme rainfall intensities
resulted in substantial flooding of urban and rural areas; about 1000 properties in the
town of Tewkesbury were affected [17]. Tewkesbury lies at the confluence of the
River Severn, flowing in from the northwest, and the River Avon, flowing in from the
northeast. Bankfull discharge is approximately 350 m3s-1 (or 4.5 m in gauged level) at
the Saxons Lode gauging station ~7km upstream of Tewkesbury. The summer 2007
event was unusual for the study site in that the majority of the flow was derived from
local rainfall. On the 20th July, two days prior to flood peak, more than 12 cm of rain
fell on the surrounding area. The flood peak of 5.43 m Ordnance Data Newlyn was
measured at Tewkesbury on July 22 with both rivers exhibiting a more rapid increase
in flow than a typical autumn or winter event that may build over many weeks, with
flows increasing from 100 m3s-1 to >500 m3s-1 in 57 hours, between the 20th of July
and the 22nd July. The river did not return to below bankfull until July 31. In the
region of interest (red box in Fig. 2) an area of 1.5 km2 was possibly flooded at the
time of the TerraSAR-X overpass, according to a 2m model simulation [18].
In order to demonstrate the applicability of our proposed SAR image

segmentation algorithm, we define the urban area as the zone inside and in the
vicinity of the built-up region of the town of Tewkesbury.
13
A. TerraSAR-X images
A unique data set consisting of numerous types of remotely sensed images over
one single event hydrograph were acquired over the selected study area [18]. From
this dataset a stripmap TerraSAR-X image acquired on July 25 2007 (at 06:34 GMT,
Wednesday) was selected (see Fig. 2). The image is a multi-look ground range
spatially enhanced scene with 1.5 m pixel spacing and has a mean incidence angle of
24. Its H/H polarization mode arguably allows for the best discrimination between a
SAR images flooded and non-flooded parts [5]. At the time of the satellite overpass
and image acquisition, there was relatively low wind speed and no rain [8]. Moreover,
no rainfall was recorded in the 30 hours preceding the TerraSAR-X acquisition, as
well as during the satellite overpass itself.
In their flood delineation study reference [8] used the single TerraSAR-X flood
image together with air-borne scanning laser altimetry (LiDAR) data. Here, we also
consider a dry reference image which is a post-flood image acquired from the same
orbit track and with the same polarization as the flood image. This way, geometric
problems related to co-registration can be limited and baseline backscatter values can
be inferred. A single scene having these imaging characteristics and covering all of
the azimuth extent of the target is available in the current TerraSAR-X image archive.
It was acquired on July 22 2008 (at 06:34 GMT, Tuesday), almost exactly a year after
the flood event had occurred. The flood and non-flood images have both been
acquired in the same month of the year. Hence it can be assumed that the state of
vegetation is similar in both images. This is important as decreases in backscatter
values between any two images are caused not only by flooding, but also by changes
in vegetation and soil moisture among others. Hence, it is also important to note that
the gauging stations close to the town of Tewkesbury did not record any rainfall in the
14
three days before the reference image was acquired. Since the weather was dry when
the reference image was acquired, the effect of soil moisture fluctuations on observed
decreases in backscatter values are considered as not significant for this test case.
The two images, whose characteristics are listed in Table I, have been georeferenced
and calibrated. In an operational context, the reference image would
ideally consist of a pre-flood satellite acquisition. However, in this particular case,
given the relative novelty of a sensor such as TerraSAR-X, it was not possible to find
a reference image, prior to the onset of flooding, acquired from the same orbital track
and with same polarization. Therefore, a post-flood image was selected.
Particular attention has been given to an adequate co-registration of the images, as
an accurate overlapping is a pre-requisite for detecting flooding-related changes in the
backscattering behaviour. The accuracy of the geo-referencing has a sub-pixel
precision. Next, the images have been filtered with a 5 x 5 Gamma-MAP filter to
decrease the speckle contribution. This filter smoothes out the speckle granularity
while preserving details, such as the contours of buildings and flooded areas [19]. It
also impacts the parameterization of the gamma PDF (1) by reducing the spread of
backscattering values associated with open water. The red box area in Fig.
2
presents the area of interest for the city of Tewkesbury: it refers to a rectangular area
of 1135x998 pixels (1.5 m pixel spacing) for a total surface of ~3 km2.
B. Validation dataset
The validation data set, consisting of very high-resolution 0.2 m aerial
photographs acquired during the flooding event in July 2007, enables a
comprehensive evaluation of the algorithms performance in terms of SARbased
15
flood delineation. An aircraft operated by the Environment Agency of England and
Wales (EA) carried out the overflights.
The flood extent was obtained through manual photo-interpretation (Fig. 3a).
Taking advantage of existing landuse maps of the area, permanent water bodies
associated with rivers and canals have been removed from the validation map. While
in general the delineation of flood boundaries from such high-resolution optical
products is relatively straightforward, it is important to note that the flooding of
densely vegetated and built-up environments can lead to some ambiguities. For
instance, in the case of bare soil fields, the accurate positioning of the separation line
between muddy flood waters and non-flooded areas is non-trivial.
In addition, it is important to bear in mind that the aerial photographs were
acquired on July 24 (at 11:30 GMT) while the TerraSAR-X image was obtained 19
hours later on July 25 (at 06:34 GMT). Although there was no significant decrease in
main river discharge between the acquisition time of aerial photographs and the
TerraSAR-X overpass [20], this time gap might be responsible for some discrepancies
between the SAR-derived and aerial photography-derived flooded areas. To illustrate
potential discrepancies, simulations with a previously calibrated hydraulic 2m
LISFLOOD-FP flood model [21], have been carried out both at the aerial photographs
acquisition time and at the TerraSAR-X overpass. These simulations show a reduction
of the flooded area of approximately 5% between the two time steps. In particular,
Fig. 3(b) shows the differences between the two simulated flood inundation maps The
most notable differences can be observed on a triangular-shaped field (see the middleright
part of the domain of interest) from which flood water was drained between the
two overpasses according to the model simulations. This location was also
problematic in terms of identifying its flooding status through photo-interpretation, as
16
explained in more details in the discussion section. These factors, all unrelated to the
processing of the SAR images, need to be taken into account during the analysis, as
all the observed differences may not be necessarily due to the inability of the
proposed algorithms for accurately extracting the flood extent from SAR imagery.
IV. RESULTS AND DISCUSSION
This section assesses the accuracy of the classification obtained with the fully
automated flood detection algorithm M2b and contrasts its performance with those of
the previously introduced M1 and M2a algorithms. The section also provides insights
into the added-value of reference images for flood delineation in urban areas.
A. Extraction of flooded areas
The flood extent has been extracted from the TerraSAR-X image using the three
methods M1, M2a and M2b.
In particular, for method M2b, Fig. 4 illustrates the optimization of the four
parameters: the mode of the open water backscatter gamma PDF, 0
m , the
radiometric threshold, 0
thr , the tolerance criterion for the region growing step, 0
rg ,
and the minimum CD value, 0. Panel (a) reports the optimization of the mode
parameter while panel (d) provides the corresponding optimized gamma PDF in red
together with the histogram of the backscatter values in the flood image. Panel (d)
displays the value of the second parameter, 0
thr (i.e. in this case study equal to 15.5
dB) as the maximum backscatter value for which there was no overlap between the
empirical histogram and the theoretical gamma PDF. The optimized 0
thr value is also
provided in Fig. 5(a), together with the backscatter histogram of pixels in the
17
reference image. This value is used to derive the reference mask of permanent water
surface-like radar response areas through thresholding of the reference image (Fig.
5(b)). From Fig. 4(d) and Fig. 5(a) it can be observed that for high backscatter values,
there is a systematic noise in the return signal of both TerraSAR-X images. This is
arguably due to the high complexity of urban topography and its considerable impacts
on the high-resolution backscattering signal. Considering the region growing step of
method M2b, it is important to mention that the parameters 0
rg and 0 are
optimized together. Therefore, the subplot in panel (b) of Fig. 4 illustrates the impact
that different 0
rg parameters (each associated with the corresponding optimal 0
value) have on the RMSE. Similarly, panel (c) provides an example of the
performance plot of parameter 0 for a given 0
rg value: Fig. 4(c) refers to the
optimal 0
rg value of the case study. The backscatter value corresponding to the
optimized 0
rg value is also displayed in panels (e) and (f). Finally, panel (f) shows the
empirical histogram of flood pixel values before and after change detection. The
histograms in panel (f) are computed only from pixels inside the SAR-derived flooded
area. The reduction of the distribution tail and the related reduction of over-detection
are indicated by the empirical histogram approaching the theoretical gamma PDF.
B. Evaluation at city level (quantitative analysis)
Three flood extent maps were obtained through the application of the three image
processing algorithms. The corresponding contingency matrices were computed using
the evidence provided by aerial photography. The binary pattern of flooded and nonflooded
pixels was compared against the reference flood map (in this case, see Fig. 3).
18
The result is a matrix (or contingency table) of four possible outcomes. With respect
to the reference flooded area, there are two ways for a remote sensing-derived flooded
area to be correct (either by correctly representing flooded or non-flooded pixels) and
two ways to be incorrect (either by erroneously under- or over-predicting the observed
inundation extent). The values of the contingency matrix for all methods are reported
in Table II (and also displayed as contingency maps in Fig. 6) for a quantitative
evaluation of the performances. Moreover, the optimized (and/or fixed) parameter
values for the region growing and the change detection are indicated.
From Table II it can be concluded that in the present case study the three
algorithms provide very similar performance levels. When the evaluation is carried
out at a regional scale (i.e. at city level), the differences seem to be marginal.
However, methods M2a and M2b still slightly outperform method M1 with respect to
all evaluation criteria listed in Table II, suggesting that change detection with respect
to a non-flood reference image does provide some advantages. The results do not
reflect the added-value that we expected from the methodological improvements of
method M2b. This result is due to the fact that, in this particular case study, the
optimized region growing threshold 0
rg equals 98%, which is very close to the predefined
99% value that reference [4] proposed for M2a. Here, the trade-off involves a
controlled growing of the seed region in order to be able to limit the over-detection of
flooded areas. While the latter can be partly removed by the subsequent change
detection, the results indicate that the reduced over-detection comes at the cost of an
increased under-detection of flooded areas. The results also indicate that method M2b,
which provides an optimal empirical distribution with respect to the targeted gamma
distribution of open water backscatter values, does not necessarily
generate a more
accurate flood inundation map than M2a. On a more positive note, it can be observed
19
that the simultaneous optimization of region growing threshold 0
rg and change
detection parameter 0, computed by minimizing the RMSE between empirical and
theoretical distribution functions, led to the maximum value of correctly detected
pixels (81.7% as reported in Table II). Moreover, from Fig. 7 it can be observed that
in this case study the optimum parameter set also yields the best performance with
respect to the validation data.
A comparison with the flood extent detected for the same test case with the semiautomatic
procedure of reference [8] cannot be carried out in a very meaningful way
due to the fact that the input data sets in both studies differ. Reference [8] took
advantage of a regional DEM so that SAR-derived water ground heights smoothly
vary along the river reach. Furthermore, reference [8] did not make use of a pre-flood
reference image. The comparison of contingency matrices reveals that the M2b
method performance of 81.7% of correctly detected pixels is rather close to the
percentage of 85.4% obtained with the flood inundation map provided by reference
[8] on a common area of interest and reference dataset. This result indicates that
topography data could be used more efficiently than pre-flood reference images for
increasing the accuracy of SAR-derived flooded areas. However, this assessment
needs to be confirmed in future studies.

To better appreciate the advantages of these methodological enhancements it is
worth analyzing the PDF of backscattering values associated with pixels located
inside the flood inundation maps, the latter corresponding either to the validation map
obtained from aerial photographs or the flood extent maps computed with the
different versions of the image-processing algorithm. The different PDFs are
displayed in Fig. 8. It becomes evident from the panels that the PDF of flood
water
pixels from high-resolution photos is reasonably close to a gamma distribution, albeit
20
characterized by a heavy tail end. M1 does enable the identification of a majority of
water pixels, but it misses out the tail of the distribution. M2a yields a better
performance, as it adds more pixels to the flood extent, thereby reducing the number
of under-detected flood pixels. However, there is still a tendency to slightly
overestimate part of the tail of the theoretical gamma PDF. Because of its enhanced
change detection procedure, the M2b method is capable of growing the seeds further
into the high percentiles of the gamma distribution, thereby further reducing the heavy
tail end while keeping the PDF of detected water pixels closer to the theoretical PDF.
It is worth noting here that the PDF of backscattering values inside the area delimited
by the high-resolution photos exhibits a particular tail that is missed by all three
versions of the flood detection algorithm. Surprisingly, the high number of pixels with
associated high backscatter values is not only due to the expected increased
backscatter response from urban structures. In fact, probability distributions from
pixels located in rural areas exhibit the same heavy tail end as pixels located in urban
areas (Fig. 9). Further research is needed to get a better understanding of the particular
shape of the PDF of backscatter values. One possible explanation, can be that some of
the pixels changed their status from flooded to non-flooded between the
two data
acquisitions, as already illustrated in Fig. 3b.
From this first overview of results, it can be concluded that the strength of change
detection is that it allows the region growing to extend further into the high
percentiles of the gamma distribution as it efficiently removes part of the resulting
over-detection. More case studies are needed before a generalization of these findings
can be done in a meaningful way. Nevertheless, these preliminary results corroborate
those reported by reference [4] in case studies dealing with coarse and moderate
resolution SAR imagery.
21
On the selected domain (i.e. rectangular area of 1135x998 pixels) the running time
of the complete process on an Intel(R) Core (TM) 2Quad CPU, 2.66Ghz and 3.24 GB
RAM is less than 30 minutes, indicating the appropriateness of the method also for
near real-time applications.
1) Analysis of errors: the problem of unseen regions
One feature that requires special attention relates to regions in a SAR image that
cannot be seen by the radar sensor because of its side-looking nature. The affected
regions are commonly referred to as shadow and layover. In the context of
urban
flood mapping shadow and layover are due to geometric distortions

caused mainly
by the presence of buildings. Their impact on SAR-based flood mapping is twofold.
First, flooding does not impact the radar response from shadow areas and,
consequently, SAR-based detection of flooded shadow areas is not possible (i.e.
problem of under-detecting floods). Second, the low radar response from shadow
regions might erroneously lead to their classification as flooded even in the
case
they are not (i.e. problem of over-detecting floods). Here, we assume that the resulting
over-detection might be addressed through change detection since urban shadow areas
do not change between two images acquired from the same track. This is confirmed
by the reduction in over-detection shown in Table II. Moreover, due to double-bounce
reflection effects, the urban layover backscatter could be different in the flood and
reference images. This phenomenon typically occurs in vegetated areas, where
flooding yields an increased backscatter due to the double-bouncing between the
flooded ground and branches or leaves, resulting in a higher return signal in the flood
image. A similar mechanism of multiple reflections between flooded streets and walls
can potentially result in a brighter backscatter in the urban areas covered by the flood
22
image. Classification errors are expected to be higher in urban areas than in forested
regions [22] and therefore the layover contribution should be taken into account when
mapping flooded urban areas. In fact, the inherent under-detection problem can only
be addressed through technological advances (e.g. look angles closer to nadir) or the
use of ancillary data (e.g. topography data).
For the imaging characteristics of the TerraSAR-X image acquired on July 25
2007, reference [8] computed a mask of areas affected by shadow and
layover in
the city of Tewkesbury (Fig. 10). They used the German Aerospace Center (DLR)
SAR end-to-end simulator (SETES) in conjunction with air-borne scanning laser
altimetry (LiDAR) data to estimate regions of the image in which water would not be
visible to the instrument. In this case study we made use of the
shadow/layover
mask from reference [8] to evaluate the risk of misclassifying pixels in areas not
visible to the SAR sensor.
In the area of interest (red box in the relevant figures), the shadow/layover mask
of reference [8] covers a total area of ~1 km2, which is a significant percentage of the
total area selected for the analysis (~3 km2). However, referring to the validation map,
the flooded area not visible to the satellite reduces to 0.25 km2, over a total flood
extent of 1.22 km2 (see Table III). Moreover, due to the 24 look angle, in this
particular case study the effect of layover is greater than shadow, as it covers a much
larger flooded area. This is mostly due to the diffuse presence of hedges along the
borders of the different fields in the rural areas.
As flooding in shadow/layover areas is undetectable for SAR, the corresponding
regions would need to be delineated a priori and considered as areas with an
unidentifiable status of flooding. In fact, even if a flood is correctly
classified in a
shadow region, this result should be viewed as an error as the right answer is obtained
23
for the wrong reason. As the objective of the developed method was to generate a
mask of surfaces that produce a radar signal response similar to that of inundated
areas in order to constrain the flood extent outside the shadow areas, in the following
we focus only on the overlap between the obtained flood extent and the shadow areas
derived by reference [8]. Note that the shadow mask itself, obtained with the SAR
simulator and the LiDAR data, might contain some degree of uncertainty. However,
we noticed that the number of such pixels was not significant in comparison to the
total number of extracted flood pixels (see Table III). In general, the overlap between
the SAR-derived flood extent and the shadow mask is restricted to the border regions
of large clusters of pixels, which were correctly classified as flooded.
Furthermore,
it can be observed from the results in Table III that M2b helps in significantly
reducing the number of pixels classified as flood water in the shadow regions.
This
is due to the fact that parts of the shadow-affected areas are included in the mask of
permanent water surface-like radar response areas described earlier. By considering a
reference image acquired from the same orbital track as the target image, the method
termed M2b reduces the risk of classifying shadow areas as flooded.
C. Evaluation at street level (qualitative/thematic analysis)
The benefits of using a reference image (including the masking of permanent
smooth areas) become obvious when looking at the spatial distribution of errors (Fig.
6). The application of algorithms M2a and M2b leads to the expected reduction of
misclassified pixels in urban areas. Numerous scattered clusters of pixels that were
initially erroneously classified as flooded could be removed, thereby
significantly
reducing over-detection. In Table IV a thematic analysis with a special focus on urban
features complements the quantitative analysis presented earlier (Table II).
24
The objective is to understand the advantages and limitations of the three variants
of the SAR-based flood delineation algorithms for correctly identifying flooding in
urban areas. From the results depicted in Table IV, it can be observed that in spite of
the high-resolution SAR imagery used in this study, the detection of flooding in builtup
environments remains a very challenging task. All algorithms struggle to recognize
the flooding status of many small-scale features that might be crucial as their state of
flooding could mean significant interruptions of everyday life. However, overall, the
enhanced algorithm M2b performs best with a slightly reduced number of
misclassified areas. In particular, M2b enables the a priori delineation of areas
characterized by specular-like reflections (i.e. areas with permanent water surface-like
radar responses). This is helpful given that smooth areas (e.g. R2 & R3) tend to be
systematically classified as flooded by M1 and, to a lesser extent, by M2a. On the
other hand, more open areas, such as the main roads R12 and R13, are correctly
classified by all three methods. It is worth mentioning that, despite these somewhat
encouraging results, M2b fails to correctly delineate flooding in many densely
vegetated and built-up environments.
These errors will be analyzed in more detail in the following sections. In this
analysis we will consider ancillary data (e.g. land use map, oral communications from
local experts) to better understand the reasons that are at the origin of the remaining
misclassifications. Moreover, the mask of the region unseen by the satellite, i.e.
shadow and layover, has also been taken into account for error detection at street
level.
25
1) Analysis of errors: the problem of over detection
As it can be seen from the urban flood maps presented in Fig. 6 both over- and
under-detection are reduced as a result of applying the M2b algorithm rather than its
predecessors. This is particularly evident in the case of the large shopping mall
labelled R3. Due to the flatness of its roof and resulting specular reflection it was
erroneously classified as flooded by M1, while M2a and M2b correctly excluded it
from the flooded area. This emblematic example best illustrates the potential added
value of reference images as they enable the a priori identification of the majority of
smooth areas.
Similarly, other wide flat regions, such as parking lots and airfields, are
recognizable in the reference image. For instance, the region labelled R1 corresponds
to large parking lot composed of three parts. M2b completely removes one of them
from the flood extent map, while the two other parts are significantly reduced in size.
The sub-optimal performance of M2b is arguably due to a difference in the number
and placement of vehicles at the time of the two satellite overpasses. In very high
resolution SAR imagery the presence or not of an object like a car inevitably impacts
the radar response. This necessarily influences the capability of the M2b algorithm to
reliably identify areas of smooth tarmac and unfortunately may not be resolvable at
all, for obvious reasons.
The region labelled R2, an area both flat and made of tarmac but not used as a
parking space, shows the capability of M2b to avoid the typical misclassifications of
smooth areas as flooded.
Finally the thematic analysis confirms the algorithms ability for identifying
permanent water bodies. The permanently flooded bed of the River Avon and some
adjacent boat marinas are removed from the flood extent map when taking into
26
account the reference image: this becomes evident when looking at the areas of over
detection in the panels (b) and (c) of Fig. 6. Clearly, this result is not achievable with
a single flood image, as M1 would invariably classify permanent water bodies as
flooded (see panel (a) of Fig. 6).
Despite the previously mentioned ability of M2b to detect areas with permanent
low backscatter values, there are still some shadow-affected areas that are erroneously
classified as flooded. A typical example is a large inclined rooftop in region R9.
This is classified as flooded by all three methods due to the fact that one
side is not
visible to the SAR sensor. Other examples of this behaviour can be found
on
various inclined rooftops in the R7 region.
To summarize, some risk of over-detecting flooded areas in built-up environments
inevitably remains. Non-flooded areas that appear smooth and water-surface like at
radar wavelengths as well as areas unseen by the satellite because of the side-looking
nature of SAR systematically produce very low signal returns and are not easily
distinguishable from flooded areas. The results of this study suggest that taking into
account the baseline backscatter values from dry reference images partly
addresses
the problem. Wide, open areas of tarmac or concrete (roads, parking lots, airfields
etc.) can be identified and removed from the final flood map (or, alternatively,
categorized as areas impossible to classify), while the situation is more problematic
with shadow areas. To check the plausibility of both types of regions to be flooded,
we expect that the use of high-resolution high-precision DEM data may be helpful.
More research on the integration of additional data sources into the image-processing
algorithm is needed for this to provide significant advantages.
27
2) Analysis of errors: the problem of under detection
With respect to the problem of under-detecting the true flood extent using SAR
observations, the results depicted in Table II indicate that method M2b leads to a
decrease in performance. In fact, the percentage of under-detected flood pixels rises
from 15.6 % obtained with the initial M2a method to its M2b-related value of 16.2 %.
However, it has to be underlined that this type of error is generally to be found on the
edge of inundated fields or in the vicinity of the main riverbed with tall vegetation
surrounding the areas. While the algorithm accurately retrieves most of the flooded
areas in wide, open areas, it can be observed that it systematically fails to retrieve
flooding under the vegetation canopy. These errors are not related to the imageprocessing
algorithm; rather they are due to the fact that with X-band radar systems
volume scattering originating from the vegetation canopy causes increased signal
return (i.e. layover). Furthermore, as already mentioned in paragraph 3.2, it cannot be
ruled out that the validation flood extent itself is affected by a slight overestimation,
as it was acquired closer to peak discharge than the satellite images. This could also at
least partly explain the under detection documented in the contingency matrix. Also,
the uncertainties in the delineation of the flood validation extent form aerial
photography are expected to have some marginal effect.
For example, an important area of apparent under-detection is the triangular
shaped field labelled R15. However, a closer look at the data reveals that due to the
time difference between the acquisitions of aerial photographs satellite imagery, it is
likely that that most of the floodwater was drained from the field in the 19 hours
preceding the TerraSAR-X acquisition. This hypothesis is confirmed by hydraulic
model simulations (see Fig. 3(b)).
28
The roughening of water surfaces due to wind is another inherent and potentially
significant limitation of the algorithm proposed in this study. When there are regular
waves on the surface of the water, Bragg resonance can result in very high signal
returns [23]. The misclassification of the area labelled R18 as non-flooded represents
a typical example. From the air photos and model simulations there can be no doubt
about the flooding of the area. However, waves are clearly identifiable on the standing
water. This renders accurate flood detection extremely difficult (if not impossible), as
it violates the algorithms main underlying assumption of flooded areas

behaving as
specular reflectors.
V. CONCLUSIONS
This study proposes a promising methodology that is shown to be capable of
providing satisfactory results in mapping flood extent in a challenging case study
However, some further improvements are still necessary before the deployment of a
fully automated SAR-based flood delineation algorithm operating in a near-real time
can be envisaged.
The proposed algorithm shares some of the characteristics advocated by reference
[5]. In fact, since no manual (and subjective) input is required from the end user, the
algorithm enables an automated, objective and repeatable flood detection. The
algorithm is computationally efficient and operates with minimum data requirements,
considering as input data a flood image and a reference image acquired before or after
the flooding.
The difficulty of detecting flooded areas in a built-up environment could be
partially addressed by a change detection approach that makes use of pre- or postflood
reference images available in the data archives of satellite data providers. In
29
particular, the shadow effect stemming from man-made structures can be taken into
account through a mask of permanent water surface-like radar response areas. This
approach overcomes the need of a high-resolution DEM and a SAR simulator for
determining shadow regions that are not visible to the satellite. On the other hand, it
requires a reference image with the same imaging characteristics as the flood image.
While the number of suitable candidate images can be very limited in case of
relatively new satellites, such as TerraSAR-X, it is important to note that image
archives are gradually being built up, which will progressively increase the likelihood
of finding adequate reference images in the online archives.
In order to further improve the method, we aim at taking advantage of topographic
and land use data, which are available at global scale, albeit with variable accuracy
and resolution. We hypothesize that such ancillary data will help to reduce the
elevation curvature along the flood edges as argued by reference [8] and to identify
parts of the under-detection caused by emerging objects such as trees of buildings.
We consider this study to be timely because there is a clear need for rapidly
acquiring, processing and distributing hydrology-related information derived from
SAR imagery. For near real-time applications in hydrology, where flood extent data is
systematically assimilated into hydrologic-hydraulic models, the value of remote
sensing data is much higher if rapidly available [24]-[25].
AKNOWLEDGMENTS
This study was part of the HYDRASENS project, financed by the National Research
Fund (FNR) of the Grand Duchy of Luxembourg and the Belgian Federal Science
Policy Office in the framework of the STEREO II research programme (Contract nr.
SR/00/100). The authors wish to thank the Environment Agency of England and
30
Wales for the aerial photography of the Tewkesbury flood and the UK Meteorological
Office for granting access to the NERC BADC for the rainfall records used in the
analysis. The TerraSAR-X flood image was received under DLR SSS project
HYD0363.
31
REFERENCES
[1] D. E. Alsdorf, E. Rodriguez, and D. P. Lettenmaier, Measuring surface water
from space, Rev. Geophys., vol. 45, no. 2, pp. 1-24, May 2007.
[2] W. A. Marcus and M. A. Fonstad, Optical remote mapping of rivers at submeter
resolutions and watershed extents, Earth Surf. Processes Landforms, vol. 33, no. 1,
pp. 4-24, Jan. 2008.
[3] G. Schumann, P. D. Bates, M. S. Horritt, P. Matgen, and F. Pappenberger,
Progress in integration of remote sensingderived flood extent and stage data and
hydraulic models, Rev. Geophys., vol. 47, Nov. 2009.
[4] P. Matgen, R. Hostache, G. Schumann, L. Pfister, L. Hoffmann, and H. H. G.
Savenije, Towards an automatic SAR-based flood monitoring system. Lessons
learned from two case studies, Physics and Chemistry of The Earth, vol. 36, no. 7-8,
pp. 241-252, 2011.
[5] S. Martinis, A. Twele, and S. Voigt, Towards operational near real-time flood
detection using a split-based automatic thresholding procedure on high resolution
TerraSAR-X data, Nat. Hazards Earth Syst. Sci., vol. 9, no. 2, pp. 303314, 2009.
[6] S. Martinis, A. Twele, and S. Voigt, Unsupervised extraction of flood-induced
backscatter changes in SAR data using Markov image modeling on irregular graphs,
IEEE Trans. Geosci. Remote Sens., vol. 49, no. 1, pp. 251-263, 2011.
[7] L. Pulvirenti, N. Pierdicca, M. Chini, and L. Guerriero, An algorithm for
operational flood mapping from Synthetic Aperture Radar (SAR) data using fuzzy
logic, Nat. Hazards Earth Syst. Sci., vol. 11, pp. 529540, Feb 2011.
[8] D. C. Mason, R. Speck, B. Devereux, G. Schumann, J. Neal, and P.D. Bates,
Flood detection in urban areas using TerraSAR-X, IEEE Trans. Geosci. Remote
Sens., vol. 48, no. 2, pp. 882-894, Feb. 2010.
32
[9] R. Speck, P. Turchi, and H. S, An end-to-end simulator for high resolution
spaceborne SAR systems, in Proc. SPIE Defense Security, 2007, vol. 6568, p. 656
80H.
[12] D. C. Mason, I. J. Davenport, J.C. Neal, G. J-P. Schumann, and P. D. Bates,
Near real-time flood detection in urban and rural areas using high resolution
synthetic aperture radar images, IEEE Trans. Geosci. Remote Sens., in press, 2012.
[13] F. Ulaby, R. Moore, and A. Fung, Microwave remote sensing, active and passive,
Vol. III: Volume scattering and emission theory, advances system and applications.
Norwood (MA), USA: Artech House, 1986.
[14] G. A. F. Seber and C. J. Wild, Nonlinear Regression. Berlin: John Wiley & Sons,
2003.
[15] R. M. Haralick and L. G. Shapiro Image segmentation techniques,
Comput.
Vision Graph. Image Process., vol. 29, no. 1, pp. 100-132, Jan. 1985.
[16] R. Hostache, P. Matgen, W. Wagner, and L. Pfister, Change detection
approaches to flood extent mapping: How to select the best pre-flood reference image
from online archives?, Geophysical Research Letters, under review.

[17] Environment Agency, Review of 2007 summer floods. Dec. 2007. Available
online at http://publications.environment-agency.gov.uk/PDF/GEHO1107BNMI-EE.
pdf
[18] G. J-P. Schumann, J. C. Neal, D. C. Mason, and P. D. Bates, The accuracy of
sequential aerial photography and SAR data for observing urban flood dynamics, a
case study of the UK summer 2007 floods, Remote Sens. Environ., vol. 115, no. 10,
pp. 2536-2546, 2011.
33
[19] A. Lops, E. Nezry, R. Touzi, and H. Laur, Structure detection and statistical
adaptive speckle filtering in SAR images, Int. J. Remote Sens., vol. 14, no. 9, pp.
1735-1758, 1993.
[20] H. Zwenzner and S. Voigt, Improved estimation of flood parameters by
combining space based SAR data with very high resolution digital elevation data,
Hydr. Earth Syst. Sci., vol. 13, no. 5, pp. 567-576, May 2009.
[21] J.C. Neal, G. Schumann, T. Fewtrell, M. Budimir, P. D. Bates, and D. C. Mason,
Evaluating a new LISFLOOD-FP formulation with data from the summer 2007 flood
in Tewkesbury, UK, J. of Flood Risk Manage., vol. 4, no. 2, pp. 88-95, Mar. 2011.
[22] E. Malnes, S. Solbo, I. Lauknes, G. J. Evertsen, T. A. Tollefsen, I. Solheim, and
M. Indregard, Floodman-global near real-time flood monitoring for hydrological
users, International conference: Innovation, advances and implementation of flood
forecasting technology, 17 to 19 October 2005, Troms, Norway, 2005.
[23] D. OGrady, M. Leblanc, and D. Gillieson, Use of ENVISAT ASAR Global
Monitoring Mode to complement optical data in the mapping of rapid broad-scale
flooding in Pakistan, Hydrol. Earth Syst. Sci. Discuss., vol. 8, no. 3, pp. 5769-5809,
2011.
[24] P. Matgen, M. Montanari, R. Hostache, L. Pfister, L. Hoffmann, D. Plaza, V. R.
N. Pauwels, G. J. M. De Lannoy, R. De Keyser, and H. H. G. Savenije, Towards the
sequential assimilation of SAR-derived water stages into hydraulic models using the
Particle Filter: proof of concept, Hydrol. Earth Syst. Sci., vol. 14, no. 9, pp. 17731785, 2010.
[25] L. Giustarini, P. Matgen, R. Hostache, M. Montanari, D. Plaza, V. R. N.
Pauwels, G. J. M. De Lannoy, R. De Keyser, L. Pfister, L. Hoffmann, and H. H. G.
34
Savenije, Assimilating SAR-derived water level data into a flood model: a case
study Hydrol. Earth Syst. Sci., vol. 15, no. 7, pp. 2349-2365, 2011.
35
FIGURE CAPTIONS
Fig. 1. General scheme of the three processing steps of the flood detection algorithm
M2b.
Fig. 2. (a1) Flood image (July 25 2007) and (b1) post-flood reference image (July 22
2008). Zoom in the area of interest (city of Tewkesbury) for (a2) the flood image and
(b2) the reference one.
Fig. 3. (a) Flood validation map obtained from high resolution aerial photography on
July 24 2007 at 11:30: the permanent water bodies are also displayed; (b) comparison
between the LISFLOOD computed flood extent at the time of aerial photographs
acquisition (July 24 2007, 11:30 GMT) and TerraSAR-X overpass (July 25 2007, 6:34
GMT).
Fig. 4. Optimization of the parameters of the automated algorithm for M2b method:
Gamma pdf mode, 0
m , backscatter threshold, 0
thr , region growing threshold (or
maximum water backscatter), 0
rg , CD threshold, 0; the backscatter histogram of
the flood image is also displayed in dark blue colour.
Fig. 5. (a) Backscatter histogram of the reference image with superimposed the
threshold value
0
, computed on the flood image; (b) reference mask: pixels in the

reference image with values lower than
thr
0
.
Fig. 6. Contingency map deriving from method: (a) M1, (b) M2a, (c) M2b. For sake
of clearness in the representation, the displayed maps have been cleaned by
neighbourhood analysis in post-processing step.
36
Fig. 7. RMSE values computed for different region growing thresholds (M2b method)
during the optimization process and corresponding performances in terms of correctly
predicted pixels (as flooded and as non-flooded).
Fig. 8. Backscatter probability density function of water pixels from the high
resolution (HR) photographs and water pixels from the method: (a) M1, (b) M2a, (c)
M2b. The histograms refer to the algorithm output, with no post-processing cleaning
step included.
Fig. 9. (a) Main rural and urban areas overlapped on the pixels in the flood image
covered by water according to the high resolution aerial photographs; (b)
corresponding backscatter probability density functions.
Fig. 10. Mask of regions unseen by TerraSAR-X due to shadow and layover, from
reference [8].
37
TABLES
TABLE I
CHARACTERISTICS OF THE AVAILABLE IMAGES (STRIPMAP MODE) FOR
THE ANALYZED FLOOD EVENT.
thr
IMAGE Date track orbit

pixel
spacing
(m)
ground
resolution
(m)
band
wave
length
(GHz)
polarization
Average mean
incidence angle
on study area
()
flood
July 25 2007
6:34
descending 109 1.5 3 X 9.6 H/H 24
reference
July 22 2008
6:34
descending 109 1.5 3 X 9.6 H/H 24
TABLE II
QUANTITATIVE EVALUATION OF TERRASAR-X DERIVED FLOOD
EXTENT.
0
RGthr 0 over
detection
under
detection
total
good
total
error
(dB) (%) (dB) (%) of the area of interest, red box in Figures
M1 -13.4 96 - 2.8 15.6 81.6 18.4
M2a -11.7 99 -3.1 2.6 15.6 81.8 18.2
M2b -12.5 98 -3.0 2.1 16.2 81.7 18.3
rg
TABLE III
QUANTITATIVE ANALYSIS OF WATER PIXELS (PIXEL SIZE 1.5M) IN THE
SHADOW REGIONS (MASON ET AL., 2010).
n of water
pixels
in shadow
n of
water pixels
water pixels
in shadow
(%)
M1 8648 374060 2.3
M2a 8738 371734 2.3
M2b 7304 357616 2.0
38
TABLE IV
IMPROVEMENT DERIVING FROM THE USE OF A REFERENCE IMAGE:
COMPARISON OF THE DIFFERENT METHODS FOR SOME REGIONS, WITH
A SPECIAL EMPHASIS ON URBAN FEATURES. SEE FIG. 6 FOR THE
LOCATION OF STREETS, CROSSINGS AND URBAN/RURAL AREAS (THE
CORRECTLY CLASSIFIED REGIONS OF EACH METHOD ARE IN BOLD
FONT).
ID region HR photographs method M1 method M2a method M2b
R1 parking lots non-flooded flooded partially flooded partially flooded
R2 tarmac area non-flooded flooded non-flooded non-flooded
R3
supermarket roof
(Morrisons Store)
non-flooded flooded non-flooded non-flooded
R4 road crossing non-flooded non-flooded non-flooded non-flooded
R5
Road
(Barton Road)
non-flooded flooded flooded flooded
R6 field on hillslope non-flooded flooded flooded flooded
R7 roof non-flooded flooded flooded non-flooded
R8 urban area non-flooded partially flooded less flooded less flooded
R9 roof non-flooded flooded flooded flooded
R10 road crossing non-flooded flooded partially flooded partially flooded
R11
Road
(East Street)
non-flooded non-flooded flooded non-flooded
R12
Road
(Chance Street)
non-flooded non-flooded non-flooded non-flooded
R13
road
(High Street)
non-flooded non-flooded non-flooded non-flooded
R14 parking lot partially flooded flooded partially flooded partially flooded
R15 field flooded (dubious) non-flooded non-flooded non-flooded
R16
road
(Ashchurch Road)
flooded partially flooded partially flooded partially flooded
R17 road crossing flooded non-flooded non-flooded non-flooded
R18 parking lot flooded partially flooded partially flooded partially flooded
R19 urban area flooded partially flooded partially flooded partially flooded
R20 urban area flooded non-flooded non-flooded non-flooded
R21
Road
(Knights Way)
flooded non-flooded non-flooded non-flooded
39
Laura Giustarini received the B.Sc. and M.Sc. degrees in environmental engineering
from the University of Perugia, Perugia, Italy, in 2003 and 2006, respectively. She has
worked for the National Research Council, Research Institute for Geo-Hydrologicl
Protection, Perugia, Italy. Since 2010, she has been with the Department of
Environment and Agro-Biotechnologies, Public Research Centre Gabriel Lippman,
Belvaux, Luxembourg. Her current research interests are the integration, through data
assimilation techniques, of radar remote sensing into coupled hydrologic and
hydraulic models for surface water management.
Renaud Hostache received the M.Sc. degree in mechanics of geophysical media and
environment from the Joseph Fourier University, Grenoble, France, in 2003 and the
Engineering degree in hydraulics and environmental sciences from Grenoble Institute
of Technology-National School of Hydraulics and Mechanism of Grenoble, Grenoble.
He has undertaken his Ph.D. research work with the Cemagref, Montepellier, Fgrance.
His Ph.D. investigated the fine 3D characterization of flood hazard, owing to datellite
imagery and its integration in flood inundation models. He received the Ph.D. degree
in water sciences from the National School of Agriculture, Water and Forest
Engineering, Montepellier, in 2006. Since 2007, he has been with the Department of
Environment and Agro-Biotechnologies, Public Research Centre Gabriel Lippman,
Belvaux, Luxembourg. His current research interests are focused on the integration of
remote-sensing observations in hydraulic models in the hydrologic and hydraulic
model uncertainty characterization and reduction.

40
Patrick Matgen earned his M.Sc. in environmental engineering at the Ecole
Polytechnique Fdrale de Lausanne (Switzerland). In 2011 he was awarded the Ph.D
degree from the water resources section of the Technical University in Delft (The
Netherlands) for his dissertation on the retrieval of surface and subsurface water from
microwave remote sensing observations and the integration of the data with flood
prediction systems. He currently holds the position of project leader responsible for
microwave remote sensing and hydrologic-hydraulic modeling projects at the Centre
de Recherche Public Gabriel Lippmann (Grand Duchy of Luxembourg).
Guy J.-P. Schumann received the M.A. degree in geography and environmental
science and the M.Sc. degree in remote sensing image processing and applications
from the University of Dundee, Dundee, U.K., in 2003 and 2005, respectively, and the
Ph.D. degree in collaboration with the Public Research CentreGabriel Lippmann,
Belvaux, Luxembourg, in 2008. He is currently with the School of Geographical
Sciences, University of Bristol, Bristol, U.K. His research interests include flood
hydrology and hydraulics and remote sensing in hydrology, in particular, radar remote
sensing for flood hydrology and management. Dr. Schumann was the recipient of a
Great Western Research Fellowship in Hydrology from the School of Geographical
Sciences, University of Bristol.
Paul D. Bates received the Ph.D. degree from the University of Bristol, Bristol, U.K.,
in 1993 with support from a Natural Environmental Research Council studentship.
Subsequently, he has worked with the University of Bristol as a Postdoctoral
Researcher and Lecturer and has been a Full Professor since 2003. He is currently the
Director of the Hydrology Research Group with the School of Geographical Sciences,
Bristol. He has been a Visiting Scientist with the Laboratoire National
DHydraulique, Princeton University, Paris, and the European Union Joint Research
Centre, Ispra, Italy. His research concerns the development and analysis of numerical
models for predicting river flood flows, principally using data derived from remote
sensing sources. He has specific interests in spatial prediction, risk, and uncertainty.
Prof. Bates is the Editor-in-Chief of the International Journal of River Basin
Management.
41
David C. Mason received the B.Sc. and Ph.D. degrees in physics from the University
of London, London, U.K., in 1963 and 1968, respectively. He has worked for the
U.K. Medical Research Council and Plessey Electronic Systems Research. Since
1984, he has been with the Natural Environment Research Council Environmental
Systems Science Centre, University of Reading, Reading, U.K., carrying out research
on the automated extraction of information from remotely sensed data and linking
these data to environmental models. His current interests include using remotely
sensed data for validation and parameterization of river flood models and assimilation
into coastal morphodynamic models.
(a)
(b)
(a)
(b)
(c)
(d)
(e)
(f)
(a) (b)
(b)
(c)
(a)
(a)
(b)
(c)
(a)
(b)
int. j. remote sensing, 2002, vol. 23, no. 18, 36813696
An eYcient method for mapping ood extent in a coastal oodplain

using Landsat TM and DEM data
Y. WANG*, J. D. COLBY and K. A. MULCAHY
Department of Geography, East Carolina University, Greenville, NC 27858,

USA
(Received 11 April 2000; in . nal form 9 July 2001)
Abstract. An eYcient and economical method for mapping ooding extent in a
coastal oodplain is described. This method was based on the re ectance features
of water versus non-water targets on a pair of Landsat 7 Thematic Mapper (TM)
images (before and during the ood event), as well as modelling inundation using
Digital Elevation Model (DEM) data. Using limited ground observation, most
ooded and nonooded areas derived from this analysis were veri. ed. Utilizing
only TM data, the total ooded areas in Pitt County, North Carolina on
30 September 1999 was 237.9 km2 or 14.0% of the total county area. This number
could be low due to the underestimation of the ooded areas beneath dense
vegetation canopies. To further investigate this underestimation, a subset of the
area covering the four central topographic quadrangles, the Greenville area, in
Pitt County was selected. Through addition of the DEM data into the ood
mapping analysis of the Greenville area revealed that the total ooded area was
98.6 km2 (out of a study area of 593.9 km2) or 16.5%. In the Greenville study
area, the three landuse and landcover categories most aVected by the ood were
bottomland forest/hardwood swamps (32.7 km2), southern yellow pine (28.8 km2),
and cultivated land (19.1 km2). Their total ooded areas were 80.6 km2 or 81.7%
of the total ooded area within this study area. The DEM data helped greatly in
identifying the ooding that occurred underneath forest canopies, especially within
bottomland forest and hardwood swamps. The method was reliable and could
be applied quickly in other coastal oodplain regions using data that are relatively
easy to obtain and analyse, and at a reasonable cost. This method should also
work well in areas of large spatial extent where topography is relative at.
1. Introduction
During an extreme ood event it is important to be able to determine quickly
the extent of ooding and the landuse and landcover types under water. This
information can be used in developing a comprehensive relief eVort (Corbley 1993).
During ooding events remotely sensed data can provide signi. cant mapping capabilities.
However, obtaining remotely sensed data that represents the ideal combination
of . ne spatial and temporal sampling, and the ability to see through clouds and/or
to discriminate ooding under forest cover is a diYcult task. In addition, accessibility
*e-mail: wangy@mail.ecu.edu
Internationa l Journal of Remote Sensing

ISSN 0143-1161 print/ISSN 1366-590 1 online 2002 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080/01431160110114484
3682 Y. Wang et al.

to the data in terms of cost, ease of acquisition, and ease in data processing and
analysis are signi. cant factors.
On 2 September 1999, Hurricane Dennis visited the Outer Banks of North
Carolina. It then returned as a tropical storm on 5 September, spreading signi. cant
precipitation across eastern North Carolina and left the ground saturated. On
15 September 1999, Hurricane Floyd made landfall near the South CarolinaNorth
Carolina border and proceeded to churn through eastern North Carolina, dumping
2546 cm of rain in many areas in less than 72 h. On 17 September the Tar, Neuse,
Roanoke and Pamlico Rivers were predicted to reach ood stage, and to continue
to rise for several days. The devastation due to ooding from these storms and two
additional precipitation events in late September was immense. Within a few days
oodwaters rose and covered over 50 000km2, causing an unprecedented disaster in
the eastern region of state as one of the worst oods in history inundated eastern
North Carolina. In addition to the loss of over 50 lives, more than 6000 homes were
destroyed and some 44 000 were damaged. Estimates indicated that losses could
exceed $6 billion (Gares 1999).
In response to the extensive ooding that occurred after Hurricane Floyd in
eastern North Carolina, we developed an eYcient method for mapping ood extent
that used Landsat 7 Thematic Mapper (TM) imagery, as well as Digital Elevation
Model (DEM) data. This method provides . ne spatial sampling and determines
ooding under forest cover within a oodplain, with data that is relatively inexpensive
and easy to obtain, process, and analyse.
Studies of mapping ood extent using Landsat TM data (e.g. Dartmouth Flood
Observatory 1999) have noted the inability of the imagery to identify ooded areas
under forest cover. Jin (1999) developed a ooding index using the Speci. c Sensor
Microwave/Imager (SSM/I ) data of the Defense Meteorologic Satellite Program
(DMSP). Synthetic Aperture Radar (SAR) data can penetrate cloud cover, and have
been applied to mapping ooded areas of the Amazon rainforest (e.g. Hess et al.
1995, Melack and Wang 1998, Miranda et al. 1998), monsoon ood damage in
Bangladesh (ImhoV et al. 1987), and river ood waves in the Great Upper Mississippi
Valley ood of 1993 (Brakenridge et al. 1998). ImhoV et al. (1987) also incorporated
the use of Landsat Multi-Spectral Scanner (MSS) data and inventoried landcover
classes inundated during ooding.
DEMs have been used in various ways to aid in ood mapping and modelling.
They have been used as an integral part of a Geographic Information System (GIS)
database applied to hydrologic ood modelling eVorts (e.g. Muzik 1996, Correia
et al. 1998). They can also be applied towards veri. cation of insurance claims after
a ood (Barnes 1996). In addition, the delineation of oodplains and the development
of ood inundation maps have relied on DEMs (e.g. Jones et al. 1998). The recognition
of error in DEMs is an important concern. However, it has been investigated
(Brown and Bara 1994), and a few examples have been provided in the literature
that is applicable to oodplain mapping (e.g. Lee et al. 1992, Hunter and
Goodchild 1995).
In this paper, we present a method formapping ood extent in a coastal oodplain
through the use of TM data, as well as DEM data. First, we describe the study area,
and the extent of damage due to ooding from the hurricane. Then we provide a
discussion of the TM, DEM, and landuse/landcover data and initial processing steps,
and describe ground observations. Next, the ood mapping eVorts are explained
using TM data for Pitt County, and TM and DEM data for a subset of Pitt County.
Flood mapping using TM and DEM data 3683
We then present the results. We also discuss the potential for applying the ood
mapping methods (using TM data alone, and combining TM and DEM data) in
coastal oodplains in general, and the limitations and cautions that should be noted
when applying these methods to mapping ood extent in areas of large spatial extent,
and areas having large topographic variation.
2. Analytical approaches
2.1. Study area and ground observation
Most of eastern North Carolina lies within the Atlantic coastal plain. Pitt County
lies in the eastern coastal plain of North Carolina, at the approximate centre of the
region. The elevation of the area drops only about 60m as it extends 120160 km
from the Piedmont region in the middle of the state towards the coast. Four large
elongated river systems drain the coastal plain in a north-westsouth-east direction.
Flat broad oodplains are usually located on the northern side of the rivers with
higher ground on the south (Gares 1999). In Pitt County the land surface has very
low relief and many parts of the region have been extensively drained, cleared and
ditched for agricultural use. The soils are primarily characterized as poorly drained
or extremely poorly drained (63.0%), with the remaining area consisting of moderately
well to well-drained soils (Gares 1999). Pitt County has a population of about
126 000 (estimated in 1998). The largest city, Greenville, is centrally located and has
a population of approximately 60 000 (estimated in 1998). The additional residents
of the county are spread throughout rural towns.
In Pitt County the majority of the 1999 ooding occurred north of the Tar River.
The Tar River has been slowly migrating southward towards the drainage divide of
the Neuse River so broad primary and secondary oodplains extend northward from
the river channel. North and immediately adjacent to the Tar River is a band
following the channel that is currently de. ned as conservation/open space landuse
in the City of Greenville (. gure 1). This conservation/open space landuse zone grades
into low and medium density residential, industrial and mixed land uses. It is clear
from the photograph, however, that there is signi. cant activity within this open
space landuse. The City of Greenville alone suVered ooding to its airport, water
treatment facility, power transmission substation and numerous residential and
industrial areas that are within or nearby current open space landuse zones. In Pitt
County, some 6000 homes were ooded. Over three-quarters of these homes were
largely uninsured. Upwards of 50 000 people were displaced. More than 6000 were
housed in emergency shelters, many for over 3 weeks.
Once the oodwaters had completely subsided, but before high water marks
faded, ground data information was gathered in the . eld. Areas both north and
south of the Tar River were examined for the extent and depth of ood waters. The
aerial photo (. gure 1) was taken on 23 September 1999 during the ood event and
is centred approximately on the City of Greenville. The ood gauge information
used for this study was taken from the gauge on the Green Street Bridge (. gure 1).
Floodwaters extend into the student housing district seen in the south-eastern section
of the photo and throughout the entire area shown north of the river. Areas of
extensive tree canopy north of the river in the primary and secondary ood plain
were completely ooded. Figure 2 shows the high water marks (reaching the middle
of windows) on a house trailer in a trailer park that is located immediately adjacent
to the north-eastern most section of the photo. These areas of tree canopy were not
3684 Y. Wang et al.
Figure 1. Aerial photo of a portion of the City of Greenville, North Carolina, taken during
the September 1999 ood. The major part of the City of Greenville is on the south
side of the Tar River.
classi. ed in the TM images as ooded but, clearly, they were ooded to a
signi. cant depth.

2.2. Remotely sensed data
For ood mapping, two sets of the remotely sensed data are required; one set
consisting of data acquired before (and as close as possible to) the ood event, and
the other acquired during the occurrence of the ood. In reality, data availability
may cause some compromise. In this study, the Tar River reached peak ood stage
in the area of Pitt County on 21 September 1999. The Landsat 7 TM data that were
available closest to this date were acquired on 30 September 1999. Due to the 16-day
repeat orbiting of the Landsat 7 satellite, the availability dates for images of pre ood data were 14 September, 29 August, 13 August and 28 July. Due to the severe
cloud coverage in the 14 September and August images, we used the image acquired
on 28 July for pre- ood analysis. (There were some thin clouds and patches of cloud
in the 28 July image that did aVect our analysis.) In summary, we ordered two TM
images, one acquired on 28 July 1999, and the other on 30 September 1999. We
geo-referenced both images and were able to determine the extent of ooding in Pitt
County using the nonooded 28 July image as a reference.
2.3. DEM data of Greenville area and river gauge readings at Greenville
DEM data were available from the United States Geological Surveys (USGS)
web site in the Spatial Data Transfer Standard (SDTS) format. The DEM has a
30m30m resolution, and in this area the elevation interval (z=height) is 0.30 m.
The accuracy of the DEM data (i.e. the uncertainty, or root mean square error,
Figure 2. Flooded mobile home in a densely vegetated trailer park on the oodplain, just
north of the Tar River, Greenville, North Carolina. The high-water mark reaching the
middle of windows is clearly shown on the side of the mobile home.
RMSE) in this area is 1 m. Four 7.5 min USGS topographic quadrangles, Greenville
NW, NE, SE and SW were downloaded, imported, and mosaiced. The four mosaiced
quads covered an area of about 600 km2. Descriptive statistics for the four-quadrangle
DEM area included min.=0 m, mode=11.9 m, median=14.0 m, mean=14.6 m,
max.=26.2 m, and standard deviation=6.4m. This study area is primarily at,
especially on the north side of the Tar River where most of the ooding occurred
(e.g. . gures 1 and 2). We then co-registered the TM and DEM data so that the same
area of interest can be easily extracted.
Flood stage on the Tar River is measured from a point 0.7m below sea level
(based on the North American Vertical Datum, NAD88). The bottom of the river
is about 2m below sea level. The Tar River leaves its banks 4m above this measuring
point. The Tar River crested at 9.2m on 21 September. On the date the imagery was
taken, 30 September, data from the USGS showed that the mean stage level for the
Tar River at the Greenville station was 6.1 m. The non- ood stage surface height of
the water in the Tar River according to the river gauge reading on 28 July 1999 was
3686 Y. Wang et al.
1.1 m. Therefore, the elevations that represented ooded areas on 30 September
ranged from 1.1 to 6.1 m. These elevations were used as a basis for classifying the
area on the Greenville topographic quadrangles into water bodies/rivers, ooded
areas, and nonooded areas, which will be discussed in detail in a later section (see
tables 5 and 6).
2.4. North Carolina landuse and landcover data
Between 1995 and 1997, the North Carolina Center for Geographic Information
and Analysis (NCCGIA), in cooperation with the NC Department of Transportation
and United States Environmental Protection Agency Region IV Wetlands Division,
contracted Earth Satellite Corporation (EarthSat) of Rockville, Maryland to generate
comprehensive landcover data for the entire state of North Carolina (Earth Satellite
Corporation 1997). There are 21 landuse and landcover type categories in the entire
state data layer. For Pitt County, only 17 categories exist ranging from highly
developed areas to unconsolidated sediment areas. To facilitate the presentation of
this paper and to provide the reader with a better understanding of the landcover
classes, we provide brief de. nitions of some categories (in table 3). The three landuse/
landcover categories that were aVected the most by the ood were bottomland forest/
hardwood swamps, southern yellow pine, and cultivated land. Bottomland forests/
hardwood swamps are areas where deciduous, dominant, woody vegetation is above
3m in height, as well as occurring in lowland and wet areas. Crown density is at
least 25%. Southern yellow pine are areas where stocking of trees is 75% evergreen
needleleaf or broad-leaf species, including the following forest types: longleaf pine,
loblolly-slash pine, other yellow pine, and pond pine. Cultivated lands are areas of
land that are occupied by row and root crops that are cultivated in distinguishable
rows and patterns. Two other important categories aVected by the oods were high
and low intensity developed areas, which contains the housing and infrastructure
for the majority of the human population in the area. High intensity developed areas
are covered by more than 80% synthetic (man-made) landcover. Low intensity
developed areas have between 50 and 80% coverage by synthetic landcover. (See
table 3 for the ooded areas for other landuse and landcover types.)
2.5. Flood mapping using TM images
The initial goal in ood mapping was to investigate the utility of the TM images
for identifying areas that were ooded or not ooded. There were two steps: (1)
identify water versus non-water areas on the TM images before and during the ood
event, respectively, and (2) compare the areas classi. ed as water or non-water on
both TM images to determine which areas represented ooding.
2.5.1. Identifying water areas versus non-water areas
There are many possible methods for identifying water versus non-water areas
using TM data (e.g. Jensen 1996). After unsuccessful trials of using supervised and
unsupervised classi. cation, and other methods, we used the addition of two
TM bands (TM4+TM7) of 28 July, and of 30 September, respectively. TM4
(0.760.90 mm, re ective infrared) is responsive to the amount of vegetation biomass,
and is useful in identifying land and water boundaries. However, it is possible to
confuse water and asphalt areas (road pavements and rooftops of buildings) in the
developed areas such as downtown, commercial/industrial areas, etc., as they appear
black on the TM4 image or they re ect little back to the sensor. On the TM7
(2.082.35 mm, mid-infrared) image, the re ectance from water, paved road surfaces,
and rooftops diVers. Thus, one can identify the water ( ooded) and non-water (non ooded) area in the developed areas, by incorporating of TM4 and TM7 into the
analysis, as detailed in table 1. This addition is done separately for the July and
September TM data. Therefore, the classi. cation rule was:
If the re ectance of pixels or areas is low in the TM4 plus TM7 image, the pixels
represented water, otherwise the pixels represented non-water or dry areas.
In the analysis, we noted that the re ectance from water, paved road surfaces,
and asphalt rooftops of buildings in the developed areas may be also distinguished
on TM5. Thus, TM5 could have been used to detect water versus non-water areas.
However, the diVerences on TM5 were slightly smaller than those on TM7 image.
Also, Banumann (1996) added two (before and during ood event) TM4 images in
his 1993 Mississippi ood analysis. He then sliced the added image into water,
ooded areas, and nonooded areas. He further added TM7 data to the combined
TM4 image to separate some confusion between the water and industrial area.
Once the representation of the re ectance values for water and non-water features
was understood, a cut-oV value could be determined to separate the two categories.
For the July TM image, the cut-oV value was 141. If a pixels DN value was less
than 141, that pixel was assigned as a water category, otherwise it would assigned
as a non-water category. For the September TM image, the cut-oV value was 109,
i.e. if a pixels DN was less than 109, that pixel was classi. ed as water, otherwise it
was classi. ed as non-water. Even though the selection of the cut-oV values may seem
to be somewhat arbitrary, we used two diVerent methods to check the cut-oV values.
One was ground truthing, and the other was the analysis of the histograms of the
TM4 plus TM7 images. In the former, ground observations along the Tar River
were used to pick the cut-oV values. These observations were made in the . eld in
early October of 1999 (e.g. . gure 2), and through the analysis of aerial photos taken
during the ood event. In the latter, the histogram of the TM4 plus TM7 image was
examined to see whether the histogram indicated the cut-oV values. This was the
case for the histogram of the July TM4 plus TM7 image; two distinct distributions
were observed from the histogram plot. There were also two identi. able distributions
in the histogram of the September TM4 plus TM7 image. However, a distinct
separation between the two distributions, was not as easily identi. ed (i.e. any DN
value ranging from 107 to 111 may be selected).
2.5.2. Determine ooded areas during the ood event
After identifying water versus non-water areas on both images (one acquired
before the ood event and the other during the ood) using the above criteria,
determination of areas that were ooded could be made. On a pixel by pixel basis,
Table 1. Re ectance of water, asphalt pavement (road surface, root of buildings, etc.), and
other non-water dry areas on TM4 and TM7 images.
Re ectance on TM4 Re ectance on TM7 Re ectance on TM4+TM7
Water low low low
Asphalt pavement low intermediate intermediate
Other dry area high high high
3688 Y. Wang et al.

there were four possible results, and the following rules were used to determine the
ooded areas:
1. If an area was classi. ed as water before and during the ood event, it was not
considered to be ooded. Water areas in the study areas are the regular river
channels, ponds, etc.
2. If an area was classi. ed as dry or non-water on the July (pre- ood) image,
and the area was classi. ed as water on the September ( ood) image, the area
was considered to be ooded.
3. If an area was dry on both the July and September images, the area was
not ooded.
4. It is possible to have an area that was classi. ed as water in the July image,
classi. ed as non-water in the September image. Possible explanations include
(a) landuse change between the dates that the imagery was acquired, and (b)
cloud eVects on the classi. cation. In our analysis, we have noticed the shadows
of some small patches of clouds in the July image. The locations shadowed
by the cloud were dry in July and September 1999.
3. Results and discussions
3.1. Flood mapping derived by TM images
Using the TM data and the method described above, a map representing ooded
areas in Pitt County on 30 September 1999, was created (. gure 3). The ooded areas
are shown in red and water (regular river channels, ponds, etc.) in blue. The small
areas shown as yellow were classi. ed as water on the July image, and non-water
category on the September image. These areas actually represent the shadows of
clouds on the July image. After ground truthing these areas that were nonooded
in September, we recoded this category as nonooded areas for further analysis (in
. gure 3). The nonooded areas are represented by the black /white image of TM
band 7 from 30 September. Table 2 summarizes the derived map by the four categories
described in 2.5.2. The major ooded areas were along the Tar River owing into
Pitt County from the north-west corner of the image and exiting the County to the
east. There were large patches of ooded areas in the north. There were ooded
areas along the tributary of the Neuse River as well (south-west side of the image,
. gure 3).
3.2. Areas ooded by each landuse and landcover type
Using the landuse and landcover data layer obtained from the NCCGIA, we
provide the following description for the derived ood map. The ooded areas along
the Tar River and along the tributary of the Neuse River were primarily bottomland
forests/hardwood swamps (e.g. . gure 1). The patches of ooded areas in the north,
north-east, and south-east of the image were mainly southern yellow pines and
cultivated lands. The three categories most aVected by the ood, in terms of size and
percentage of the total ooded area in Pitt County, were southern yellow pine
(91.7 km2 or 5.4% out of the total areas in the County), bottomland forest/hardwood
swamps (73.6 km2 or 4.3%), and cultivated lands (40.0km2 or 2.4%). It should be
noted that most of the high and low intensity developed areas were not ooded on
30 September 1999. (The oodwater from the Tar River had receded about 3m from
its crest on 21 September. The main developed areas are near the banks of the Tar
River.) It should be also noted that there was a slight diVerence of the sizes of the
Figure 3. The ood extent in Pitt County, North Carolina on 30 September 1999, derived
from a pair of Landsat TM images of 28 July and 30 September 1999. The grey
rectangle indicates the Greenville study area.
Table 2. Flood mapping of Pitt County, North Carolina, as of 30 September 1999.
Area (km2) Area (%)
Water bodies/Rivers 13.3 0.8
Flooded areas 237.9 14.0
Cloud shadows 1.2 0.1
Non- ooded areas 1444.8 85.1
Total 1697.2 100
water bodies/river channels derived from the TM image pair of 1999 and NCCGIA
landuse data layer (13.3 km2 vs 12.2 km2, cf. Tables 2 and 3). This diVerence could
be due to diVerent ways to identify water, due to the errors in our analysis and/or
3690 Y. Wang et al.
Table 3. Total areas from the landuse and landcover type data layer and ooded areas of
each landuse and landcover type in Pitt County (NC) derived from TM data.
Total areas from the
landuse data layer Flooded areas on 30
(km2) Sept. 1999 (km2) Overall (%)
High intensity developed 12.6 1.1 0.1
Low intensity developed 22.8 0.9 0.1
Cultivated 612.8 40.0 2.4
Managed herbaceous cover 40.6 3.1 0.2
Unmanaged herbaceous
coverupland 13.4 0.5 0.0
coverwetland 0.1 0.0 0.0
Evergreen shrubland 177.1 16.0 0.9
Deciduous shrubland 19.5 1.9 0.1
Mixed shrubland 26.0 2.3 0.1
Mixed upland hardwoods 0.1 0.0 0.0
Bottomland forest/hardwood
swamps 369.3 73.6 4.3
Needleaf deciduous 0.1 0.0 0.0
Southern yellow pine 350.5 91.7 5.4
Mixed hardwoods/conifers 39.1 3.4 0.2
Oak/gum/cypress 0.3 0.1 0.0
Water bodies/rivers 12.2 3.1 0.2
Unconsolidated sediment 0.7 0.1 0.0
Total 1697.2 237.8 14.0
in the landuse data layer, due simply to landuse change since the creation of the
landuse data layer in 1996, or all three.
3.3. Addition of the DEM data into the ood mapping analysis
Due to the dense or continuous canopy coverage in bottomland forest/hardwood
swamps and in some dense southern yellow pine stands, and due to the lack of
canopy penetration of the TM data, ooded areas under the canopies were not
detected by classi. cation of the TM data. This underestimation of ooded areas was
veri. ed through ground truthing and visual interpretation of low-altitude oblique
aerial photos taken during the 1999 ood. On the ood map, these undetected
ooded areas show up as patches or holes along the primary oodplain near the
riverbanks. It is important to point out this underestimation, because oods in the
coastal oodplains of North Carolina, as well as the entire East Coast and the coast
of the Gulf of Mexico often occur from the mid-summer to fall, and trees in the
oodplain are almost fully leaf-on during this period of time. Radar data (especially
radar data from a long wavelength system) can penetrate the (dense) canopies and
identify whether the areas underneath the canopies were ooded or not. However,
due to the cost of the radar (such as ERS SAR or Radarsat SAR) data we did not
incorporate them in the ood mapping analysis. One alternative is to integrate DEM
data into the analysis. There are several advantages to this integration. In the USA,
the DEM data are widely available and can be directly download from USGS/EROS
web site. (It should be noted that the availability of DEM data in other countries
might be limited.) Also, most of the bottomland forest and hardwood swamps in the
oodplain are located in places of low elevation or along the banks of rivers. By
using river gauge reading to inundate the DEM, one can map the ood underneath
tree canopies in the bottomland forest, and hardwood swamps, as well as in some
southern yellow pine stands in low elevation areas. Additionally, high-quality DEM
data work well for ood mapping in areas of relatively at terrain, as exists in this
study area and other coastal oodplains along the East Coast and the Gulf of
Mexico. The following discussion describes the integration of the DEM and TM
data for the ood mapping study.
We used four 7.5 min quadrangle s of USGS DEM data for the Greenville areas.
A grey box shown in . gure 3 indicates the coverage of the four quads, which includes
most of the Tar River in Pitt County. We then extracted the area (of the block,
. gure 4(a)) from the ood map of Pitt County (. gure 3) derived from the TM data.
Black areas on the image represent water bodies/rivers, grey represents ooded areas,
and white represents nonooded areas. A detailed statistical summary of ooded
areas for each landuse and landcover type is provided in table 4. The three categories
having the largest areas and highest percentage of ooding were southern yellow
pine, bottomland forest/hardwood swamps, and cultivated land.
We then inundated the DEM based on the river gauge readings before the ood
event and on 30 September 1999. The river gauge station is near the centre of the
four quads (. gure 1). These readings were 1.1m preceding the ood event, and 6.1m
on 30 September. Using the rules found in table 5, we reclassi. ed the DEM into
water bodies/rivers, ooded areas, and nonooded areas (. gure 4(b)). The simple
threshold used for inundating the DEM was possible due to the relatively at terrain
(no sinks) away from the river channel within our study area. The ooded areas
were located in the oodplain of the Tar River, and ooded areas did not exhibit
the canopy holes found on the TM images. By inundating the DEM data, ooded
areas under the canopies of the bottomland forest and hardwood swamps could be
identi. ed. At higher elevations and away from the river, the DEM suggested that
those areas were dry or there was no ooding. The DEM does not identify water
bodies and/or ooded areas at higher elevations. The ooded areas of each landuse
and landcover type derived from the DEM data were also tabulated (table 4), and
the largest area (22.6km2) and highest percentage (3.8%) of ooded landcover type
was bottomland forest/hardwood swamps.
The . nal ood map for the Greenville areas was derived by using the logical
OR operator to combine the ooded areas from either the TM data or the DEM
data (. gure 4(c)). Flooded areas located away from the river and at high elevation
were identi. ed by the TM data. Flooded areas near the river and its tributaries were
determined primarily by the DEM and partially by the TM data. No patches or

holes were visible on the combined ood map. This OR logic allows us to extract
the best of the TM and DEM data in the ood mapping analysis, and overcame
some of the de. ciencies of using the TM data or the DEM data alone. The use of
both TM and DEM data in ood mapping was straightforward and eYcient.
Furthermore, based on our limited ground observation and analysis of aerial photos
taken in the study area during the ood, the results were fairly accurate and reliable.
3.4. L imitation of the integration of the DEM data into the ood mapping analysis
Although the results derived from the integrated TM and DEM data were very
promising, we would like to oVer two cautions regarding the accuracy of the DEM
data and inundation of the DEM data using the river stage data.
The DEM data, created by USGS, have an estimated accuracy of 1m (RMSE).
3692 Y. Wang et al.
Figure 4. Integration of the TM and DEM data in the 1999 ood mapping study for areas
of four Greenville quadrangles. (a) Extracted ood map (from . gure 3) derived from
TM data alone, (b) inundated DEM data based on the river gauge reading on
30 September 1999, and (c) . nal ood map by combining the TM and DEM data
through a logical OR for ooded areas.
Table 4. Total and ooded areas of each landuse and landcover type in Greenville (NC) derived from
TM data alone, DEM data alone, and TM and
DEM data combined.
Total areas Flooded areas Flooded areas Flooded areas
from the derived from derived from derived from
landuse data TM data DEM data TM and DEM
layer (km2) (km2) Overall (%) (km2) Overall (%) data (km2) Overall (%)
High intensity developed 9.2 0.7 0.1 0.2 0.0 0.8 0.1
Low intensity developed 11.9 0.3 0.1 0.1 0.0 0.4 0.1
Cultivated 208.8 15.1 2.6 8.5 1.4 19.1 3.2
Managed herbaceous cover 18.9 2.1 0.4 1.3 0.2 2.5 0.4
coverupland 0.5 0.0 0.0 0.0 0.0 0.0 0.0
coverwetland 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Evergreen shrubland 74.3 7.0 1.2 2.9 0.5 8.0 1.3
Deciduous shrubland 7.8 1.1 0.2 0.7 0.1 1.3 0.2
Mixed shrubland 9.6 1.0 0.2 0.4 0.1 1.1 0.2
Mixed upland hardwoods 0.1 0.0 0.0 0.0 0.0 0.0 0.0
swamps 112.7 23.2 3.9 22.6 3.8 32.7 5.5
Southern yellow pine 123.9 28.2 4.8 2.0 0.3 28.8 4.9
Mixed hardwoods/conifers 11.4 1.2 0.2 2.0 0.3 2.6 0.4
Oak/gum/cypress 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Water bodies/rivers 4.7 1.3 0.2 2.0 0.3 1.3 0.2
Unconsolidated sediment 0.1 0.0 0.0 0.0 0.0 0.0 0.0
Total 593.9 81.2 13.9 42.7 7.0 98.6 16.5
3694 Y. Wang et al.
Table 5. Classi. cation rules based on the gauge readings before the ood event and on
30 September 1999. (The interval of the DEM data is 0.3m.)
Min. Max.
Water bodies/rivers 1
Flooded areas >1 6
Non- ooded areas >6
In some areas the RMSE may be higher, for example, under canopies due to the
DEM generation procedure used by the USGS. We re-ran the analysis with the
DEM dataset at 1m of the river gauge reading (with the integration of TM data) .
Recoding the elevation to represent ooding at 1m less than the river gauge reading
did not signi. cantly change the pattern of the ood (i.e. 95.3km2 vs 98.6 km2 total
ooded areas and 16.5% vs 16.1%, tables 6 and 4). The elevation data recoded at
1m above the gauge reading did expand the ood to the north of the river considerably,
as this is an area of very low relief (110.4km2 vs 98.6 km2, tables 6 and 4). Due
to the estimated accuracy of the DEM data, the total ooded areas derived from
the combination of TM and DEM data in the Greenville areas could vary from
95.3 km2 to 110.4 km2, and the ooded areas from 16.1% to 18.6% of the total study
area (table 6).
Flooding of the DEM data only works for a reasonable distance from the river
gauge from which you measure stage height. This can be a signi. cant distance in
Table 6. Flooded areas of each landuse and landcover type in Greenville derived from TM
and DEM data. The DEM data were set at 1m of the river gauge reading of Tar
River on 30 September 1999.
At 1m less than the At 1m above the river
river gauge data gauge data
Flooded Flooded
areas (km2) Overall (%) areas (km2) Overall (%)
High intensity developed 0.7 0.1 1.4 0.2
Low intensity developed 0.3 0.1 1.0 0.2
Cultivated 17.7 3.0 23.6 4.0
Managed herbaceous cover 2.4 0.4 4.0 0.7
Unmanaged herbaceous cover
upland 0.0 0.0 0.0 0.0
Unmanaged herbaceous cover
wetland 0.0 0.0 0.0 0.0
Evergreen shrubland 7.6 1.3 9.3 1.6
Deciduous shrubland 1.3 0.2 1.4 0.2
Mixed shrubland 1.1 0.2 1.2 0.2
Mixed upland hardwoods 0.0 0.0 0.0 0.0
swamps 31.8 5.4 35.1 5.9
Southern yellow pine 28.6 4.8 29.3 4.9
Mixed hardwoods/conifers 2.5 0.4 2.8 0.5
Oak/gum/cypress 0.0 0.0 0.0 0.0
Water bodies/rivers 1.3 0.2 1.3 0.2
Unconsolidated sediment 0.0 0.0 0.0 0.0
Total 95.3 16.1 110.4 18.6

areas of low relief such as the coastal plain of eastern North Carolina. This is a
relatively at region, as shown above by the summary of the statistics of the fourquad
DEM data of Greenville. To work in an area of larger spatial extent or large
variation of topography (even in a relatively small spatial extent) , stage height from
other river gauges should be incorporated and an interpolation method developed
to adequately represent ood elevation upstream and downstream. This was one
reason that we limited the use of the river gauge data collected in Greenville to leave
out inundation in the north-west corner and eastern portion of the Tar River in our
DEM data (. gure 3). However, in areas where few river gauges exist estimates have
to be made.
4. Conclusions
A simple and eYcient method for mapping ood extent in a coastal oodplain
has been presented. With limited ground observation, most ooded and nonooded
areas derived from the analysis were veri. ed. This method was based on a comparison
of the re ectance feature of the water versus non-water targets on a pair of TM
images (one acquired before and the other during the ood event), as well as by
incorporating DEM data into the analysis. The objective of incorporating the DEM
data into the analysis was to overcome the limitation of the TM data in distinguishing
between ooded areas and forest canopies. Due to the lack of penetration through
the vegetation canopies in the forested areas such as in bottomland forest and
hardwood swamps, TM data alone could not identify those ooded areas and led
to an underestimation of the ooding. This method was reliable and could be used
in the other coastal oodplains (such as the East Coast, and the coast of the Gulf
of Mexico of the USA), using similar TM images, DEM data, and river stage data.
This method should work well for areas of large spatial extent if the ( local )
topography is relatively at, as demonstrated in this study.
The total ooded areas derived from the TM data alone, on 30 September 1999
for Pitt County, North Carolina were 237.9km2 or 14.0% of the total county area.
This number may be low due to the underestimation of the ooded areas beneath
dense vegetation canopies. The landuse/landcover categories most aVected by the
ood were the southern yellow pine (91.7 km2), bottomland forest/hardwood swamps
(73.6 km2), and cultivated land (40.0 km2). Their total ooded areas were 205.3km2
or 86.3% of the total ooded areas in the County.
Through integrating the classi. cation of ooded areas from TM imagery and
from inundating a DEM to represent ooding, the following results were obtained
for the Greenville area on 30 September 1999. The total ooded area was 98.6km 2
(out of the total study area of 593.9km2) or 16.5%. The landuse/landcover categories
most aVected by ooding were bottomland forest/hardwood swamps (32.7km2),
southern yellow pine (28.8km2), and cultivated land (19.1 km2). Their total ooded
areas were 80.6km2 or 81.7% of the total ooded area in the studied area.
Incorporating the DEMdata assisted greatly in identifying the ooding that occurred
underneath the forest canopies, especially under the canopies of bottomland forest
and hardwood swamps. However, it should be noted that there were two main
limitations regarding the integration of DEM data with TM data for ood mapping.
One was the use of river gauge readings to inundate the DEM, and the other was
the handling of error in the DEM. The authors intend to investigate methods of
error representation (e.g. Hunter and Goodchild 1995) in the future. In addition, the
US Army Corp of Engineers is currently surveying high-water marks from the
3696 Flood mapping using TM and DEM data
ooding in this area, and this data will be used to assess the accuracy of the DEM
in modelling ood extent.
Acknowledgements
This research was partially funded by a grant to East Carolina University from
the Natural Hazards Research and Applications Information Center at the University
of Colorado, Boulder, CO, and partially supported by East Carolina University.
References
Banumann, P. R., 1996, Flood analysis: 1993 Mississippi ood. URL http://umbc7.umbc.edu/
~tbenja1/baumann/baumann.html, Vol. 4 of Remote Sensing Core Curriculum.
Barnes, S. B., 1996, Northwest Flood 96: GIS in the face of disaster. Geographic Information
Systems, 2225.
Brackenridge, G. R., Tracy, B. T., and Knox, J. C., 1998, Orbital SAR remote sensing of
a river ood wave. International Journal of Remote Sensing, 19, 14391445.
Brown, D. G., and Bara, T. J., 1994, Recognition and reduction of systematic error in
elevation and derivative surfaces from 7 -minute DEMs. Photogrammetric Engineering
and Remote Sensing, 60, 189194.
Corbley, K., 1993, Remote sensing and GIS provide rapid response for ood relief. Earth
Observation Magazine, September, 2, 2830.
Correia, F. N., Rego, F. C., Saraiva, M. D. S, and Ramos, I., 1998, Coupling GIS with
hydrologic and hydraulic ood modeling. Water Resources Management, 12, 229249.
Earth Satellite Corporation, 1997, Comprehensive land cover mapping for the state of
North Carolina. Final report, March 1997, Rockville, Maryland.
Dartmouth Flood Observatory, 1999, DFO-1999-076 ooding from Hurricane Floyd.
NASA-supported Dartmouth Observatory.
Gares, P., 1999, Climatology and hydrology of Eastern North Carolina and their eVects on
creating the ood of the century. North Carolina Geographer, 7, 311.
Hess, L. L., Melack, J. M., Filoso, S., and Wang, Y., 1995, Realtime mapping of inundation
on the Amazon oodplain with the SIR-C/X-SAR synthetic aperture radar. IEEE
T ransactions on Geoscience and Remote Sensing, 33, 896904.
Hunter, G. J., and Goodchild, M. F., 1995, Dealing with error in spatial databases: a simple
case study. Photogrammetric Engineering and Remote Sensing, 61, 529537.

Imhoff, M. L., Vermillon, C., Story, M. H., Choudhury, A. M., and Gafoor, A., 1987,
Monsoon ood boundary delineation and damage assessment using space borne
imaging radar and Landsat data. Photogrammetric Engineering and Remote Sensing,
4, 405413.
Jensen, J. R., 1996, Introductory Digital Image Processing: a Remote Sensing Perspective, 2nd
edn (Englewood CliVs, New Jersey: Prentice Hall ).
Jin, Y. -Q., 1999, A ooding index and its regional threshold value for monitoring oods in
China from SSM/I data. International Journal of Remote Sensing, 20, 10251030.
Jones, J. L., Haluska, T. L., Williamson, A. K., and Erwin, M. L., 1998, Updating ood
inundation maps eYciently: building on existing hydraulic information and modern
elevation data with a GIS. US Geological Survey Open-File Report 98200.
Lee, J., Snyder, P. K., and Fisher, P. F., 1992, Modeling the eVect of data errors on feature
extraction from digital elevation models. Photogrammetric Engineering and Remote
Sensing, 58, 14611467.
Melack, J. M., and Wang, Y., 1998, Delineation of ooded area and ooded vegetation in
Balbina Reservoir (Amazonas, Brazil ) with synthetic aperture radar. Verh Internat
American Journal of Geographic Information System 2013, 2(1): 1-5 DOI: 10.5923/j.ajgis.20130201.01
Application of Remote Sensing and GIS for Flood

Hazard Management: A Case Study from Sindh
Province, Pakistan Kabir Uddin , Deo Raj Gurung, Amarnath Giriraj, Basanta
*
Shrestha Mountain Environment and Natural Resources Information System, International Centre for Integrated Mountain
Development, GPO Box 3226, Kathmandu, Nepal Abstract Floods are one of the most common hazards in the world,
affecting peoples lives and livelihoods. Flood hazard mapping and flood shelters suitability analysis are vital elements in
appropriate land use planning for flood-prone areas. This paper describes application of Remote Sensing (RS) and
Geographical Information Systems (GIS) in identifying flood hazard zones and flood shelters and are therefore important
tools for planners and decision makers. The purpose of this article is to describe a simple and efficient methodology to
accurately delineate flood inundated areas, flood-hazard areas, and suitable areas for flood shelter to minimize flood
impacts. Possible extent of flooding and suitable location flood shelter sites were modeled and mapped for Sindh Province
in Pakistan, using the software ArcGIS model builder. The output was validated using inundation maps based on flood
events that took place in 2010 in Pakistan. These were mapped using object-based image analysis (OBIA) implemented in
eCognition software. The catastrophic flood of 2010 inundated a total area of 7579 km2, while the modeled result
indicated the hazard area to be 6216 km2 out of 46138 km2. Discrepancies in modeled and mapped results are
insignificant and acceptable considering the manual flood management interventions which are beyond the capability of
models to represent. Thus, this method is robust enough to develop flood hazard zoning maps and map shelter sites for
flood management. Keywords Remote Sensing, GIS, OBIA, Flood, Hazard, Sindh River, Pakistan
1. Introduction
More than half the world's population lives in Asia, which is approximately one-fifth of the earth's land area. As natural
disasters increase in both intensity and severity around the world, the Asian region continues to suffer a disproportionate
number of hazard events and related losses in lives, infrastructure, stability, and economic progress. In the period 1994
2004 alone, Asia accounted for one third of 1,562 flood disasters worldwide and nearly 60,000people were killed in
floods[1]. A flood is an overflow of an expanse of water that submerges the land. Flood hazards are the most common and
destructive of all natural disasters and are a constant threat to life and property. Each year, flood disasters result in
tremendous losses and social disruption worldwide. Over the past 30 years, floods have been the most catastrophic natural
disaster affecting, on average, about 80 million people per year or half of the total population affected by any natural
disaster, causing * Corresponding author: kuddin@icimod.org (Kabir Uddin) Published online at http://journal.sapub.org/ajgis Copyright 2013
Scientific & Academic Publishing. All Rights Reserved
economic damage worth over US$11 million annually around the world[2]. As increasing human activity downstream of rivers results in
greater flood damage; floods themselves are also increasing in size and frequency due to human activities in the upstream section of the
river system[3] In recent years, risk-based approaches have received increasing attention as a viable means to manage flood hazards.
Hazard risk and vulnerability assessment (HRVA) examines the hazards that may affect a community in order to determine the risk that
each hazard event poses to both the community as a whole and to vulnerable elements in the community.
In Pakistan, floods have been recognized as a major natural calamity. The country has a long history of flooding from the
Indus River and its tributaries and the floods of 1928, 1929, 1955, 1957, 1959, 1973, 1976, 1988, 1992, 1995, 1996 and
1997 attest to its destructive nature and its adverse impact on lives and property in Pakistan. More than three million
homes were destroyed and 160 persons lost their lives in the great flood of 1973. Three years later in 1976, floods
demolished over 10 million houses while 425 lives were lost with other losses amounting to a Rs. 6 billion. Towards the
end of September 1988, an unprecedented 2 Kabir Uddin et al.: Application of Remote Sensing and GIS for Flood Hazard
Management: A Case Study from Sindh Province, Pakistan
flood occurred, inflicting damage nationwide in Pakistan worth about Rs. 17 billion. The catastrophic flood of 1992
surpassed all previous records with damages estimated at Rs. 50 billion. Sindh is the south eastern province of Pakistan.
The province is associated with the longest river of Pakistan which stretches to about 1976 miles (3180 km) long.
Originating in the Tibetan Plateau in the vicinity of Lake Mansarovar in the Tibet Autonomous Region, the river runs a
course through the Ladakh district of Jammu and Kashmir, then enters the Northern Areas (Gilgit-Baltistan) flowing
through the North in a southerly direction along the entire length of the country to merge into the Arabian Sea near the port
city of Karachi in Sindh. It is the major source of irrigation for Pakistans fertile agricultural lands. The rivers estimated
annual flow stands at around 207 cubic kilometers (cu km), making it the 21st largest river in the world in terms of annual
flow. Climate is not uniform over the Indus Basin. It varies from subtropical arid and semi-arid to temperate sub-humid in
the plains of the Sindh and Punjab provinces, and alpine in the mountainous highlands to the north. Annual precipitation
ranges between 100 mm and 500 mm in the lowlands to a maximum of 2000 mm (water equivalent) on mountain slopes.
Snowfall at higher altitudes (above 2500 m) accounts for most of the river runoff. Alluvial land in Pakistan is known to be
a flood-prone area.
Earth observation techniques can contribute to finding out more accurately what causes floods. Together with flood hazard
mapping, earth observation techniques can be used to assessing damage to property, infrastructure, and agricultural crops.
The uses of remotely sensed data in identifying the trends of river channel migration and as source of input data for
determining river behaviour has been gaining popularity in recent years. One of the main characteristics of remote sensing
is its capability to generate a large amount of information frequently and spatially, becoming a powerful tool for
monitoring changing aquatic environments. Remote sensing data has been used to document water quality estimate water
depths and monitor river channel changes and aquatic habitat[4-6]. Object-based analysis of multispectral imagery was
introduced early on in remote sensing literature[7]; however, the object-based approach has largely been ignored in favour
of pixel-based methods which have been easier to implement[8]. One of the advantages of object-based image analysis is
the multitude of additional information that can be derived from image objects compared to the amount of information
available from individual pixels[9]. This study aims to determine flood hazard, flood extended areas, and proposed flood
shelters using remote sensing and geographic information systems tools. An effort was made to prepare a flood hazard
zone map and a proposed flood shelter map using GIS and remote sensing techniques and incorporating a satellite image,
topographic map. This study also attempts to exploit the capabilities of remote sensing and GIS techniques to suggest an
appropriate methodology for flood hazard mapping.
2. Methodology
2.1. Study Area
The study area is located in the south-eastern part of Nepal and seven districts of Sindh province (Figure 1). Sindh lies in a
tropical to subtropical region; it is hot in the summer and mild to warm in winter. Temperatures frequently rise above 46C
between May and August, and drops to a minimum average of 2C during December and January in the northern and
elevated regions. Annual rainfall averages about seven inches, falling mainly during July and August. The province is
mostly arid, with scant vegetation except for the irrigated Indus Valley.
Figure 1. Location Map of Study Area American Journal of Geographic Information System 2013, 2(1): 1-5 3
2.2. Data Used

For this study moderate resolution imaging spectroradiometer (MODIS) imagery was used to generate a flood inundation
map. A topographic map and Shuttle Radar Topography Mission (SRTM) DEM data from various sources were used to
analyse flood hazard area and flood shelter areas.
2.3. Image analysis
Flood hazard depends on many factors such as flood inundation and topography of the area and etc. Enumerate the flood
extended area mapping were followed few steps. The acquired MODIS reprojected UTM, Zone 47. After that eCognition
Developer software was used for object-based image analysis (OBIA). Figure 2 presents the flow diagram of object-based
image analysis for flood extended mapping.
Compared with pixel-based methods, this approach shows better classification results with higher accuracy as it uses both
spectral and spatial information[4,10,11]. The basic step in eCognition image analysis is to segmentation a scene where
information of single pixels is aggregated into homogeneous image objects. Multiresolution segmentation, which was used
in this study for object-based image analysis, groups areas of similar pixel values into objects; homogeneous areas result in
larger objects, heterogeneous areas in smaller ones. Therefore several segmentations were tested with different parameters
until the result was satisfying.. For each object information on average spectral values, average Normalized Difference
Vegetation Index (NDVI), and land and water mask were derived. In the pre-processing stage, an NDVI image was created
using the customised feature in the form: NDVI = (RED IR) / (RED + IR ), NDVI indices values range from -1.0 to 1.0,
non-vegetation area negative and vegetation area will have positive value. Land and Water Mask created by = IR) /
(Green) * 100. Land and water mask index values can range from 0 to 255, but water values typically range 0 to 50. This
information was used to develop suitable classification algorithms for flood inundation mapping.
Figure 2. Flow diagram of flood extended mapping
Figure 3. Flow diagram of flood shelter analysis 4 Kabir Uddin et al.: Application of Remote Sensing and GIS for Flood Hazard
Management: A Case Study from Sindh Province, Pakistan
The next step was to define classes and insert class hierarchy. Image objects were linked to class objects and each
classification link was stored in the membership value of the image object to the linked class. Image objects were
classified using user-defined rules. Objects with an area smaller than the defined minimum mapping unit were merged
with other objects. The classified inundation map was exported to shape file format for further processing and flood hazard
processing and flood shelters and modeling. Several types of factor/parameter were required to do flood hazard
assessment. Acquired datasets of stream order, river, digital elevation model and land cover including all of the data that a
flood under such circumstances would include. One of these is slope which is an essential topographic factor that guides
the flood potential of spatial areas. Slope data layers were generated using the digital elevation model of the study area.
Vegetation poses a major restraint to flooding; vegetation reduces runoff and helps in percolation. Euclidea n allocation
calculates for each cell the nearest source based on Euclidean distance and calculates the direction in degrees to the nearest
source. Normally, all nearest source areas are flood-prone areas. Euclidean allocations of stream order were generated.
Euclidean distances of rivers were generated and short disease from river ranked flood potential. In order to combine the
datasets, they must be set to a common scale. Each dataset data is then reclassified to rank the potential hazard location.
Then reclassify date sets came to a common scale ranges from 1 to 5. Each raster was assigned ranks at the time of
reclassification: a ranking of 1 means with the most flood potential, and 5, least flood potential. Once, all the four raster
layers are created using the models developed, they are overlaid together to produce the final flood hazard map of Shind
province. The weighted layers were created in order to reduce the number of inputs in the final weighted overlay. The final
output of a flood potential index was the result of equally weighting 20% each the weighted layers. The results were
validated with the recent flood map of Sindh province.
Providing flood shelter in flood-affected areas is very important. Flood shelters provide flood-affected people with areas to
run to for shelter and keeps them safe while the flood rages Digital elevation model, land cover, settlement, road and flood
map were used to model suitable locations for flood shelter. Preparing the shelters suitability slope from the elevation
dataset, point density from the settlement dataset, Euclidean distance from the road and flood dataset were used. Each
datasets data were then reclassified to rank the suitability of locations for flood shelter. Each raster was assigned a rank at
the time of reclassification, with rank 1 as less suitable for flood shelter and 5 as most suitable. Once all four raster
layers have been created using the models developed, they were overlaid together to produce the final suitability flood
shelters location map of Shind province (Figure 3). They were then weighted according to a percentage of influence and
combined to produce a map displaying suitable locations for the proposed flood shelter. Results were selected and the
optimum site identified for the new flood shelter from alternative sites.
Figure 4. Flood hazard zoning map
3. Results
Figure 5. Flood shelter suitability map Table 1. District wise flood hazard area
Restricted
Not suitable
Most Suitable
Legend
3
4
5
6
7
Distric
t
Larkan
a
Khairp
ur
Jacoba
bad
Sukkur
Naush
ahro
Feroze
Shikar
pur
Ghotki
Very
high
High
Moder
ate
Low
Very
low
1335
3087
1706
945
439
1079
1218
4182
9316
944
2788
1840
41
941
792
905
1572
1380
666
1976
6
0
0
710
1064
770
20
355
1647
2351
2070
Verein L imnol, 26, 23742377.

Miranda,F. P., Fonseca, L. E. N., and Carr, J. R., 1998, Semivariogram textural classi. cation
of JERS-1 (Fuyo-1) SAR data obtained over a ooded area of the Amazon rainforest.
International Journal of Remote Sensing, 19, 549556.
Muzik, I., 1996. Flood modeling with GIS-derived distributed unit hydrographs. Hydrologic
Processes, 10, 14011409.
Evaluating Flood Hazard for Land-Use Planning

in Greater Dhaka of Bangladesh Using Remote
Sensing and GIS Techniques
Ashraf M. Dewan & M. Monirul Islam &
T. Kumamoto & M. Nishigaki
Received: 3 February 2006 / Accepted: 13 October 2006 / Published online: 8 November 2006
# Springer Science + Business Media B.V. 2006
Abstract Floods are a common feature in rapidly urbanizing Dhaka and its adjoining areas.
Though Greater Dhaka experiences flood almost in every year, flood management policies
are mostly based on structural options including flood walls, dykes, embankments etc.
Many shortcomings of the existing flood management systems are reported in numerous
literatures. The objective of this paper is to assess flood hazard in Greater Dhaka for the
historical flood event of 1998 using Synthetic Aperture Radar (SAR) data with GIS data.
Flood-affected frequency and flood depth calculated from the multi-date SAR imageries
were used as hydrologic parameters. Elevation heights, land cover classification,
geomorphic division and drainage network data generated from optical remote sensing
and analogue maps were used through GIS approach. Using a ranking matrix in three
dimensional multiplication mode, flood hazard was assessed. All possible combination of
flood hazard maps was prepared using land-cover, geomorphology and elevation heights for
flood-affected frequency and floodwater depth. Using two hazard maps which produced the
highest congruence for flood frequency and flood depth, a new flood hazard map was
developed by considering the interactive effect of flood-affected frequency and floodwater
depth, simultaneously. This new hazard map can provide more safety for flood countermeasures
because pixels belonging to higher hazard degrees were increased due to the
Water Resour Manage (2007) 21:16011612
DOI 10.1007/s11269-006-9116-1
A. M. Dewan (*)
Department of Geography and Enviornment, University of Dhaka,
Dhaka 1000, Bangladesh
e-mail: amd1971@univdhaka.edu
M. M. Islam
Department of Civil Engineering, International University of Business Agriculture and Technology,
4 Embankment Drive Road, Sector 10 Uttara, Dhaka 1230, Bangladesh
T. Kumamoto
Department of Earth Science, Faculty of Science, Okayama University,
3-1-1 Tsushima Naka, Okayama 700-8530, Japan
M. Nishigaki
Department of Environmental Design and Civil Engineering, Okayama University,
3-1-1 Tsushima Naka, Okayama 700-8530, Japan
consideration of higher degrees of ranks. The estimation of flood hazard areas revealed that
a major portion of Greater Dhaka comprised moderate to very high hazard zone. Only a
little portion (8.04%) was found to be the least vulnerable to potential flood hazard.
Conversely, 28.70% of Greater Dhaka was found within very high hazard zone. Based on
this study, comprehensive flood hazard management strategies for land use planning
decision were proposed for the efficient management of future flood disasters.
Key words greater Dhaka . GIS . flood hazard . SAR . flood frequency . floodwater depth
1 Introduction
During the last two decades, Remote Sensing (RS) has been played an important role in the
fields of hydrology and water resources management (Bastiaansen 1998). Geographic
Information System (GIS) on the other hand has also been used extensively to model
surface water, particularly flood and associated damage (Boyle et al. 1998; Green and
Cruise 1995; Werner 2001). Recently, the integration capabilities of satellite data with GIS
have opened up opportunities for quantitative analysis of hydrological events, such as
flood, at all geographic and spatial scales. Remote sensing application is however,
considered imperative for third world countries because it is difficult for government to
update their database frequently with the ground observation techniques because of time
and cost associated with traditional methods (Dong et al. 1997). The potential of this
facility for improving hazard evaluation and risk reduction is continually being explored
(Carrara and Guzzetti 1993).
Floods are a regular feature in Greater Dhaka, the capital of Bangladesh during the
monsoon season. Dhaka has experienced many disastrous floods in the past of which the
1998 flood is said to be the worst in memorable records (Faisal et al. 1999; Dewan et al.
2004). The 1998 flood resulted in severe damage and untold sufferings to the inhabitants of
Dhaka and adjoining areas. For example, damage on water supply and sanitation sector
alone caused massive monetary loss worth US $ 10 million (Institute of Flood Control and
Drainage Research (IFCDR) 1998). Furthermore, 262,000 houses were fully or partially
collapsed (Islam 1998) and 3.7 million people were directly affected by the 1998 flood
(Jahan 2000). Though various measures including flood walls, embankments have been
constructed, the loss of properties and causalities are on the rise (Mohit and Akhter 2000).
Available data suggests that the Bangladesh government, in association with international
agencies, have spent 141.6 million US $ on flood control measures to solve the perennial
flood problem in Dhaka, but the annual flood damage and flood affected areas increased
considerably (Mohit and Akhter 2000). For example, a total of 4.55 million people were
affected by the 1998 flood (Hye 2000) and most of the part of Greater Dhaka was under
water of various depths for more than 8 weeks (Jahan 2000). It is necessary to mention here
that the construction of embankments fosters to the encroachment of floodplains in Dhaka
(Chowdhury et al. 1998). In addition, Rasid and Mallik (1996) identified several
environmental problems linked with the embankment in Dhaka.
Even though the occurrences of floods cannot be prevented, the negative consequences
can be lessened by an integrated approach to disaster management. Disaster management
includes four elements such as mitigation, preparedness, response and recovery (Quarantelli
1991). Literature suggests that flood control works and post disaster recovery are the
primary approaches for flood management in Greater Dhaka. It has become apparent in
1602 Water Resour Manage (2007) 21:16011612
1998 that such an approach is inadequate to combat recurrent flood problems. It is also
clear that the growing flood hazards of Dhaka have not been analyzed for purposes of
disaster mitigation (Mitchell 1999). As Dhaka contributes 16.7% of GDP to the national
development (Bangladesh Bureau of Statistics (BBS) 1998), the potential loss of GDP due
to flood is very high. Experts fear that the flood vulnerability of Dhaka is likely to
exacerbate in the effect of the probable climatic change. Thus, following the 1998
devastation flood monitoring, hazard assessment and disaster relief have been identified to
be the major issues by the water experts, urban planners and key policy makers of the
country. They have called for the accurate mapping of flood prone areas (Islam 1998),
assessment of flood hazard (Nishat 1998) and the development flood risk maps for land-use
planning (Hossain 1998).
Considering the above fact, the objective of this article is to assess flood hazard in
Greater Dhaka of Bangladesh for the historical flood event of 1998 using Synthetic
Aperture Radar (SAR) data with GIS. A land-use planning decision for the mitigation of
potential flood disaster is also put forward on the basis of this study.
2 The Study Area
The study area was the Greater Dhaka of Bangladesh (Figure 1). The latitudes and
longitudes of the study area are 2368 N 9033 E and 2390 N, 9050 E, respectively.
The study area is located mainly on an alluvial terrace, known as the Modhupur terrace of
the Pleistocene period. Topographically, Dhaka is a flat land. The surface elevation of the
area ranges between 1 and 14 m (Flood Action Plan (FAP) 8A 1991). It lies in the subtropical
monsoon zone and experiences humid climatic conditions. The City experiences
about 2000 mm annual rainfall, of which more than 80% occurs during the monsoon
season. Historically, the city is endowed with rivers, numerous khals (ephemeral water
bodies) and canals that drain water from the upper reaches during the monsoon season. As
population increased, these areas have been encroached on, thus compounding flood
problems. In the Figure 1, it is depicted that the urban areas in Greater Dhaka have
increased in 1996 and 2000 over the year 1990.
3 Data Preparation
Six RADARSAT SAR (Synthetic Aperture Radar) (Narrow and Wide Beam mode)
imageries were acquired that covered the entire flood season of 1998 (JulySept). The
selected SAR data were justified by analyzing the water level records of five monitoring
stations in and around Greater Dhaka (Dewan et al. 2005). The SAR data were despeckled
using the GAMA-MAP filter with 55 window sizes. After suppressing the speckle
inherent to SAR images, geometric correction was carried out using a referenced Landsat
TM image of 1999 covered the same area until the root mean square errors resulted in less
than one pixel. A second order polynomial fit was applied and the pixel values were
resampled. A pixel of SAR data covers 50 m for narrow and 100 m for wide beam mode in
beneath. This is the ground resolution of the SAR data.
The major impediment to study flood hazard in the study area was the unavailability of
GIS data in digital format. Therefore, in this study, a number of GIS layers were generated.
A Digital elevation model (DEM) of Greater Dhaka area was obtained from the Institute of
Water Resour Manage (2007) 21:16011612 1603
Figure 1 Extent of the study area.
Water Modeling, Bangladesh (IWM). Using the DEM, an elevation map was constructed.
The elevation data were grouped into 14 categories, at 1 m intervals. The land-cover map
was generated in a GIS platform, using one IRS-1D PAN data which was acquired during
the dry season of 29 February 2000 (path 110 and row 055). Field data and the recent
topographic map (scale 1:10,000) were aided to develop land-cover data. A geomorphic
map was also developed by using a Landsat TM of 1999. Training areas were selected by
on the TM Image by using the available analogue maps and field investigation. An
administrative division map for Greater Dhaka was also digitized using the recent
topographic map (scale 1:10,000). A total of 34 administrative boundaries were obtained
in the study area. In addition, the drainage network of Greater Dhaka was generated from
IRS-1D PAN data. Finally, all GIS data were integrated with geometrically corrected SAR
data within a GIS approach. The Bangladesh Transverse Mercator System (BTM) was used
as the coordinate system. Each of the GIS and SAR images produced 166,560 pixels in a
common coordinate system and each pixel covered 50 m on the ground.
4 Analysis of Flood-Affected Frequency
The concept of flood-affected frequency (Islam and Sado 2000) was used to develop flood
hazard maps. A flood frequency map based on flood duration was developed using the
multi-date SAR imagery. The classified imageries obtained by the threshold algorithm
(Dewan and Nishigaki 2004) were used to generate a flood frequency map. Total six
available SAR images during the flood, which were acquired on July 7 and 31 1998;
August 10 and 25 1998; and September 10 and 17 1998, were classified into water and
non-water areas. Initially, the classified images of July 7 and 31 were combined to construct
a single classified image of July; similarly two images of August and two images of
September were combined to construct the single image for August and September,
respectively. These combinations provided an opportunity to get a common boundary
needed to develop a flood-affected frequency map. Finally, these three images were
superimposed to characterize the inundated areas. The inundated area that appeared in all
the images was considered the highly damaged area. The common inundated areas that
appeared in two and one of the three images were deemed as medium damaged and low
damaged areas, respectively. Non-inundated area in all the images was considered as
non-damaged areas. Thus, four flood frequency categories were obtained, corresponding
to damage rankings of class 4, class 3, class 2 and class 1 as high, medium, low and
non-damaged areas, respectively.
5 Analysis of Floodwater Depth
Multi-temporal SAR imageries were also used to estimate the floodwater depth. It is very
difficult to obtain flood-depth information from SAR data. However, in this study we used
the technique suggested by Islam and Sado (2000), to estimate floodwater depth from SAR
images. First, a land level map of Greater Dhaka was extracted from the DEM that consist
four height classes. This land level map was used to collect representative ground truth data
for each unit. Using a Global Positioning System (GPS), a total of 100 signatures (25 for
each category) were collected and plotted on the land level map. Thus, a ground truth map
was created and subsequently brought into GIS. These signatures were then used to extract
training pixels. To do so, 7 July, 25 August and 10 September 1998 images were considered
that comprised early to peak flooding of 1998. Signatures were collected by superimposing
the ground truth map and a digital elevation model on each SAR image. In addition, visual
interpretation and grey level differences of digital number (DN) values were also helpful to
determine the selection of training pixels. A minimum distances algorithm of supervised
classification was then applied to classify SAR data into four flood depth categories, i.e.,
deep, medium, shallow and no-flooding. If deep floodwater depth observed in a single
image then it was regarded as deep depth, and if medium depth appeared in any single
image for the area but that was represented as shallow depth area by other two images, then
it was considered as medium depth. After the classification of deep and medium, the remaining
inundated pixels were considered as shallow depth. Non-inundated areas were considered as
noflooding
areas. Thus, four flood depth categories were obtained, corresponding to depth rankings
of class 1, class 2, class 3 and class 4 as no-flooding, shallow, medium and deep depth,
respectively. It is necessary to mention here that the DEM was not considered for the estimation
of flood depth because of its coarse spatial resolution (300 m). However, to derive flood depth,
we checked and used benchmark points to get training areas from the DEMand field data, which
was eventually used to develop flood depth categories from SAR data.
In order to evaluate the classified results of floodwater depth, derived depth map was
further authenticated by a flood depth map which was developed in Institute of Water
Modeling (IWM) (2000) using a hydrodynamic model. A comparison was made to
understand the agreement between two floodwater depth maps. It is found that deep flooded
category, for instance, covered 20.95% area in the depth map produced by IWM, whilst the
same category covered 21.21% area in the floodwater depth map developed in this study.
6 Flood Hazard Assessment
Even though satellite remote sensing has shown great potentiality for hazard assessment
(McKean
et al. 1991), the big challenge is the lack of generally accepted methods for producing hazard
maps or even on the scope of producing such maps (Rhoads 1986). Moreover, many well
developed methods usually applied in developed nations may not be accessible to developing
country particularly in Bangladesh, due to the lack of digital data or restriction to access to data
sources. For example, DEM based detail flood mapping is not possible for the entire Bangladesh
since high resolution digital elevation data of the whole country or part of it, is not available.
Therefore, a simple procedure is adopted in this paper for flood hazard assessment.
To assess flood hazard, a model was constructed in Model Maker utility of ERDAS
IMAGINE (v. 8.6) software. In order to assess the flood hazard for each category of land cover,
geomorphology and elevation units, a weighted score was estimated and hazard ranks were
decided using the technique suggested by Rahman et al. (1991). Following steps were involved
in the development of flood hazard maps. Step 1: Flood-affected frequency and floodwater
depth maps were superimposed on thematic layers to estimate the percentage of area occupied
by each category of flood-affected frequency and floodwater depth. We used land cover data
with elevation height and geomorphic data to signify the different types of land covers. Because
of the importance of the land use, value and the significance uses of different types of land cover
are not same. Hydrological parameters like flood frequency and flood depth maps obtained from
remote sensing data were superimposed onto the land cover, elevation height and
geomorphologic
maps to view the interactive effect between hydrological parameters and thematic
information.
Step 2: A weighted score for the acquired area percentage for each category of land
cover, elevation, and geomorphic unit was estimated using the following equation:
Weighted Score Class 1 _ 1:0 Class 2 _ 3:0 Class 3 _ 5:0 Class 4 _ 7:0 1
Where, class 1 to class 4 represents the area percentage occupied by the categories of each
hydraulic component i.e., flood-affected frequency and floodwater depth, for each ID
category of the respective GIS component. The coefficients 1.0, 3.0, 5.0 and 7.0 in Eq. 1
were used to describe the weight for the flood loss. Calculated weighted score and their
corresponding hazard ranks for each geomorphic category for flood-affected frequency is
presented in Table I. Step 3: After calculating the weighted score, points for each category
of land cover, elevation and geomorphology was estimated on the basis of a linear
interpolation between 0 and 10, where 0 represents the lowest score and 10 represents the
highest score. This represents the degree of flood hazard. Step 4: On the basis of the
estimated points, hazard ranks were determined. In order to quantify flood hazard, three
hazard ranks were established. For example, hazard ranks were fixed by assigning 13.5 for
hazard rank 1, 3.56.5 for hazard rank 2 and 6.510.0 as hazard rank 3. The higher the
ranking value, the more susceptible that particular class is considered to be to the
occurrence of flood. Step 5: Using three dimensional multiplication mode suggested by
Islam and Sado (2000), assessment of flood hazard was performed for flood-affected
frequency and floodwater depth. The resulting flood hazard maps were further reclassified
according to the intensity of hazards. Thus, five distinct hazard zones corresponded to least,
less, moderate, high and very high were obtained. To illustrate flood hazard on the basis of
mapping unit, three dimensional multiplication mode applied for three thematic maps.
Two hazard maps were developed for flood-affected frequency and floodwater depth by
considering three dimensional multiplication mode. Although the two flood hazard maps
showed only 34% congruence, these maps provide hazard information for flood frequency
and floodwater depth, independently. They cannot represent higher hazard ranks for two
hydraulic parameters at the same time. Moreover, these maps may not be perceptible to
many people since they represent two hydrologic parameters. Therefore, a new hazard map
was constructed using the interactive effect of flood-affected frequency and floodwater
depth simultaneously, which can provide more safety for flood countermeasures because
Table I Flood hazard ranks for geomorphological division by flood-affected frequency (ID: geomorphic
identification number; HR: hazard rank)
ID Class 1 Class 2 Class 3 Class 4 Score Point HR 13
1 38.94 23.86 13.18 24.02 344.58 5.01 2
2 83.08 10.01 3.26 3.66 154.99 2.25 1
3 28.63 17.86 15.01 38.50 426.76 6.20 2
4 0.44 0.80 2.93 95.82 688.27 10.00 3
5 56.58 13.67 8.74 21.01 288.37 4.19 2
6 63.35 17.26 9.52 9.88 231.85 3.37 1
7 47.83 15.69 11.03 25.46 328.24 4.77 2
8 0.55 0.95 3.18 95.32 686.55 9.98 3
9 23.18 16.45 12.73 47.63 469.64 6.82 3
10 42.79 23.41 12.53 21.27 324.55 4.72 2
11 5.91 5.40 6.20 82.49 630.55 9.16 3
12 9.66 7.43 7.59 75.32 597.13 8.68 3
13 55.36 7.14 5.36 32.14 328.57 4.77 2
Figure 2 Flood hazard map developed using land-cover, elevation and geomorphic categories for floodaffected
frequency and floodwater depth simultaneously.
the pixels belonging to higher ranks were increased due to the consideration of higher
hazard degrees. If one hazard map shows the higher rank than the other for a pixel, then the
higher rank was assigned for that pixel. The final hazard map is shown in Figure 2.
The hazard area estimation using the final flood hazard map developed by the interactive
effect of flood frequency and floodwater depth concurrently revealed that only 8.04% areas
were within the least hazard zone (Table II) that corresponded to the non-flooded areas in
Greater Dhaka. Highly elevated lands (>13 m) of the study area represented the least hazard
zone. Moderately high lands were within the group of less hazard zone and constituted
21.65%. These are the most extensively developed area in Greater Dhaka. Urban peripheral
zones of Greater Dhaka were characterized by moderate hazard zone (15.05%). High
hazard (26.56%) and very high hazards (28.70%) were assigned to cells where higher
degrees of ranks observed. Extreme lowlands and areas adjacent to river banks are
designated as very high hazard zone.
7 Disaster Management Policy for Land-Use Planning
The developed flood hazard map further can be incorporated to land-use planning
decisions. This may help to reduce potential flood damage in Greater Dhaka. Micro scale
flood management plan can be of little help in the effort of flood damage abatement. A
comprehensive plan addressing flood hazard management is therefore, necessary. This plan
should combine land use strategies for each zone with the careful consideration of certain
structural controls. This can be achieved by the minimal disruption of natural environments.
Table III presents an example of general management strategies based on this study and
rigorous field investigation. These strategies could serve as basic components in a
comprehensive flood management plan for Greater Dhaka.
Infilling of natural channels and lowlands is the primary cause for increasing flood
hazard in Greater Dhaka. Therefore, preservation of natural channels and infiltration
processes should be the top priority when formulating policy. It is postulated that during
urban developments small channels are often viewed as being insignificant and are
destroyed, but in reality they have very vital role in easing flood peaks (Dunne and Leopold
1978 cited in Rhoads 1986). Greater Dhaka is endowed with natural channels and canals
that used to drain floodwater. Due to the increasing population pressure, those canals have
lost their ability to drain floodwater during the monsoon. Therefore, further infilling of
those channels must be prohibited and every effort should be made to prevent the
destruction of any channel regardless of size. Human settlements should not be allowed in
those areas comprised of high and very hazard zones. If necessary, structures must be flood
Table II Areas in Greater Dhaka with various degree of flood hazard
Hazard zone (s) Area (ha) Percentage (%)
Least 3348.50 8.04
Less 9013.75 21.65
Moderate 6266.75 15.05
High 11061.00 26.56
Very high 11950.00 28.70
Total 41640.00 100.00
proofed and must be put in elevated places. Proper maintenance of the existing storm water
facilities in the least and less hazard zones should be another priority.
8 Conclusions
In this paper, flood hazard assessment for the historical flood event of 1998 in Greater
Dhaka of Bangladesh was examined using an integrated approach of GIS and remote
sensing. The study demonstrated a simple and cost effective way to use geographical
information system for creating flood hazard maps from the available dataset. Although it
was very time consuming to derive the GIS database because of the size of the study area,
GIS was invaluable in reducing the complexity associated with hazard assessment.
Flood hazard maps by considering the interactive effect of land cover, elevation, and
geomorphic units for flood affected frequency and floodwater depth were constructed. The
hazard maps showed that a major portion of Dhaka were within moderate to very high
hazard zone, especially fringe areas. It is projected that these areas will be fully urbanized
by the year 2010, therefore higher priorities must be paid for the development of apposite
flood countermeasures for areas having higher hazard potential. Comprehensive land-use
policies for flood disaster management have been proposed on the basis of this study. It is
anticipated that proposed land-use policies can contribute to effective flood forecasting,
relief and emergency management for future flood event.
Table III General flood hazard management strategies Greater Dhaka of Bangladesh
General management strategies
Least hazard zone
(i) High development densities allowable provided that stormwater retention basins are adequately
constructed. It also requires of their proper maintenance.
Less hazard zone
(i) Moderate development densities can be maintained.
(ii) Require structures should be put on higher places.
(iii) Should have separate drainages facilities wherever possible.
(iv) Existing natural channels should be kept for flood flow.
(v) Require proper maintenance and cleaning of storm water facilities on timely basis.
Moderate and high hazard zone
(i) Detail hydrologic analysis of flood characteristics must be performed prior to allocate for the new
development.
(ii) Low development densities may be allowed on condition.
(iii) Structures must be flood-proofed.
(iv) Adequate drainage facilities of floodwaters must be secured.
(v) Require maintenance of drainage facilities, natural channels, canals and khals.
(vi) Detail flood zoning should be conducted and require to introduce proper land development rules.
(vii) Should have emergency evacuation facilities and adequate flood shelters.
(viii) Community based flood management program should be introduced.
Very high hazard zone
(i) Must be kept open for flood conveyance as permanent natural open space.
(ii) Any kind of developments must be prohibited.
(iii) Structures must not be allowed.
(iv) Require preservation of natural drainage systems, existing floodplains.
As observed in Dhaka, increasing population pressure is forcing many people to enter
the vacant land of the City by filling up of natural channels and floodplains. Consequently,
flood risk is increasing. In order to ameliorate flood induced damage, the developed flood
hazard map would be invaluable. Government can use hazard maps for ensuring the proper
development planning of the very high and high hazard zones which is suppose to be
urbanized by the year 2010. Urban planners can use this information to make
environmentally sound land-use decisions. One more advantage is that the hazard maps
can be quickly updated if modification, such as changes in land use, occurs subsequent to
the original study. Furthermore, this type of study will provide information about flood
protection measure such as construction and development of infrastructure and preparedness
of aid and relief operation for high hazard areas for future flood event.
References
Bangladesh Bureau of Statistics (BBS) (1998) Statistical yearbook sof Bangladesh. Ministry of Planning,
Dhaka, p 224
Bastiaansen WGM (1998) Remote sensing in water resources management: the state of the art, International
Water Management Institute (IWMI), Colombo
Boyle SJ, Tsanis IK, Kanaroglou PS (1998) Developing geographic information system for land use impact
assessment in flooding condition. J Water Resour Plan Manage 124:8998
Carrara A, Guzzetti F (eds) (1993) GIS in assessing natural hazards. Kluwer, Italy
Chowdhury JU, Rahman R, Bala SK, Islam AKM Saiful (1998) Impact of 1998 flood on Dhaka City and
performance of flood control works. Institute of Flood Control and Drainage Research, Dhaka, p 76
Dewan AM, Nishigaki M (2004) Using Synthetic Aperture Radar with GIS for flood damage estimation, in
Proceedings of the 6th Annual Meeting of the Japan Society for Disaster Information Studies, Japan
Society for Disaster Information Studies, Tokyo
Dewan AM, Nishigaki M, Komatsu M (2004) Use of multi-temporal RADARSAT-1 ScanSAR data to
differentiate between flooded and non-flooded areas, in Annual Meeting of the Chugoku Branch of the
Japan Society of Civil Engineers, Tottori, Japan
Dewan AM, Yeboah KK, Nishigaki M (2005) Flood mapping and damage evaluation in Greater Dhaka,
Bangladesh with remote sensing. Asian Profile 33(5):495512
Dong Y, Forster B, Ticehurst C (1997) Radar backscatter analysis for urban environments. Int J Remote Sens
18:13511364
Faisal IM, Kabir MR, Nishat A (1999) Non-structural flood mitigation measures for Dhaka City. Urban
Water 1:145153
Flood Action Plan (FAP) 8A (1991) Master plan study for Greater Dhaka protection project. Japan
International Cooperation Agency, Dhaka
Green RG, Cruise JF (1995) Urban watershed modeling using geographic information system. J Water
Resour Plan Manage 121:318325
Hossain MM (1998) Bangladesh floods and its management initiatives. Steps Towards Development 3:1219
Hye A (2000) Effects on livelihood. In: Nishat A, Reazuddin M, Amin R, Khan AR (eds) The 1998 flood:
impact on environment of Dhaka City. Department of Environment and IUCN, Dhaka
Institute of Flood Control and Drainage Research (IFCDR) (1998) Impact of 1998 flood on Dhaka City and
performance of flood control works, Dhaka
Institute of Water Modeling (IWM) (2000) Baseline information study of the Dhaka combined flood control
cum eastern bypass road project, Final Report, Dhaka. p 86
Islam N (1998) Flood 98 and the future of urban settlements in Bangladesh. National Seminar on Flood 98
and Management of Floods in Future, 8th December 1998, Dhaka
Islam MM, Sado S (2000) Development of flood hazard maps of Bangladesh using NOAA-AVHRR images
with GIS. Hydrol Sci J 45:337355
Jahan S (2000) Coping with flood: the experience of the people of Dhaka during the 1998 flood disaster.
Aust J Emerg Manag 15(3):1620 (Spring)
McKean J, Buechel S, Gaydo L (1991) Remote sensing and landslide hazard assessment. Photo Eng Rem
Sen 57(9):11851193
Mitchell JK (1999) Megacities and natural disasters: a comparative analysis. GeoJournal 49:137142
Mohit MA, Akhter S (2000) Delineation of flood damaged zones of Dhaka City based on the 1998 flood by
using GIS. In: Ali MA, Seraj SM, Ahmed S (eds) Engineering concerns of flood. Bangladesh University
of Engineering and Technology, Dhaka, pp303318
Nishat A (1998) A discussion on flood management in Bangladesh. The 1998 Deluge-Developing Coping
Capacities, Special Workshop on the 1998 flood, Nagorik Durjog-Mokabela Uddogh, Dhaka,
Bangladesh
Quarantelli EL (1991) Some aspects of disaster planning in developing countries, in disaster management
and regional development planning with Peoples Participation, Vol. II, United Nations Center for
Regional Development, Dhaka
Rahman NM, Ochi S, Murai S, Hashimoto T, Kakiuchi H (1991) Flood risk mapping in Bangladesh-flood
disaster management using remote sensing and GIS, in application of remote sensing in Asia and
Oceania-environmental change monitoring, Asian Association of Remote Sensing, Tokyo
Rasid H, Mallik A (1996) Living on the edge of stagnant water: an assessment of environmental impacts of
construction-phase drainage congestion along Dhaka City Flood Control Embankment, Bangladesh.
Environ Man 20:8998

Rhoads BL (1986) Flood hazard assessment for land-use planning near desert mountains. Environ Man
10:97106
Werner MGF (2001) Impact of grid size in GIS based flood extent mapping using 1D flow model. Phys
Chem Earth 26:517522
http://ppg.sagepub.com/
Progress in Physical Geography

http://ppg.sagepub.com/content/33/2/183
The online version of this article can be found at:
DOI: 10.1177/0309133309339563
Progress in Physical Geography 2009 33: 183 originally published online 10 June 2009
Karen E. Joyce, Stella E. Belliss, Sergey V. Samsonov, Stephen J. McNeill and Phil J. Glassey
mapping natural hazards and disasters

A review of the status of satellite remote sensing and image processing
techniques for
Published by:
http://www.sagepublications.com
Additional services and information for Progress in Physical Geography can be found at:
Email Alerts: http://ppg.sagepub.com/cgi/alerts
Subscriptions: http://ppg.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://ppg.sagepub.com/content/33/2/183.refs.html
Downloaded from ppg.sagepub.com by Giyanto Giyanto on October 11, 2010
Progress in Physical Geography 33(2) (2009) pp. 183207

DOI: 10.1177/0309133309339563
A review of the status of satellite remote

sensing and image processing
techniques
for mapping natural hazards and
disasters
Karen E. Joyce,1* Stella E. Belliss,2 Sergey V.
Samsonov,1 Stephen J. McNeill2 and Phil J. Glassey1
1
2
GNS Science, PO Box 30368, Lower Hutt, 5040, New Zealand

Landcare Research, PO Box 40, Lincoln 7640, New Zealand
Abstract: In the event of a natural disaster, remote sensing is a valuable source of spatial information
and its utility has been proven on many occasions around the world. However, there are many
different types of hazards experienced worldwide on an annual basis and their remote sensing
solutions are equally varied. This paper addresses a number of data types and image processing
techniques used to map and monitor earthquakes, faulting, volcanic activity, landslides, fl ooding, and
wildfi re, and the damages associated with each. Remote sensing is currently used operationally for
some monitoring programs, though there are also diffi culties associated with the rapid acquisition
of data and provision of a robust product to emergency services as an end-user. The current
status of remote sensing as a rapid-response data source is discussed, and some perspectives given
on emerging airborne and satellite technologies.
Key words: image processing, natural hazards, optical, remote sensing, SAR, thermal.
_
*Author for correspondence. Email: k.joyce@gns.cri.nz
The Author(s), 2009. Reprints and permissions:
http://www.sagepub.co.uk/journalsPermissions.nav
I Natural hazards and disasters
The use of remote sensing within the
domain of natural hazards and disasters has
become increasingly common, due in part to
increased awareness of environmental issues
such as climate change, but also to the increase
in geospatial technologies and the
ability to provide up-to-date imagery to the
public through the media and internet. As
technology is enhanced, demand and expectations
increase for near-real-time monitoring
and visual images to be relayed to emergency
services and the public in the event of a
natural disaster. Recent improvements to
earth monitoring satellites are paving the way
to supply the demand. Techniques needed
to exploit the available data effectively and
rapidly must be developed concurrently,
to ensure the best possible intelligence is
reaching emergency services and decisionmakers
in a timely manner.
A comprehensive review of remote sensing
for some natural hazards was completed by
Tralli et al. (2005), and Gillespie et al. (2007)
184 Progress in Physical Geography 33(2)

reviewed natural hazard prediction and assessment
by remote sensing, with a focus on
the types of sensors available. The research
presented in this paper complements these
works by directing attention to the manner
in which the mapping was achieved with
respect to different image processing techniques.
Further, we also report on the reality
of remotely sensed image acquisition and
processing requirements with the view that
timely intelligence and information extraction
is critical in an emergency response to a
hazard or disaster situation.
For reference, the details of a number of
satellites and sensors that are commonly used
or have the potential for hazards mapping
are collated in Table 1.
The four phases of the disaster management
cycle include reduction (mitigation),
readiness (preparedness), response and
recovery (Cartwright, 2005). Remote sensing
has a role to play in each of these phases,
though this paper focuses primarily on its
contribution to the response phase. Several

different types of natural hazards and disasters
are presented in the following sections
to determine the commonly used image processing
techniques (summarized in Table 2),
their advantages and disadvantages, and to
review the reality of applying them in a rapidresponse
environment.
Through this review it became apparent
that many mapping operations are still using
manual interpretation techniques to achieve
(Continued)
Table 1 Summary of the characteristics of some sensors used in hazards mapping

and
monitoring
Satellite Sensor Swath (km) Nadir spatial
resolution (m)
Revisit capability
Airborne sensors variable variable > 0.1 Mobilized to order
CASI variable 12
Hymap 100225 210
Worldview Panchromatic 16.4 0.46 1.1 days
Multispectral 16.4 1.85
Quickbird Panchromatic 16.5 0.6 1.53 days
Multispectral 16.5 2.4
Ikonos Panchromatic 11 1 1.53 days
Multispectral 11 4
RapidEye^ Multispectral 77 x 1500 6.5 1 day
EO-1 ALI 60 30 Every 16 days
Hyperion 7.5 30
Terra ASTER 60 15,30,90 416 days
Terra / Aqua MODIS 2300 250, 500, 1000 At least twice daily for
each satellite
ALOS PRISM 35 4 Several times per year
as per JAXA acquisition
plan
AVNIR 70 10
PALSAR (Fine) 4070 10
PALSAR (ScanSAR) 250350 100
SPOT-4 Panchromatic 6080 10 11 times every 26 days
Multispectral 6080 20
Karen E. Joyce et al.: A review of satellite remote sensing and image processing techniques
185
Satellite Sensor Swath (km) Nadir spatial
resolution (m)
Revisit capability
SPOT-5 Panchromatic 6080 5 11 times every 26 days
Multispectral 6080 10
Kompsat Panchromatic 15 1 23 days
Multispectral 15 1
Landsat-5 TM Multispectral 185 30 Every 16 days
TM Thermal 185 120
Landsat-7* ETM+ Panchromatic 185 15 Every 16 days
ETM+ Multispectral 185 30
ETM+ Thermal 185 60
NOAA AVHRR 2399 1100 Several times per day
Envisat MERIS 575 300 23 days
Radarsat-2 Ultra-fi ne 20 3 Every few days
Radarsat-1/-2 Fine 50 8
Radarsat-2 Quad-pol fi ne 25 8
Radarsat-1/-2 Standard 100 25
Radarsat-2 Quad-pol standard 25 25
Radarsat-1 Wide 150 30
Radarsat-1/-2 ScanSAR narrow 300 50
Radarsat-1/-2 ScanSAR wide 500 100
Radarsat-1/-2 Extended high 75 25
Radarsat-1 Extended low 170 35
ERS-2 100 30 35-day repeat cycle
Envisat ASAR standard 100 30 36-day repeat cycle
ASAR ScanSAR 405 1000
TerraSAR-X Spotlight 10 1 11-day repeat cycle;
Stripmap 30 3 2.5-day revisit capability
ScanSAR 100 18
Cosmo-Skymed^ Spotlight 10 <1 ~37 hours
Stripmap 40 315
ScanSAR 100-200 30100
*Landsat-7 nearing the end of its useful life; problems with scan line corrector resulting in data gaps
^Figures quoted for one satellite in constellation
Table 1 Continued
high accuracy, but are disadvantaged by

speed of reproduction. It is considerably faster
and more sustainable to have automatically
implemented algorithms for hazard detection
and monitoring. Several programs already
exist globally for providing near-real-time
information to monitor thermal hotspots
associated with volcanic activity or fi re (for
example, see University of Hawaii and
Geoscience Australia). Ultimately it would
be useful to have similar automated methods
and algorithms in place for the acquisition
and processing of imagery and provision of
geospatial information about other hazard
events. To achieve this goal, robust techniques
need to be developed and thoroughly

tested in an operational environment. The
following sections document the testing of
various methods for scientifi c and monitoring
purposes, while examples of some current
operational programs, of the role of remote
sensing in rapid response systems, and of
emerging developments are given in later
parts of the paper.
II Earthquakes and faulting
There are several aspects involved in the detection
of earthquakes, faulting, and damages
associated with each. DInSAR (Differential
Interferometric Synthetic Aperture Radar)
is generally accepted as the best method for
earthquake associated deformation mapping;
LiDAR (Light Detection and Ranging)
provides the highest resolution DEM available
for fault interpretation; and very high
resolution optical data will provide the best
imagery for damage assessment of buildings
and infrastructure.
1 Optical detection of earthquakes and faulting
The technique of choice in the use of remote
sensing for fault mapping with optical data
is manual interpretation, regardless of the
data source (Fu et al., 2004; 2005; Walker,
2006; Walker et al., 2007). Frequently, the
effects of earthquake activity and faulting are
not manifested in spectral variations within
image data, but in topographical changes.
Image interpretation therefore relies on the
expertise of the analyst, rather than spectral
classifi ers. It is possible that this fi eld could
benefit from the use of filters specifically
designed to detect linear features. Note that
fault detection is more of an exercise in preparedness
than rapid response.
A number of different techniques have
been reported in the literature to map the
extent of earthquake damage, particularly in
urban areas. Image differencing of multidate
spectral ratios demonstrated better results
than synthetic aperture radar (SAR) in Turkey
and Iran, though a combination of optical and
SAR coherence was reported to give the
most accurate result (Stramondo et al.,
2006). Sertel et al. (2007) used semivariogram
analysis of SPOT panchromatic imagery
obtained both before and after the Izmit
earthquake in Turkey. This technique demonstrated
the possibility of mapping earthquake
severity based on changes in the shape
of semivariograms, although further research
was suggested before the relationship to
a quantifi able amount of damage could be
determined. It may also be of use where the
spatial resolution of the image data is sufficient
to detect textural changes, though
insuffi cient to detect specifi c damages. See
also section IV for details of landslide mapping
as a result of earthquake damage.
2 Thermal and microwave detection of
earthquakes and faulting
As an alternative to mapping earthquake
damage, several studies have sought to characterize
short-term temperature increases
immediately prior to earthquakes. While the
detection of thermal anomalies has thus far
been conducted retrospectively, refi nement
of this technique and routine investigation
may hold information key to earthquake
prediction and warnings. A normal temperature
for a region can be calculated using
a time series of image data and an image of
interest compared with this to determine
areas of anomaly (Choudhury et al., 2006).

This technique, as well as the split-window
method, can be used with various multiband
thermal sensors. Temperature anomalies
have been observed over both land and sea
in this manner (Ouzounov et al., 2006).
Recently, attempts were undertaken to
measure the microwave signal produced by
rock failures during earthquakes with passive
microwave sensors such as Advanced
Microwave Scanning Radiometer for Earth
Observation System (AMSR-E) aboard
the satellite Aqua. Some initial results are
promising (Takano and Maeda, 2009) but
more work needs to be done in this direction.
3 SAR detection of earthquakes and faulting
High resolution Synthetic Aperture Radar
(SAR) intensity data is used for mapping
ground changes and infrastructure damages
187
Table 2 Remotely sensed data types and image processing techniques for
information extraction about natural hazards
Data type Sensor examples Technique Application Advantages Disadvantages
Multispectral
high to
moderate
resolution
Ikonos,
Quickbird,
SPOT, ASTER,
ALOS
Manual
interpretation
Infrastructure and
property damage
due to fl ooding,
earthquakes,
landslides, etc
Benefi ts from analysts knowledge
of the area in addition to other
interpretation cues such as context,
site, association, shape, size;
immediate vector output fi le
Can be subjective, time-consuming
for widespread events, and nonrepeatable
Spectral
classifi cation
Location and
extent of fl ooding,
landslides, volcanic
debris, fi re scars
Relatively rapid to apply over a large
area
Non-unique spectral response
values, may require additional
manual editing, appropriate
algorithm must be selected for
optimal result
Semivariogram
analysis and
other textural
classifi ers
Damage due to
earthquakes; location
of landslides
May be useful when spatial resolution
is lower than desired
Only returns relative damage
estimates
Image
thresholding
(including band
ratios)
Location and
debris, fi re scars
Simple and rapid to apply, band ratios
reduce illumination variability, can be
applied with panchromatic data
Determination of threshold values
may be subjective
Image
differencing
Location and
debris, fi re scars
Can be conducted on panchromatic
data, band ratios or SAR backscatter
imagery
Requires before and after imagery
that is accurately co-registered
and radiometrically balanced, only
takes the spectral information from
a single band (though this may be
a ratio combination), all changes
will be identifi ed regardless of their
relevance to the particular natural
hazard (eg, crop rotations); still
need to determine a threshold of
change
(Continued)
Data type Sensor examples Technique Application Advantages Disadvantages

Postclassifi
cation
change
detection
Location and
debris, fi re scars
Does not require radiometric
calibration between multiple images
Requires before and after imagery
that is accurately co-registered, all
changes will be identifi ed regardless
of their relevance to the particular

natural hazard (eg, crop rotations),
requires classifi cation to also be
completed on before image
DEM
generation
DEM is used as
a supplementary
information in variety
of studies
Photogrammetric methods can
provide very high resolution DEMs in
the absence of LiDAR
Stereo imaging is not automatically
acquired so may not be available;
DEM creation software is not
standard in image processing
packages ie, costs extra, derived
elevation is based on vegetation
rather than ground height, no data
in cloudy areas
SWIR GOES, TOMS,
MODIS,
ASTER
Split window Height, extent and
volume of volcanic
ash clouds
Commonly used and well tested Relatively low spatial resolution of
thermal sensors. Unable to derive
sub-pixel components
Thermal ASTER,
MODIS,
AVHRR
Split window Crater lake
temperatures, lava
fl ow, precursor to
earthquake activity,
temperature and size
of fi re hotspots
Commonly used and well tested Relatively low spatial resolution of
thermal sensors. Unable to derive
subpixel components
Dual band Crater lake
temperatures, lava
fl ow, precursor to
earthquake activity,
temperature and size
of fi re hotspots
Can derive subpixel components Assumes only two thermal
components
UV or
thermal
GOES, TOMS,
MODIS,
ASTER
Absorption in
UV or TIR
Height, extent and
volume of SO2 and
other gas emissions
Commonly used and well tested Low spatial resolution of
geostationary satellites, requires

very high temporal resolution to
monitor changes
Table 2 Continued
189
SAR JERS-1, ERS1/2, ENVISAT,
ALOS
PALSAR,
TerraSAR-X,
Radarsat-1/2,
Cosmo-SkyMed
Coherence Change detection
due to landsliding,
fl ooding, fi re, etc
Provides quantitative estimation of
ground changes
Does not work well in densely
vegetated regions, affected by
seasonal changes, accuracy
decreases with time
Backscatter
intensity
Change detection
due to landsliding,
fl ooding, fi re, etc
Can be used in cloudy conditions,
side-looking acquisition geometry is
benefi cial for certain applications
Quantitative analysis is complicated
and varies signifi cantly for different
regions, may be diffi cult to interpret
for non-experienced end-users
Interferometry/
DEM
generation
DEM is used as
supplementary
information in variety
of studies
Independent of weather conditions Accuracy depends on acquisition
geometry, wave-length and
coherence, side-looking acquisition
geometry creates distortion and
shadowed areas
Differential
interferometry
Surface deformation
due to volcanic or
tectonic activity;
velocity and extent
of slow moving
landslides
High precision, high resolution of
some new sensors
Dependent on spatial baseline and
DEM accuracy; cannot determine
difference between vertical and
horizontal components, high
accuracy only available in areas

without dense vegetation
Polarimetry Land-cover
classifi cation and
change detection
Ability to detect features that are not
visible on optical images, side-looking
acquisition geometry
Dependent on type of land cover
and seasonal changes
DEM PALSAR,
LiDAR,
TerraSAR-X
DTM
differencing
Volume of
landslide related
earth movement,
fault locations
and elevation
displacement
Provides quantitative estimation of
volumetric depositions and ground
change
Requires imagery both before and
after event to be accurately coregistered
Ikonos,
Quickbird,
SPOT
Photogrammetric methods can
provide very high resolution DEMs in
the absence of LiDAR
Stereo imaging is not automatically
acquired so may not be available;
DEM creation software is not
standard in image processing
packages ie, costs extra, derived
elevation is based on vegetation
rather than ground height, no data
in cloudy areas
Airborne
LIDAR sensors,
SEASAT
Manual
interpretation
Very high horizontal and vertical
resolution, can give accurate surface
elevation (rather than tree heights)
Acquisition of LiDAR is expensive
and takes a considerable amount of
time to process

by calculating a ratio or difference between
multitemporal images and then applying
supervised or unsupervised classifi cation in
the same way as is done with optical data
(Matsuoka and Yamazaki, 2005). The main
limitation of this approach is a significant
variability of backscatter intensity for different
regions, lack of quantitative estimations
and dependence on incidence angle.

A few modern SAR satellites such
as TerraSAR-X, Radarsat-2 and ALOS
PALSAR are capable of providing data of
various polarizations simultaneously. Phase
shift and intensity difference between images
of various polarizations are dependent on
land cover and, therefore, can be successfully
used for its classifi cation (Czuchlewski
et al., 2003). For example, due to side-looking
acquisition geometry, urban constructions
often produce distinct signal caused by the
double bounce mechanism (Guillaso et al.,
2005) and this pattern changes when buildings
are damaged by an earthquake. For this
particular case and many other applications,
SAR polarimetry will produce valuable results
and complement optical observations.
Differential SAR interferometry is possibly
one of the best techniques used for
mapping ground deformation produced by
earthquakes. Differential interferometry
(DInSAR) calculates the phase difference
between SAR images acquired before and
after an event or some other period when deformation
has occurred (Massonnet and
Feigl, 1998; Rosen et al., 2000). The accuracy
of this technique depends on data type and its
quality: wave-band, perpendicular and temporal
baselines, ground conditions (such as
vegetation and snow coverage), tropospheric
and ionospheric noise. In the most favourable
conditions it is possible to achieve accuracy
better than one quarter of SAR wavelength,
about 0.51 cm for X-band, 12 cm for Cband,
and 23 cm for L-band. This accuracy
is suffi cient for mapping ground deformation
of a moderate earthquake (M5 and up)
depending on the depth of the epicentre.
The success of DInSAR depends on the
degree of phase correlation between the
various scenes, which in turn depends on
the relative timing and geometry of the various
scenes, as well as decorrelation due to
the atmosphere, the relative accuracy of the
orbit knowledge, and the precise conditions
of image acquisition. Decorrelation occurs
when surface conditions are signifi cantly different
between two acquisitions, or when
they appear different in case of large spatial
baselines. The effect of decorrelation is less
significant for L-band than for C- and Xband.
In early SAR satellites, images that
might otherwise be suitable for InSAR or
DInSAR were sometimes incoherent due to
lack of yaw steering, although this is not a signifi

cant issue for SAR satellites at the time
of writing. Various techniques have been
developed in order to minimize atmospheric
noise (stacking; Small Baseline Subeset
SBAS; Permanent Scatterers PS) but
usually atmospheric noise is not a problem
for earthquake mapping because of a large
magnitude of coseismic signal. However, in
close proximity to a fault, displacements are
too large and interferometric phase cannot
be reconstructed due to ambiguity in phase
unwrapping (Hanssen, 2001).
4 LiDAR detection of faulting
Airborne LiDAR surveys are increasingly
useful for mapping surface expressions of
faulting. The extremely high vertical and
horizontal resolution is ideal for observing
previously undetected faults. Cunningham
et al. (2006) demonstrated the utility of mapping
active faults after applying a tree removal
algorithm to the derived digital elevation
model (DEM). Subsequent analysis was completed
manually by visual interpretation.
Manual interpretation was also used in
New Zealand to extend the length of
known faults and identify and map new fault
scarps (Begg and Mouslpoulou, 2007). Topographical
profi les were used to assist analysis
and quantify vertical deformations. However,
191
this technique of hazard mapping and
monitoring is not appropriate for providing
information in a rapid-response emergency
situation due to the time it takes to acquire
and process the data to a point where it can
be manually interpreted.
III Volcanic activity
Detection and monitoring of volcanic activity
spans a number of different data types and
processing methods. Thermal anomalies are
commonly detected by comparing a location
with its background or average temperature;
volcanic deposits are best detected with
optical data, often using the normalized
difference vegetation index (NDVI) for
spectral enhancement; the split-window
method is used for detecting ash composition
within clouds; and InSAR is best for
volcanic deformations.
1 Optical detection of volcanic activity
A variety of sensors are available to provide
data suitable for debris mapping, with a
preference noted towards the higher spatial
resolution satellites such as SPOT, Landsat

and ASTER (Kerle et al., 2003; Joyce et al.,
2008c). The use of additional spatial information
such as pre-event images has been
considered important for accurate detection
of the deposit and vital for damage assessment.
Quickbird and IKONOS could be used
for this application, dependent on the extent
of debris and fi nancial constraints on the
project. Lower spatial resolution satellites
such as AVHRR have been found to be inadequate
(Kerle et al., 2003). ASTER has
the added advantage of providing data that
can be used to develop a DEM of the region,
which can be useful in volumetric analysis of
debris deposits (Huggel et al., 2008).
Few methods of automatic detection of
volcanic debris using remote sensing have
been reported in the literature, and it appears
that the favoured technique is manual digitization.
However, the normalized difference
vegetation index (NDVI) is commonly used
for its ability to enhance the difference between
volcanic deposits and surrounding
vegetated areas (Castro and Carranza, 2005;
Harris et al., 2006). The NDVI can also be
combined with a threshold value to delineate
the deposit. Other techniques that have been
used to aid visual interpretation of changes
due to volcanic activity include a multiband
display incorporating different input dates
(Calomarde, 1998; Castro and Carranza,
2005), principal components analysis, and
image subtraction (Torres et al., 2004).
2 Thermal detection of volcanic activity
Of the more common applications, thermal
monitoring of crater lake temperatures, lava,
and thermal anomalies have been conducted
since the 1980s, initially using the thermal
sensor on board the Landsat series of sensors
(Francis, 1989; Oppenheimer, 1996; Kaneko,
1998; Kaneko and Wooster, 1999; 2005;
Donegan and Flynn, 2004; Harris et al.,
2004). The number of thermal applications
has since increased with the development of
more sophisticated techniques such as subpixel
analysis with multiband thermal sensors
like ASTER (Pieri and Abrams, 2004; 2005;
Lombardo and Buongiorno, 2006), MODIS
(Wright and Flynn, 2003; Wright et al.,
2004; 2005) and AVHRR (Mouginis-Mark
et al., 1994; Carn and Oppenheimer, 2000;
Patrick et al., 2005). The dual-band method
is commonly applied to derive subpixel-level
thermal anomalies, though is considered an
oversimplifi cation of reality as it assumes that

only two features of different temperatures
exist within a pixel, whereas realistically there
may be up to seven components (Wright and
Flynn, 2003). Alerts for thermal anomalies
observed on a global scale are available freely
on the internet, utilizing both MODIS and
GOES image data (Wright and Flynn, 2003;
Wright et al., 2004; Rothery et al., 2005).
These need to be viewed with caution,
however, as the alerts are not sensitive to
merely warm features (eg, minor crater lake
heating), and will also be triggered by other
non-volcanic sources of heat such as a
bushfi re (Rothery et al., 2005).

Remote sensing is commonly used to
monitor ash clouds and volcanic gaseous
emissions (SO2) from both explosive and non
explosive eruptions. Large-scale eruptions
can eject gases and ash into the stratosphere,
which may be monitored using geostationary
satellites that measure absorption in the UV,
SWIR and TIR wavelengths (eg, TOMS,
GOES). However, they often lack the
spatial resolution and radiometric sensitivity
to monitor smaller fl uctuations in volcanic
activity. Such plumes may be monitored with
higher resolution using ASTER (Realmuto
et al., 1997; Pieri and Abrams, 2004; Pugnaghi
et al., 2006), though temporal variability monitoring
is compromised with the less frequent
overpasses of this sensor. SO2 is measured
using diagnostic absorption features in either
the UV or TIR region of the spectrum. Ash
composition is commonly measured using
the split-window technique, and requires
subtraction of the refl ectance in one SWIR
or TIR band from another (Prata, 1989). This
technique allows distinction between volcanic
ash clouds and those of meteorological
origin, based on threshold values (Dean et al.,
2004; Tupper et al., 2004; Gu et al., 2005).
Ash monitoring is considered operational and
useful for aviation and warnings to volcanic
ash advisory centres (VAACs).
3 SAR detection of volcanic activity
SAR has successfully been used for mapping
volcanic deformations (Berardino et al., 2002;
Kwoun et al., 2006) as well as for monitoring
of pyroclastic fl ows and lahars (Terunuma
et al., 2005). It is particularly useful in eruptions
where there is a lot of smoke obscuring
the target and preventing the effective use
of optical data. Three products derived from
SAR data are usually used: SAR backscatter

intensity, InSAR coherence, and DInSAR
interferometry.
Various ground conditions (eg, roughness,
soil moisture, slope) affect intensity of refl ection.
Therefore, calculating a difference or
a ratio of intensity images from before and
after the event should produce an image
where changes caused by pyroclastic fl ows
or lahars are easily observable. However,
Terunuma et al. (2005) showed that changes
of intensity signal are not clearly observed
for either C- or L-band SAR sensors. But
coherence calculated for the pair of images
before and after an event was signifi cantly
lower in regions where pyroclastic fl ows or
lahars had occurred. However, estimation of
coherence variation is only possible in regions
with a high initial degree of coherence, ie,
those not covered by dense vegetation or
snow, and not being eroded too quickly.
SAR interferometry can be used to derive
DEMs before and after an eruption. By subtracting
these models it is possible to observe
large-scale deformations caused by lahars or
pyroclastic fl ows. Alternatively, differential
interferometry can estimate deformation
with subcentimetre accuracy caused by
thermal and pressure sources underlying the
surface. By monitoring volcanoes over a
long period it is possible to reconstruct
temporal patterns of deformation using
SBAS (Berardino et al., 2002) or permanent
scatterers techniques (Ferreti et al., 2001;
Hooper et al., 2004).
IV Landslides
It is possible to use both SAR and optical data
for landslide detection, but it appears that
optical data provides better results, most
likely due to spatial resolution and sensor
look angle. It can, however, suffer from misclassifi
cation with other areas of bare groud.
Multitemporal analysis is preferable and
spectral enhancement is often required. SAR
data would be most useful in the event of
storm-induced landsliding, where cloud cover
impeded optical acquisition. InSAR could be
useful for measuring the rate of slow-moving
landslides. Difficulties can arise with SAR
data in areas of high slopes due to layover and
shadowing effects.
1 Optical detection of landslides
Visual interpretation, with and without onscreen
digitizing of both two- and threedimensional
data, has been commonly used
in the past and is still an effective method of

193
landslide mapping (Singhroy, 1995; Singhroy
et al., 1998; Ostir et al., 2003; Nichol and
Wong, 2005a; Kumar et al., 2006; Voigt et al.,
2007). Orthophotography in particular has
demonstrated its utility for mapping landslides
in detail and IKONOS with pansharpening
has been suggested to be of
equivalent if not superior use (Nichol and
Wong, 2005b). Manual techniques benefi t
from the analysts knowledge of the area,
though they cannot be automated, and are
impracticably time-consuming when mapping
widespread numerous landslides. Some
emerging studies are attempting to use more
automated extraction techniques, utilizing
band ratios (Cheng et al., 2004; Rau et al.,
2007) and unsupervised (Dymond et al.,
2006) and supervised classifi cation (Nichol
and Wong, 2005b; Joyce et al., 2008a; 2008b)
to reduce the level of manual interpretation,
while still providing reasonably high accuracy
levels (up to 80%). In a study to test the most
accurate method for mapping landslides with
SPOT-5 imagery, it was determined that
the spectral angle mapper (SAM) supervised
classifi cation and NDVI thresholding were
the most accurate semi-automated techniques
compared with the results achieved
with parallelepiped, minimum distance to
means, principal components, and multitemporal
image differencing (Joyce et al.,
2008a; 2008b). This study also noted that
manual digitizing produced a higher accuracy
than any of the aforementioned techniques,
but was considerably more time-consuming
for a widespread event.
Multitemporal image analysis is a valuable
technique that can be used if imagery is
available both before and after a landsliding
event. This is perhaps the most promising
option for rapid response. The process
applied then relies on digital change detection
a number of methods of which are available
(Singh, 1989; Jensen, 1996). An overall accuracy
of approximately 70% was achieved
using the postclassification comparison
technique for landslide detection in Hong
Kong (Nichol and Wong, 2005b). While
apparently effective, and commonly used in
other areas of interest, this method is not
documented frequently in landslide detection
literature. An alternative technique requires
multidate image differencing, as demonstrated

with SPOT imagery in Taiwan (Cheng
et al., 2004), using an infrared and red band
ratio, image differencing, and thresholding.
The results were later confined to slopes
greater than 22 to reduce misclassifi cation
due to human-induced change. Image differencing
has also proven effective using
panchromatic data for landslide mapping in
Italy (Hervas et al., 2003; Rosin and Hervas,
2005), though while the authors report success
an accuracy assessment is not given. Less
success was reported when using SPOT-5
data in New Zealand, and diffi culties with
this technique were associated with cloud
cover affecting the radiometric calibration
between the two dates (Joyce et al., 2008a;
2008b). This technique is also heavily reliant
on accurate co-registration between images.
Unsupervised classification without
change detection was used by Dymond et al.
(2006) with SPOT-5 data for mapping the
combined erosion scar and debris of a landslide.
The classification was restricted to
slopes greater than 5 in an effort to reduce
misclassifi cation of bright pixels that may
otherwise be areas of bare ground. While
80% accuracy was reported, only landslides
greater than 10 000 m2 were checked, and
an independent data set was not used for
verification. Given the spatial resolution
of SPOT (10 m), it is possible that there
were many undetected small landslides that
were not represented in this accuracy statement.
Due to the high contrast of landslides
with background features, subpixel identification
is possible, though accurate
boundary delineation is not (Nichol and
Wong, 2005b).
An alternative option to the use of spectral
classifiers with optical data is to include
textural layers (Whitworth et al., 2005). The
image roughness with respect to its surrounds
caused by landslide debris and shadow effects

can add additional information as a layer in a
classifi cation. A principal components layer
can also be used to assist distinction between
land-cover types that appear similar in the
textural layer (Whitworth et al., 2005).
The calculation of change in a DEM surface
is one of the more frequently used
techniques for landslide detection and monitoring
using remotely sensed data (Kaab,
2002; Ostir et al., 2003; Singhroy and Molch,
2004; Casson et al., 2005; Chen et al., 2006;

Nichol et al., 2006; Tsutsui et al., 2007). This
technique is only useful for large landslides
with considerable vertical change in the topography.
The stereo viewing capability of several
contemporary sensors (SPOT, Ikonos,
Quickbird) makes this a viable technique
for acquiring imagery for the use of change
detection and potentially for rapid response.
In addition, the very high spatial resolution of
the panchromatic stereo imagery from the
Ikonos and Quickbird satellite sensors produce
very detailed elevation models that are
considerably more cost-effective than the
equivalent areal coverage of airborne LiDAR
or SAR. The technique of DEM differencing
also allows volumetric calculation of erosion
and debris. However, stereo imaging is not
automatically acquired with all sensors and
may not be available for the specifi c area of
interest, especially as rapid response type
imagery. In addition, DEM extraction tools do
not come as standard with image processing
software, thus require separate add-on
modules or software to create (eg, Leica
photogrammetry suite, DEM extraction
tool in ENVI).
2 SAR detection of landslides
As with optical remote sensing, there is
no distinct backscatter signature that can
uniquely be associated with the mixed targets
in a landslide. Instead, it is necessary to either
use expert interpreter knowledge on a single
scene, or estimate the backscatter difference
from a pre- and post-landslide event,
and apply some threshold of change (Belliss
et al., 1998).
While detecting backscatter difference is
theoretically a straightforward task, there
are some complications involved. First, if
the images have slightly different viewing
positions or different ground local incidence
angles, then the scenes will exhibit an apparent
change in brightness due to the difference in
local incidence angle. This topographic difference
can be corrected, provided a good
DEM is available (Pairman et al., 1997).
Second, the inherent radar brightness of a
target depends to a certain extent on surface
environmental conditions on each date.
Therefore, a difference-between-dates
backscatter image might still show an overall
brightness difference different from zero (ie,
false positive). This effect is exacerbated with
shorter wavelengths. Finally, unless the preand
post-landslide images tightly embrace the

times of the landslide event, the differencebetweendates backscatter image will tend
to falsely detect landslides that are simply
due to land-cover change a problem that is
similarly recognized with optical imagery.
Differential interferometry is used to
measure velocities and extent of slowmoving
landslides. In Rott and Nagler (2006)
ERS SAR was used to map a landslide that
is moving with 2.4 cm per year velocity.
Unfortunately, such a high accuracy is only
possible in regions not covered by dense
vegetation that are coherent for extended
periods of time.
V Flooding
Flooding is readily apparent in both optical
and SAR data, providing there is knowledge
of water body location prior to the event.
The use of SAR is preferable due to the
likelihood of associated cloud cover. Simple
techniques of image thresholding are easy to
implement.
1 Optical detection of fl ooding
Mapping and monitoring water inundation can
be a challenging process for passive remote
sensing due to the often coincident cloud
cover, and the fact that water is generally not
195
visible under a closed vegetation canopy. The
utility of high temporal resolution sensors
such as AVHRR is realized by Sandholt
et al. (2003), who state that although the
spatial resolution is coarser than many other
satellite sensors the frequent revisit time
offers a greater probability of obtaining cloudfree
imagery. They used linear spectral unmixing
with thermal imagery to determine
inundated areas, but were faced with the
difficulty of selecting pure endmembers.
Alternatively, they also tested supervised
maximum likelihood and ISODATA clustering
classifi cations with the higher spatial
resolution Landsat ETM+, concluding that
no technique is necessarily better than the
other, rather that each has its advantages and
disadvantages depending on the flooding
extent, cloud cover and temporal variability.
Manual analysis of MISR imagery was
completed to determine quantitative characteristics
of the 2004 tsunami development
along the eastern coast of India (Garay and
Diner, 2007). This provided information about
wave propagation and behaviour, but was
not used to estimate the extent of damage.

Potentially one of the most useful studies
for rapid-response flood mapping was
conducted to create on-board satellite processing
algorithms for Hyperion imagery (Ip
et al., 2006). The algorithm utilizes three
narrow spectral bands for a classifi cation and
is then compared to a base scene to extract
fl ood detail rather than just wet regions (eg,
rivers, lakes). However, Hyperion has limited
global coverage, and obtains imagery in relatively
small segments that would be useful
for localized fl ooding but not necessarily large
events.
The extreme fl ooding events associated
with several tropical storms in recent years
(Hurricane Katrina, Cyclone Nargis) have
been successfully and rapidly mapped using
a variety of sensors to take advantage of differences
in spatial and temporal resolution.
Geoscience Australia is actively acquiring
imagery of fl ooding events in Australia and
attempting to develop semi-automated
techniques for extracting inundated areas
(Lymburner et al., 2008; Thankappan et al.,
2008). Flooding events of 2007 and 2008
were successfully mapped using Landsat-5
TM and ALOS AVNIR-2.
2 SAR detection of fl ooding
SAR would appear to be an ideal sensor for
the detection of extensive flooded areas,
since the backscatter signature of water is so
distinctive compared with that of vegetation
(Lewis et al., 1998). Spectacular examples
of the use of SAR include the April 1997
Red River flood near Winnipeg, Canada
(Bonn and Dixon, 2005), and the Mississippi
fl ood of 1993 (Nazarenko et al., 1995). The
basic underlying assumption in these cases
is that the fl oodwater remains visible for a
suffi ciently long period of time to allow for
acquisition of imagery and subsequent delineation
of the flood boundary, which was
certainly the case in these two major fl ood
events.
SAR backscatter intensity and InSAR
coherence can be successfully used together
for mapping of regions affected by fl ooding.
In Oberstadler et al. (1997) it was shown that
flooded areas appear darker on ERS SAR
intensity images and, therefore, comparing
two images before and during fl ooding it is
possible to map flooded areas with a high
degree of accuracy. By combining SAR with
other geospatial data such as DEMs, it is also
possible to estimate the depth of water in

fl ooded regions. InSAR coherence can also
be used for the same purposes (Geudtner
et al., 1996). This technique maps coherence
of a SAR pair of images acquired before and
during fl ooding and comparing it to a pair of
SAR images that are both acquired before
flooding. Areas affected by flooding have
signifi cantly lower coherence than dry areas
and by subtracting both coherence maps it is
possible to identify these areas easily.
One of the unique features of SAR is
the ability to detect areas of fl ooding under
closed-canopy vegetation. Areas of fl ooded
vegetation show with enhanced backscatter,

due to the corner-refl ector effect formed from
the vegetation and the smooth water surface.
The effect is wavelength- and vegetationdependent,
with short-wavelength (X- and
C-band) sensitive to fl ooding under grasses
(MacDonald et al., 1980), mid-wavelength
(S-band) sensitive to flooding under reed
and brush vegetation (Lewis et al, 1998), and
long-wavelength (L- and P-band) sensitive
to fl ooding under trees (Imhoff et al., 1987).
This phenomenon has been long known with
wavelengths as short as K-band (Waite and
MacDonald, 1971), and has been explained
by a comprehensive model (Ormsby et al.,
1985). As in the case of the SAR detection
of landslides above, successful detection of
flooded areas under vegetation requires
some visual interpretation experience, or the
assistance of a scene gathered before the
fl ooding in order to make a comparison.
VI Wildfi re
Monitoring fi res using remote sensing is done
in one of two ways: applications are based on
either near-real-time monitoring of the fi re
itself and/or smoke, or on mapping the extent
and severity of burnt areas. MODIS and
AVHRR are the most commonly used data
types for this application due to their high
temporal resolution. The use of their NIR
wavelengths also allows for some penetration
through smoke to view the fi re scar.
1 Optical detection of wildfi re
Mapping and monitoring of fi re scars can use
similar methods to those described for landslides,
as the characteristic of these features
is cleared vegetation. However, while landsliding
results in exposed soils appearing
as brighter features than their surrounds,
burnt vegetation and organic matter in soils
generally results in optically darker scar

features in imagery due to the presence of
ash. The degree of difference in refl ectance
values between burnt and unburnt surfaces
is dependent on the type of vegetation.
Satellite-based remote sensing has been
used to extract information about the combustion
completeness and also the fraction
of an area that contains burns, through
developing regression models with in situ
information (Roy and Landman, 2005). While
texture was considered to vary with burn
scars in low spatial resolution (1 km) imagery,
the isodata spectral classifi er produced higher
accuracy levels for detection of scars in
savannas. Band ratios and normalized indices
are considered to provide better separation
between burnt and non-burnt areas than
individual spectral bands, particularly when
combining wavelengths in the SWIR region
of the spectrum (Trigg et al., 2005). Other
ratios, such as the normalized difference
vegetation index (NDVI) and the normalized
burn ratio (NBR), were found to be less
effective (Trigg et al., 2005; Hoy, 2007).
Unsupervised classification has also been
successfully used to delineate large scar
areas in mosaicked AVHRR imagery. Most
recently, the bushfi res in Victoria, Australia
(February 2009), have been captured
by optical sensors like MODIS, ASTER,
Hyperion and ALI. It is too soon to report on
automated or semi-automated techniques
for extraction of the burn extent.
2 Thermal detection of wildfi re
As with estimating hotspots due to volcanic
activity, the MODIS sensor is often used.
Two techniques are commonly cited for
derivation of surface temperatures and
hotspots. The dual-band technique derives
the temperature of two components and
their portions within a pixel (Dozier, 1981),
though problems with the oversimplifi cation
of this technique were explored by Giglio
and Kendall (2001). As an alternative, the
split-window technique has proven useful as
a basis for global monitoring of fi res, through
detecting changes from a specified background
value. However, it is unable to derive
subpixel components. These methods are
197
similar to those used for detection of thermalbased
volcanic activity. Thermal differences
have also been noticed between burnt and
unburnt areas, thus temperature can also be

used to map and monitor fi re scars.
3 SAR detection of wildfi re
Synthetic aperture radar has been successfully
used for identification of burnt areas
over the boreal forest of Alaska, Canada
(Bourgeau-Chavez and Kasischke, 2002), and
Siberia (Ranson et al., 2003) as well as over
the tropical rain forest of Indonesia (Siegert
et al., 1995) and semi-arid Mediterranean
forest (Gimeno and San-Miguel-Ayanz,
2004). In these studies it was shown that
the intensity of SAR backscatter signal was
noticeably different between predominantly
undisturbed and fi re-disturbed forest. This
was due to an increase in soil moisture, increased
surface roughness exposed to the
incoming microwave radiation, and damage
to the vegetation canopy by fi re.
In Gimeno and San-Miguel-Ayanz
(2004), various SAR products with different
resolution and various incidence angles were
tested and novice incidence-angle normalization
was introduced. It was also shown
that acquisitions with low incidence angle
are the most successful in identifi cation of
burnt areas in hilly regions. In Couturier
et al. (2001) the relationship between SAR
backscatter intensity and fi re-related Daily
Drought Index (DDI) was investigated over
rain forest of Indonesia. A strong correlation
between these two characteristics was
observed, which suggested that SAR data
can be successfully used not only for identifi
cation of burnt areas but also as a proxy
for the susceptibility of forest to burn.
VII Operationally active hazard
monitoring programs using remotely
sensed imagery
Various programs are becoming available
and more accessible for serving both imagery
and information with respect to hazards and
disasters. The USGS hosts a Hazards Data
Distribution System (HDDS), where it is
possible to download pre- and post-event
imagery from recent hazard events. It is also
possible to retrieve baseline imagery from
other areas in the United States through
this interface. However, this is not an online
operational portal for analysed imagery.
The University of Hawaii operates two
websites for near-real-time monitoring of
thermal hotspots. The first site contains
GOES imagery of selected sites in the
Western Hemisphere that is updated every
15 minutes, displaying combinations of

visible, mid-infrared and thermal radiation.
The second, more interactive site is based
on MODIS imagery and displays hotspots as
determined by a threshold in thermal values
(Wright et al., 2004). The user can select a
particular area of interest and retrieve a text
fi le containing the location of the hotspot.
Geoscience Australia has a similar hotspot
identifying GIS interface based on MODIS
and AVHRR thermal imagery. This covers
Australia, New Zealand and the South
Pacifi c. Again, a threshold is used to identify
locations hotter than a background value.
The University of Maryland hosts a global
fi re mapper that is linked to a system to provide
email alerts of fi res within a specifi ed
area of interest. The option is also available
at several of these sites to export the current
hotspots to Google Earth for near-real-time
viewing by emergency responders. Due to the
underlying data, each of these applications
experiences diffi culties in detecting hotspots
in areas of high cloud and/or smoke.
To ensure safe navigation and monitor
possible climatic impact, NOAA records
global historical volcanic eruptions, tracks
volcanic ash eruptions affecting the United
States, issues volcanic ash advisories, and
provides ash cloud forecasts. The USGS
Volcano Hazards CAP Alert provides daily
updates of volcanic activity for sites identifi ed

as requiring a watching brief (for example,
see http://volcano.wr.usgs.gov/cap/cap_
display.php?releaseid=3886).
In the Asia/Pacifi c region, Sentinel Asia
is an on-demand network of information
delivery websites, largely using free-to-air
satellite imagery, developed to provide online
information in near-real-time. Based upon the
Australian bushfi re tracking system, Sentinel
Hotspots, and supported by the Japanese
Government via the Japan Aerospace Exploration
Agency (JAXA), Sentinel Asia can
detect and then monitor natural disasters
in the region (Kaku et al., 2006). Thus far,
it has been activated a number of times by
several countries and contains information
on a various events such as the cyclone and
fl ooding in Myanmar and earthquake damage
in China (May 2008), as well as other fl ooding
events throughout Asia and the bushfi res in
Australia (February 2009).
The International Charter Space and
Major Disasters has a membership of several

international space agencies and space
system operators and aims to provide remotely
sensed imagery and data to member
countries affected by a disaster. In January
and February 2009 alone, the charter was
activated eight times for an earthquake and
landslides (Costa Rica); fl oods (USA, Morocco,
Argentina, Namibia); fires (Australia); a
hurricane (France); and volcanic eruption
(Chile). The type of imagery acquired will
be specific to the scale, location and type
of hazard and could be a combination of
optical and SAR. There does not appear to
be a standard for image processing, and this
may depend on the organization enlisted for
project management. The service is provided
free to member organizations.
VIII Remote sensing for providing rapid
response information
Remote sensing satellites have frequently
been used to contribute to disaster management.
The most common, best understood,
and operational of these uses is that of
weather satellites for cyclones, storms and,
in some cases, fl ash fl oods. These systems
have certain clear advantages. For instance,
there are many orbiting and geostationary
satellite services available, and coverage of
almost any part of the world is available in
small timescales ranging from hours to a few
days. Further, imagery from these satellites
is relatively cheap or freely available, and
the scale of the events roughly matches the
resolution of the satellite imagery. Spatial
resolution, image extent and spectral characteristics
play a large role in determining
whether or not a particular sensor or data
type is capable of detecting individual
hazards, irrespective of the ability to acquire
or process these data (see Table 3, derived
from similar work in coastal environments
CRSSIS, 2006). Many of these data types
have been discussed in previous sections with
specifi c examples.
There are a number of other provisos on
the ability of a satellite sensor to monitor a
disaster. Where imagery cannot be recorded
on board the satellite, and where there is no
local receiving station coverage or where a
local receiving station is not licensed for a
particular satellite, data cannot be collected.
For many parts of the world, medium to
high resolution remote sensing satellites will
only acquire data after the satellite has been
programmed to do so. In these circumstances,

coverage of the affected area is likely to
be delayed and possibly missed. However,
when major disasters unfold, most satellite
operators will schedule imagery collection,
even without confirmed programming requests,
either on humanitarian grounds or in
the hope of data sales.
In a country with national reception capabilities,
programming satellite acquisition
may not be required but data acquisition will
still depend upon satellite orbit constraints,
For example, when the World Trade Center
in New York was attacked, the French
SPOT satellite was the fi rst able to acquire
imagery, on 11 September, just hours after
the event. The American Ikonos satellite
collected imagery the following day (Huyck
and Adams, 2002). In this situation, the
satellites were the only platforms capable
199
of acquiring early imagery since there was
an air traffi c ban in force over the USA that
was not lifted until 13 September. From
then on, aircraft-mounted sensors were
able to provide daily multispectral imagery,
thermal imagery, photographs, hyperspectral
imagery, and LiDAR. Some time later,
some aerial photographs were converted to
three-dimensional information by use of proprietary
software from Pictometery, which
can make all angles of a structure visible and
measurable. Such images would have been
useful for assessing hazards, eg, overhanging
debris, during the clean-up (Huyck and
Adams, 2002).
There are still many countries that do
not have access to direct reception stations
within their territory for medium to high
spatial resolution imagery. In an emergency,
commercial satellite services can be tasked
to collect data. Depending upon the position
of the particular satellite within its orbit, the
time interval between an urgent data programming
request and the fi rst acquisition
attempt could be as short as 24 hours or
as long as several days. As a general rule,
satellite services more commonly used for
emergencies are better at rapid response.
A prime example of such a satellite service
is Radarsat, which can typically schedule a
data acquisition at very short notice and then
supply the data to the data requester within
hours of a successful acquisition. Other
optical commercial satellite services could

accept a programming request at similar
short notice. Successful data acquisition
would then rely predominantly upon orbital
constraints plus, for optical satellites, cloudfree
conditions over the area of need.
IX Emerging systems
Every year, more earth observation satellites
are launched than go out of commission.
Currently, both Radarsat and SPOT have
two satellites in orbit, and many other
services with one satellite are planning a replacement
and/or a second vehicle. Just as
there are constellations of communications
satellites to allow continual coverage, a
number of national/international initiatives
are in the process of launching constellations
of satellites to enable daily or more frequent
observation of anywhere in the world. One
of the main drivers/justifications of these
services is national security and this includes
hazard monitoring. Systems that are under
way include Cosmo-Skymed, a constellation
of two X-band SAR satellites and two optical
satellites phased at 90 degrees from one
another. Another constellation under way is
one developed by Surrey Satellite Technology
(SST) and known as the Disaster Monitoring
Constellation (DMC). This is a series of microsatellites
carrying multispectral sensors,
each of which will be owned by individual
countries/partners that will cooperate in data
acquisition and distribution in the event of a
disaster. When complete, the constellation
should enable daily data coverage at the
equator and hourly at higher latitudes (da
Silva Curiel et al., 2005).
The German satellite TerraSARX
(launched in June 2007) is capable of
acquiring imagery with up to 1 m spatial resolution.
Again the literature on processing
and applications of this sensor is limited, but
the sensor is expected to be another useful
option for disaster monitoring. TanDEM-X
is scheduled for launch in 2009 and will fl y
in orbit close to TerraSAR-X, allowing the
generation of high resolution DEMs that
could potentially be used for change analysis,
particularly with respect to volumetric
analysis of landslide-related earth movement.
Also German, the RapidEye constellation
of fi ve satellites launched in August
2008 boasts a daily revisit time with a 6.5 m
pixel size. Their online kiosk is also designed
for rapid access to data. The combination of
high spatial and temporal resolution in optical

sensors holds great potential for disaster
monitoring.
In addition to satellite platforms, the high
level of fl exibility afforded by some airborne
platforms is proving to be of real utility
for disaster mapping and monitoring. With
state of the art GPS and IMU (inertial measurement
units) on board, it is possible to
Table 3 Utility of various data types for providing information about natural hazards
Spectral Visible NIR SWIR Hyperspectral Thermal SAR LiDAR
Spatial (very high = <5 m;
high = 520 m; medium =
20250 m; coarse = >250 m)
Very high High Medium Coarse High Medium Coarse Very high Medium Coarse Medium Coarse Single
polarization
Polarimetric DEM
Sensor example Quickbird,
Ikonos
ASTER,
SPOT,
ALOS
Landsat MODIS,
AVHRR
ASTER,
SPOT
Landsat MODIS,
AVHRR
CASI,
Hymap
Hyperion MODIS ASTER,
Landsat
MODIS,
AVHRR
Radarsat-1,
ERS-1/2
Radarsat-2,
TerraSAR-X
Airborne
Sensor
Volcano Thermal
anomaly
< 100C
EEEEEEEEEEAAEEE
Thermal
anomaly
>100C
EEEEBBBBBBAAEEE
Thermal
anomaly
>1000C
BBBBBBBBBBAAEEE
Lahar B A A E E E E B B E B E B D B
Ash clouds
detection B B B B B B A E * B B B B E E E
Ash clouds
quantifi cationE E E E B B A E * B B B B E E E
Gas clouds
detection B B B B B B B E * B B B A E E E
Gas clouds
quantifi cationE E E E B B B E * B B B A E E E
Deformation E E E E E E E E E E E E A D A
Debris B B B C E E E B B C E E C D B
Lava fl ow B B B C B B C B B C B B C D B
Pyroclastic
fl ow B B B C B B C B B C B B C D B
201
Earthquakes
and faulting
(see also
fi res, fl ooding
and
landslides)
Fault location B C C E C C E B C E E E B E A
Deformation E E E E E E E E E E E E A C A
Aftermath
building
and property
damage
BBBCEEEBBCEECCB
Landslides Scar + debris
fl ow B A B C E E E B B E E E C D B
Isolate scar
from debris
fl ow
CCEEEEECCEEECDB
Flooding Inundated areaA A B C B B C B B C B C A A B
Aftermath
building
and property
damage
ABBCCCEBBEEECDC
Wildfi re Fire front B B B B B B A B B B B A E E E
Aftermath
building
and property
damage
BBBCBBCBBCEECDB
Landscape
scars B A A C E E E B B C E E C A B
A: Clearly demonstrated to work using standard image processing systems and is openly available in the literature
B: Shown to work with experimental image data sets or over limited areas with very small pixels or over global scales with large
pixels
C: If extent is bigger than several pixels
D: Not widely available in literature but theoretically should be a potential use
E: Not feasible
* Listed as not feasible because of aircraft restrictions on fl ying over volcanic ash/gas clouds, rather than sensor inability

achieve relatively accurate geopositioning
for acquired imagery and to develop automated
procedures for orthorectifi cation. Onboard
processing can also facilitate anomaly
detection such as that used in fi re mapping
so that a useful product can be served to
the end-user as a shapefi le, image, or layer
on Google Earth within minutes of capture
(Ambrosia and Hinkley, 2008). Systems
are being developed for both manned and
unmanned vehicles that have the added
advantage of being able to fl y under clouds in
some areas and can incorporate a variety of
sensors including optical and thermal.
X Summary
The use of remote sensing for mapping and
monitoring natural hazards has diversified
in recent years owing to an increase in data
availability and technological advances in
their interpretation. Remote sensing has
proven useful for a range of applications including
the detection of earthquakes, faulting,
volcanic activity, landslides, fl ooding, wildfi re,
and the damages associated with each. A
number of options for processing different
data types collected in response to each of
these hazards have been reviewed, with the
commonality being the strong preference
for baseline data to be acquired and maintained

pre-event. It has been noted on several
occasions that damage assessment is not
possible without an understanding of the
initial-state environmental characteristics.
Automated techniques are well established
for identifi cation of fi re and volcanic activity
associated with excessive heat, but the
operationalization of mapping other hazard
and disaster events requires more robust
and generic techniques to be developed and
implemented.
As the importance of good spatial data is
becoming increasingly recognized, remote
sensing in the fi eld of hazard assessment and
disaster management is likely to grow in the
future. New earth observation satellites are
continually being launched, recognizing the
prospective market in disaster management,
but the provision of acquired image data in a
rapid response situation remains a challenge
both technically and fi nancially. There is also
the potential for increased use of airborne
platforms to provide the fi rst level of image
data in an emergency situation by acquiring,
processing and serving imagery in near-realtime
to the end-user.
It is not possible to recommend a single
data type or processing solution that will
work under all conditions. This is a broad
fi eld of applications where some techniques
will work better under some circumstances
than another. While manual interpretation
of many data types for various applications
provides a popular mapping solution, this
is unlikely to be the way forward for rapid
response and emergency events. SAR data
and InSAR techniques, for example, are
of considerable value for mapping fl ooding
extent and earth deformations due to volcanic
or tectonic activity, but are unable to
detect thermal anomalies or concentrations
of volcanic gaseous emissions. Optical data
offers several advantages over SAR, but is
inherently affected by cloud cover, smoke
or haze at the time of satellite overpass. The
fl exibility provided by a multisensor, multiplatform
approach is likely to give the most
comprehensive coverage of a disaster event.
Acknowledgements
This work was supported by the New
Zealand Foundation for Research, Science
and Technology under contract C05X0403.
Thank you to anonymous reviewers for
helpful comments on the original manuscript.
References
Ambrosia, V.G. and Hinkley, E. 2008: NASA
science serving society: improving capabilities for fi re
characterization to effect reduction in disaster losses.
IEEE International Geoscience and Remote Sensing
Symposium, Boston, MA, 4, IV-628-631.
Begg, J.G. and Mouslpoulou, V. 2007: Rangitaiki
Plains: another veil is lifted. New LiDAR data from
Bay of Plenty, New Zealand. Geological Society of
New Zealand Miscellaneous Publication 123A, 9.
Belliss, S.E., Pairman, D. and McNeill, S.J. 1998:
Use of Radarsat data to map Landslide erosion in
203
steep landforms. Application Development and
Research Opportunity (ADRO) Final Symposium,
1315 October, Montreal, Quebec, Canada,
Electronic Proceedings.
Berardino, P., Fornaro, G., Lanari, R. and Sansoti,
E. 2002: A new algorithm for surface deformation
monitoring based on small baseline differential SAR
interferometry. IEEE Transactions on Geoscience and
Remote Sensing 41, 237583.
Bonn, F. and Dixon, R. 2005: Monitoring fl ood extent
and forecasting excess runoff risk with Radarsat-1
data. Natural Hazards 35, 37793.
Bourgeau-Chavez, L. and Kasischke, E. 2002:
Mapping fire scars in global boreal forest using
imaging radar data. International Journal of Remote
Sensing 23, 421134.
Calomarde, R. 1998: An application of remote sensing
and geographic information system (GIS): a case
study in Sto. Tomas-Marela River, Mt Pinatubo,
Zambales, Philippines. Asian Conference on Remote
Sensing, Manila.
Carn, S.A. and Oppenheimer, C. 2000: Remote
monitoring of Indonesian volcanoes using satellite
data from the Internet. International Journal of
Cartwright, S. 2005: National Civil Defence Emergency
Management Plan Order 2005. Published
under the authority of the New Zealand Government,
Wellington, New Zealand, 68 pp.
Casson, B., Delacourt, C. and Allemand, P. 2005:
Contribution of multi-temporal remote sensing
images to characterize landslide slip surface application
to the La Clapiere landslide (France). Natural
Hazards and Earth System Sciences 5, 42537.
Castro, O.T. and Carranza, E.J.M. 2005: Remote
sensing of temporal variations in spatial distributions
of lahar and pyroclastic-fl ow deposits, West Mount
Pinatubo, Philippines. In Erasmi, S., Cyffka, B. and
Kappas, M., editors, Remote sensing and GIS for
environmental studies, volume 113, Gottingen:
Gottinger Geographische Abhandlungen, 22329.
Centre for Remote Sensing and Spatial Information
Science (CRSSIS) 2006: Coastal remote
sensing mapping capability matrix. Retrieved 17 May
2009 from http://www.gpem.uq.edu.au/CRSSIS/
tools/rstoolkit/pdf/Mapping%20capability%20mat
rix.pdf
Chen, R.F., Chang, K.J., Angelier, J., Chan, Y.C.,
Deffontaines, B., Lee, C.T. and Lin, M.L. 2006:
Topographical changes revealed by high-resolution
airborne LiDAR data: the 1999 Tsaoling landslide
induced by the Chi-Chi earthquake. Engineering
Geology 88, 16072.
Cheng, K.S., Wei, C. and Chang, S.C. 2004:

Locating landslides using multi-temporal satellite
images. Advances in Space Research 33, 296301.
Choudhury, S., Dasgupta, S., Saraf, A.K. and
Panda, S. 2006: Remote sensing observations of
pre-earthquake thermal anomalies in Iran. International
Journal of Remote Sensing 27, 438196.
Couturier, S., Taylor, D., Siegert, F., Hoffmann,
A. and Bao, M. 2001: ERS SAR backscatter. A
potential real-time indicator of the proness of modified
rainforest to fire. Remote Sensing of Environment
76, 41017.
Cunningham, D., Grebby, S., Tansey, K., Gosar,
A. and Kastelic, V. 2006: Application of airborne
LiDAR to mapping seismogenic faults in forested
mountainous terrain, southeastern Alps, Slovenia.
Geophysical Research Letters 33, L20308, DOI:
10.1029/2006GL027014.
Czuchlewski, K., Weissel, R.J.K. and Kim, Y.
2003: Polarimetric synthetic aperture radar study of
the Tsaoling landslide generated by the 1999 Chi-Chi
earthquake. Journal of Geophysical Research 108(F1),
6006, DOI: 10.1029/2003JF000037.
da Silva Curiel, A., Boland, L., Cooksley, J., Bekhti,
M., Stephens. P., Suna. W. and Sweeting, M.
2005: First results from the disaster monitoring constellation
(DMC). Acta Astronautica 56, 26171.
Dean, K., Dehn, J., Papp, K.R., Smith, S., Izbekov,
P., Peterson, R., Keamey, C. and Steffke, A.
2004: Integrated satellite observations of the
2001 eruption of Mt Cleveland, Alaska. Journal of
Volcanology and Geothermal Research 135, 5173.
Donegan, S.J. and Flynn, L.P. 2004: Comparison of
the response of the Landsat 7 Enhanced Thematic
Mapper Plus and the Earth Observing-1 Advanced
Land Imager over active lava flows. Journal of
Volcanology and Geothermal Research 135, 10526.
Dozier, J. 1981: A method for satellite identifi cation
of surface temperature fi elds of subpixel resolution.
Remote Sensing of Environment 11, 22129.
Dymond, J.R., Ausseil, A.G., Shepherd, J.D. and
Buettner, L. 2006: Validation of a region-wide
model of landslide susceptibility in the ManawatuWanganui region of New Zealand. Geomorphology
74, 7079.
Ferreti, A., Prati, C. and Rocca, F. 2001: Permanent
scatterers in SAR interferometry. IEEE Transactions
on Geoscience and Remote Sensing 39, 820.
Francis, P.W. 1989: Remote sensing of volcanoes.
Advances in Space Research 9, 8992.
Fu, B., Awata, Y., Du, J., Ninomiya, Y. and He,
W. 2005: Complex geometry and segmentation of
the surface rupture associated with the 14 November
2001 great Kunlun earthquake, northern Tibet,
China. Tectonophysics 407, 4363.
Fu, B., Ninomiya, Y., Lei, X., Toda, S., and
Awata, Y. 2004: Mapping active fault associated
with the 2003 Mw 6.6 Bam (SE Iran) earthquake
with ASTER 3D images. Remote Sensing of Environment
92, 15357.
Garay, M.J. and Diner, D.J. 2007: Multi-angle
Imaging SpectroRadiometer (MISR) time-lapse
imagery of tsunami waves from the 26 December

assessment of fi re severity in Alaskan black spruce
forests. Masters thesis, Department of Geography,
University of Maryland. Retrieved 17 May 2009 from
http://hdl.handle.net/1903/6960
Huggel, C., Schneider, D., Miranda, P.J., Delgado

Granados, H. and Kaab, A. 2008: Evaluation of
ASTER and SRTM DEM data for lahar modeling:
a case study on lahars from Popocatepetl Volcano,
Mexico. Journal of Volcanology and Geothermal
Research 170, 99110.
Huyck, C. and Adams, B. 2002: Emergency response
in the wake of the World Trade Center attack: the
remote sensing perspective. Engineering and Organizational
Issues Related to The World Trade Center
Terrorist Attack, volume 3, Report MCEER-02SP05, 58 pp.
Imhoff, M.L., Vermillion, C., Story, M.H.,
Choudhury, A.M., Gafoor, A. and Polcyn,
F. 1987: Monsoon fl ood boundary delineation and
damage assessment using space borne imaging radar
and Landsat data. Photogrammetric Engineering and
Ip, F., Dohm, J.M., Baker, V.R., Doggett, T.,
Davies, A.G., Castano, R., Chien, S., Cichy,
B., Greeley, R., Sherwood, R., Tran, D. and
Rabideau, G. 2006: Flood detection and monitoring
with the autonomous sciencecraft experiment onboard
EO-1. Remote Sensing of Environment 101, 46381.
Jensen, J.R. 1996: Introductory digital image processing:
a remote sensing perspective. Upper Saddle River, NJ:
Prentice-Hall, 316 pp.
Joyce, K.E., Dellow, G.D. and Glassey, P.J.
2008a: Assessing image processing techniques for
mapping landslides. Proceedings of IEEE International
Geoscience and Remote Sensing Symposium, Boston,
MA 2, 123134.
2008b: Methods for mapping landslides in New
Zealand using satellite optical remote sensing. 14th
Australasian Remote Sensing and Photogrammetry
Conference, Darwin, electronic proceedings.
Joyce, K.E., Samsonov, S., Jongens, R., Lee, J.M.
and Glassey, P.J. 2008c: Using remote sensing for
mapping the effects of natural hazards in New Zealand.
Proceedings of IEEE International Geoscience and
Remote Sensing Symposium, Boston, MA 2, 125154.
Kaab, A. 2002: Monitoring high-mountain terrain
deformation from repeated air- and spaceborne
optical data: examples using digital aerial imagery and
ASTER data. ISPRS Journal of Photogrammetry and
Kaku, K., Held, A.A., Fukui, H. and Arakida, M.
2006: Sentinel Asia initiative for disaster management
support in the Asia-Pacifi c region. Proceedings
of SPIE 6142, 641203, DOI: 10.1117/12.697822.
Kaneko, T. 1998: Thermal observation of the 1986
eruption of Izu-Oshima volcano (Japan) using
Landsat TM data. Advances in Space Research 21,
51720.
2004 Sumatra-Andaman earthquake. Remote
Sensing of Environment 107, 25663.
Geudtner, D., Winter, R. and Vachon, P. 1996:
Flood monitoring using ERS-1 SAR interferometry
coherence maps. Proceedings of IEEE International
Geoscience and Remote Sensing Symposium 2,
96668.
Giglio, L. and Kendall, J.D. 2001: Application of
the Dozier retrieval to wildfi re characterization: a
sensitivity analysis. Remote Sensing of Environment
77, 3449.
Gillespie, T.W., Chu, J., Frankenberg, E. and
Thomas, D. 2007: Assessment and prediction of
natural hazards from satellite imagery. Progress in
Physical Geography 31, 45970.
Gimeno, M. and San-Miguel-Ayanz, J. 2004:

Evaluation of Radarsat-1 data for identifi cation of
burnt areas in Southern Europe. Remote Sensing of
Environment 92, 37075.
Gu, Y.X., Rose, W.I., Schneider, D.J., Bluth,
G.J.S. and Watson, I.M. 2005: Advantageous
GOES IR results for ash mapping at high latitudes:
Cleveland eruptions 2001. Geophysical Research
Letters 32, L02305, DOI: 02310.01029/02004GL0
21651.
Guillaso, S., Ferro-Famil, L., Reigber, A. and
Pottier, E. 2005: Building characterization using
L-band polarimetric interferometric SAR data. IEEE
Geoscience and Remote Sensing Letters 2, 34751.
Hanssen, R.F. 2001: Radar interferometry: data
interpretation and error analysis. Dordrecht: Kluwer,
308 pp.
Harris, A.J.L., Flynn, L.P., Matias, O., Rose,
W.I. and Cornejo, J. 2004: The evolution of an
active silicic lava fl ow fi eld: an ETM+ perspective.
Journal of Volcanology and Geothermal Research 135,
14768.
Harris, A.J.L., Vallance, J.W., Kimberly, P.,
Rose, W.I., Matias, O., Bunzendahl, E.,
Flynn, L.P. and Garbeil, H. 2006: Downstream
aggradation owing to lava dome extrusion and
rainfall runoff at Volcan Santiaguito, Guatemala. In
Rose, W.I., Bluth, G.J.S., Carr, M.J., Ewert, J.W.,
Patino, L.C. and Vallance, J.W., editors, Volcanic
hazards in Central America, Geological Society of
America, Special Paper 412, 85104.
Hervas, J., Barredo, J.I., Rosin, P.L., Pasuto, A.,
Mantovani, F. and Silvano, S. 2003: Monitoring
landslides from optical remotely sensed imagery: the
case history of Tessina landslide, Italy. Geomorphology
54, 6375.
Hooper, A., Zebker, H., Segail, P. and Kampes,
B. 2004: A new method for measuring deformation
on volcanoes and other natural terrains using InSAR
persistent scatterers. Geophysical Research Letters 31,
L23611, DOI: 23610.21029/22004GL021737.
Hoy, E.E. 2007: Evaluating the potential of the
normalized burn ratio and other spectral indices for
205
Kaneko, T. and Wooster, M.J. 1999: Landsat infrared
analysis of fumarole activity at Unzen Volcano:
time-series comparison with gas and magma fl uxes.
5764.
2005: Satellite thermal analysis of the 1986 IzuOshima lava flows. Journal of Volcanology and
Geothermal Research 148, 35571.
Kerle, N., Froger, J.L., Oppenheimer, C. and
de Vries, B.V. 2003: Remote sensing of the 1998
mudfl ow at Casita volcano, Nicaragua. International
Kumar, K.V., Martha, T.R. and Roy, P.S. 2006:
Mapping damage in the Jammu and Kashmir caused
by 8 October 2005 M-w 7.3 earthquake from the
Cartosat-1 and Resourcesat-1 imagery. International
Kwoun, O.I., Lu, Z., Neal, K. and Wicks, C. 2006:
Quiescent deformation of the Aniakchak Caldera,
Alaska, mapped by InSAR. Geology 34, 58.
Lewis, A.J., Henderson, F.M. and Holcomb,
D.W. 1998: Radar fundamentals: the geoscience
perspective. In Henderson, F.M. and Lewis, A.J.,
editors, Principles and applications of imaging radar,

New York: Wiley, 13181.
Lombardo, V. and Buongiorno, M.F. 2006: Lava
fl ow thermal analysis using three infrared bands of
remote-sensing imagery: a study case from Mount
Etna 2001 eruption. Remote Sensing of Environment
101, 14149.
Lymburner, L., Mueller, N. and Islam, A. 2008:
Assessing the accuracy of fl ood mapping techniques
using coincident MODIS and Landsat TM scenes. 14th
Australasian Remote Sensing and Photogrammetry
Conference, Darwin, electronic proceedings.
MacDonald, H.C., Waite, W.P. and Demarcke,
J.S. 1980: Use of Seasat satellite radar imagery
for the detection of standing water beneath forest
vegetation. Proceedings of the American Society of
Photogrammetry Annual Technical Meeting RS-3(B),
113.
Massonnet, D. and Feigl, K. 1998: Radar interferometry
and its application to changes in the Earth surface.
Reviews of Geophysics 36, 441500.
Matsuoka, M. and Yamazaki, F. 2005: Building
damage mapping of the 2003 Bam Iran Earthquake
using Envisat/ASAR intensity imagery. Earthquake
Spectra 21, S285, DOI: 10.1193/1191.2101027.
Mouginis-Mark, P., Garbeil, H. and Flament, P.
1994: Effects of viewing geometry on AVHRR observations
of volcanic thermal anomalies. Remote
Sensing of Environment 48, 5160.
Nazarenko, D.M., Martenson, D., Rossignol, S.
and Staples, G. 1995: RADARSAT image characteristics
and application requirements. Record
of the IEEE 1995 International Radar Conference,
35155, DOI: 10.1109/RADAR.1995.522571.
Nichol, J. and Wong, M.S. 2005a: Detection and
interpretation of landslides using satellite images.
Land Degradation and Development 16, 24355.
2005b: Satellite remote sensing for detailed landslide
inventories using change detection and image fusion.
International Journal of Remote Sensing 26, 191326.
Nichol, J.E., Shaker, A. and Wong, M.S. 2006:
Application of high-resolution stereo satellite
images to detailed landslide hazard assessment.
Geomorphology 76, 6875.
Oberstadler, R., Honsch, H. and Hutch, D. 1997:
Assessment of the mapping capabilities of ERS1 SAR data for flood mapping: a case study in
Germany. Hydrological Processes 11, 141525.
Oppenheimer, C. 1996: Crater lake heat losses estimated
by remote sensing. Geophysical Research
Letters 23, 179396.
Ormsby, J.P., Blanchard, B.J. and Blanchard, A.J.
1985: Detection of lowland flooding using active
microwave systems. Photogrammetric Engineering
and Remote Sensing 51, 31728.
Ostir, K., Veljanovski, T., Podobnikar, T. and
Stancic, Z. 2003: Application of satellite remote
sensing in natural hazard management: the Mount
Mangart landslide case study. International Journal
of Remote Sensing 24, 39834002.
Ouzounov, D., Bryant, N., Logan, T., Pulinets,
S. and Taylor, P. 2006: Satellite thermal IR phenomena
associated with some of the major earthquakes
in 19992003. Physics and Chemistry of the
Earth 31, 15463.
Pairman, D., Belliss, S.E. and McNeill, S.J. 1997:
Terrain influences on SAR backscatter around
Mt. Taranaki, New Zealand. IEEE Transactions on
Geoscience and Remote Sensing 35, 92432.
Patrick, M.R., Dehn, J. and Dean, K. 2005:

Numerical modelling of lava fl ow cooling applied to
the 1997 Okmok eruption: comparison with advanced
very high resolution radiometer thermal imagery.
Journal of Geophysical Research Solid Earth 110,
B02210, DOI: 02210.01029/02003JB002538.
Pieri, D. and Abrams, M. 2004: ASTER watches
the worlds volcanoes: a new paradigm for volcanological
observations from orbit. Journal of Volcanology
and Geothermal Research 135, 1328.
2005: ASTER observations of thermal anomalies
preceding the April 2003 eruption of Chikurachki
volcano, Kurile Islands, Russia. Remote Sensing of
Prata, A.J. 1989: Infrared radiative transfer calculations
for volcanic ash clouds. Geophysical Research Letters
16, 129396.
Pugnaghi, S., Gangale, G., Corradini, S. and
Buongiorno, M.F. 2006: Mt Etna sulfur dioxide fl ux
monitoring using ASTER-TIR data and atmospheric
observations. Journal of Volcanology and Geothermal
Research 7490.

Ranson, K., Kovacs, K., Sun, G. and Kharuk, V.
2003: Disturbance recognition in the boreal forest
using radar and Landsat-7. Canadian Journal of
Rau, J.Y., Chen, L.C., Liu, J.K. and Wu, T.H.
2007: Dynamics monitoring and disaster assessment
for watershed management using time-series
satellite images. IEEE Transactions on Geoscience and
Realmuto, V.J., Sutton, A.J. and Elias, T. 1997:
Multispectral thermal infrared mapping of sulfur
dioxide plumes: a case study from the East Rift Zone
of Kilauea Volcano, Hawaii. Journal of Geophysical
Research Solid Earth 102, 1505772.
Rosen, P., Hensley, P., Joughin, I., Li, F., Madsen,
S., Rodriguez, E. and Goldstein, R. 2000:
Synthetic aperture radar interferometry. Proceedings
of the IEEE 88, 33382.
Rosin, P.L. and Hervas, J. 2005: Remote sensing
image thresholding methods for determining landslide
activity. International Journal of Remote Sensing
26, 107592.
Rothery, D.A., Coppola, D. and Saunders, C. 2005:
Analysis of volcanic activity patterns using MODIS
thermal alerts. Bulletin of Volcanology 67, 53956.
Rott, H. and Nagler, T. 2006: The contribution
of radar interferometry to the assessment of
landslide hazards. Advances in Space Research 37,
71019.
Roy, D.P. and Landman, T. 2005: Characterizing
the surface heterogeneity of fi re effects using multitemporal
refl ective wavelength data. International
Sandholt, I., Nyborg, L., Fog, B., Lo, M., Bocoum, O.
and Rasmussen, K. 2003: Remote sensing
techniques for fl ood monitoring in the Senegal River
Valley. Danish Journal of Geography 103, 7181.
Sertel, E., Kaya, S. and Curran, P.J. 2007: Use of
semivariograms to identify earthquake damage in
an urban area. IEEE Transactions on Geoscience and
Siegert, F., Kuntz, S., Streck, C. and Bergbauer,
B. 1995: Land use planning and monitoring in
Indonesia using ERS-1 RADAR data. Proceedings
of International Conference on Remote Sensing and
GIS for Environmental Resources Management The

Indonesian-European Experience, Jakarta, Indonesia,
11017.
Singh, A. 1989: Digital change detection techniques
using remotely-sensed data. International Journal of
Singhroy, V. 1995: SAR integrated techniques for
geohazard assessment. Advances in Space Research
15, 6778.
Singhroy, V. and Molch, K. 2004: Characterizing
and monitoring rockslides from SAR techniques.
Advances in Space Research 33, 29095.
Singhroy, V., Mattar, K.E. and Gray, A.L.
1998: Landslide characterization in Canada using
interferometric SAR and combined SAR and TM
images. Advances in Space Research 21, 46576.
Stramondo, S., Bignami, C., Chini, M., Pierdicca,
N. and Tertulliani, A. 2006: Satellite radar and
optical remote sensing for earthquake damage detection:
results from different case studies. International
Takano, T. and Maeda, T. 2009: Experiment and
theoretical study of earthquake detection capability
by means of microwave passive sensors on a
satellite. IEEE Geoscience and Remote Sensing Letters
6, 10711.
Terunuma, T., Nishida, K., Amada, T.,
Mizuyama, T., Sato, I. and Urai, M. 2005:
Detection of traces of pyroclastic fl ows and lahars
with satellite synthetic aperture radars. International
Thankappan, M., Reddy, S., Ackerly, S. and
Forghani, A. 2008: Sentinel fl oods prototype: a
system for rapid dissemination of satellite derived
information on fl ooding. 14th Australasian Remote
Sensing and Photogrammetry Conference, Darwin,
electronic proceedings.
Torres, R., Mouginis-Mark, P., Self, S., Garbeil,
H., Kallianpur, K. and Quiambao, R. 2004:
Monitoring the evolution of the Pasig-Potrero alluvial
fan, Pinatubo Volcano, using a decade of remote
sensing data. Journal of Volcanology and Geothermal
Research 138, 37192.
Tralli, D.M., Blom, R.G., Zlotnicki, V., Donnellan,
A. and Evans, D.L. 2005: Satellite remote sensing
of earthquake, volcano, fl ood, landslide and coastal
inundation hazards. ISPRS Journal of Photogrammetry
and Remote Sensing 59, 18598.
Trigg, S.N., Roy, D.P., and Flasse, S. 2005: An in
situ study of the effects of surface anisotropy on the
remote sensing of burned savannah. International
Tsutsui, K., Rokugawa, S., Nakagawa, H.,
Miyazaki, S., Cheng, C.T., Shiraishi, T. and
Yang, S.D. 2007: Detection and volume estimation
of large-scale landslides based on elevation-change
analysis using DEMs extracted from high-resolution
satellite stereo imagery. IEEE Transactions on
Geoscience and Remote Sensing 45, 168196.
Tupper, A., Carn, S., Davey, J., Kamada, Y., Potts,
R., Prata, F. and Tokuno, M. 2004: An evaluation
of volcanic cloud detection techniques during recent
significant eruptions in the western Ring of Fire.
Remote Sensing of Environment 91, 2746.
Voigt, S., Kemper, T., Riedlinger, T., Kiefl , R.,
Scholte, K. and Mehl, H. 2007: Satellite image
analysis for disaster and crisis-management support.
IEEE Transactions on Geoscience and Remote Sensing
45, 152028.
207
Waite, W.P. and MacDonald, H.C. 1971: Vegetation
penetration with K-band radars. IEEE Transactions
on Geoscience and Remote Sensing GE-9, 14755.
Walker, R.T. 2006: A remote sensing study of active
folding and faulting in southern Kerman province,
SE Iran. Journal of Structural Geology 28, 65468.
Walker, R.T., Nissen, E., Molor, E. and Bayasgalan,
A. 2007: Reinterpretation of the active faulting in
central Mongolia. Geology 35, 75962.
Whitworth, M.C.Z., Giles, D.P. and Murphy, W.
2005: Airborne remote sensing for landslide hazard
assessment: a case study on the Jurassic escarpment
slopes of Worcestershire, UK. Quarterly Journal of
Engineering Geology and Hydrogeology 38, 285300.
Wright, R. and Flynn, L.P. 2003: On the retrieval of
lava-fl ow surface temperatures from infrared satellite
data. Geology 31, 89396.
Wright, R., Carn, S. and Flynn, L.P. 2005: A
satellite chronology of the MayJune 2003 eruption
of Anatahan volcano. Journal of Volcanology and
Geothermal Research 146, 10216.
Wright, R., Flynn, L.P., Garbeil, H., Harris,
A.J.L. and Pilger, E. 2004: MODVOLC: nearrealtime thermal monitoring of global volcanism.
2949.
Wilson &Wilson's
COMPREHENSIVE ANALYTICAL CHEMISTRY
VOLUME XLII
NON-DESTRUCTIVE MICROANALYSIS OF
CULTURAL HERITAGE MATERIALS
Edited by
K. J'ANSSENS
R. VAN GRIEKEN
University ofAntwerp
Department of Chemistry
Universiteitsplein, I
B-2610 Antwerp
Belgium
~~
a -c tir_ t
2004
ELSEVIER
AMSTERDAM -- BOSTON - HEIDELBERG -- LONDON - NEW YORK - OXFORD - PARIS
SAN DIEGO - SAN FRANCISCO - SINGAPORE - SYDNEY- TOKYO
Contents
Contributors to Volume XLII . . . . . . . . . . . . . . . . . . . . . vi
Volumes in the Series . . . . . . . . . . . . . . . . . . . . . . . . . xi
Series Editor's Preface . . . . . . . . . . . , , . . . . . . . . . . . xxv
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxvii
Chapter 1. Introduction and overview . . . . . . . . . . . . . . . . . 1
K. Janssens and R. Van Grieken
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Overview of the analytical reference section . . . . . . . . 5
1.3 Overview of the case studies section . . . . . . . . . . . . 7
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
PART I. ANALYTICAL REFERENCE SECTION
Chapter 2. UV-, IR- and X-ray imaging . . . . . . . . . . . . . . . . 15

Franz Mairinger
2 .1 Scientific investigations of works of arts and crafts . . . . 15
2 .2 Application of electromagnetic radiation for the
examination of cultural heritage objects . . . . . . . . . . 16
2 .3 Instrumental basis . . . . . . . . . . . . . . . . . . . . . 16
2.3.1 Light and radiation sources . . . . . . . . . . . . 17
2.3.2 Imaging . . . . . . . . . . . . . . . . . . . . . . 25
2.3.3 Sensor systems . . . . . . . . . . . . . . . . . . . 33
2.3.4 Sensor subsystems . . . . . . . . . . . . . . . . . 41
2.4 Surface examinations . . . . . . . . . . . . . . . . . . . 44
2.4.1 Surface examinations with ultraviolet radiation . . 45
2.4.2 Instrumental techniques for UVfluorescence
photography . . . . . . . . . . . . . . . . . . . . 46
2.4.3 Instrumental techniques for reflected UV
photography . . . . . . . . . . . . . . . . . . . . 46
2.4.4 Application ofUVfluorescence photography . . . . 47
2.4.5 Application ofUV photography . . . . . . . . . . 48
2.5 Depth examinations . . . . . . . . . . . . . . . . . . . . 49
2.5.1 Depth examinations with infrared radiation . . . . 50
2.5.2 Depth examinations with X-rays and gamma-rays . 54
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Contents
Chapter 3. Electron microscopy and its role in cultural

heritage studies . . . . . . . . . . . . . . . . . . . . . . . . . . 73
A. Adriaens andM.G. Dowsett

3.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 73
3 .1.1 Why use electron microscopy? . . . . . . . . . . . 73
3.1 .2 Imaging with electrons , . . . . . . . . . . . . . . 73
3,1 .3 "Varieties ofelectron microscopy . . . . . . . . . . 75
3 .1.4 Recent developments in commercial SEM . . . . . 77
3.2 The interaction of electrons with
a solid-contrast mechanisms . . . . . . . . . . . . . . . 79
3,2.1 Scattering . . . . . . . . . . . . . . . . . . . . . 80
3.2.2 Secondary electron emission . , . . . . . . . . . . 81
3.2.3 Backscattered electrons . . . . . . . . . . . . . . 82
3.2.4 Cathodoluminescence . . . . . . . . . . . . . . . 83
3 .2.5 Core-level excitation and X-ray or
Auger emission . . . . . . . . . . . . . . . . . . . 84
3.2.6 Electron energy loss spectroscopy . . . . . . . . . 86
3.2.7 Diffraction in TEM . . . . . . . . . . . . . . . . . 88
3.2.8 Image contrast in TEM . . . . . . . . . . . . . . 91
3.3 Components and optics of electron
microscopes . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.3.1 Basic optics . . . . . . . . . . . . . . . . . . . . . 92
3.3.2 The electron gun . . . . . . . . . . . . . . . . . . 92

3,3.3 Focusing an electron beam . . . . . . . . . . . . . 93
3.3.4 Newtonian lens model . . . . . . . . . . . . . . . 95
3 .3.5 The magnetic prism . . . . . . . . . . . . . . . . 97
3,3.6 Detectors . . . . . . . . . . . . . . . . . . . . . . 98
3.4 Sample preparation techniques . , . . . . . . . . . . . . 102
3,5 Origin/ provenance studies . . . . . . . . . . . . . . . . . 104
3.5.1 Ceramics . . . . . . . . . . . . . . . . . . . . . . 104
3.5.2 Glass . . . . . . . . . . . . . . . . . . . . . . . . 105
3.6 Technology and techniques of manufacture . . , . . . . . 106
3.6.1. Seals . . . . . . . . . . . . . . . . . . . . . . . . 106
3.6.2 Ceramics . . , . . . . . . . . . . . . . . . . . . . 107
3.6.3 Glass . . . . . . . . . . . . . . . . . . . . . . . . 113
3.6.4 Metals . . . . . . . . . . . . . . . . . . . . . . . 114
3 .7 Use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.8 Degradation processes, corrosion and
weathering . . . . . . . . . . . . . . . . . . . . . . . . . 118
3.8.1 Metals . . . . . . . . . . . . . . . . . . . . . . . 119
Contents
3 .8.2 Glass . . . . . . . . . . . . . . . . . . . . . . . . 119
3 .8.3 Ceramics . . . . . . . . . . . . . . . . . . . . . . 120
3 .9 Authenticity and authentication . . . . . . . . . . . . . . 121
3 .10 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . 123
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . 124
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Chapter 4. X-ray based methods of analysis . . . . . . . . . . . . . 129
K. Janssens
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 129
4.2 Basic principles . . . . . . . . . . . . . . . . . . . . . . . 131
4.2.1 X-ray wavelength and energy scales . . . . . . . . 131
4.2.2 Interaction of X-rays with matter . . . . . . . . . 131
4.2.3 The photoelectric effect; X-ray
fluorescence . . . . . . . . . . . . . . . . . . . . . 133
4.2.4 Scattering and diffraction . . . . . . . . . . . . . 137
4.2.5 X-ray absorption fine structure and
spectroscopy . . . . . . . . . . . . . . . . . . . . 143
4.3 Instrumentation for X-ray investigations . . . . . . . . . 147
4.3.1. X-ray sources . . . . . . . . . . . . . . . . . . . . 148
4.3 .2 X-ray detectors . . . . . . . . . . . . . . . . . . . 155
4.3 .3 X-ray fluorescence instrumentation . . . . . . . . 164
4.3 .4 XRD instrumentation . . . . . . . . . . . . . . . 186
4.3 .5 XAS instrumentation at SR beamlines . . . . . . . 193
4.4 A survey of applications of X-ray methods in the cultural
heritage sector . . . . . . . . . . . . . . . . . . . . . . . 194
4.4.1 Compositional analysis of historic glass . . . . . . 195
4.4.2 Pigments . . . . . . . . . . . . . . . . . . . . . . 195
4.4.3 Lustre ware . . . . . . . . . . . . . . . . . . . . 200
4.4.4 Metallic artefacts . . . . . . . . . . . . . . . . . . 202

4.4.5 Analysis of graphic documents . . . . . . . . . . . 206
4.4.6 Mn oxidation in odontolites . . . . . . . . . . . . 210
4.4.7 Therapeutic and cosmetical chemicals
of Ancient Egypt . . . . . . . . . . . . . . . . . . 211
4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 215
A4.1 Figures-of-merit for XRF spectrometers . . . . . . . . . . 217
A4.1.1 Analytical sensitivity . . . . . . . . . . . . . . . . 217
A4.1.2 Detection and determination limits . . . . . . . . 217
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
Chapter 5. Ion beam microanalysis . . . . . . . . . . . . . . . . . . 227
T. Calligaro, J.-C. Dran and J. Salomon

5.1 Historical background and motivation . . . . . . . . . . . 227
5.2 Fundamentals of ion beam analysis . . . . . . . . . . . . 228
5 .2.1 Interaction of radiations with matter . . . . . . . 230
5 .2.2 Particle-induced X-ray emission . . . . . . . . . . 237
5 .2.3 Elastic scattering of particles . . . . . . . . . . . 244
5.2.4 Nuclear reaction analysis . . . . . . . . . . . . . 246
5 .3 Specific arrangements for the study of art and
archaeological objects . . . . . . . . . . . . . . . . . . . 251
5.3.1 External beams . . . . . . . . . . . . . . . . . . . 251
5.3.2 Nuclear microprobes . . . . . . . . . . . . . . . . 253
5.3.3 Micro and macro-imaging . . . . . . . . . . . . . 255
5.3 .4 Portable systems . . . . . . . . . . . . . . . . . . 255
5.4 Applications in the field of art and archaeology . . . . . . 257
5 .4.1 Materials' identification . . . . . . . . . . . . . . 257
5.4.2 Provenance of the materials . . . . . . . . . . . . 263
5.4.3 Alteration phenomena . . . . . . . . . . . . . . . 266
5.4.4 Authentication and relative dating . . . . . . . . . 267
5.5 Survey of worldwide IBA activity in the field of
cultural heritage . . . . . . . . . . . . . . . . . . . . . . 268
5.6 Conclusion and future prospects . . . . . . . . . . . . . . 268
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271.
Chapter 6. X-ray photoelectron and Auger electron spectroscopy . . . 277
Annick Hubin and Herman Terryn
6.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 277
6.2 The basic concepts of XPS and AES . . . . . . . . . . . . 279
6.2.1 Principle of X-ray photoelectron spectroscopy . . . 279
6.2.2 Principle of Auger electron spectroscopy . . . . . . 282
6.3 XPS and AES instruments . . . . . . . . . . . . . . . . . 286
6.3.1 General set-up . . . . . . . . . . . . . . . . . . . 286
6.3.2 The vacuum system . . . . . . . . . . . . . . . . 287
6.3.3 The X-ray source for XPS . . . . . . . . . . . . . 289
6.3.4 The electron gun for AES . . . . . . . . . . . . . 290
6.3.5 Detection of electron energy . . . . . . . . . . . . 291
6.3.6 The ion gun . . . . . . . . . . . . . . . . . . . . 294
6.3.7 The sample holder and stage . . . . . . . . . . . . 295

6.4 Sample requirements . . . . . . . . . . . . . . . . . . . 296
Contents
Contents
6.5 Information in XPS and AES spectra . . . . . . . . . . . 297

6.5.1 Surface analysis . . . . . . . . . . . . . . . . . . 297
6.5.2 Qualitative analysis . . . . . . . . . . . . . . . . 297
6.5 .3 Quantitative analysis . . . . . . . . . . . . . . . 298
6.5.4 Chemical analysis . . . . . . . . . . . . . . . . . 301
6.5.5 In-depth analysis . . . . . . . . . . . . . . . . . 303
6.5.6 Data analysis . . . . . . . . . . . . . . . . . . . 306
6.5.7 Imaging . . . . . . . . . . . . . . . . . . . . . . 306
6.6 Comparison of XPS, AES and other surface
analytical techniques . . . . . . . . . . . . . . . . . . . . 308
6.7 XPS and AES for chemical analysis of
cultural heritage materials . . . . . . . . . . . . . . . . . 308
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
Chapter 7. Laser ablation inductively coupled plasma mass
spectrometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
Teresa E. Jeffries
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 313
7.2 The inductively coupled plasma mass spectrometer . . . . 314
7.2.1 Historical account . . . . . . . . . . . . . . . . . 314
7.2.2 Operational rationale . . . . . . . . . . . . . . . . 315
7.2.3 The inductively coupled plasma . . . . . . . . . . 317
7.2.4 The plasma sampling interface . . . . . . . . . . 321
7.2.5 Ion focusing . . . . . . . . . . . . . . . . . . . . 323
7.2 .6 Quadrupole mass analyser . . . . . . . . . . . . . 323
7.2 .7 The vacuum system . . . . . . . . . . . . . . . . 327
7.2 .8 Ion detection and signal handling . . . . . . . . . 329
7 .3 Laser ablation : essential components . . . . . . . . . . . 329
7.3 .1 Development of the laser . . . . . . . . . . . . . . 329
7.3.2 The association of lasers with ICP-MS . . . . . . . 331
7.3.3 Stimulated emission . . . . . . . . . . . . . . . . 333
7.3.4 Nd:YAG laser (resonator) cavity . . . . . . . . . . 334
7.3.5 Harmonic generation . . . . . . . . . . . . . . . . 337
7.3.6 Harmonic separation . . . . . . . . . . . . . . . . 338
7.3.7 Energy attenuation. and control . . . . . . . . . . . 340
7.3 .8 Beam delivery and viewing optics . . . . . . . . . 341
7.3 .9 Ablation cell and sample transport , . . . . . . . 342
7.4 Analytical concepts and factors affecting analysis . . . . . 343
7.4.1 Why use the technique? . . . . . . . . . . . . . . 343
xvu
Contents
7 .4.2 Sample preparation and mounting . . . . . . . . . 344
7 .4.3 Analysis of transient signals . . . . . . . . . . . . 345
7.4.4 Factors affecting analysis . . . . . . . . . . . . . 347
7.4 .5 Optimization and calibration . . . . . . . . . . . . 350
7.4.6 Figures of merit and analytical performance
targets . . . . . . . . . . . . . . . . . . . . . . . 352
7.5 Continuing developments and final remarks . . . . . . . . 355
7.5.1. Continuing developments . . . . . . . . . . . . . 355
7.5.2 Final remarks . . . . . . . . . . . . . . . . . . . 356
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
Chapter 8. Infrared, Raman microscopy and fibre-optic
Raman spectroscopy (FORS) . . . . . . . . . . . . . . . . . . . 359
Howell G.M. Edwards and Dalva L.A. de Faria
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 359
8.2 Comparison of the potential use of IR and Raman
spectroscopies for the non-destructive analysis
of art works . . . . . . . . . . . . . . . . . . . . . . . . 360
8.3 Some theoretical aspects of IR and Raman
spectroscopies . . . . . . . . . . . . . . . . . . . . . . . 367
8.4 Instrumentation . . . . . . . . . . . . . . . . . . . . . . 373
8.5 Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . 387.
8 .6 Resonance Raman . . . . . . . . . . . . . . . . . . . . . 387
8 .7 SERS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
8 .8 Intensity measurements in Raman scattering . . . . . . . 390
8 .9 Raman spectroscopy with fibre optics . . . . . . . . . . . 392
8.9.1 Sampling considerations . . . . . . . . . . . . . . 392
8.9.2 Probe design . . . . . . . . . . . . . . . . . . . . 392
8.9.3 Probe background . . . . . . . . . . . . . . . . . 393
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
Chapter 9. Secondary ion mass spectrometry. Application to
archaeology and art objects . . . . . . . . . . . . . . . . . . . . 397
Evelyne Darque-Ceretti and Marc Aucouturier
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 397
9.2 Principles and equipment . . . . . . . . . . . . . . . . . 399
9.2.1 Principles . . . . . . . . . . . . . . . . . . . . . . 399
9.2.2 Equipment and choice of analytical parameters . . 47.0
9.3 Analysis procedures . . . . . . . . . . . . . . . . . . . . 420
9.3.1 Elemental identification, sensitivity , . . . . . . . 420
9.3.2 Quantitative analysis . . . . . . . . . . . . . . . 425
9.3.3 In-depth analysis and depth resolution . . . . . . 429
9.3.4 Surface analysis . .
-'---. . . 435
9.3.5 Imaging, lateral resolution . . . . . . . . . . . . . 438
9.3 .6 Chemical compound analysis and distribution . . . 439
9.4 Examples of applications for cultural heritage . . . . . . . 440
9.4.1 Dating and/or provenance studies based
on isotopic analysis . . . . . . . . . . . . . . . . 440
9.4.2 Dating (not based on isotopic analysis) . . . . . . 444
9.4.3 Provenance studies not based on isotopic
analysis . . . . . . . . . . . . . . . . . . . . . . 447
9 .4.4 Surface layer analysis on artefacts . . . . . . . . 448
9 .4.5 Interface studies on coated layers . . . . . . . . . 453
9 .4.6 ToF-SIMS applications . . . . . . . . . . . . . . . 455

9.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 458
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
PART II. CASE STUDIES SECTION
Contents
Chapter 10. The non-destructive investigation of copper

alloy patinas . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
David A. Scott
10.1 A brief historical account . . . . . . . . . . . . . . . . . 465
10.2 Optical examination . . . . . . . . . . . . . . . . . . . . 467
10.3 Environmental scanning electron microscopy . . . . . . . 470
10.4 X-ray fluorescence analysis . . . . . . . . . . . . . . . . 474
10.5 Scanning X-ray fluorescence microanalysis . . . . . . . . 477
10.6 XRD analysis . . . . . . . . . . . . . . . . . . . . . . . . 480
10.7 FTIR spectroscopy . . . . . . . . . . . . . . . . . . . . . 485
10.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . 488
Aclmowledgements . . . . . . . . . . . . . . . . . . . . . . . . 490
References . . . . . . . . . . . . . . . . . . . . . . . . . . 490
Chapter 11 . Precious metals artefacts . . . . . . . . . . . . . . . . . 493
G. Demortier
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 493
11.2 Non-destructive analysis of gold jewellery items . . . . . . 493
11.2.1 Contribution of atomic and nuclear
(but non-radioactive) methods to the analysis
of ancient gold jewellery items . . . . . . . . . . . 493
Contents
11.2.2 Illustration of the analytical performances of
non-vacuum PIXE for gold artefacts . . . . . . . . 496
11.3 The soldering of gold . . . . . . . . . . . . . . . . . . . . 498
11 .3 .1 Ancient recipes for gold soldering . . . . . . . . . 498
11.3.2 Iranian goldsmithery from. the 4th century BC . . . 499
11.3.3 Tartesic gold artefacts . . . . . . . . . . . . . . . 502
11.3 .4 Later Iranian goldsmithery . . . . . . . . . . . . . 506
11.3 .5 Preparations of low-melting brazing alloys . . . . . 509
11.3 .6 A new reading of Elder Pliny's Natural History . . 517
11.3 .7 Italian jewellery . . . . . . . . . . . . . . . . . . 520
11.3 .8 Gold artefacts from Slovenia . . . . . . . . . . . . 521
x1.3 .9 The Guarrazar treasure . . . . . . . . . . . . . . 525
11.3 .10 Merovingian and late Byzantine jewellery . . . . . 528
11.4 Pre-Hispanic gold artefacts of Mesoamerica . . . . . . . . 530
11.4.1 Archaeological context . . . . . . . . . . . . . . . 530
11.4.2 A selection of typical artefacts . . . . . . . . . . . 534
11.4.3 Differential PIXE . . . . . . . . . . . . . . . . . . 536
11.4.4 Application to the measurement of the gold
enhancement at the surface of tumbaga . . . . . . 538
11.5 Characterization of complex items . . . . . . . . . . . . . 544
11.5.1 XRF induced by a -y-ray source . . . . . . . . . . . 544
11.5.2 Gamma-ray transmission measurements . . . . . 544
11 .5.3 Study of a composite gold jewellery artefact . . . . 545
11.6 Gold coins . . . . . . . . . . . . . . . . . . . . . . . . . 548

11 .6.1 Fineness measurements of gold coins . . . . . . . 549
11 .6.2 Gold coins from the ancient world . . . . . . . . . 548
11 .6.3 Gold coins from the new world . . . . . . . . . . . 556
11.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . 558
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 560
Chapter 12. Diagnostic methodology for the examination of Byzantine
frescoes and icons. Non-destructive investigation
and pigment identification . . . . . . . . . . . . . . . . . . . . . 565
Sister Daniilia, Sophia Sotiropoulou, Dimitrios Bikiaris,

Christos Salpistis, Georgios Karagiannis and
Yannis Chryssoulakis
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 565
12.2 The Entry of the Mother of God into the Temple . . . . . . 566
12 .2.1 Description . . . . . . . . . . . . . . . . . . . . . 566
Contents
12.2 .2 The preparation of the plaster: materials

and technique . . . . . . . . . . . . . . . . . . . 567
12.2 .3 The drawing . . . . . . . . . . . . . . . . . . . . 568
12.2 .4 Materials and painting techniques . . . . . . . . . 570
12.2.5 Study of the colour palette . . . . . . . . . . . . . 581
12.2.6 Conclusions . . . . . . . . . . . . . . . . . . . . 584
12.3 Mother of God Hodegetria . . . . . . . . . . . . . . . . . 585
12.3.1 Description . . . . . . . . . . . . . . . . . . . . . 585
12.3.2 Construction and state of preservation
of the support . . . . . . . . . . . . . . . . . . . 587
12.3.3 State of preservation of the surface . . . . . . . . 588
12.3.4 The ground . . . . . . . . . . . . . . . . . . . . . 592
12.3.5 The drawing . . . . . . . . . . . . . . . . . . . . 592
12.3.6 Materials and technique of the painting . . . . . . 593
12.3.7 Conclusions . . . . . . . . . . . . . . . . . . . . 601
A12.1 Experimental details . . . . . . . . . . . . . . . . . . . . 602
A12.1.1 Non-destructive analysis . . . . . . . . . . . . . 602
A12.1.2 Micro-sampling analysis . . . . . . . . . . . . . 603
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 604
Chapter 13. The provenance of medieval silver coins: analysis with
EDXRF, SEMIEDX and PIXE . . . . . . . . . . . . . . . . . . . 605
Robert Linke, Manfred Schreiner, Guy Demortier,

Michael Alram and Heinz Winter
13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 605
13.2 The Friesacher Pfennig . . . . . . . . . . . . . . . . . . 606
13.2.1 Introduction . . . . . . . . . . . . . . . . . . . 606
13.2.2 Experimental . . . . . . . . . . . . . . . . . . . 608
13.2.3 Results . . . . . . . . . . . . . . . . . . . . . . . 613
13.3 The Tiroler Kreuzer . . . . . . . . . . . . . . . . . . . . 622
13.3.1 Introduction . . . . . . . . . . . . . . . . . . . . 622
13.3.2 Experimental . . . . . . . . . . . . . . . . . . . 624

13.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 629
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 630
Chapter 14. Pigment identification in illuminated manuscripts . . . 635
Peter Vandenabeele and Luc Moens

14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 635
14.2 Combined method approach . . . . . . . . . . . . . . . . 636
1.4.2.1 Analysis of manuscripts . . . . . . . . . . . . . . 639
14.2.2 Sources of impurities . . . . . . . . . . . . . . . . 652
14.3 Analysis of the manuscripts from the collection
of Raphael De Mercatellis . . . . . . . . . . . . . . . . . 654
14.3.1 Introduction . . . . . . . . . . . . . . . . . . . . 654
14.3.2 Pigment identification with TXRF and MRS , . . . 655
14.3.3 Intra-manuscript comparison ofExpositio
problematum Aristotelis . . . . . . . . . . . . . . 657
14.3.4 Analysis of Decretum Gratiani . . . . . . . . . . . 658
14.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 659
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 660
Chapter 15. Provenance analysis ofglass artefacts . . . . . . . . . . 663
Bernard Gratuze and Koen Janssens

15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 663
15.2 Obsidian, a natural glass used since the Paleolithic era . . 665
15.3 Bronze and Iron age glasses . . . . . . . . . . . . . . . . 610
15.3.1 Neolithic first artificial glassy materials and
the discovery of glass during Bronze Age . . . . . 670
15.3.2 When trade beads reached Europe . . . . . . . . . 671
15 .3 .3 Middle Bronze Age plant ash soda-lime glasses . . 672
15.3 .4 Late Bronze Age mixed soda-potash glasses . . . 674
15.3.5 Iron Age and Antiquity natron soda-lime glasses . 675
15 .3.6 Protchistoric glass trade routes . . . . . . . . . . 678
15 .3.7 Glass chrono-typo-chemical models:
a dating tool? . . . . . . . . . . . . . . . . . . . . 6'78
15.4 Glass trade towards and from Central Asia and
the Indian world during Antiquity . . . . . . . . . . . . . 680
15.5 Carolingian. glass production: some unusual lead
glass composition smoothers . . . . . . . . . . . . . . . . 687
15.6 Late Middle Age recycled glass . . . . . . . . . . . . . . 691
15.7 Glass technology transfer during the 16th-17th century
to and from Antwerp . . . . . . . . . . . . . . . . . . . . 699
15.8 Trade beads : the glass trade internationalization,
during the Post-Medieval period . . . . . . . . . . . . . . 702
15.9 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . 705
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 707
Contents
Chapter 16. Corrosion of historic glass and enamels . . . . . . . . . 713

Manfred Schreiner
16.1 Introduction . . . . . . . . . . . . . , . . . . . . . . . . 713
16.2 The weathering of medieval stained glass . . . . . . . . . 716
16.2.1 SEM investigations of the corrosion phenomena
on naturally weathered medieval glass . . . . . . 719
16.2.2 The determination of hydrogen in the leached
surface layer by SIMS and NRA . . . . . . . . . . 725
16.2.3 Leaching studies of glass with medieval
composition . . . . . . . . . . . . . . . . . . . . 731
16.2.4 IRRAS investigations on leached glass with
medieval composition . . . . . . . . . . . . . . . 735
16.2.5 Weathering phenomena on glass with medieval
composition studied with TM-AFM . . . . . . . . 738
16.3 The degradation of medieval enamels . . . . . . . . . . . 742
16.3.1 SEM investigations of the enamel of the
medieval goblets . . . . . . . . . . . . . . . . . . 745
16.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 750
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 753
Chapter 17. A study of ancient manuscripts exposed to iron-gall
ink corrosion . . . . . . . . . . . . . . . . . . . . . . . . . . . 755
Ewa Bulska and Barbara Wagner
17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 755
17.1.1 Iron-gall ink . . . . . . . . . . . . . . . . . . . . 756
17.1.2 Iron-gall ink corrosion . . . . . . . . . . . . . . 758
17.1.3 Investigated artefacts , . . . . . . . . . . . . . . 760
17.2 Analytical methods . . . . . . . . . . . . . . . . . . . . 763
17.2.1 Inspection by scanning electron microscopy . . . . 766
17.2.2 Compositional analysis by X-ray fluorescence
spectrometry , . . . . . . . . . . . . . . . . . . . 766
17.2.3 Electron probe micro-analysis . . . . . . . . . . . 768
17.2.4 Laser ablation inductively coupled plasma
mass spectrometry . . . . . . . . . . . , . . . . . 770
17.2.5 Elemental analysis by inductively coupled plasma
mass spectrometry . . . . . . . . . . . . . . . . . 773
17.2.6 Graphite furnace atomic absorption
spectrometry . . . . . . . . . . . . . . . . . . . . 774
17,2.7 Mossbauer spectrometry . . . . . . . . . . . . . . 775
Contents
1.7 .2.8 Investigation ofFe(II)/Fe(III) by X-ray absorption
near edge spectroscopy . , . . . . . . . . . . . . . 778
17.3 Searching for the conservation treatment . . . . . . . . . 780
17.3.1 Reconstitution of manuscript by model samples . . 780
17.3.2 Requirement for conservation treatment . . . . . 782
17.3.3 Investigation of the model samples . . . . . . . . 784
17.4 Concluding comments . . . . . . . . . . . . . . . . . . . 784
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 785
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 789
See discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/41806088
New Methodologies for the Conservation of Cultural

Heritage:
Micellar Solutions, Microemulsions, and Hydroxide
Nanoparticles
ARTICLE in ACCOUNTS OF CHEMICAL RESEARCH MARCH 2010
Impact Factor: 22.32 DOI: 10.1021/ar900193h Source: PubMed
CITATIONS
49
READS
510
4 AUTHORS:
Rodorico Giorgi
University of Florence
SEE PROFILE
Michele Baglioni
SEE PROFILE
Debora Berti
SEE PROFILE
Piero Baglioni
SEE PROFILE
Available from: Piero Baglioni
Retrieved on: 11 December 2015
New Methodologies for the Conservation of

Cultural Heritage: Micellar Solutions,
Microemulsions, and Hydroxide Nanoparticles
RODORICO GIORGI, MICHELE BAGLIONI, DEBORA BERTI, AND
PIERO BAGLIONI*,
Department of Chemistry and CSGI, University of Florence, via della Lastruccia

3 - Sesto Fiorentino, 50019 Florence, Italy
RECEIVED ON JULY 4, 2009
CON SPECTUS
odern civilizations inherited artworks have a powerful impact on society,
from political, sociological, and anthropological points of view, so

the conservation of our Cultural Heritage is fundamental for conveying to
future generations our culture, traditions, and ways of thinking and behaving.
In the conservation of cultural artifacts, scientists intervene in the degradation
of often unique handcrafts, resulting from a delicate balance of
aging, unpredicted events, environmental conditions, and sometimes incorrect
previous restoration treatments, the details of which are often not precisely
known.
Nanoscience and nanotechnology are revolutionizing materials science
in a pervasive way, in a manner similar to polymer chemistrys revolution
of materials science over the preceding century. The continuous
development of novel nanoparticle-based materials and the study of
physicochemical phenomena at the nanoscale are creating new
approaches to conservation science, leading to new methodologies that
can revert the degradation processes of the works of art, in most cases
restoring them to their original magnificent appearance.
Until recently, serendipity and experiment have been the most frequent
design principles of formulations for either cleaning or consolidation
of works of art. Accordingly, the past has witnessed a number of actively detrimental treatments, such
as the
application of acrylic and vinyl resins to wall paintings, which can irreversibly jeopardize the
appearance (or even the
continued existence) of irreplaceable works of art. Current research activity in conservation science is
largely based
on the paradigm that compatibility of materials is the most important prerequisite for obtaining
excellent and durable
results.
The most advanced current methodologies are (i) the use of water-based micelles and microemulsions
(neat or combined
with gels) for the removal of accidental contaminants and polymers used in past restorations and (ii)
the application
of calcium hydroxide nanoparticles for the consolidation of works of art. In this Account, we highlight
how
conservation science can benefit from the conceptual and the methodological background derived
from both soft (microemulsions
and micelles for cleaning) and hard (nanoparticles for consolidation) nanoscience.
A combination of different nanotechnologies allows todays conservators to provide, in each restoration
step, interventions
respectful of the physicochemical characteristics of the materials used by artists. The palette of
methods
provided by nanoscience is continuously enriching the field, and the development of novel
nanomaterials and the study
of nanoscale physicochemical phenomena will further improve the performance of restoration
formulations and our
comprehension of degradation mechanisms.
Published on the Web 03/05/2010 www.pubs.acs.org/acr Vol. 43, No. 6 June 2010 695-704 ACCOUNTS OF CHEMICAL RESEARCH 695
10.1021/ar900193h 2010 American Chemical Society
Introduction
It is well accepted that modern Conservation Science originated

from the tragic floods that devastated Florence and Venice
in 1966, imposing the search for new methodologies to
restore and conserve the immense Cultural Heritage heavily
damaged by the flood. A scientific method for conservation of
carbonatic materials was proposed for the first time by Enzo
Ferroni at the International ICOM Conference in Amsterdam
(1969).1,2 This new method for the in situ consolidation of
wall paintings was able to restore, in a two-step series of
chemical reactions, the degradation of calcium carbonate, the
principal constituent of wall paintings, drastically damaged by
the flood. The new Science for Conservation has developed in
a fast and healthy way following two main streams: (i) the
analytical characterization of the materials constituting the
works of art, the characterization of the pictorial technique
used by the artists, and the chemical reactions involved in
their degradation; (ii) the search for new scientific methods for
the restoration/conservation, allowing the transfer to future
generations of our Cultural Heritage. The last point will be the
object of this Account, where we will review some of the most
important methodologies developed, mainly using the cultural
framework of soft-condensed matter and nanoscience.
Generally speaking, the restoration of a work of art consists
of two distinct aspects:
(1) cleaning, which is a transient treatment, meant to remove
the materials not originally belonging to the work of art.
(2) consolidation, which is a durable intervention that should
remediate, prevent, or slow down further degradation due
to aging or external agents.
All of the current research activity in Conservation Science
is mainly based on the paradigm that compatibility of materials

is the most important prerequisite in order to obtain excellent
and durable results. In view of this seemingly intuitive, yet
novel, awareness, it is clear that the use of polymers for the
consolidation of inorganic materials should be avoided as
much as possible in the restoration of the original features of
the unaltered material.
Currently this is mainly achieved thanks to the tremendous
advancements witnessed by nanoscience and nanotechnology
that provide theoretical and technical backgrounds to formulate
innovative systems for restoration.3,4
We pioneered the application of nanoscience to Cultural
Heritage conservation, and we devised several methods for
the consolidation of wall paintings, for paper and wood deacidification,
and for the removal of grime, dirt, polymers, and
other organic materials soiling the original artifacts.5-12 In this
Account, we will highlight (as an example of the potential of
these new nanotools) the use of soft microemulsions and
micelles for cleaning and hard nanoparticles for consolidation
of works of art.
Degradation of Polymer Resins Used in

Past Restoration
Reversibility of the restoration treatment has been presented
for decades as a milestone in conservation.13 This principle
has supported several conservators in the usage of polymer
resins, made available by the industry since 1960, which were
considered the panacea for the restoration of works of art.
Consolidation and protection of paintings, stone, wood, paper,
glass, ceramic, and bones was made possible by simple application
of organic solutions, through immersion, injection, or
brushing, of acrylic, vinylic, and silicone-based polymers and
(their) copolymers. Most of these polymers were considered
reversible, that is, the solvent used for the application could
be later used for the removal (if desired) of the applied resins.
The large use of polymers was also favored by the the brilliant
water-white color of the polymers that makes it possible
to secure masses of high light transmission and great
optical clarity - because of the remarkable stability of the polymers
to the action of heat and light, these properties are permanent
and by the many striking and beautiful effects can
be produced by the incorporation of dyes and pigments.14
Unfortunately, the application of polymer coatings induces
severe modifications of the main physicochemical properties
of artifacts, particularly for inorganic matrices as wall
paintings.15,16 Investigations performed on Paraloid B72 (or
Acryloid) (EMA/MA 70:30 w/w; ethyl methacrylate/methyl
acrylate; weight average molecular weight 88 000 g/mol; supplier
Rohm and Haas (USA)), Elvacite 2046 (nBMA/iBMA; n-butyl
methacrylate/i-butyl methacrylate), and Primal AC33 (or
Rhoplex, EA/MMA 60:40 w/w; ethyl acrylate/methyl methacrylate;
viscosity (Pa s) 6, Tg (C) 16; supplier Rohm and Haas
(USA)) showed that water capillarity, water vapor permeability,
and surface wettability of lime mortar change drastically.
The application of polymers leads to a strong increase of surface
hydrophobicity, to a decrease of about 70-85% of the
capillary rise coefficients and 40-50% of the vapor permeability,
and to an increase of contact angle from about 30 to
100.17
When the surface is coated with a polymer, salt crystallization

occurs within the pores of the wall paintings, generating
strong mechanical stress beneath the painted layer
(50-100 m) and leading to the complete disruption of the
pictorial layer in a period of time from about 5 to 30-40
Nanoscience for the Conservation of Cultural Heritage Giorgi et al.
696 ACCOUNTS OF CHEMICAL RESEARCH 695-704 June 2010 Vol. 43, No. 6
years, depending on the environmental conditions. The alteration

of the natural breathing of the surface due to the presence
of the polymer increases consistently the deleterious
mechanical effect of salt crystallization from the interior of the
wall with the complete detachment of the painting from its
support, that is, the surface of the work of art. In addition to
the degradation of the painted layer promoted by the drastic
change of the physicochemical characteristics, polymers themselves
degrade, and upon aging they are subjected to molecular
weight changes due to cross-linking reactions and chain
scissions. The main consequence of cross-linking is the loss of
solubility, which makes the application irreversible.18-20
Photoaging under UV light was examined by some authors to
detect the molecular weight distribution change due to chain scissions
or coupling of macroradicals.18 The overall stability of the
polymers was shown to be influenced by the alkyl side groups,
particularly in the case of paraloid B66 (nBMA; poly(methyl,nbutyl
methacrylate)) and paraloid B67 (iBMA; poly(isobutyl methacrylate)),
where relatively longer ester chains are present. In
these compounds, the oxidation is favored by hydrogen atoms,
either on the methylenes of the n-butyl groups in B66 or on the
tertiary carbon of the isobutyl groups in B67. Polymers containing
n-butyl ester groups undergo fast and extensive cross-linking
together with some chain scissioning.
Paraloid B72 and paraloid B82 (MMA, poly(methyl methacrylate))
show a good stability toward oxidation due to the presence
of the EMA and MMA methacrylic units19 and seem to be
suitable as consolidants and protective agents for applications to
wall paintings and other artifacts. However, it is important to
notice that the composition of commercially available products
contains almost 1% i-BMA, formed during the polymerization
reaction. Even tiny amounts of i-BMA may initiate radical formation
leading to extensive cross-linking reactions, similarly to paraloid
B66 and B67. The difference in paraloid B72 with the other
member of this polymer family is mainly related to the slower
degradation process but is quite common the evidence of
enhanced degradation generated by paraloid B72 (as well as
B66 and B67) used in past restorations. The degradation effects
are particularly evident in locations where the environmental
conditions are severe due to fluctuations of temperature and relative
humidity or in very polluted urban areas where photochemical
smog favors polymer degradation, as demonstrated in a
recent paper by Favaro and co-workers.20
Conservation treatments based on resins have produced
irreversible degradation of most of Mesoamerican paintings.21
Recent investigations in the archeological site of Mayapan
(Mexico, Post-Classic Age 1200-1450 a.C.) have demonstrated
the dramatic conditions of these paintings recently submitted
to consolidation and protection treatments with
mowilith (vinyl acetate/n-butyl acrylate 65:35 w/w copolymer)
resins. A photographic documentation reveals that in less
than 10 years a relevant portion of the original painting was
lost (see Figure 1). This was due to the combination of the
mowilith film with extensive salt crystallization processes that
quickly destroyed the plaster beneath the paint layer.
In addition to the mechanical degradation, the natural aging
of polymer resins generates discoloration and loss of solubility,
due to the increase of the polymer molecular weight. At the end
of the degradation process, these polymers are very difficult or
impossible to remove with conventional neat solvents. Considering
the severe side effects generated by the use of polymers
and the necessity of their removal, we proposed the use of nanostructured
fluids; some examples are highlighted in the following.
Cleaning and Removal of Polymeric

Coatings by Means of Amphiphile-Based
Systems
Historically, the first application of soft matter concepts was

performed in Florence at the end of the 1980s during the restoration
of the Renaissance masterpiece by Masaccio,
FIGURE 1. Mural paintings in Templo de los Nichos Pintados in Mayapan (Yucatan). The pictures show the
degradation that occurred to the
paintings from the last restoration in 1999 due to the mowilith DM5 coating.
Vol. 43, No. 6 June 2010 695-704 ACCOUNTS OF CHEMICAL RESEARCH 697
Masolino, and Lippi in the Brancacci Chapel in Florence

(1984-1990).22 Investigations under UV light of the paintings
revealed the presence of a multitude of wax spots due to
the extinction of votive candles kept close to the wall painting.
The removal of this hydrophobic material from the hydrophilic
fresco painting was obtained using microemulsions.23-26
The microemulsion used in Brancaccis paintings was very
similar to the so-called French microemulsion and was
formed by nanosized dodecane droplets, dispersed in an
aqueous solution of ammonium27 dodecylsulfate (surfactant)
and 1-butanol (cosurfactant).
The oil in water microemulsion ensured low aggressiveness
with respect to the original components of the painted
layer, due to the presence of water as a dispersing medium
that remained in direct contact with the hydrophilic surface of
the wall painting.24
Since this pioneering example, several other surfactantbased
micellar and microemulsive systems have been used by
our group to clean wall paintings and canvases both from accidental
contamination and from polymers that have been
extensively used as consolidants and protectives in past
restorations.
Nanocompartmentalized systems specifically tailored for
the removal of paraloid B72, primal AC33, and mowilith DM5
(vinyl acetate/n-butyl acrylate 65:35 w/w copolymer; viscosity
(Pa s) 3.5, Tg (C) 2; supplier Hoechst, Germany) resins have
been recently formulated by some of us. Both micelles and
swollen micelles work as nanocontainers of solvents that can
dissolve polymers to obtain their complete removal from the
surface and from the porous texture of the work of art. The
detergency capability mainly depends on the very large surface
area of micelles and nanodroplets available for interaction
with the polymer coating. The water-based system
reduces the penetration into the artifact porous matrix that
occurs with organic solvents and the associated polymer redissolution
into the artifact but also reduces the toxicity of the
formulation, offering better and faster performance than pure

organic solvents. Table 1 reports the composition of some of
the most successful formulations used for cleaning.
Oil in water microemulsion made with sodium dodecyl sulfate
(SDS), 1-pentanol (PeOH) as a cosurfactant, and a small
amount of a mixture of p-xylene and nitro diluent (a mixture
of 62% toluene, 15% butyl acetate, 15% ethyl acetate, 6%
n-butyl alcohol, 2% cellosolve acetate) was shown to be effective
for paraloid B72 removal (Classical system; see Table 1).
Polar polymer resins such as poly(vinyl acetate), not soluble in
the oil mixture above-reported, were successfully removed by
using a four-component micellar solution with a high amount
of propylene carbonate (PC), which replaced the nonpolar mixture
of xylene and nitro diluent (Conegliano system). Previous
investigations have shown that PC acts as a cosurfactant
inducing a rod-to-sphere micelle transition reducing the size
and aggregation number of SDS micelles by increasing the
headgroup area.28,29
The Conegliano system could also be used for paraloid
B72 removal since it favors the swelling of paraloid B72 films,
which can be then removed by mechanical action for a complete
cleaning of the surface.
Very often paraloid B72 and mowilith DM5 have been
used in mixtures or sequentially applied to wall paintings. An
ideal system should remove simultaneously both resins. A
major improvement in this direction has been recently
obtained by adding ethyl acetate (EA) to the Conegliano formulation
to obtain the Mayapan micelles, whose composition
is reported in Table 1. The aromatic and aliphatic solvents
of the Classical formulation are replaced by less toxic solvents.
Pure ethyl acetate and propylene carbonate are known to
solubilize paraloid B72. They are partially water-soluble (EA
8% w/w, PC 20% w/w), but a saturated aqueous solution of
EA and PC provides only partial swelling of the polymer. The
addition of SDS and 1-PeOH either to the EA or PC aqueous
solutions provides a drastic improvement in the efficiency of
the removal process, due to the presence of micelles that provide
the large interfacial area necessary for polymer uptake.
It is important to stress here that micellar solutions formed by
SDS or SDS/PeOH alone are practically ineffective in polymer
removal; therefore it is the synergistic action of the presence
of micelles and PC that causes the polymer removal. Even better
performance was obtained by adding simultaneously 8%
EA and 8% PC to aqueous SDS/1-PeOH micelles (Mayapan
system). This system allowed a complete cleaning of the surfaces
treated with paraloid B72 in a few minutes. The Classical
system is very efficient in removing acrylic polymers, but
the polymer removal is quite slow compared with the Mayapan
micelles. This is probably due to the lower amount of
TABLE 1. Composition (% w/w) of Systems Used for Polymer
Removal from Wall Paintingsa
Classical Conegliano Mayapan
component composition component composition component composition
water 86.2 water 69 water 73.3
SDS 3.9 SDS 5.1 SDS 3.7
1-PeOH 6.5 1-PeOH 3.9 1-PeOH 7
p-xylene 1.8 PC 22 PC 8
nitro diluent (ND) 1.6 ethyl acetate (EA) 8
Classical refers to the first o/w microemulsion formulated for paraloid B72
removal; the other formulations are named after the places where they were
a
used for the first time.

organic solvents contained in the former leading to different

microstructural features of the system.
It is clear from the above examples that the design of efficient
formulations for the tailored removal of polymers
requires a deep comprehension of the solubilization mechanisms,
which can be more complex than traditional detergency,
due to the presence of partially water-soluble organic
solvents (cosolvents). To better understand the polymer
removal process, we performed swelling tests of polymer films
deposited in glass vials and detergency tests on paraloid films
on models of wall paintings.
Quasi-elastic light scattering (QELS) is an efficient and noninvasive
tool to characterize size and size distribution of complex
systems in the nanometer domain. Figure 2 shows that
the intensity weighted size distribution of Mayapan micelles is
centered around 4 nm with a relatively broad size distribution,
while the average droplet diameter of the Classical sysFIGURE 2. Size distribution of the Mayapan micellar system
obtained from QELS.
FIGURE 3. Schematic representation (bottom panel) of the differences in the removal mechanisms of paraloid B72
performed by
Conegliano and Mayapan formulations. Conegliano system only swells the polymer layer, while the application of
Mayapan micelles to a
paraloid B72-coated surface results in a liquid-liquid phase separation. SAXS spectra (upper panel) obtained after
the application of
Mayapan micelles to a paraloid B72-coated surface. The more dense phase (green line) still contains micelles, while
the upper phase (red
line) contains the polymer dissolved in a blend of EA/PC/PeOH.
tem is about 17 nm with polydispersity of 25%, typical of a

common o/w microemulsion. The smaller micellar size,
together with the presence of specific organic solvents, results,
for a given surfactant content in a very high interfacial area
that is correlated to the kinetics of the cleaning process.
The Conegliano micelles are characterized by very small
aggregates with a hydrodynamic diameter of about 2 nm28,29
and refer to clusters formed by 7-10 SDS monomers and not
to conventional micelles. For the sake of clarity, we indicated
these clusters as micelles. Mayapan micelles are similar but
present a higher effectiveness toward polymer removal that is
nicely illustrated by the results of a semiquantitative test on
model systems that we used to compare the removal efficiency
of these two systems. Coating the bottom of two vials
with a known amount of paraloid B72 (whose thickness is
comparable to that spread on art handcraft) and putting the
film in contact with Conegliano and Mayapan micellar solutions
(see Figure 3, bottom), we found that the Conegliano
micelles are able to swell the polymer, but the swollen polymer
remains at the bottom of the vial. Mayapan micelles were
also able to swell the polymer but in addition a spontaneous
detaching of the polymer from the glass was observed with a
liquid-liquid phase separation in the micellar phase. This
behavior is ideal for polymer removal from artifacts.
In order to understand this mechanism, both phases have
been investigated with static and dynamic scattering techiniques.
Figure 3 reports the SAXS spectra obtained from the
polymer-rich and the surfactant-rich phases. The less dense
phase contains the polymer and a mixture of H2O, PeOH, PC,

and EA solvents, enriched in the less polar components. The
more dense and abundant phase is still a micellar solution but
is depleted of the organic phase.
It is clear that the mixture of organic solvents, dispersed in
the aqueous medium thanks to the presence of micelles, guarantees
specific interactions with the polymer (see Figure 4)
suggesting a mechanism where the Mayapan micelles work as
a carrier of solvents during the phase separation process that
is the final step of the cleaning procedure.
Figure 5 shows a brick with a 1 cm thick lime plaster layer,
painted with fresco technique, treated with paraloid B72. Visual
FIGURE 4. Schematic representation of the hypothesized
detergency mechanism of the Mayapan (upper panel) and Classical
(lower panel) systems. While in the first case the partially watersoluble
solvents (EA and PC) partition among the aqueous and the
micellar phase, in the second case, we have a traditional
microemulsion where the water-insoluble oil core is dispersed in
water by the presence of a surfactant film. The naked micelles
(SDS/PeOH) do not detatch paraloid B72 from the surfaces. In this
cartoon, the Mayapan micelles act as solvent carriers and quickly
mediate their interaction with the polymer, which chooses the
right amounts of solvents to undergo a swelling process. It is worth
noticing that the micelles of the Mayapan system, after the
interaction with paraloid B72, change about 10% in size, while the
microdroplets of the Classical system get sensibly smaller; for
microemulsions, the kinetics is slower, due to the oil confinement in
the micellar core. After the interaction, the micelles decrease in size
because of the solvent release.
FIGURE 5. Experimental cleaning test on model fresco paintings
treated with paraloid B72: (A) water/SDS/1-PeOH/(8%)EA; (B) water/
PC; (C) Classical system; (D) Conegliano system; (E) water/EA/PC; (F)
Mayapan system.
analysis of this model painting after cleaning showed a complete

recovery of the optical features of the surface in the area
treated with the Mayapan system (see Figure 5F). The Classical
and Conegliano systems gave slightly worse results, as is clearly
observable from the polymer residues still present in the treated
areas as shown in Figure 5, areas C and D, respectively.
As already mentioned, the Mayapan micelles are also
active for mowilith DM5 removal. Figures 6 and 7 report illustrative
examples of a fresco painting treated with the Mayapan
system.
Confinement of Micellar Solutions and

Microemulsions in Chemical Gels
Microemulsions and micellar solutions are usually applied

over a painted surface by using cellulose-pulp compresses or
physical gels soaked with them.30 In many cases, as in the
case of canvas or wood painting cleaning, a more controlled
release of the microemulsion is required to avoid the swelling
of inorganic and organic binding materials by water, that
is, the most abundant component in an oil in water
microemulsion.
The advantage of using gels resides in the easy manipulation,
the possibility of microemulsion and micellar solution
confinement into the gel network, and the possibility to
make them responsive to external stimuli (i.e., pH, magnetic
fields, temperature, light).31-35
Gel structure may result from formation of relatively
weak physical bonds (i.e., hydrogen bonds or van der Waals

interactions) or covalent chemical bonds; both can be
responsible for the cross-linking of particles or polymer
chains to create the gelled state. In the first case, we refer
to a physical gel, in the second to a chemical gel.
FIGURE 6. Mural painting sample treated, on the right side, with mowilith DM5 under glazing light (center), and
SEM pictures of the
uncoated (left) and coated (right) surface.
FIGURE 7. Details of the painting surface after removal of mowilith DM5 coating by using micellar solution
(Mayapan system).
FIGURE 8. Linen canvas glued with mowilith DMC2 for a relining
treatment. In the red box, the appearance of the treated area after
micellar solution application.
Among the large number of polymers that can be used for

gelation, cellulose derivatives, polyacrylic acids, and
polyamines are the most used in the field of conservation science.
Unfortunately, these physical gels present some difficulties
in their removal and may release residues on the surface
of the work of art after the application. Besides, the gels mesh
size is difficult to control during gelation.36,37
Chemical gels are characterized by stronger interactions in
the gel molecular network. This type of gel behaves like a soft
solid, and the mesh size is tunable by controlling the crosslinking
during the polymerization reaction.
Chemical gels, based on an acrylamide/bis-acrylamide network
and PEG or silane cross-linkers, loaded with the detergent
solutions have been recently synthesized and applied to
Conservation.11,38
The microemulsion release or uptake can be modulated by
controlling the gel mesh size or applying a stimulus. In a previous
paper,38 it was shown by SAXS that a microemulsion
retains its structure when loaded into the gel. The uptake/
release capability of the gel was quantified by analysis of
hydration/dehydration curves. Microemulsion-loaded gel was
shown to be effective for the removal of paraloid B72 from
limestone and wall paintings.
Cleaning tests on canvases, previously treated with adhesives
(i.e., mowilith DMC2 (vinylacetate/dibutylmaleate copolymers)
and acrylate/methacrylate copolymers) that are
normally used for the relining of canvas paintings, have been
performed with a chemical gel, tailored in order to have good
water retention. The gel was loaded with the Mayapan system
specifically developed for the swelling and the removal of
acrylate/methacrylate and vinylacetate/butylacrylate copolymers.
After a 2-hour application, the capillary sorption of water
through the ground and paint layer was restricted to the contact
area of gel over the canvas. The swollen polymers were
partly removed by the gel, and the complete cleaning was
accomplished by gentle mechanical action with a blade (see
Figure 8).39
Inorganic Nanoparticles for the

Consolidation of Wall Paintings
Very often the cleaning process (particularly for wall paintings)
cannot be completely performed without a concomitant
consolidation of the painted layer. Consolidation is often
forced by the degradation of the painted layer enhanced by
the polymer, as occurs in Mexico where the unfavorable climate
conditions40,41 increase the rates of polymer degradation

generating severe flaking of surfaces and paint
detachments in a very short time.
Consolidation of wall paintings and limestone by using
compatible material is now achieved by using calcium hydroxide
nanoparticle dispersions that are completely compatible
with the original material constituting the works of art. They
present increased reactivity and dramatically different transport
properties in porous media.42 The high surface area of
nanoparticles influences their chemical reactivity, producing
just few days after the application a consistent consolidation
of the treated surfaces, due to the reaction of hydroxide with
CO2 from the air to give crystalline calcium carbonate (calcite).
These positive effects are particularly important in archeological
sites where the conservation in situ usually requires
an immediate intervention after discovery.
Nanoparticles have been used in several restoration workshops
in Europe and are extensively used in Mexico for the in
situ consolidation of paintings (see Figure 9) and limestone in
La Antigua Ciudad Maya de Calakmul, a UNESCO World Heritage
Site since 2002 (Campeche, Mexico).43,44 Nanoparticles
FIGURE 9. Wall paintings from Mayan Classic period in Calakmul (Mexico). Calcium hydroxide nanoparticles were
used for consolidation.
of calcium and barium hydroxide have been used in Calakmul

for in situ consolidation of the wall paintings (from the
Classic period (600-700 AD)) discovered during the excavations
from 2004 to 2008 in the Structure I of the Acro polis
Chik Naab. Figure 9 reports an image of one of the main
buildings in the Calakmul area, and some parts of the mural
paintings were consolidated by using calcium hydroxide
nanoparticles.
The use of hydroxide nanoparticles is not restricted to consolidation
of wall paintings but can be used to control the pH
in cellulosic material (paper) or wood to avoid the catalytic
cleavage of the _-glycosidic bond or the Fenton reaction usually
occurring in the presence of transition metal ions, opening
new perspective in the conservation of an enormous
number of documents and artifacts.6,7
Conclusions and Outlook

Conservation science has matured using plenty of methodological
tools borrowed from nanoscience. Soft matter, surface
science, detergency, and polymer and organic chemistry
contribute to this new branch of science. Micelles, microemulsions,
and nanoparticle systems are valuable systems for the
optimal conservation of artifacts. These systems have many
avenues for additional improvements at the hands of
researchers.
Improvements of these technologies will be possible
through accurate investigations aiming at the knowledge of
the following:
extraction capability and transport mechanisms in microemulsions
and micellar solutions
transport mechanisms and structure of the physical and
chemical gels
extraction capability and transport mechanisms of confined
microemulsion and micellar systems
synthesis of new organic and inorganic nanoparticles and
nano- or microcontainers for repairing or self-repairing

These new smart materials add a new palette to the classical
methodologies for the cleaning of works of art. Paraphrasing
a well-known sentence we could say that there is
plenty of room for nanoscience in Conservation.
Thanks are due to Massimo Bonini, Doris Rengstl, Ilaria Lapini,
Giacomo Pizzorusso, and David Chelazzi (CSGI) for the assistance
in laboratory experiments, Michel Menu and Aurelia
Chevalier (Centre de Recherche et de Restauration des Muse es
de France C2RMF) for the experiments on canvas paintings,
Carla Giovannone, Tiziana DellOmo, and Lucia Di Paolo (ISCR,
Rome) for assistance with the application tests on model wall
paintings, and Ramon Carrasco Vargas (director of the
Proyecto Arqueologico Calakmul-Mexico), Diana Magaloni
(UNAM-Mexico and director of the Museo Nacional de Antropologia),
Maria Del Carmen Castro and Yareli Jaidar Benavides
(CNCPC-Mexico), and Claudia Garcia Solis, Conservator of the
Proyecto Arqueologico Mayapan (Mexico) for experiments in
situ. Financial support from CSGI and Bilateral project ItalyMexico is acknowledged.
BIOGRAPHICAL INFORMATION
Rodorico Giorgi, Ph.D. in Science for Cultural Heritage Conservation
at the University of Florence and master in Chemistry in
1996, is currently research fellow at the Department of Chemistry
and CSGI. Giorgis main research activity is in physicochemical
characterization of materials, investigation of degradation
processes and the development of methodologies for the conservation
of works of art materials (wall and canvas paintings, stone,
paper, photographs, and wood). Giorgi is author of about 60
publications.
Michele Baglioni is a doctoral candidate in Science for Cultural
Heritage Conservation at the University of Florence and holds a
Master in Technology for the Conservation of Cultural Heritage.
He is currently working in the development of soft matter systems
for conservation.
Debora Berti obtained her Laurea Degree in Chemistry in
1993 and a Ph.D. in Physical Chemistry in 1997. She has been
Visiting Scientist at the ETH (Zurich) during 1994-1995 and
postdoctoral fellow at CSGI until 2000. In 2000, she joined the
University of Florence, where she coordinates a research group
working on phospholipid and nucleolipid self-assembly. She is
the author of more than 60 publications and has presented
more than 100 invited and contributed lectures to international
conferences. Her background in physical chemistry of surfactant
self-assembly has gradually shifted throughout the years
to self-assembly of bioinspired and biorelevant functional
amphiphiles, using scattering techniques.
Piero Baglioni is a Full Professor of Physical Chemistry at the
Department of Chemistry and CSGI of the University of Florence.
He is the author of about 300 publications in the field
of colloids and interfaces and a pioneer in the application of
soft matter to the conservation of Cultural Heritage. He has produced
several innovative methods applied worldwide.
FOOTNOTES
*To whom correspondence should be addressed. E-mail: piero.baglioni@unifi.it. Phone:
+39 0555253033. Fax: +39 0555253032.
No kinship exists among the authors.
REFERENCES
1 Ferroni, E.; Malaguzzi, V.; Rovida, G. Experimental Study by Diffraction of
Heterogeneous Systems as a Preliminary to the Proposal of a Technique for the
Restoration of Gypsum Polluted Murals. ICOM-CC Plenary Meeting - The
International Council of Museums-Committee for Conservation, Amsterdam,

Netherlands, 1969 [Paper read at the Amsterdam ICOM meeting in 1969; copy in
ICCROM library, Rome].
2 Ferroni, E.; Baglioni, P. Experiments About the Method for the Restoration of
Sulfated Frescoes. In Proceedings of the Symposium Scientific Methodologies
Applied to works of Art, Florence, May 2-5, 1984; Parrini, P. L., Ed.; Montedison
Progetto Cultura: Milan, 1986, pp 108-109.
3 Baglioni, P.; Giorgi, R. Soft and Hard Nanomaterials for Restoration and Conservation
of Cultural Heritage. Soft Matter 2006, 2, 293303.
4 Baglioni, P.; Giorgi, R.; Dei, L. Soft Condensed Matter for the Conservation of
Cultural Heritage. C. R. Chim. 2008, 12, 6169.
5 Ambrosi, M.; Dei, L.; Giorgi, R.; Neto, C.; Baglioni, P. Colloidal Particles of Ca(OH) 2:
Properties and Application to Restoration of Frescoes. Langmuir 2001, 17, 4251
4255.
6 Giorgi, R.; Dei, L.; Ceccato, M.; Schettino, C. V.; Baglioni, P. Nanotechnologies for
Conservation of Cultural Heritage: Paper and Canvas Deacidification. Langmuir
2002, 18, 81988203.
7 Giorgi, R.; Bozzi, C.; Dei, L.; Gabbiani, C.; Ninham, B. W.; Baglioni, P. Nanoparticles
of Mg(OH)2: Synthesis and Application to Paper Conservation. Langmuir 2005, 21,
84958501.
8 Giorgi, R.; Chelazzi, D.; Baglioni, P. Nanoparticles of Calcium Hydroxide for Wood
Conservation. The Deacidification of the Vasa Warship. Langmuir 2005, 21, 10743
10748.
9 Carretti, E.; Dei, L.; Baglioni, P. Solubilization of Acrylic and Vinyl Polymers in
Nanocontainer Solutions. Application of Microemulsions and Micelles to Cultural
Heritage. Langmuir 2003, 19, 78677872.
10 Carretti, E.; Giorgi, R.; Berti, D.; Baglioni, P. Oil-in-Water Nanocontainers as Low
Environmental Impact Cleaning Tools for Works of Art: Two Case Studies. Langmuir
2007, 23, 63966403.
11 Bonini, M.; Lenz, S.; Giorgi, R.; Baglioni, P. Nanomagnetic Sponges for the Cleaning
of Works of Art. Langmuir 2007, 23, 86818685.
12 Carretti, E.; Grassi, S.; Cossalter, M.; Natali, I.; Caminati, G.; Weiss, R. C.; Baglioni,
P.; Dei, L. Poly(vinyl alcohol)-Borate Hydro/Cosolvent Gels: Viscoelastic Properties,
Solubilizing Power, and Application to Art Conservation. Langmuir 2009, 25, 8656
8662.
13 Mora, P.; Mora, L.; Philipot, P. Conservation of Wall Painting; Butterworths: London,
1984.
14 Neher, H. T. Acrylic Resins. Ind. Eng. Chem. 1936, 28, 267271.
15 Resins in Conservation, Proceedings of the Symposium, Edinburgh, 21 and 22 May,
1982; Tate, J. O., Tennent, N. H., Eds.; Scottish Society for Conservation and
Restoration: Edinburgh, 1983.
16 Horie, C. V. Materials for Conservation; Butterworths: London, 1987.
17 Carretti, E.; Dei, L.; Baglioni, P. Aqueous Polyacrylic Acid Based Gels:
Physicochemical Properties and Applications in Cultural Heritage Conservation. Prog.
Org. Coat. 2004, 49, 282289.
18 Lazzari, M.; Chiantore, O. Thermal-Ageing of Paraloid Acrylic Protective Polymers.
Polymer 2000, 41, 64476455.
19 Chiantore, O.; Lazzari, M. Photo-Oxidative Stability of Paraloid Acrylic Protective
Polymers. Polymer 2001, 42, 1727.
20 Favaro, M.; Mendichi, R.; Ossola, F.; Rosso, U.; Simon, S.; Tomasin, P.; Vigato,
P. A. Evaluation of Polymers for Conservation Treatments of Outdoor Exposed Stone
Monuments. Part I: Photo-Oxidative Weathering. Polym. Degrad. Stab. 2006, 91,
30833096.
21 Orea, H.; Magar, V. A Brief Review of the Conservation of Wall Paintings in Mexico.
In Preprints of the 13th Triennial Meeting ICOM Committee for Conservation, ICOMCC,
Rio de Janeiro, 22-27 September 2002; Vontobel, R., Ed.; James and James
(Science Publishers): London, 2002; p 176.
22 Borgioli, L.; Caminati, G.; Gabrielli, G.; Baglioni, P. Removal of Hydrophobic
Impurities from Pictorial Surfaces by Means of Heterogeneous Systems. Sci. Tech.
Cult. Herit. 1995, 4, 6774.
23 De Gennes, P. G.; Taupin, C. Microemulsions and the Flexibility of Oil/Water
Interfaces. J. Phys. Chem. 1982, 86, 22942304.
24 Ferroni, E.; Gabrielli, G.; Caminati, G. Asportazione di Materiali Idrofobi da Superfici
Pittoriche Murali Mediante Solubilizzazione in Sistemi Dispersi. In La Cappella
Brancacci, la scienza per Masaccio, Masolino e Filippino Lippi, Quaderni del
restauro; Olivetti: Milan, 1992; pp 162-171.
25 Friberg, S. E.; Flaim, T. Surfactant Association Structures. In Inorganic Reactions In
Organized Media; Holt, S. L., Ed.; ACS Symposium Series 177; American Chemical
Society: Washington, DC, 1982; pp 1-17.
26 Fontell, K.; Ceglie, A.; Lindman, B.; Ninham, B. W. Some Observations on Phase
Diagrams and Structure in Binary and Ternary Systems of
Didodecyldimethylammonium Bromide. Acta Chem. Scand. 1986, A40, 247256.
27 The ammonium counterion was chosen because the painting, after the cleaning
from the wax, was consolidated using the Ferroni-Dini method, which uses the
ammonium carbonate in the first step of the consolidation process.
28 Palazzo, G.; Fiorentino, D.; Colafemmina, G.; Ceglie, A.; Carretti, E.; Dei, L.; Baglioni,
P. Nanostructured Fluids Based on Propylene Carbonate/Water Mixtures. Langmuir
2005, 21, 67176725.
29 Colafemmina, G.; Fiorentino, D.; Ceglie, A.; Carretti, E.; Fratini, E.; Dei, L.; Baglioni,
P.; Palazzo, G. Structure of SDS Micelles with Propylene Carbonate as Cosolvent: A
PGSE-NMR and SAXS Study. J. Phys. Chem. B 2007, 111, 71847193.
30 Carretti, E.; Fratini, E.; Berti, D.; Dei, L.; Baglioni, P. Nanoscience for Art
Conservation: Oil-in-Water Microemulsions Embedded in a Polymeric Network for
the Cleaning of Works of Art. Angew. Chem., Int. Ed. 2009, 48, 89668969.
31 Baglioni, P.; Dei, L.; Carretti, E.; Giorgi, R. Gels for the Conservation of Cultural
Heritage. Langmuir 2009, 25, 83738374, and references therein.
32 Carretti, E.; Dei, L.; Weiss, R. G. Soft Matter and Art Conservation: Rheoreversible
Gels and Beyond. Soft Matter 2005, 1, 1722.
33 Peng, F.; Li, G.; Liu, X.; Wu, S.; Tong, Z. Redox-Responsive Gel-Sol/Sol-Gel
Transition in Poly(acrylic acid) Aqueous Solution Containing Fe(III) Ions Switched by
Light. J. Am. Chem. Soc. 2008, 130, 1616616167.
34 Liu, Q.; Wang, Y.; Li, W.; Wu, L. Structural Characterization and Chemical Response
of a Ag-Coordinated Supramolecular Gel. Langmuir 2007, 23, 82178223.
35 Carretti, E.; Dei, L.; Baglioni, P.; Weiss, R. G. Synthesis and Characterization of Gels
from Polyallylamine and Carbon Dioxide as Gellant. J. Am. Chem. Soc. 2003, 125,
51215129.
36 Wolbers, R. C. Cleaning Painted Surfaces: Aqueous Methods; Archetype
Publications: London, 2000.
37 Stulik, S.; Miller, D.; Khanjian, H.; Khandekar, N.; Wolbers, R.; Carlson, J.; Petersen,
W. C. In Solvent Gels for the Cleaning of Works of Art: The Residue Question; Dorge,
V., Ed.; The Getty Conservation Institute: Los Angeles, 2004; pp 18-83 and
references therein.
38 Bonini, M.; Lenz, S.; Falletta, E.; Ridi, F.; Carretti, E.; Fratini, E.; Wiedenmann, A.;
Baglioni, P. Acrylamide-Based Magnetic Nanosponges: A New Smart
Nanocomposite Material. Langmuir 2008, 24, 1264412650.
39 Chevalier, A.; Chelazzi, D.; Baglioni, P.; Giorgi, R.; Carretti, E.; Stuke, M.; Menu, M.;
Duchamp, R. Extraction dAdhe sifs de Rentoilage en Peinture de Chevalet: Nouvelle
Approche. In Proceedings of the 15th Triennal Conference ICOM Committee for
Conservation, New Delhi, September 22-26; Bridgland, J., Ed.; Allied Publishers
Pvt. Ltd: New Delhi, 2008; Vol. II, pp 581-589.
40 Riederer, J. The Restoration of Archaeological Monuments in the Tropical Climate. In
Proceedings of 7th Triennial ICOM Meeting, Copenhagen, 10-14 September 1984;
De Froment, D., Ed.; ICOM Committee for Conservation: Paris, 1984; pp 21-22.
41 Espinosa, A. Conservation and Restoration of the Murals of the Temple of the
Paintings in Bonampak. In Proceedings of in situ Archaeological Conservation, 613 April 1986, Mexico City; Hodges, H. W. M., Ed.; INAH: Mexico City, 1987; pp
84-89.
42 Odom, T. W.; Pileni, M.-P. Nanoscience (guest editorial). Acc. Chem. Res. 2008,
41, 1565 and articles collected in this special issue.
43 Giorgi, R.; Chelazzi D.; Carrasco, R.; Colon, M.; Desprat, A.; Baglioni, P. The Maya
Site of Calakmul: In situ Preservation of Wall Paintings and Limestone by Using
Nanotechnologies. Proceedings of the IIC Congress 2006, Munich - The Object in
Context: Crossing Conservation Boundaries; Saunders, D., Townsend, J. H.,
Woodcock, S., Eds.; James and James: London, 2006; pp 162-169.
44 Giorgi, R.; Dei, L.; Baglioni, P. A New Method for Consolidating Wall Paintings Based
on Dispersions of Lime in Alcohol. Stud. Conserv. 2000, 45, 154161.
New Methodologies for the Conservation of Cultural

Heritage:
Micellar Solutions, Microemulsions, and Hydroxide
Nanoparticles
ARTICLE in ACCOUNTS OF CHEMICAL RESEARCH MARCH 2010
Impact Factor: 22.32 DOI: 10.1021/ar900193h Source: PubMed
CITATIONS
49
READS
510
4 AUTHORS:
Rodorico Giorgi
SEE PROFILE
Michele Baglioni
SEE PROFILE
Debora Berti
SEE PROFILE
Piero Baglioni
SEE PROFILE
Available from: Piero Baglioni
New Methodologies for the Conservation of

Cultural Heritage: Micellar Solutions,
Microemulsions, and Hydroxide Nanoparticles
RODORICO GIORGI, MICHELE BAGLIONI, DEBORA BERTI, AND
PIERO BAGLIONI*,
Department of Chemistry and CSGI, University of Florence, via della Lastruccia

3 - Sesto Fiorentino, 50019 Florence, Italy
RECEIVED ON JULY 4, 2009
CON SPECTUS
odern civilizations inherited artworks have a powerful impact on society,
from political, sociological, and anthropological points of view, so

the conservation of our Cultural Heritage is fundamental for conveying to
future generations our culture, traditions, and ways of thinking and behaving.
In the conservation of cultural artifacts, scientists intervene in the degradation
of often unique handcrafts, resulting from a delicate balance of
aging, unpredicted events, environmental conditions, and sometimes incorrect
previous restoration treatments, the details of which are often not precisely
known.
Nanoscience and nanotechnology are revolutionizing materials science
in a pervasive way, in a manner similar to polymer chemistrys revolution
of materials science over the preceding century. The continuous
development of novel nanoparticle-based materials and the study of
physicochemical phenomena at the nanoscale are creating new
approaches to conservation science, leading to new methodologies that
can revert the degradation processes of the works of art, in most cases
restoring them to their original magnificent appearance.
Until recently, serendipity and experiment have been the most frequent
design principles of formulations for either cleaning or consolidation
of works of art. Accordingly, the past has witnessed a number of actively detrimental treatments, such
as the
application of acrylic and vinyl resins to wall paintings, which can irreversibly jeopardize the
appearance (or even the
continued existence) of irreplaceable works of art. Current research activity in conservation science is
largely based
on the paradigm that compatibility of materials is the most important prerequisite for obtaining
excellent and durable
results.
The most advanced current methodologies are (i) the use of water-based micelles and microemulsions
(neat or combined
with gels) for the removal of accidental contaminants and polymers used in past restorations and (ii)
the application
of calcium hydroxide nanoparticles for the consolidation of works of art. In this Account, we highlight
how
conservation science can benefit from the conceptual and the methodological background derived
from both soft (microemulsions
and micelles for cleaning) and hard (nanoparticles for consolidation) nanoscience.
A combination of different nanotechnologies allows todays conservators to provide, in each restoration
step, interventions
respectful of the physicochemical characteristics of the materials used by artists. The palette of
methods
provided by nanoscience is continuously enriching the field, and the development of novel
nanomaterials and the study
of nanoscale physicochemical phenomena will further improve the performance of restoration
formulations and our
comprehension of degradation mechanisms.
Published on the Web 03/05/2010 www.pubs.acs.org/acr Vol. 43, No. 6 June 2010 695-704 ACCOUNTS OF CHEMICAL RESEARCH 695
10.1021/ar900193h 2010 American Chemical Society
Introduction
It is well accepted that modern Conservation Science originated

from the tragic floods that devastated Florence and Venice
in 1966, imposing the search for new methodologies to
restore and conserve the immense Cultural Heritage heavily
damaged by the flood. A scientific method for conservation of
carbonatic materials was proposed for the first time by Enzo
Ferroni at the International ICOM Conference in Amsterdam
(1969).1,2 This new method for the in situ consolidation of
wall paintings was able to restore, in a two-step series of
chemical reactions, the degradation of calcium carbonate, the
principal constituent of wall paintings, drastically damaged by
the flood. The new Science for Conservation has developed in
a fast and healthy way following two main streams: (i) the
analytical characterization of the materials constituting the
works of art, the characterization of the pictorial technique
used by the artists, and the chemical reactions involved in
their degradation; (ii) the search for new scientific methods for
the restoration/conservation, allowing the transfer to future
generations of our Cultural Heritage. The last point will be the
object of this Account, where we will review some of the most
important methodologies developed, mainly using the cultural
framework of soft-condensed matter and nanoscience.
Generally speaking, the restoration of a work of art consists
of two distinct aspects:
(1) cleaning, which is a transient treatment, meant to remove
the materials not originally belonging to the work of art.
(2) consolidation, which is a durable intervention that should
remediate, prevent, or slow down further degradation due
to aging or external agents.
All of the current research activity in Conservation Science
is mainly based on the paradigm that compatibility of materials
is the most important prerequisite in order to obtain excellent
and durable results. In view of this seemingly intuitive, yet
novel, awareness, it is clear that the use of polymers for the
consolidation of inorganic materials should be avoided as
much as possible in the restoration of the original features of
the unaltered material.

Currently this is mainly achieved thanks to the tremendous
advancements witnessed by nanoscience and nanotechnology
that provide theoretical and technical backgrounds to formulate
innovative systems for restoration.3,4
We pioneered the application of nanoscience to Cultural
Heritage conservation, and we devised several methods for
the consolidation of wall paintings, for paper and wood deacidification,
and for the removal of grime, dirt, polymers, and
other organic materials soiling the original artifacts.5-12 In this
Account, we will highlight (as an example of the potential of
these new nanotools) the use of soft microemulsions and
micelles for cleaning and hard nanoparticles for consolidation
of works of art.
Degradation of Polymer Resins Used in

Past Restoration
Reversibility of the restoration treatment has been presented
for decades as a milestone in conservation.13 This principle
has supported several conservators in the usage of polymer
resins, made available by the industry since 1960, which were
considered the panacea for the restoration of works of art.
Consolidation and protection of paintings, stone, wood, paper,
glass, ceramic, and bones was made possible by simple application
of organic solutions, through immersion, injection, or
brushing, of acrylic, vinylic, and silicone-based polymers and
(their) copolymers. Most of these polymers were considered
reversible, that is, the solvent used for the application could
be later used for the removal (if desired) of the applied resins.
The large use of polymers was also favored by the the brilliant
water-white color of the polymers that makes it possible
to secure masses of high light transmission and great
optical clarity - because of the remarkable stability of the polymers
to the action of heat and light, these properties are permanent
and by the many striking and beautiful effects can
be produced by the incorporation of dyes and pigments.14
Unfortunately, the application of polymer coatings induces
severe modifications of the main physicochemical properties
of artifacts, particularly for inorganic matrices as wall
paintings.15,16 Investigations performed on Paraloid B72 (or
Acryloid) (EMA/MA 70:30 w/w; ethyl methacrylate/methyl
acrylate; weight average molecular weight 88 000 g/mol; supplier
Rohm and Haas (USA)), Elvacite 2046 (nBMA/iBMA; n-butyl
methacrylate/i-butyl methacrylate), and Primal AC33 (or
Rhoplex, EA/MMA 60:40 w/w; ethyl acrylate/methyl methacrylate;
viscosity (Pa s) 6, Tg (C) 16; supplier Rohm and Haas
(USA)) showed that water capillarity, water vapor permeability,
and surface wettability of lime mortar change drastically.
The application of polymers leads to a strong increase of surface
hydrophobicity, to a decrease of about 70-85% of the
capillary rise coefficients and 40-50% of the vapor permeability,
and to an increase of contact angle from about 30 to
100.17
When the surface is coated with a polymer, salt crystallization
occurs within the pores of the wall paintings, generating
strong mechanical stress beneath the painted layer
(50-100 m) and leading to the complete disruption of the
pictorial layer in a period of time from about 5 to 30-40
years, depending on the environmental conditions. The alteration

of the natural breathing of the surface due to the presence
of the polymer increases consistently the deleterious
mechanical effect of salt crystallization from the interior of the
wall with the complete detachment of the painting from its
support, that is, the surface of the work of art. In addition to
the degradation of the painted layer promoted by the drastic
change of the physicochemical characteristics, polymers themselves
degrade, and upon aging they are subjected to molecular
weight changes due to cross-linking reactions and chain
scissions. The main consequence of cross-linking is the loss of
solubility, which makes the application irreversible.18-20
Photoaging under UV light was examined by some authors to
detect the molecular weight distribution change due to chain scissions
or coupling of macroradicals.18 The overall stability of the
polymers was shown to be influenced by the alkyl side groups,
particularly in the case of paraloid B66 (nBMA; poly(methyl,nbutyl
methacrylate)) and paraloid B67 (iBMA; poly(isobutyl methacrylate)),
where relatively longer ester chains are present. In
these compounds, the oxidation is favored by hydrogen atoms,
either on the methylenes of the n-butyl groups in B66 or on the
tertiary carbon of the isobutyl groups in B67. Polymers containing
n-butyl ester groups undergo fast and extensive cross-linking
together with some chain scissioning.
Paraloid B72 and paraloid B82 (MMA, poly(methyl methacrylate))
show a good stability toward oxidation due to the presence
of the EMA and MMA methacrylic units19 and seem to be
suitable as consolidants and protective agents for applications to
wall paintings and other artifacts. However, it is important to
notice that the composition of commercially available products
contains almost 1% i-BMA, formed during the polymerization
reaction. Even tiny amounts of i-BMA may initiate radical formation
leading to extensive cross-linking reactions, similarly to paraloid
B66 and B67. The difference in paraloid B72 with the other
member of this polymer family is mainly related to the slower
degradation process but is quite common the evidence of
enhanced degradation generated by paraloid B72 (as well as
B66 and B67) used in past restorations. The degradation effects
are particularly evident in locations where the environmental
conditions are severe due to fluctuations of temperature and relative
humidity or in very polluted urban areas where photochemical
smog favors polymer degradation, as demonstrated in a
recent paper by Favaro and co-workers.20
Conservation treatments based on resins have produced
irreversible degradation of most of Mesoamerican paintings.21
Recent investigations in the archeological site of Mayapan
(Mexico, Post-Classic Age 1200-1450 a.C.) have demonstrated
the dramatic conditions of these paintings recently submitted
to consolidation and protection treatments with
mowilith (vinyl acetate/n-butyl acrylate 65:35 w/w copolymer)
resins. A photographic documentation reveals that in less
than 10 years a relevant portion of the original painting was
lost (see Figure 1). This was due to the combination of the
mowilith film with extensive salt crystallization processes that
quickly destroyed the plaster beneath the paint layer.
In addition to the mechanical degradation, the natural aging
of polymer resins generates discoloration and loss of solubility,
due to the increase of the polymer molecular weight. At the end

of the degradation process, these polymers are very difficult or
impossible to remove with conventional neat solvents. Considering
the severe side effects generated by the use of polymers
and the necessity of their removal, we proposed the use of nanostructured
fluids; some examples are highlighted in the following.
Cleaning and Removal of Polymeric

Coatings by Means of Amphiphile-Based
Systems
Historically, the first application of soft matter concepts was
performed in Florence at the end of the 1980s during the restoration
of the Renaissance masterpiece by Masaccio,
FIGURE 1. Mural paintings in Templo de los Nichos Pintados in Mayapan (Yucatan). The pictures show the
degradation that occurred to the
paintings from the last restoration in 1999 due to the mowilith DM5 coating.
Masolino, and Lippi in the Brancacci Chapel in Florence

(1984-1990).22 Investigations under UV light of the paintings
revealed the presence of a multitude of wax spots due to
the extinction of votive candles kept close to the wall painting.
The removal of this hydrophobic material from the hydrophilic
fresco painting was obtained using microemulsions.23-26
The microemulsion used in Brancaccis paintings was very
similar to the so-called French microemulsion and was
formed by nanosized dodecane droplets, dispersed in an
aqueous solution of ammonium27 dodecylsulfate (surfactant)
and 1-butanol (cosurfactant).
The oil in water microemulsion ensured low aggressiveness
with respect to the original components of the painted
layer, due to the presence of water as a dispersing medium
that remained in direct contact with the hydrophilic surface of
the wall painting.24
Since this pioneering example, several other surfactantbased
micellar and microemulsive systems have been used by
our group to clean wall paintings and canvases both from accidental
contamination and from polymers that have been
extensively used as consolidants and protectives in past
restorations.
Nanocompartmentalized systems specifically tailored for
the removal of paraloid B72, primal AC33, and mowilith DM5
(vinyl acetate/n-butyl acrylate 65:35 w/w copolymer; viscosity
(Pa s) 3.5, Tg (C) 2; supplier Hoechst, Germany) resins have
been recently formulated by some of us. Both micelles and
swollen micelles work as nanocontainers of solvents that can
dissolve polymers to obtain their complete removal from the
surface and from the porous texture of the work of art. The
detergency capability mainly depends on the very large surface
area of micelles and nanodroplets available for interaction
with the polymer coating. The water-based system
reduces the penetration into the artifact porous matrix that
occurs with organic solvents and the associated polymer redissolution
into the artifact but also reduces the toxicity of the
formulation, offering better and faster performance than pure
organic solvents. Table 1 reports the composition of some of
the most successful formulations used for cleaning.
Oil in water microemulsion made with sodium dodecyl sulfate
(SDS), 1-pentanol (PeOH) as a cosurfactant, and a small
amount of a mixture of p-xylene and nitro diluent (a mixture

of 62% toluene, 15% butyl acetate, 15% ethyl acetate, 6%
n-butyl alcohol, 2% cellosolve acetate) was shown to be effective
for paraloid B72 removal (Classical system; see Table 1).
Polar polymer resins such as poly(vinyl acetate), not soluble in
the oil mixture above-reported, were successfully removed by
using a four-component micellar solution with a high amount
of propylene carbonate (PC), which replaced the nonpolar mixture
of xylene and nitro diluent (Conegliano system). Previous
investigations have shown that PC acts as a cosurfactant
inducing a rod-to-sphere micelle transition reducing the size
and aggregation number of SDS micelles by increasing the
headgroup area.28,29
The Conegliano system could also be used for paraloid
B72 removal since it favors the swelling of paraloid B72 films,
which can be then removed by mechanical action for a complete
cleaning of the surface.
Very often paraloid B72 and mowilith DM5 have been
used in mixtures or sequentially applied to wall paintings. An
ideal system should remove simultaneously both resins. A
major improvement in this direction has been recently
obtained by adding ethyl acetate (EA) to the Conegliano formulation
to obtain the Mayapan micelles, whose composition
is reported in Table 1. The aromatic and aliphatic solvents
of the Classical formulation are replaced by less toxic solvents.
Pure ethyl acetate and propylene carbonate are known to
solubilize paraloid B72. They are partially water-soluble (EA
8% w/w, PC 20% w/w), but a saturated aqueous solution of
EA and PC provides only partial swelling of the polymer. The
addition of SDS and 1-PeOH either to the EA or PC aqueous
solutions provides a drastic improvement in the efficiency of
the removal process, due to the presence of micelles that provide
the large interfacial area necessary for polymer uptake.
It is important to stress here that micellar solutions formed by
SDS or SDS/PeOH alone are practically ineffective in polymer
removal; therefore it is the synergistic action of the presence
of micelles and PC that causes the polymer removal. Even better
performance was obtained by adding simultaneously 8%
EA and 8% PC to aqueous SDS/1-PeOH micelles (Mayapan
system). This system allowed a complete cleaning of the surfaces
treated with paraloid B72 in a few minutes. The Classical
system is very efficient in removing acrylic polymers, but
the polymer removal is quite slow compared with the Mayapan
micelles. This is probably due to the lower amount of
TABLE 1. Composition (% w/w) of Systems Used for Polymer
Removal from Wall Paintingsa
Classical Conegliano Mayapan
component composition component composition component composition
water 86.2 water 69 water 73.3
SDS 3.9 SDS 5.1 SDS 3.7
1-PeOH 6.5 1-PeOH 3.9 1-PeOH 7
p-xylene 1.8 PC 22 PC 8
nitro diluent (ND) 1.6 ethyl acetate (EA) 8
Classical refers to the first o/w microemulsion formulated for paraloid B72
removal; the other formulations are named after the places where they were
used for the first time.
a

organic solvents contained in the former leading to different

microstructural features of the system.
It is clear from the above examples that the design of efficient
formulations for the tailored removal of polymers

requires a deep comprehension of the solubilization mechanisms,
which can be more complex than traditional detergency,
due to the presence of partially water-soluble organic
solvents (cosolvents). To better understand the polymer
removal process, we performed swelling tests of polymer films
deposited in glass vials and detergency tests on paraloid films
on models of wall paintings.
Quasi-elastic light scattering (QELS) is an efficient and noninvasive
tool to characterize size and size distribution of complex
systems in the nanometer domain. Figure 2 shows that
the intensity weighted size distribution of Mayapan micelles is
centered around 4 nm with a relatively broad size distribution,
while the average droplet diameter of the Classical sysFIGURE 2. Size distribution of the Mayapan micellar system
obtained from QELS.
FIGURE 3. Schematic representation (bottom panel) of the differences in the removal mechanisms of paraloid B72
performed by
Conegliano and Mayapan formulations. Conegliano system only swells the polymer layer, while the application of
Mayapan micelles to a
paraloid B72-coated surface results in a liquid-liquid phase separation. SAXS spectra (upper panel) obtained after
the application of
Mayapan micelles to a paraloid B72-coated surface. The more dense phase (green line) still contains micelles, while
the upper phase (red
line) contains the polymer dissolved in a blend of EA/PC/PeOH.
tem is about 17 nm with polydispersity of 25%, typical of a

common o/w microemulsion. The smaller micellar size,
together with the presence of specific organic solvents, results,
for a given surfactant content in a very high interfacial area
that is correlated to the kinetics of the cleaning process.
The Conegliano micelles are characterized by very small
aggregates with a hydrodynamic diameter of about 2 nm28,29
and refer to clusters formed by 7-10 SDS monomers and not
to conventional micelles. For the sake of clarity, we indicated
these clusters as micelles. Mayapan micelles are similar but
present a higher effectiveness toward polymer removal that is
nicely illustrated by the results of a semiquantitative test on
model systems that we used to compare the removal efficiency
of these two systems. Coating the bottom of two vials
with a known amount of paraloid B72 (whose thickness is
comparable to that spread on art handcraft) and putting the
film in contact with Conegliano and Mayapan micellar solutions
(see Figure 3, bottom), we found that the Conegliano
micelles are able to swell the polymer, but the swollen polymer
remains at the bottom of the vial. Mayapan micelles were
also able to swell the polymer but in addition a spontaneous
detaching of the polymer from the glass was observed with a
liquid-liquid phase separation in the micellar phase. This
behavior is ideal for polymer removal from artifacts.
In order to understand this mechanism, both phases have
been investigated with static and dynamic scattering techiniques.
Figure 3 reports the SAXS spectra obtained from the
polymer-rich and the surfactant-rich phases. The less dense
phase contains the polymer and a mixture of H2O, PeOH, PC,
and EA solvents, enriched in the less polar components. The
more dense and abundant phase is still a micellar solution but
is depleted of the organic phase.
It is clear that the mixture of organic solvents, dispersed in
the aqueous medium thanks to the presence of micelles, guarantees

specific interactions with the polymer (see Figure 4)
suggesting a mechanism where the Mayapan micelles work as
a carrier of solvents during the phase separation process that
is the final step of the cleaning procedure.
Figure 5 shows a brick with a 1 cm thick lime plaster layer,
painted with fresco technique, treated with paraloid B72. Visual
FIGURE 4. Schematic representation of the hypothesized
detergency mechanism of the Mayapan (upper panel) and Classical
(lower panel) systems. While in the first case the partially watersoluble
solvents (EA and PC) partition among the aqueous and the
micellar phase, in the second case, we have a traditional
microemulsion where the water-insoluble oil core is dispersed in
water by the presence of a surfactant film. The naked micelles
(SDS/PeOH) do not detatch paraloid B72 from the surfaces. In this
cartoon, the Mayapan micelles act as solvent carriers and quickly
mediate their interaction with the polymer, which chooses the
right amounts of solvents to undergo a swelling process. It is worth
noticing that the micelles of the Mayapan system, after the
interaction with paraloid B72, change about 10% in size, while the
microdroplets of the Classical system get sensibly smaller; for
microemulsions, the kinetics is slower, due to the oil confinement in
the micellar core. After the interaction, the micelles decrease in size
because of the solvent release.
FIGURE 5. Experimental cleaning test on model fresco paintings
treated with paraloid B72: (A) water/SDS/1-PeOH/(8%)EA; (B) water/
PC; (C) Classical system; (D) Conegliano system; (E) water/EA/PC; (F)
Mayapan system.
analysis of this model painting after cleaning showed a complete

recovery of the optical features of the surface in the area
treated with the Mayapan system (see Figure 5F). The Classical
and Conegliano systems gave slightly worse results, as is clearly
observable from the polymer residues still present in the treated
areas as shown in Figure 5, areas C and D, respectively.
As already mentioned, the Mayapan micelles are also
active for mowilith DM5 removal. Figures 6 and 7 report illustrative
examples of a fresco painting treated with the Mayapan
system.
Confinement of Micellar Solutions and

Microemulsions in Chemical Gels
Microemulsions and micellar solutions are usually applied
over a painted surface by using cellulose-pulp compresses or
physical gels soaked with them.30 In many cases, as in the
case of canvas or wood painting cleaning, a more controlled
release of the microemulsion is required to avoid the swelling
of inorganic and organic binding materials by water, that
is, the most abundant component in an oil in water
microemulsion.
The advantage of using gels resides in the easy manipulation,
the possibility of microemulsion and micellar solution
confinement into the gel network, and the possibility to
make them responsive to external stimuli (i.e., pH, magnetic
fields, temperature, light).31-35
Gel structure may result from formation of relatively
weak physical bonds (i.e., hydrogen bonds or van der Waals
interactions) or covalent chemical bonds; both can be
responsible for the cross-linking of particles or polymer
chains to create the gelled state. In the first case, we refer
to a physical gel, in the second to a chemical gel.
FIGURE 6. Mural painting sample treated, on the right side, with mowilith DM5 under glazing light (center), and
SEM pictures of the
uncoated (left) and coated (right) surface.
FIGURE 7. Details of the painting surface after removal of mowilith DM5 coating by using micellar solution
(Mayapan system).
FIGURE 8. Linen canvas glued with mowilith DMC2 for a relining
treatment. In the red box, the appearance of the treated area after
micellar solution application.
Among the large number of polymers that can be used for

gelation, cellulose derivatives, polyacrylic acids, and
polyamines are the most used in the field of conservation science.
Unfortunately, these physical gels present some difficulties
in their removal and may release residues on the surface
of the work of art after the application. Besides, the gels mesh
size is difficult to control during gelation.36,37
Chemical gels are characterized by stronger interactions in
the gel molecular network. This type of gel behaves like a soft
solid, and the mesh size is tunable by controlling the crosslinking
during the polymerization reaction.
Chemical gels, based on an acrylamide/bis-acrylamide network
and PEG or silane cross-linkers, loaded with the detergent
solutions have been recently synthesized and applied to
Conservation.11,38
The microemulsion release or uptake can be modulated by
controlling the gel mesh size or applying a stimulus. In a previous
paper,38 it was shown by SAXS that a microemulsion
retains its structure when loaded into the gel. The uptake/
release capability of the gel was quantified by analysis of
hydration/dehydration curves. Microemulsion-loaded gel was
shown to be effective for the removal of paraloid B72 from
limestone and wall paintings.
Cleaning tests on canvases, previously treated with adhesives
(i.e., mowilith DMC2 (vinylacetate/dibutylmaleate copolymers)
and acrylate/methacrylate copolymers) that are
normally used for the relining of canvas paintings, have been
performed with a chemical gel, tailored in order to have good
water retention. The gel was loaded with the Mayapan system
specifically developed for the swelling and the removal of
acrylate/methacrylate and vinylacetate/butylacrylate copolymers.
After a 2-hour application, the capillary sorption of water
through the ground and paint layer was restricted to the contact
area of gel over the canvas. The swollen polymers were
partly removed by the gel, and the complete cleaning was
accomplished by gentle mechanical action with a blade (see
Figure 8).39
Inorganic Nanoparticles for the

Consolidation of Wall Paintings
Very often the cleaning process (particularly for wall paintings)

cannot be completely performed without a concomitant
consolidation of the painted layer. Consolidation is often
forced by the degradation of the painted layer enhanced by
the polymer, as occurs in Mexico where the unfavorable climate
conditions40,41 increase the rates of polymer degradation
generating severe flaking of surfaces and paint
detachments in a very short time.
Consolidation of wall paintings and limestone by using
compatible material is now achieved by using calcium hydroxide
nanoparticle dispersions that are completely compatible

with the original material constituting the works of art. They
present increased reactivity and dramatically different transport
properties in porous media.42 The high surface area of
nanoparticles influences their chemical reactivity, producing
just few days after the application a consistent consolidation
of the treated surfaces, due to the reaction of hydroxide with
CO2 from the air to give crystalline calcium carbonate (calcite).
These positive effects are particularly important in archeological
sites where the conservation in situ usually requires
an immediate intervention after discovery.
Nanoparticles have been used in several restoration workshops
in Europe and are extensively used in Mexico for the in
situ consolidation of paintings (see Figure 9) and limestone in
La Antigua Ciudad Maya de Calakmul, a UNESCO World Heritage
Site since 2002 (Campeche, Mexico).43,44 Nanoparticles
FIGURE 9. Wall paintings from Mayan Classic period in Calakmul (Mexico). Calcium hydroxide nanoparticles were
used for consolidation.
of calcium and barium hydroxide have been used in Calakmul

for in situ consolidation of the wall paintings (from the
Classic period (600-700 AD)) discovered during the excavations
from 2004 to 2008 in the Structure I of the Acro polis
Chik Naab. Figure 9 reports an image of one of the main
buildings in the Calakmul area, and some parts of the mural
paintings were consolidated by using calcium hydroxide
nanoparticles.
The use of hydroxide nanoparticles is not restricted to consolidation
of wall paintings but can be used to control the pH
in cellulosic material (paper) or wood to avoid the catalytic
cleavage of the _-glycosidic bond or the Fenton reaction usually
occurring in the presence of transition metal ions, opening
new perspective in the conservation of an enormous
number of documents and artifacts.6,7
Conclusions and Outlook
Conservation science has matured using plenty of methodological

tools borrowed from nanoscience. Soft matter, surface
science, detergency, and polymer and organic chemistry
contribute to this new branch of science. Micelles, microemulsions,
and nanoparticle systems are valuable systems for the
optimal conservation of artifacts. These systems have many
avenues for additional improvements at the hands of
researchers.
Improvements of these technologies will be possible
through accurate investigations aiming at the knowledge of
the following:
extraction capability and transport mechanisms in microemulsions
and micellar solutions
transport mechanisms and structure of the physical and
chemical gels
extraction capability and transport mechanisms of confined
microemulsion and micellar systems
synthesis of new organic and inorganic nanoparticles and
nano- or microcontainers for repairing or self-repairing
These new smart materials add a new palette to the classical
methodologies for the cleaning of works of art. Paraphrasing
a well-known sentence we could say that there is
plenty of room for nanoscience in Conservation.
Thanks are due to Massimo Bonini, Doris Rengstl, Ilaria Lapini,

Giacomo Pizzorusso, and David Chelazzi (CSGI) for the assistance
in laboratory experiments, Michel Menu and Aurelia
Chevalier (Centre de Recherche et de Restauration des Muse es
de France C2RMF) for the experiments on canvas paintings,
Carla Giovannone, Tiziana DellOmo, and Lucia Di Paolo (ISCR,
Rome) for assistance with the application tests on model wall
paintings, and Ramon Carrasco Vargas (director of the
Proyecto Arqueologico Calakmul-Mexico), Diana Magaloni
(UNAM-Mexico and director of the Museo Nacional de Antropologia),
Maria Del Carmen Castro and Yareli Jaidar Benavides
(CNCPC-Mexico), and Claudia Garcia Solis, Conservator of the
Proyecto Arqueologico Mayapan (Mexico) for experiments in
situ. Financial support from CSGI and Bilateral project ItalyMexico is acknowledged.
BIOGRAPHICAL INFORMATION
Rodorico Giorgi, Ph.D. in Science for Cultural Heritage Conservation
at the University of Florence and master in Chemistry in
1996, is currently research fellow at the Department of Chemistry
and CSGI. Giorgis main research activity is in physicochemical
characterization of materials, investigation of degradation
processes and the development of methodologies for the conservation
of works of art materials (wall and canvas paintings, stone,
paper, photographs, and wood). Giorgi is author of about 60
publications.
Michele Baglioni is a doctoral candidate in Science for Cultural
Heritage Conservation at the University of Florence and holds a
Master in Technology for the Conservation of Cultural Heritage.
He is currently working in the development of soft matter systems
for conservation.
Debora Berti obtained her Laurea Degree in Chemistry in
1993 and a Ph.D. in Physical Chemistry in 1997. She has been
Visiting Scientist at the ETH (Zurich) during 1994-1995 and
postdoctoral fellow at CSGI until 2000. In 2000, she joined the
University of Florence, where she coordinates a research group
working on phospholipid and nucleolipid self-assembly. She is
the author of more than 60 publications and has presented
more than 100 invited and contributed lectures to international
conferences. Her background in physical chemistry of surfactant
self-assembly has gradually shifted throughout the years
to self-assembly of bioinspired and biorelevant functional
amphiphiles, using scattering techniques.
Piero Baglioni is a Full Professor of Physical Chemistry at the
Department of Chemistry and CSGI of the University of Florence.
He is the author of about 300 publications in the field
of colloids and interfaces and a pioneer in the application of
soft matter to the conservation of Cultural Heritage. He has produced
several innovative methods applied worldwide.
FOOTNOTES
*To whom correspondence should be addressed. E-mail: piero.baglioni@unifi.it. Phone:
+39 0555253033. Fax: +39 0555253032.
No kinship exists among the authors.
REFERENCES
1 Ferroni, E.; Malaguzzi, V.; Rovida, G. Experimental Study by Diffraction of
Heterogeneous Systems as a Preliminary to the Proposal of a Technique for the
Restoration of Gypsum Polluted Murals. ICOM-CC Plenary Meeting - The
International Council of Museums-Committee for Conservation, Amsterdam,
Netherlands, 1969 [Paper read at the Amsterdam ICOM meeting in 1969; copy in
ICCROM library, Rome].
2 Ferroni, E.; Baglioni, P. Experiments About the Method for the Restoration of
Sulfated Frescoes. In Proceedings of the Symposium Scientific Methodologies

Applied to works of Art, Florence, May 2-5, 1984; Parrini, P. L., Ed.; Montedison
Progetto Cultura: Milan, 1986, pp 108-109.
3 Baglioni, P.; Giorgi, R. Soft and Hard Nanomaterials for Restoration and Conservation
of Cultural Heritage. Soft Matter 2006, 2, 293303.
4 Baglioni, P.; Giorgi, R.; Dei, L. Soft Condensed Matter for the Conservation of
Cultural Heritage. C. R. Chim. 2008, 12, 6169.
5 Ambrosi, M.; Dei, L.; Giorgi, R.; Neto, C.; Baglioni, P. Colloidal Particles of Ca(OH) 2:
Properties and Application to Restoration of Frescoes. Langmuir 2001, 17, 4251
4255.
6 Giorgi, R.; Dei, L.; Ceccato, M.; Schettino, C. V.; Baglioni, P. Nanotechnologies for
Conservation of Cultural Heritage: Paper and Canvas Deacidification. Langmuir
2002, 18, 81988203.
7 Giorgi, R.; Bozzi, C.; Dei, L.; Gabbiani, C.; Ninham, B. W.; Baglioni, P. Nanoparticles
of Mg(OH)2: Synthesis and Application to Paper Conservation. Langmuir 2005, 21,
84958501.
8 Giorgi, R.; Chelazzi, D.; Baglioni, P. Nanoparticles of Calcium Hydroxide for Wood
Conservation. The Deacidification of the Vasa Warship. Langmuir 2005, 21, 10743
10748.
9 Carretti, E.; Dei, L.; Baglioni, P. Solubilization of Acrylic and Vinyl Polymers in
Nanocontainer Solutions. Application of Microemulsions and Micelles to Cultural
Heritage. Langmuir 2003, 19, 78677872.
10 Carretti, E.; Giorgi, R.; Berti, D.; Baglioni, P. Oil-in-Water Nanocontainers as Low
Environmental Impact Cleaning Tools for Works of Art: Two Case Studies. Langmuir
2007, 23, 63966403.
11 Bonini, M.; Lenz, S.; Giorgi, R.; Baglioni, P. Nanomagnetic Sponges for the Cleaning
of Works of Art. Langmuir 2007, 23, 86818685.
12 Carretti, E.; Grassi, S.; Cossalter, M.; Natali, I.; Caminati, G.; Weiss, R. C.; Baglioni,
P.; Dei, L. Poly(vinyl alcohol)-Borate Hydro/Cosolvent Gels: Viscoelastic Properties,
Solubilizing Power, and Application to Art Conservation. Langmuir 2009, 25, 8656
8662.
13 Mora, P.; Mora, L.; Philipot, P. Conservation of Wall Painting; Butterworths: London,
1984.
14 Neher, H. T. Acrylic Resins. Ind. Eng. Chem. 1936, 28, 267271.
15 Resins in Conservation, Proceedings of the Symposium, Edinburgh, 21 and 22 May,
1982; Tate, J. O., Tennent, N. H., Eds.; Scottish Society for Conservation and
Restoration: Edinburgh, 1983.
16 Horie, C. V. Materials for Conservation; Butterworths: London, 1987.
17 Carretti, E.; Dei, L.; Baglioni, P. Aqueous Polyacrylic Acid Based Gels:
Physicochemical Properties and Applications in Cultural Heritage Conservation. Prog.
Org. Coat. 2004, 49, 282289.
18 Lazzari, M.; Chiantore, O. Thermal-Ageing of Paraloid Acrylic Protective Polymers.
Polymer 2000, 41, 64476455.
19 Chiantore, O.; Lazzari, M. Photo-Oxidative Stability of Paraloid Acrylic Protective
Polymers. Polymer 2001, 42, 1727.
20 Favaro, M.; Mendichi, R.; Ossola, F.; Rosso, U.; Simon, S.; Tomasin, P.; Vigato,
P. A. Evaluation of Polymers for Conservation Treatments of Outdoor Exposed Stone
Monuments. Part I: Photo-Oxidative Weathering. Polym. Degrad. Stab. 2006, 91,
30833096.
21 Orea, H.; Magar, V. A Brief Review of the Conservation of Wall Paintings in Mexico.
In Preprints of the 13th Triennial Meeting ICOM Committee for Conservation, ICOMCC,
Rio de Janeiro, 22-27 September 2002; Vontobel, R., Ed.; James and James
(Science Publishers): London, 2002; p 176.
22 Borgioli, L.; Caminati, G.; Gabrielli, G.; Baglioni, P. Removal of Hydrophobic
Impurities from Pictorial Surfaces by Means of Heterogeneous Systems. Sci. Tech.
Cult. Herit. 1995, 4, 6774.
23 De Gennes, P. G.; Taupin, C. Microemulsions and the Flexibility of Oil/Water
Interfaces. J. Phys. Chem. 1982, 86, 22942304.
24 Ferroni, E.; Gabrielli, G.; Caminati, G. Asportazione di Materiali Idrofobi da Superfici
Pittoriche Murali Mediante Solubilizzazione in Sistemi Dispersi. In La Cappella
Brancacci, la scienza per Masaccio, Masolino e Filippino Lippi, Quaderni del
restauro; Olivetti: Milan, 1992; pp 162-171.
25 Friberg, S. E.; Flaim, T. Surfactant Association Structures. In Inorganic Reactions In
Organized Media; Holt, S. L., Ed.; ACS Symposium Series 177; American Chemical
Society: Washington, DC, 1982; pp 1-17.
26 Fontell, K.; Ceglie, A.; Lindman, B.; Ninham, B. W. Some Observations on Phase
Diagrams and Structure in Binary and Ternary Systems of
Didodecyldimethylammonium Bromide. Acta Chem. Scand. 1986, A40, 247256.
27 The ammonium counterion was chosen because the painting, after the cleaning
from the wax, was consolidated using the Ferroni-Dini method, which uses the
ammonium carbonate in the first step of the consolidation process.
28 Palazzo, G.; Fiorentino, D.; Colafemmina, G.; Ceglie, A.; Carretti, E.; Dei, L.; Baglioni,
P. Nanostructured Fluids Based on Propylene Carbonate/Water Mixtures. Langmuir
2005, 21, 67176725.
29 Colafemmina, G.; Fiorentino, D.; Ceglie, A.; Carretti, E.; Fratini, E.; Dei, L.; Baglioni,
P.; Palazzo, G. Structure of SDS Micelles with Propylene Carbonate as Cosolvent: A
PGSE-NMR and SAXS Study. J. Phys. Chem. B 2007, 111, 71847193.
30 Carretti, E.; Fratini, E.; Berti, D.; Dei, L.; Baglioni, P. Nanoscience for Art
Conservation: Oil-in-Water Microemulsions Embedded in a Polymeric Network for
the Cleaning of Works of Art. Angew. Chem., Int. Ed. 2009, 48, 89668969.
31 Baglioni, P.; Dei, L.; Carretti, E.; Giorgi, R. Gels for the Conservation of Cultural
Heritage. Langmuir 2009, 25, 83738374, and references therein.
32 Carretti, E.; Dei, L.; Weiss, R. G. Soft Matter and Art Conservation: Rheoreversible
Gels and Beyond. Soft Matter 2005, 1, 1722.
33 Peng, F.; Li, G.; Liu, X.; Wu, S.; Tong, Z. Redox-Responsive Gel-Sol/Sol-Gel
Transition in Poly(acrylic acid) Aqueous Solution Containing Fe(III) Ions Switched by
Light. J. Am. Chem. Soc. 2008, 130, 1616616167.
34 Liu, Q.; Wang, Y.; Li, W.; Wu, L. Structural Characterization and Chemical Response
of a Ag-Coordinated Supramolecular Gel. Langmuir 2007, 23, 82178223.
35 Carretti, E.; Dei, L.; Baglioni, P.; Weiss, R. G. Synthesis and Characterization of Gels
from Polyallylamine and Carbon Dioxide as Gellant. J. Am. Chem. Soc. 2003, 125,
51215129.
36 Wolbers, R. C. Cleaning Painted Surfaces: Aqueous Methods; Archetype
Publications: London, 2000.
37 Stulik, S.; Miller, D.; Khanjian, H.; Khandekar, N.; Wolbers, R.; Carlson, J.; Petersen,
W. C. In Solvent Gels for the Cleaning of Works of Art: The Residue Question; Dorge,
V., Ed.; The Getty Conservation Institute: Los Angeles, 2004; pp 18-83 and
references therein.
38 Bonini, M.; Lenz, S.; Falletta, E.; Ridi, F.; Carretti, E.; Fratini, E.; Wiedenmann, A.;
Baglioni, P. Acrylamide-Based Magnetic Nanosponges: A New Smart
Nanocomposite Material. Langmuir 2008, 24, 1264412650.
39 Chevalier, A.; Chelazzi, D.; Baglioni, P.; Giorgi, R.; Carretti, E.; Stuke, M.; Menu, M.;
Duchamp, R. Extraction dAdhe sifs de Rentoilage en Peinture de Chevalet: Nouvelle
Approche. In Proceedings of the 15th Triennal Conference ICOM Committee for
Conservation, New Delhi, September 22-26; Bridgland, J., Ed.; Allied Publishers
Pvt. Ltd: New Delhi, 2008; Vol. II, pp 581-589.
40 Riederer, J. The Restoration of Archaeological Monuments in the Tropical Climate. In
Proceedings of 7th Triennial ICOM Meeting, Copenhagen, 10-14 September 1984;
De Froment, D., Ed.; ICOM Committee for Conservation: Paris, 1984; pp 21-22.
41 Espinosa, A. Conservation and Restoration of the Murals of the Temple of the
Paintings in Bonampak. In Proceedings of in situ Archaeological Conservation, 613 April 1986, Mexico City; Hodges, H. W. M., Ed.; INAH: Mexico City, 1987; pp
84-89.
42 Odom, T. W.; Pileni, M.-P. Nanoscience (guest editorial). Acc. Chem. Res. 2008,
41, 1565 and articles collected in this special issue.
43 Giorgi, R.; Chelazzi D.; Carrasco, R.; Colon, M.; Desprat, A.; Baglioni, P. The Maya
Site of Calakmul: In situ Preservation of Wall Paintings and Limestone by Using
Nanotechnologies. Proceedings of the IIC Congress 2006, Munich - The Object in
Context: Crossing Conservation Boundaries; Saunders, D., Townsend, J. H.,
Woodcock, S., Eds.; James and James: London, 2006; pp 162-169.
44 Giorgi, R.; Dei, L.; Baglioni, P. A New Method for Consolidating Wall Paintings Based
on Dispersions of Lime in Alcohol. Stud. Conserv. 2000, 45, 154161.
int. j. remote sensing, 2002, vol. 23, no. 3, 429441
Integration of remote sensing data and GIS for accurate mapping of

ooded areas
P. A. BRIVIO, R. COLOMBO, M. MAGGI and R. TOMASONI
Remote Sensing Department, IRRS, National Research Council, Via Bassini 15,
20133-Milano, Italy
(Received 8 April 1999; in . nal form 11 July 2000)
Abstract. This paper describes a synergetic use of satellite radar images and
ancillary information to detect ooded areas at their peak and evaluates its
potential with mapping. The procedure was tested on the catastrophic ood that
occurred in Regione Piemonte in Italy in November 1994. Two ERS-1 synthetic

aperture radar (SAR) images were processed, one acquired one month before the
ood and the other acquired three days after the event. Visual interpretation and
two diVerent thresholding techniques were performed. The ood map derived
shows only a small fraction (20%) of the actually ooded lands because of the
time delay between the ood peak and the satellite overpass. To overcome this
limitation, the authors developed a new procedure to estimate the ooded area
at the peak time by integrating the ooded area from SAR imagery with digital
topographic data from a GIS technique. This method allowed inundated areas
covering 96.7% of the ooded area oYcially recorded by the local government
to be mapped. The proposed procedure is suitable for mapping ooded areas
even when satellite data are acquired some days after the event, thus overcoming
the constraint of temporal resolution in the application of SAR imagery in
hydrology.
1. Introduction
In densely populated areas, both in tropical and temperate regions, oods are
one of the most important natural disasters as they inundate large areas causing
damage to agricultural crops and property, road and rail links, and the loss of
human lives. Appropriate land use planning, which requires accurate knowledge of
ood extent for locating ood prone areas, is a key tool to improve ood management
and to mitigate its catastrophic eVects. The knowledge of the spatial extent of
inundated areas is essential both during the emergency phase, when it is necessary
to have an overall view of the phenomenon in order to plan relief eVorts, and after
the event for the assessment of damaged areas by authorities and insurance
companies.
Traditional methods of ood mapping are based on ground surveys and aerial
observations, but when the phenomenon is widespread, such methods are time
consuming and expensive; furthermore timely aerial observations can be impossible
e-mail: brivio@irrs.mi.cnr.it
International Journal of Remote Sensing

ISSN 0143-1161 print/ISSN 1366-5901 online 2002 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080/01431160010014729
430 P. A. Brivio et al.

due to prohibitive weather conditions. An alternative option is oVered by satellite
remote sensing (RS) technology. In recent decades optical data acquired by sensors
onboard spacecraft have been used in many studies to map inundated areas over
regions characterized by very diVerent conditions in climate, morphology and land
use. These approaches for ood extent evaluation have involved digital or visual
analyses of Syste`me Probatoire pour lObservation de la Terre Multispectral (SPOT
XS) scenes of Europe and Asia (Robin and Le Rhun 1989, Blasco et al. 1992),
Landsat Thematic Mapper (TM) data in North America (Lougeay et al. 1994) and
Landsat Multispectral Scanner (MSS) images of West Africa (Brivio et al. 1984,
Zilioli et al. 1993). Thermal data collected during the night-time by NOAA-9 AVHRR
(National Oceanic and Atmospheric Administration Advanced Very High Resolution
Radiometer) were also used by Barton and Bathols (1989) to map the extent of
ooding in central Australia. An estimate of the extent of ooded areas was conducted
by Berg et al. (1981) using the NOAA satellite thermal infrared sensor. Despite their
coarse spatial resolution, AVHRR images have the advantage of high temporal
resolution, and sometimes are the only available source of information.
Bad weather conditions during and after ood events can represent a strong
constraint to the utilization of optical remotely sensed data. For this reason, optical
sensors are generally used to assess inundated . elds only some days after the event,
either by recognition of uvial sediments left on the land (Rosso 1995) or by the
detection of vegetation stress (Michener and Houhoulis 1997). On the contrary,
spaceborne radar systems, because of their exclusive cloud penetration capacity, oVer
a primary tool for real-time assessment of ooded areas. DiVerent methodologies
have been developed using special sensor microwave imager SSM/I (Tanaka et al.
2000) or single-band (Sharma et al. 1996) and multi-frequency (Ormsby et al. 1985,
Hess et al. 1995) synthetic aperture radar (SAR) data. The disadvantages of the use
of radar sensors lie in the diYcult classi. cation of the acquired signal because of the
in uence of complex ground and system variables. This is particularly true when
single-band and single-polarization data are available, as is the case in the ERS-1/2
satellites. To improve this situation, multi-temporal techniques based on detection
of changes between radar images acquired before and after the inundation event are
usually recommended (Badji et al. 1994, Wang et al. 1995, Profeti and Macintosh
1997, Takeuchi et al. 1999). Moreover if a high temporal resolution dataset is
available, change detection analysis can be used to produce a ood evolution map
(Noyelle et al. 1995, Oberstadler et al. 1995, Laugier et al. 1997).
If radar images are acquired some days after the event, when only a few areas
are still submerged by water, it is advisable to use a multi-sensor approach. In this
case ooded areas as derived from radar data are complemented with information
extracted from optical images, such as the areal extent of . ne material left during
the ood (ImhoV et al. 1987, Bonansea 1995). Another possibility is the integration
in a Geographical Information System (GIS) of radar imagery acquired during the
ood event with information derived from digital topography (Giacomelli and
Mancini 1996, Brakenridge et al. 1998).
This paper proposes a new methodology which, integrating results obtained from
ERS SAR images and topography in an appropriate GIS methodology, allows a
satisfactory mapping of inundated areas, even when satellite data are acquired some
days after the ooding event. The catastrophic ood event that occurred in Regione
Piemonte (northern Italy) on November 1994 is used to illustrate the methodology.
Integration of SAR data and GIS for ood mapping 431
2. Mapping ooded areas through an integrated RS and GIS approach
The magnitude of the ood and the extent of ood-aVected areas depend on the
intensity of rainfall, its duration, the watershed topography and its conditions at the
time of the heavy rainfall. Observation from a RS satellite provides a timeinstantaneous
portrait of the ood stage over a wide area. However, the period of
submergence can vary from hours to weeks and the time delay between the ood
peak phase and the satellite observation may severely limit the capabilities of mapping
the widest spread of ooding. Figure 1 illustrates the relationships between ooded
area extent, ood peak phase and SAR image acquisition.
2.1. Flood mapping on SAR imagery
The fundamental characteristic recorded on a radar image is the backscattering
coeYcient, which may vary from surface to surface. The strength of the returned
signal from the surface is in uenced by combined system and ground parameters,
including the average surface roughness and soil dielectric properties. Horizontal
smooth surfaces, such as water bodies, re ect nearly all incident radiation away from
the sensor and the weak return signal is represented by dark tonality on radar
images with the result that standing water areas are easily recognizable. This specular
re ection can be decreased by bad weather conditions and /or the presence of vegetation,
roughening the surface and making the detection of ooded areas more di Ycult
(Laugier et al. 1997). Moreover, SAR data are subject to speckle, a multiplicative
random noise that considerably reduces the interpretability of the images and limits
classi. cation techniques, and SAR images have to be . ltered in order to increase
the signal-to-noise ratio. The enhanced Lee adaptive . lter is an eYcient tool for
reducing SAR image speckle by removing high-frequency noise while preserving
edges or sharp features in the image (Zhengao and Fung 1994). Shadows at the
Figure 1. Temporal relationships between ood event, ooded area and SAR observation
highlighting the diVerence between ooded area at the peak (T p) and ooded area
after the event (T p+DT ) seen on an SAR image.

foothills, paved roads and the hydrological network itself can be confused with
inundated areas since they also appear in a dark grey tone. The determination of
paved roads and river streams can be established with the help of auxiliary
information derived from digitized topographic maps.
DiVerent kinds of approaches can be adopted to map the extent of ooded areas.
The . rst one is based on segmentation techniques applied to SAR imagery acquired
after the ood. A second approach consists of comparing two SAR images, one
taken before and the other after the ood, through visual interpretation or change
detection analysis. A third approach, which is not considered in this study, involves
the use of coherence information deduced from an interferometric pair of SAR images
(Nico et al. 2000).
When a segmentation approach is adopted to detect ooded areas on the after ood SAR image, optimal threshold de. nition is required. The threshold value can
be determined on a frequency histogram of the areas corresponding to dark grey
tones in the SAR image and to ooded areas in the aerial photographs. A possible
alternative is to use an adaptive thresholding algorithm. Ober et al. (1998) developed
an iterative process which, assuming a Gaussian distribution of the data, calculates
the mean value (m) and the standard deviation (s) and assigns to the Digital Number
(DN) values out of the interval [m s; m s], the extreme value of the same interval.
This iterative process continues until all pixels are one of only two values (resulting
in a binary image ( ooded and nonooded areas)) or when all the image pixels
have a value belonging to the interval [m s; m s]; in a such case m is the searched
threshold value.
When using a multi-temporal approach, principal component analysis (PCA) of
SAR images, taken at diVerent times, oVers great opportunities for ooded area
detection (Calabresi 1995, Henebry 1997). In this approach a ood map is drawn
up directly on a computer screen by means of visual interpretation of the false colour
composite represented in RGB by images taken after and before the ood event.
These approaches allow satisfactory determination of ooded zones where water
is still remaining at the time of space observation. However, this situation does not
always represent the widest spread of the ooding. Often, ash ooding occurs and
then the water level decreases very quickly; in this case observation from remote
sensing satellites cannot meet the peak ooding phase.
2.2. An integrated RS and GIS approach
In order to overcome these limitations, a new procedure was developed in a GIS
environment with the purpose of estimating the ooded area at the peak using
standing water area observed after the ood event seen on SAR imagery and ancillary
information extracted from the digital elevation topography. The main steps of the
overall methodology are presented in . gure 2. The basic assumption of this approach
is that water has to ow out from the main river channel up to reach at least the
areas recognized as ooded in the SAR data. The technique, based on the computation
of least accumulative costdistance surface (Douglas 1994), . lls up the ooded
place by tracing backwards from the main river stream towards still remaining water
observed on SAR image.
From the digital topography of the study area a least accumulative costdistance
matrix is derived. In such a raster image, each cell is characterized by a value equal
to the least accumulative cost that water has to spend to get to the cell departing
from main stream of the river. Cost corresponds to the work necessary for the water
Figure 2. Schematic diagram of the combined methodology: 1, detect ooded areas from
SAR data; 2, generate least accumulative costdistancematrix from digital topography
of the study area, where the origin is the river; 3, a cell in the study area is ooded if
its cost is lower than the cost of the cell detected from SAR data.
to overcome resistance formed by natural and anthropic features such as topography,

surface roughness, vegetation cover, land useall factors that aVect water ow over
the surface.
The least accumulative costdistance matrix is modelled as a continuous, cumulative

function from the main river stream up to the limits of the study area using a
recursive procedure common to many GIS environments. An automatic procedure
implemented in ARC/INFO (ESRI 1992) was used for this study.
First the costdistance, i.e. the cost required for water to move from the source
cell to each of the neighbour cells, is calculated. Initially the source cells correspond
to the cells of the main stream. This process is then repeated considering the least
costly neighbouring cell (i.e. the cell that presents the least resistance to the water
ow) as the new source cell. The iterative process stops when all the cells have been
assigned an accumulative cost.
In this procedure the costdistance between two any adjacent cells A and B
(costdistA_B) is calculated according to the following formula:
cost distA_B=
(costA+costB)distA_B
2
(VF)(HF) (1)
where costA and costB are values associated with each cell A and B, distA_B is the
distance between cells A and B, and VF and HF are vertical and horizontal factors
accounting for topography and surface typology respectively.
The cost value (costA and costB) associated with each cell is a weight factor
proportional to the resistance incurred by water when owing over the cell. It is a
per-unit distance measure and in this case is represented by the elevation of the
digital topography. The Pythagorean theorem, using a representation in which the
centres of two adjacent cells are connected by links, calculates the distance between
two cells (distA_B).
The vertical factor (VF) takes into account the cost necessary to overcome the
height diVerence between two cells. Its computation is based on the slope map and
values are assigned using a linear function relating the slope to the interval 0 to 2,
as required by the used ARC/INFO procedure.
The horizontal factor (HF) accounts for any friction encountered by water owing
over the surface and is linked to the surface characteristicsroughness, land use,
vegetation cover, etc. Because this factor is very complex to model, the typology of
the surface was assumed constant over the entire area and set equal to one
everywhere.
The least accumulative costdistance matrix is then integrated with a base layer
consisting of the ooded area map derived from the previous analysis of SAR image.
Cost matrix cells corresponding to ooded points detected on the SAR image are
considered to be border limits of the inundation event: the cost value assumed by
the cell at the largest distance from main river stream is selected as a threshold.
Application of these criteria allows selection of all those cells of the cost matrix
where water could be expected. In fact, because the cost value is a function of
topography and distance, the energy necessary for water to reach higher cost cells
must also be suYcient to get water to all those cells with lower cost value. The . nal
result is a map of all zones aVected by the widest spread of the inundation event
within the study area.
3. A case study: Tanaro ood in 1994
Our study is concerned with the ood that occurred in Regione Piemonte
(northern Italy) on the days of 4, 5 and 6 November 1994. The persistence of a wide
atmospheric depression over north-west Europe caused heavy rains all over the Po
river plain. The Tanaro basin, located in southern Piemonte, was the most aVected
by this extreme rainfall. Over 200mm of rain fell on 4 and 5 November in the upper
and middle part of the Tanaro catchment and in the upper portions of its tributaries,
the Belbo, Bormida and Orba rivers. Precipitation reached a maximum hourly
intensity of 55mm, with maximum cumulative value of 264.4mm over 24 h, causing
the saturation of both the drainage capacity of the slopes and the impoundment of
the hydrographic network. Rainfall produced exceptional discharge of the rivers and
hydrometric levels never recorded before (unit discharge 0.67.3m3 s 1km 2).
Floods caused widespread damage to 38 urban areas, with the creation of 44 victims
and 2000 homeless (Luino 1999). A comparable case in the Tanaro basin has not
been recorded since 1879.
3.1. Study area and data used
The study area is a reach of the Tanaro basin included between the towns of
Asti and Felizzano. This part of the catchment is an almost at area with an average
slope of 1% and is mainly devoted to agricultural activities. The land cover as
derived from supervised classi. cation of Landsat 5 TM data taken in July 1994 is
as follows: 79% cropland (wheat, sugar beet, grass), 8% poplar plantations, 6%
urban areas and 7% water bodies.
The satellite data used were two ERS-1 SAR images, the . rst one acquired on 4
October 1994, one month before the ood and the second one acquired on
9 November 1994, three days after the event (. gure 3). The SAR mounted onboard
ERS-1 is a C-band system (5.6cm wavelength) with a nominal incidence angle of
23 at mid-swath, which transmits and receives vertically polarized microwave
electromagnetic
energy. ERS-1 Precision Image (PRI) products used are three-look
digital images, corrected for antenna elevation gain pattern and range spreading
loss, with pixel dimensions of 12.5m12.5 m.
In addition to the satellite images, ancillary data were used. Contour lines were
extracted from a digital topographic map in raster format (1:10 000) and used to
derive the digital elevation model (DEM); the main streams of the river network
and paved roads were extracted from the same map. Colour aerial photographs
taken on 9 and 10 November 1994 over some areas were also available and used to
de. ne ground truth areas. Moreover, maps (1:10 000 scale) of the actual inundated
areas, acquired from ground surveys by the Settore Prevenzione del Rischio
Geologico, Meteorologico e Sismico of Regione Piemonte, were used as reference
data for accuracy evaluation. The areal extension of the actual inundated areas was
estimated to be 54km2 from this reference map.
Both radar images and ancillary data were pre-processed and arranged in a GIS
environment. A set of ground control points (GCPs) taken from the 1:10 000 topographic
map was used to rectify the 9 November ERS-1 image to the Universal
Transverse Mercator (UTM) map projection, producing a root means square error
(RMSE) of 11.6 m. The pre- ood image of 4 October was then registered to the
after- ood scene (RMSE 8.75m) through a . rst-degree polynomial transformation,
to form a perfectly overlapping multi-temporal dataset. A nearest-neighbour resampling
technique was adopted for both images in order to retain radiometric integrity.
Figure 3. Tanaro river basin study area where a catastrophic ood occurred in November
1994. ERS-1 SAR image acquired on 9 November, three days after the ood event.

The digital topography was generated from contour lines having a vertical interval
of 5m over the mountain areas and 10m along the oodplain. The interpolation
process was carried out using an iterative . nite diVerence technique coupled with a
drainage enforcement algorithm (Hutchinson 1989) speci. cally developed to generate
hydrologically corrected digital topography. The grid size was de. ned at the same
resolution of the SAR data (12.5m12.5m). Areas below an elevation threshold
value were extracted and used to mask (on the radar images) the area (151km2) on
which to apply the mapping procedures.
3.2. Analysis of the results
Processing of a single radar image acquired some days after the ood peak and
multi-temporal analysis of radar data taken before and after the inundation event
allowed only a limited assessment of the ooded areas. The two thresholding procedures
and visual interpretation were able to map only a small fraction of the total
inundated area, amounting to 54km2.
Visual interpretation of a false colour composite obtained from multi-temporal
SAR images and Second Principal Component (. gure 4) gave larger extension of
ooded areas (10.23km2) compared to maps determined by thresholding approaches
applied to a single-date SAR imagery. An iterative threshold algorithm applied to a
single-date SAR image allowed us to map an area of 7.08km2; using a simple
threshold value derived from the cumulative frequency curve mapped only 3.96km2
as ooded zones.
Figure 5 shows a map of ooded areas derived from visual interpretation of the
false colour composite of the multi-temporal dataset. In the same . gure, the study
area extracted from digital topography of the Tanaro watershed is also presented.
Areas classi. ed as inundated are patches spread around the main river stream. The
water surfaces in this map cover only 18.9%of the ooded area recorded by Regione
Piemonte. These large diVerences between the ooded zone detected on SAR imagery
Figure 4. False colour composite of multi-temporal ERS-1 radar images (RGB: 9 November
1994; 4 October 1994; Second Principal Component). Areas still submerged by water
three days after the ood event are easily recognizable due to their intense blue colour.
Figure 5. Flood map derived from visual interpretation of false colour composite of multitemporal
ERS-1 radar images (RGB: 9 November 1994; 4 October 1994; Second
Principal Component), compared with the reference ood map compiled by the Settore
Prevenzione del Rischio Geologico, Meteorologico e Sismico of Regione Piemonte.
and reference data can be explained by considering the time delay between the ERS-1
passage and the ood peak.
On 9 November, when the satellite image was acquired, most of the water had
already own back into the river or in. ltrated into the soil. Aerial photographs
(. gure 6) taken on 9 and 10 November 1994 over some sites con. rmed that, at the
time of the ERS-1 overpass, most of the water was not actually still standing on
land, and areas estimated as inundated corresponded to few zones still under water
three days after the ood event. Moreover, the presence of a layer of uvial wet . ne
sediments made the recognition of areas aVected by the ood event diYcult. In fact,
in radar images, wet sediments produce intermediate backscattering values between
low-intensity signals of water and the higher intensity signals of agricultural areas.
In some cases poplar plantations close to the Tanaro river were also a constraint to
ERS-1 in monitoring standing water, because of the rather low penetration capacity
of the C microwave band at which it operates.
In contrast to the previous results, which re ect the situation three days after the
ood peak, the ood map of . gure 7 obtained by application of the combined
methodology describes the maximum ooded area (60km2). The ooded zones are
seen to extend both sides of the river and form a continuous belt around the
main stream.
This map was obtained using the ooded area map given by the simple threshold
approach as a base layer to be integrated with the least accumulative costdistance
matrix. The Tanaro river basin was subdivided into diVerent sections re ecting
Figure 6. Enlargement of ood map of . gure 5 and oblique aerial photograph taken on
9 November 1994 over the same area (Courtesy of Fabio Luino, CNR-IRPI Torino).
diVerent morphological situations, each one corresponding to an area mapped as

ooded in ERS data.
The map of ooded areas at the peak estimated by the new integrated approach
was compared pixel-to-pixel with the digitized reference map of inundated areas
compiled by Regione Piemonte. From this analysis a coincidence of 96.7%(52.2km2)
was found between the ooded areas estimated and those o Ycially recorded by the
local government oYce.
The results obtained appear very promising, especially because the methodology
allows mapping of inundated areas accurately over a large area and without the
need for a complex simulations approach. Even though hydraulic models may
provide a more correct description of the ooding process, they are also characterized
by a more critical parameterization.
Figure 7. Map of the maximum extent of ooded area estimated by the new approach based
on integration of the ood map derived from ERS-1 post- ood image and the least
accumulate costdistance matrix.
4. Conclusions
Techniques based on visual interpretation of multi-temporal satellite data and
thresholding algorithms applied to single-date satellite data show serious limitations
in determination of inundated areas at the peak of the ood. This is because of the
inappropriate time of acquisition of satellite radar data with respect to the ood
peak event. In fact, in the case study presented here, most of the water had already
own back into the river or had been in. ltrated into the soil at the time of the
satellite overpass.
To overcome these limitations, a new approach was proposed which integrates
results obtained from SAR data with ancillary information using a least accumulative
costdistance matrix. The catastrophic ood that occurred in northern Italy in
November 1994 was used to test the obtained results through detailed comparison
with a reference map of the ooded area compiled by Regione Piemonte. When
compared with the reference map showing the area actually ooded, the ooded
area map obtained from the new integrated RS and GIS approach shows an accuracy
of 96%. This result indicates that images collected by ERS-1 SAR some days after
an event can be useful in generating ood maps of the peak if used in conjunction
with an appropriate methodology, whereas without such an approach SAR data do
not allow such an accurate delimitation of ooded areas.
Acknowledgments
This study was supported by the National Group for Prevention from
Hydrogeological Hazards (GNDCI) of the National Council of Research (CNR) of
Italy. We are also grateful to Giovanna Ober for software implementation of the
iterative threshold technique. Suggestions of two anonymous referees greatly
improved the quality of the manuscript. The 4 October 1994 and the 9 November
1994 ERS-1 PRI images were kindly provided by ESA-ESRIN, Frascati, Italy.
References
Badji, M., Dautrebande, S., Mokaden, A. I., and Dewez, A., 1994, ERS-1 SAR imagery
applied to rural basins hydrological studies: the ood inundation mapping and monitoring.
Proceedings of the First Workshop on ERS-1 Pilot Projects, T oledo, Spain,
2224 June 1994, SP-365 (Noordwijk, The Netherlands: ESA-ESTEC), pp. 117124.
Barton, I. J., and Bathols, J. M., 1989, Monitoring oods with AVHRR. Remote Sensing of
Environment, 30, 8994.
Berg, C. P., Matson, M., and Wiesnet, D. R., 1981, Assessing the Red River of the North
1978 ooding from NOAA satellite data. In Satellite Hydrology, edited byM. Deutsch,
D. R. Wiesnet and A. Rango (Minneapolis, MN: American Water Resources
Association), pp. 309315.
Blasco, F., Bellan, M. F., and Chaudhury, M. U., 1992, Estimating the extent of oods in
Bangladesh using SPOT data. Remote Sensing of Environment, 39, 167178.
Bonansea, E., 1995,Mapping of 1994 oods in Piemonte Region, Italy: an example of remote
sensing and GIS application. Proceedings of First ERS T hematic Working Group
Meeting on FloodMonitoring, ESA-ESRIN, Frascati, Italy, 2627 June 1995 (Frascati,
Italy: ESA-ESRIN).
Brakenbridge, G. R., Tracy, B. T., and Knox, J. C., 1998, Orbital SAR remote sensing of
a river ood wave. International Journal of Remote Sensing, 19, 14391445.
Brivio, P. A., Gregoire, J. M., and Zilioli, E., 1984, The detection of hydrological indicators
in the study of Niger River regime by means of Landsat imageries. ITC Journal,
3, 191199.
Calabresi, G., 1995, The use of ERS data for ood monitoring: an overall assessment.
Proceedings of the 2nd ERS Applications Workshop, L ondon, UK, 68 December 1995,
SP-383 (Noordwijk, The Netherlands: ESA-ESTEC), pp. 237241.
Douglas, D. H., 1994, Least-cost-path in GIS using an accumulated cost surface and slope
lines. Cartographica, 31, 3751.
ESRI, Environmental System Research Institute Inc., 1992, ARC/INFO Users Guide
6.1GRID Command References. Environmental System Research Institute Inc.,
Redlands, California, USA.
Giacomelli, A., and Mancini, M., 1996, Integrazione di immagini satellitari ERS-1 PRI e
modelli digitali del terreno per lindividuazione di aree esondate. Proceedings of XXV
Convegno di Idraulica e Costruzioni Idrauliche, T orino, Italy, 1618 September 1996,
Vol. I (Torino, Italy: MAF Servizi), pp. 182192.
Henebry, G. M., 1997, Advantages of principal components analysis for land cover segmentation
from SAR images series. Proceedings of T hirs ERS Symposium on Space at the
Service of our Environment, Florence, Italy, 1421March 1997, SP-414, Vol. I
(Noordwijk, The Netherlands: ESA-ESTEC), pp. 175178.
Hess, L. L., Melack, J. M., Filoso, S., and Wang, Y., 1995, Delineation of inundated areas
and vegetation along the Amazon oodplain with the SIR-C SyntheticAperture Radar.
IEEE T ransactions on Geoscience and Remote Sensing, 33, 896904.
Hutchinson, M. F., 1989, A new procedure for gridding elevation and stream line data with
automatic removal of spurious pits. Journal of Hydrology, 106, 211232.
Imhoff, M. L., Vermillion, C., Story, M. H., Choudhury, A. M., Gafoor, A., and Polcyn,
F., 1987, Monsoon ood boundary delineation and damage assessment using space
borne imaging radar and Landsat data. Photogrammetric Engineering and Remote
Sensing, 53, 405413.
Laugier, O., Fellah, K., Tholey, N., Meyer, C., and De Fraipont, P., 1997, High temporal
detection and monitoring of ood zone dynamic using ERS data around catastrophic
natural events: the 1993 and 1994 Camargue ood events. Proceedings of T hird ERS
Symposium on Space at the Service of our Environment, Florence, Italy, 1421March
1997, SP-414, Vol. I (Noordwijk, The Netherlands: ESA-ESTEC), pp. 559564.
Lougeay, R., Baumann, P., and Nellis, M. D., 1994, Two digital approaches for calculating
the area of regions aVected by the great American ood of 1993. Geocarto International,
9, 5359.
Luino, F., 1999, The ood and landslide event of November 46 1994 in Piedmont Region
(Northwestern Italy): causes and related eVects in Tanaro Valley. Proceedings of XXII
General AssemblyEuropean Geophysical Society, V ienna, 2125 April 1997, Vol. 24,
N. 2 (Amsterdam: Elsevier), pp. 123129.
Michener, W. K., and Houhoulis, P. F., 1997, Detection of vegetation changes associated
with extensive ooding in a forested ecosystem. Photogrammetric Engineering and
Remote Sensing, 63, 13631374.
Nico, G., Pappalepore, M., Pasquariello, G., Refice, A., and Samarelli, S., 2000,
Comparison of SAR amplitude vs. coherence ood detection methodsa GIS
application. International Journal of Remote Sensing, 21, 16191631.
Noyelle, J., Demeire, S., and Marinelli, L., 1995, Identi. cation of ooded areas in the Rhone
and Var river, France, using ERS-1 data. Proceedings of First ERS T hematic Working
Group Meeting on Flood Monitoring, ESA-ESRIN, Frascati, Italy, 2627 June 1995
(Frascati, Italy: ESA-ESRIN), pp. 7080.
Ober, G., Cella, F., Meroni, F., Tomasoni, R., and Zonno, G., 1998, Characterisation of
urban areas through texture analysis. Proceedings SPIE Earth Surface Remote Sensing
II, Vol. 3496, edited by G. Cecchi and E. Zilioli (Bellingham, Washington: SPIE),
pp. 94101.
Oberstadler, R., Honsch, H., and Hutch, D., 1995, Assessment of the mapping capabilities
of ERS-1 SAR data for ood mapping: a case study in Germany. Proceedings of the
2ndERS ApplicationsWorkshop, L ondon,UK, 68 December 1995, SP-383 (Noordwijk,
The Netherlands: ESA-ESTEC), pp. 247255.
Ormsby, J. P, Blanchard, B. J., and Blanchard, A. J., 1985, Detection of lowland ooding
using active microwave systems. Photogrammetric Engineering and Remote Sensing,
51, 317328.
Profeti, G., and Macintosh, H., 1997, Flood management through Landsat-TM and ERS
SAR data: a case study. Hydrological Processes, 11, 13971408.

Robin, M., and Le Rhun, J., 1989, Linodation de Fevrier 1988 dans le marais de Monts
(Vendee) a` partir dune image Spot-1 multibande. Photo-interpretation, 1, (2), 916.
Rosso, C., 1995, Analisi di dati LandsatTM per lo studio e la classi. cazione di aree alluvionate.
Un esempio sul bacino del Tanaro. Telerilevamento, GIS e cartogra. a al servizio
dellinformazione territoriale. Proceedings of VIII Convegno Nazionale AIT , Chieri
(To), Italy, 1720 October 1995 (Torino, Italy: CSEA Bonafus), pp. 636643.
Sharma, P. K., Chopra, R., Verma, V. K., and Thomas, A., 1996, Flood management using
remote technology: the Punjab (India) experience. International Journal of Remote
Sensing, 17, 35113521.
Takeuchi, S., Konishi, T., Suga, Y., and Kishi, S., 1999, Comparative study for ood
detection using JERS-1 SAR and Landsat TM data. Proceedings of IGARSS99,
Hamburg, Germany, 28 June2 July 1999 (Piscataway, NJ: IEEE), pp. 873875.
Tanaka, M., Sugimura, T., and Tanaka, S., 2000, Monitoring water surface ratio in the
Chinese oods of summer 1998 by DMSP-SSM/I. International Journal of Remote
Sensing, 21, 15611569.
Wang, Y., Koopmans, B. N., and Pohl, C., 1995, The 1995 ood in The Netherlandsmonitored
from spacea multisensor approach. International Journal of Remote Sensing, 16,
27352739.
Zhengao, S., and Fung, K. B., 1994, A comparison of digital speckle . lters. Proceedings of
IGARSS94, Pasadena, California, USA, 812 August 1994, Vol. IV (Piscataway, NJ:
IEEE), pp. 21292133.
Zilioli, E., Cerutti, P., Colombo, R., Dessena, M. A., and Fussi, F., 1993, Studio geomorfologico
idrologico del delta interno del Fiume Niger (Mali ). Rivista Italiana di
T elerilevamento (AIT ), 1, 4348.
The nanolimes in Cultural Heritage conservation:

Characterisation and analysis of the carbonatation
process
ARTICLE in JOURNAL OF CULTURAL HERITAGE JULY 2008
Impact Factor: 1.57 DOI: 10.1016/j.culher.2007.10.007
CITATIONS
43
READS
163
3 AUTHORS:
Valeria Daniele
Universit degli Studi dell'Aquila
SEE PROFILE
Giuliana Taglieri
SEE PROFILE
Raimondo Quaresima
SEE PROFILE

Available from: Giuliana Taglieri
Original article
The nanolimes in Cultural Heritage conservation:

Characterisation
and analysis of the carbonatation process
Valeria Daniele, Giuliana Taglieri*, Raimondo Quaresima
Department of Chemistry, Chemical Engineering and Materials, University of LAquila, Monteluco di Roio, I-67040
LAquila, Italy
Received 28 June 2007; accepted 18 October 2007
Abstract
Water and milk of lime are usually adopted for conservative surfaces treatments, thanks to the
conversion of lime into calcium carbonate.
Calcium carbonate is, as a matter of fact, very compatible with many carbonatic lithotypes and
architectonic surfaces, because its characteristics
are very similar to those of the materials to be restored. But there are some limiting aspects to
treatments effectiveness: the reduced penetration
depth, the binder concentration and the incompleteness carbonatation process. In order to improve
lime treatments, Ca(OH)2 particles with
submicrometric dimensions (nanolimes) are recently introduced in Cultural Heritage conservation.
Lime nanoparticles are typically produced
by a chemical precipitation process in supersaturated aqueous solutions of the reactants (calcium
chloride and sodium hydroxide). The aim
of the present work is to analyse the nanolime carbonatation process in relation to some parameters,
like time and the relative humidity
conditions. For this scope, lime nanoparticles are therefore synthesised and characterised by X-ray
diffraction (XRD), scanning and transmission
electron microscopy (SEMeTEM), electron diffraction measurements (ED) and dark field images (DFI).
The possibility to improve the nanolime
carbonatation process is investigated using an alcoholic suspension and by adding a baking soda
solution in order to disaggregate particles
and to increase CO2 content in the suspension respectively. The efficiency of the nanolime
carbonatation process is reported too.
After that the lime nanoparticles are applied on natural lithotypes (Estoril and Pietra Serena) and
some tests are performed in order to
estimate the superficial consolidating and protective effect of the treatment: Scotch Tape Test,
capillarity and imbibition tests. SEM analyses
are performed to evaluate penetration depth and surface adhesion of nanolime treatments.
_ 2008 Elsevier Masson SAS. All rights reserved.
Keywords: Calcium hydroxide; Consolidation; Lime; Nanoparticles; Protection
1. Introduction
The use of lime (Ca(OH)2), in building industry as in Cultural
Heritage conservation, is based on the well-known carbonation
reaction and on the characteristics of calcium carbonate
(CaCO3) obtained. The low solubility and the compatibility
between the latter compound and material substrates offer a favourable
use in many lime-based conservative treatments (preconsolidation
[1], cleaning [2], consolidation and protection [3]).
The applicationsemploy lime solutions (lime milk or limewater).
Lime water consolidation is generally obtained by spraying
the lime solution on the cleaned surface. To reach a good
penetration, the treatment is repeated several times until the surface
is able to absorb limewater [4]; some authors indicate that it
could be necessary to repeat the application for 30e40 times [5].
Lime milk is used on the same basis as lime water [6]. Compared
with limewater, lime milk treatments involve greater amounts of
lime with the same water volume; this represents an advantage
due to a reduced water percentage brought to the stone.
Lime water and lime milk treatments are characterised by
some limitations due to:
- the incomplete conversion of lime into calcium carbonate
that leaves free particles on the surfaces [6];
- the binder concentration due to the low water solubility of
lime, giving chromatic alteration to stone surfaces;
- a reduced penetration depth [4].
* Corresponding author. Tel.: 39 0862 434212; fax: 39 0862 434203.
E-mail address: taglieri@ing.univaq.it (G. Taglieri).
1296-2074/$ - see front matter _ 2008 Elsevier Masson SAS. All rights reserved.
doi:10.1016/j.culher.2007.10.007
Available online at www.sciencedirect.com
Journal of Cultural Heritage 9 (2008) 294e301

http://france.elsevier.com/direct/CULHER/
In order to obviate to these limitations, Ca(OH)2 particles

with submicrometric dimensions (nanolimes) are synthesised
[7e11]; the innovative nanolime treatments constitute a valid
alternative to the lime traditional ones [12e18]. The particles
are synthesised at high temperature in supersaturated aqueous
solutions of reactants; the precipitate phase is then dispersed
into an alcoholic medium to improve particles disagglomeration
and stability [19].
Concerning the above limitation due to the conversion of
lime, this paper could represent a contribution in analysing
the carbonatation process and in finding a way to promote
the completeness of the reaction itself.
According to the mentioned literature, a 2-propanol alcoholic
nanolime suspension is prepared [20]. The conversion
of lime into calcium carbonate is estimated by taking into account
parameters like the time you need to complete the reaction,
the dispersing medium (water or alcohol), and the relative
humidity conditions. To complete the carbonatation reaction
of the alcoholic nanolime produced, the addition of a baking
soda (NaHCO3) solution is considered. The completeness of
the carbonatation reaction is, as a matter of fact, a relevant issue
in order to enhance the effectiveness of the application of
a pure, crystalline and nanometric lime to be used in Cultural
Heritage applications.
To correlate the nanolime produced to its properties, morphological
and microstructural characterisation are performed
by X-ray diffraction (XRD) and profile analysis, electron
microscopy (SEM and TEM), electron diffraction (ED) and
dark field images (DFI).
The produced lime nanoparticles are applied on two natural
lithotypes (Estoril, Pietra Serena) in order to evaluate the consolidating
and protective effectiveness of the treatment, and
some tests are performed: Scotch Tape Test [21], capillarity
[22] and imbibition [23] tests. With scanning electron microscopy
(SEM), the penetration depth of the treatment is evaluated
too.
Fig. 1. TEM micrograph of the lime nanoparticles agglomerate (WS sample).
Fig. 2. (a) TEM image of a group of nanolime particles (WS sample); (b) electron diffraction image on A particle; (c
and d) TEM dark field images for two different
crystalline planes orientation of A particle.
V. Daniele et al. / Journal of Cultural Heritage 9 (2008) 294e301 295
2. Experimental
2.1. Materials
Calcium chloride (CaCl2), sodium hydroxide (NaOH), sodium
bicarbonate (NaHCO3) and 2-propanol pro analysis
products, supplied by Merck, are used without further purification.
Water is purified by a Millipore Organex system
(R _ 18 MU cm).
2.2. Synthesis of the particles
To obtain about 20 g of Ca(OH)2 nanoparticles, two different
aqueous solutions of 900 ml, containing 0.3 mol/l of CaCl 2
and 0.6 mol/l of NaOH, respectively, are prepared. The NaOH
alkaline solution (used as precipitator) is added dropwise into
the CaCl2 solution (speedz4 ml/min, temperature of 90 _C).
After about 24 h, two distinct phases are observed: a limpid

supernatant solution and a white precipitated phase. In order
to remove the produced NaCl, several deionised water washings
are performed; the chloride content is measured by means
of a UV spectrophotometer (Dr. Lange Mod. CASAS 50).
Aqueous nanolime suspension (with a concentration of
15 mg/ml) is defined as sample WS.
An alcoholic nanolime suspension is prepared in rotavapor
at a temperature of 80 _C at low pressure (100 mbar). When
a ratio of Ca(OH)2/H2Oz0.6 is reached, 1000 ml of
Fig. 3. (a) TEM micrograph of a small single particle (A 0); (b) electron diffraction images on A0 particle; (c) A0 particle
dark field image.
Fig. 4. SEM micrograph of the aqueous nanolime suspension (sample WS).
25 30 35 40 45
2-theta
Intensity (a.u.)
Ca(OH)2
* CaCO3
*
***
Fig. 5. XRD pattern of the aqueous nanolime suspension (sample WS, concentration
of 15 mg/ml) after 30 min of air exposition time.
296 V. Daniele et al. / Journal of Cultural Heritage 9 (2008) 294e301
2-propanol is added in order to obtain a final suspension concentration

of 15 mg/ml (sample ALS).
Sample ALS is then exposed, at room temperature, to different
relative humidity conditions (RH 40%, 70%, 90%), depositing
0.2 ml of suspension on a silica sample holder in
a climatic chamber.
ALBKS sample is prepared by adding to ALS sample an
aqueous solution (0.16 mol/l) of NaHCO3, corresponding to
a concentration of 15 mg/ml. In a preliminary instance, the ratio
between lime and baking soda solution is considered 1:1;
this addiction doesnt seem to alter the dispersion stability.
2.3. Morphological and microstructural characterisation
To correlate the characteristics of the obtained nanolime to
the carbonatation process, the suspensions are characterised by
TEM (Philips CM200) and SEM (Philips XL30CP) techniques.
In order to analyse morphology, particles dimensions
and crystallinity, 0.2 ml of WS sample is deposited on a grid.
The carbonatation process in air is investigated by XRD
(Philips XPert PW 1830). XRD measurements allow, as a matter
of fact, to determine the crystalline phases (data from
JCPDS), to estimate crystallites dimension (D) (Scherrer equation),
and to understand the completeness of the carbonatation
reaction. The sample is prepared by depositing 0.2 ml of the
nanolime suspension on a silica sample holder; the measures
are performed on dry samples. In particular, WS and ALS
samples are left in the laboratory environmental conditions
(T 20 _C and RH 40%), and investigated after the following
times: 30 min, 1 day and 24 h, 30 days. After that the samples
are left 24 h in relative humidity (RH) conditions of 70%
and 90% (ALS70 and ALS90, respectively), and tests are
performed in order to quantify the humidity influence on nanolime
carbonatation process.
Each experimental diffraction spectrum is elaborated by
a Profile Fit Software (Philips PROFIT v.1.0). The ratio between

the CaCO3 peak area and the spectrum total area is assumed
as the carbonatation process efficiency (yield); the ratio
between CaCO3 and Ca(OH)2 areas is indicated as
ICaCO3=CaOH2
. All peak profile fittings are carried out retaining
the Ka2 wavelength.
*
*
*
*
*
25 30 35 40 45
2-theta
Ca(OH)2
* CaCO3
Intensity (a.u.)
Fig. 6. XRD pattern of the alcoholic nanolime suspension (sample ALS).
Table 1
Relative humidity influence on the carbonatation process
Sample RH (%) Yield (%)
ALS 40 65
ALS70 70 80
ALS90 90 93
*
***
Intensity (a.u.)
25 30 35 40 45
2-theta
Ca(OH)2
* CaCO3
Fig. 7. XRD pattern of the alcoholic nanolime suspension mixed with the baking
soda solution (sample ALBKS).
Fig. 8. Porosimetric distribution of: (a) Estoril; (b) Pietra Serena.
3. Results and discussion

TEM micrographs, obtained on WS sample, are reported.
Fig. 1 shows a typical Ca(OH)2 nanoparticles agglomerate,
where the particles range from 50 to 600 nm; in Fig. 2a a particle
with a hexagonal and a regular shape is shown (particle
A), while a smaller irregular single particle is reported in
Fig. 3a (particle A0). As concerns their crystallinity, these particles
reveal typically crystalline features (circular symmetrylike),
with spots and few marked circles (Fig. 2b) or with
a spot size distribution (Fig. 3b). The difference can be related
to the particles dimension: smaller (<100 nm) or greater
(600 nm) particles seem to reveal a single or several crystallites,
respectively.
Dark field images confirm these considerations. By tilting
the incident beam, regions corresponding to crystalline grains
with planes oriented along the chosen direction become light;
thus, the variation or not in bright marks the presence of different
crystallites (Fig. 2c and d) or of a single crystal
(Fig. 3c).
SEM micrograph (Fig. 4) shows the effect of the carbonatation
process on Ca(OH)2 particles in WS samples; it is
possible to observe only a partial growth of CaCO3 crystallites
on lime particles surfaces. The partial CaCO3 growth can be
a result of the incomplete carbonatation process and/or of

the SEM technique itself. As a matter of fact, the liquid phase
is quickly eliminated during the specimen gold sputtering
(metallization), realised under vacuum conditions.
XRD measurement on WS sample, performed after 30 min,
is reported in Fig. 5. It is possible to recognize Ca(OH)2 and
CaCO3 phases (84-1276 and 85-1108 JCPDS patterns, respectively).
The parameter ICaCO3=CaOH2
is 1.2. For this sample, I
ratio remains the same also for the spectra recorded after 1 h,
24 h and 30 days: the carbonatation process, in the environmental
experimental conditions, stops after 30 min. The corresponding
WS yield is about 50%.
XRD pattern of ALS sample, still obtained after 30 min, is
shown in Fig. 6. It is possible to note a light improvement in
the carbonatation process corresponding to an ICaCO3=CaOH2
of
1.9 and a yield of 65%. In Table 1 the influence of the relative
humidity conditions in the carbonatation process is shown. It
is possible to note the expected improvement of the yield,
but the completeness of the reaction can be reached only at
the limiting conditions.
In Fig. 7 the XRD result, obtained on ALBKS sample after
30 min of air exposure time, is shown. In this case the main
crystalline phase is CaCO3 and Ca(OH)2. A very weak signal
at about 32.3_ could be attributed to the presence of Na2CO3$2H2O (08-0448 JCPDS pattern) as a reaction product.
In this case the ICaCO3=CaOH2
ratio is 18.7 while the yield reaches
the value of 95% already in the first 30 min of air exposure
time.
Table 2
Porosimetric data
Lithotypes Total pore
volume (mm3/g)
Average pore
radius (mm)
Total porosity (%)
Estoril 61.54 0.46 14.95
Pietra Serena 34.43 6.15 7.05
Fig. 9. SEM micrographs of Estoril sample: (a) untreated material surface; (b) material surface after nanolime
treatment; (c) penetration depth of the treatment in
a typical cross-section.
4. Nanolimes treatments on natural stones

The obtained nanolimes, together with the baking soda solution,
are then applied on some natural stones (Estoril and Pietra
Serena), in order to evaluate the protective effectiveness of the
treatment. The typical porosimetric distribution graphs, obtained
by a Porosimeter 2000 Series e Carlo Erba, are shown
in Fig. 8, and in Table 2 the quantitative results are summarised.
The pore size distribution data relative to Estoril sample exhibit
an average pore radius of 0.46 mm and a total porosity of about
15%. On the contrary, Pietra Serena lithotype appears very different
as in the pore size distribution as in porosity.
The nanolime treatment is realised, with a brush, by applying
the alcoholic suspension on the dry and clean stone surface.
SEM analyses are performed in order to evaluate the
reached penetration depth and the surface adhesion of nanolime

treatments.
To investigate the superficial consolidation of the treated
stones, a Scotch Tape Test (STT) is performed. Capillarity
and imbibition tests, according to literature procedures, are finally
executed, before and after the treatment, to understand
the material behaviour towards water.
In Fig. 9 SEM micrographs of Estoril sample are reported: it
is possible to note the partial filling of the pores due to the treatment
(Fig. 9b) and a penetration depth of about 30 mm (Fig. 9c).
SEM analysis on the Pietra Serena stone is reported in
Fig. 10. In particular, Fig. 10a shows that the nanolime treatment
seems to reach a penetration depth of about 1 mm. This
result could be attributed to the pore size distribution and the
high porosity value of this lithotype (see Table 2). In Fig. 10b
the adhesion of the treatment on grains surface is reported too.
In Table 3 the STT results, in terms of the materials removed
from the surface before and after the treatment, are reported;
the percentage variation is indicated as DV. The
treatment seems to be effective, relative to the superficial consolidation
properties of the considered materials. The best
Table 3
Scotch Tape Test (STT): experimental results
Lithotypes Untreated materials
(mg/cm2)
Treated materials
(mg/cm2)
DV (%)
Estoril 0.54 0.15 _70%
Pietra Serena 0.60 0.40 _35%
Fig. 10. SEM micrographs of Pietra Serena stone: (a) penetration depth; (b)
adhesion of the treatment on grains surface.
Fig. 11. Capillarity curves: untreated stones (continued line), treated stones
(dots line).
result is obtained for Estoril (DVz70%), while Pietra Serena

reached a DV value of about 35%.
Capillarity and imbibition tests are reported in Fig. 11 and
Fig. 12, respectively. The treatment effectiveness on water absorption
in both the samples is evident. Results are particularly
interesting for Pietra Serena stone, probably due to a partial
filling of macropores and their transformation into smaller
pores, as shown in SEM investigation.
5. Conclusions
In this work the carbonatation of Ca(OH)2 nanoparticles,
synthesised starting from stoichiometric ratio of the initial reactants
at a synthesis temperature of 90 _C, is analysed. The
morphological and microstructural characterisations show
that crystalline and regularly shaped particles, with dimensions
ranging between 50 and 600 nm, are obtained.
An aqueous nanolime suspension of 15 mg/ml concentration
stops its carbonatation within 30 min, reaching a carbonatation
yield of about 50%, while the use of 2-propanol alcohol,
as de-agglomerating medium, produces a yield of 65%.
An attempt to complete the carbonation process by adding
an external CO2 source is then realised using a baking soda solution
(NaHCO3) as a promoter of the process itself: a conversion
factor of 95% is so reached.
After that the lime nanoparticles, together with the baking

soda solution, are applied on two natural stones (Estoril and
Pietra Serena) and three different tests seem to confirm the
protective and consolidating effectiveness of the treatment:
Scotch Tape Test, capillarity and imbibition tests. With
scanning electron microscopy (SEM), the penetration depth
of the treatment is evaluated too.
Further and more complete study on better promoter agents
should be carried out in order to avoid surfaces alteration on
treated stones. As a matter of fact, the introduction of sodium
salt within the porous structure of the stones can produce efflorescence
phenomena.
Acknowledgments
The authors thank Eng. Giovanna Di Tommaso and Dr.
Lorenzo Arrizza and Dr. Maria Gianmatteo, of the InterDepartmental
Centre of Electron Microscopy of the University
of LAquila, for their collaboration. A special thanks is directed,
last but not least, to Prof. Luigi Dei (University of Florence)
for his very kind and helpful cooperation.
References
[1] G.G. Amoroso, M. Camaiti, Trattamento alla calce, in: Scienza dei Materiali
e restauro. La pietra: dalle mani degli artisti e degli scalpellini
a quelle dei chimici macromolecolari, Alinea Editrice, Firenze, 1997,
pp. 165e167.
[2] N. Ashurst, Cleaning Historic Buildings, Donhead, London, 1994.
[3] C. Arcolao, Trattamento allacqua di calce, In: Tecniche di restauro architettonico,
tomo II, P.B. Torsello, UTET, Torino, 1996, 698e699.
[4] M.G. Vinardi, M.H. Cully, A. Brunetto, La reversibilita` nel restauro. Riflessioni,
esperienze, percorsi di ricerca, In: XI National Congress Scienza
e Beni Culturali: La Pulitura delle Superfici dellArchitettura,
Bressanone, 1e4 Luglio, vol. XIX (2003) 399e405.
[5] P. Sten, Lime Water Consolidation, ICCROM, 1981, pp. 53e61.
[6] R. Turriziani, I leganti ed il calcestruzzo, Edizioni Sistema, Roma, 1972,
pp 70e82.
[7] R. Giorgi, L. Dei, P. Baglioni, A new method for consolidating wall
paintings based on dispersion of lime in alcohol, Studies in Conservation
45 (2000) 154e161.
[8] B. Salvadori, L. Dei, Synthesis of Ca(OH)2 nanoparticles from diols,
Langmuir 17 (2001) 2371e2374.
[9] P. Baglioni, C. Cesari, L. Dei, R. Giorgi, R. Grassi, M. Lorenzetti,
M. Mauro, D. Pinzani, P. Ruffo, G. Schonhaut, Stucco forte veneziano
con Tecniche della Tradizione: proprieta` chimico-fisiche, degrado in
ambiente lagunare e conservazione mediante nanofasi cristalline di
Ca(OH)2, In: National Congress Scienza e Beni Culturali: Lo Stucco.
Cultura, Tecnologia, Conoscenza, Bressanone, 10e13 Luglio, vol.
XVII (2001) 289e298.
Fig. 12. Imbibition curves: untreated stones (continued line), treated stones
(dots line).
[10] M. Ambrosi, L. Dei, R. Giorgi, C. Neto, P. Baglioni, Colloidal particles
of Ca(OH)2: properties and applications to restoration of frescoes, Langmuir
17 (2001) 4251e4255.
[11] A. Nanni, L. Dei, Ca(OH)2 nanoparticles from W/O microemulsions,
Langmuir 19 (2003) 933e938.
[12] M. Ambrosi, L. Dei, R. Giorgi, C. Neto, P. Baglioni, Stable dispersions of
Ca(OH)2 in aliphatic alcohols: properties and application in cultural heritage
conservation, Progress in Colloid and Polymer Science 118 (2001)
68e72.
[13] R. Giorgi, L. Dei, M. Ceccato, C. Schettino, P. Baglioni, Nanotechnologies
for conservation of cultural heritage: paper and canvas deacidification,
Langmuir 18 (2002) 8198e8203.
[14] M. Ambrosi, P. Baglioni, P.R. David, L. Dei, R. Giorgi, C. Lalli,
G. Lanterna, A. Mairani, M. Matteini, M. Rizzi, G. Schonhaut, Inorganic
Consolidants and Protectives for Architectonic Surfaces: Experimental
Tests on Santa Prisca Church Apse in Rome, In: Second International

Congress on Science and Technology for the safeguard of Cultural Heritage
in the Mediterranean Basin, vol. 2 (1992) 873e877.
[15] L. Dei, B. Salvadori, E. Arlango, F. Pietropoli, C. Scardellato, Gli affreschi
del XIII e XIV secolo nella Cripta di San Zeno a Verona: la sperimentazione
della nanocalce dispersa in alcol iso-propilico durante
lintervento conservative, In: International Congress Scienza e Beni Culturali:
Sulle pitture murali. Riflessioni, conoscenze, interventi, Bressanone,
12e15 Luglio, vol. XXI (2005) 293e302.
[16] L. Dei, B. Radicati, B. Salvadori, Sperimentazione di un consolidante
a base di idrossido di calcio nanofasico sugli affreschi della Cappella
del Podesta` al Museo del Bargello di Firenze: aspetti chimico-fisici e
prove di colore, In: International Congress Scienza e Beni Culturali:
Sulle pitture murali. Riflessioni, conoscenze, interventi; Bressanone,
12-15 Luglio, vol. XXI (2005) 99e108.
[17] L. Dei, M. Favaro, R. Giorgi, R. Portieri, Gli stucchi della Cappella della
Madonna di Lourdes nella Chiesa di S.Giovanni Evangelista in Venezia:
caratterizzazione del degrado e pre-consolidamento mediante dispersioni
alcoliche di calce, , In: National Congress Scienza e Beni Culturali: Lo
Stucco. Cultura, Tecnologia, Conoscenza; Bressanone, 10e13 Luglio,
vol. XVII (2001) 557e566.
[18] P. Baglioni, R. Giorgi, Soft and hard nanomaterials for restoration and
conservation of cultural heritage, Soft Matter 2 (2006) 293e303.
[19] M. Ambrosi, R. Giorgi, B. Salvadori, L. Dei, Nanotecnologie per la conservazione
di affreschi e materiali lapidei a base carbonatica, in: First International
Workshop on Science, Technology and Cultural Heritage,
Venezia, 29 Giugnoe1 Luglio (2004), pp. 29e33.
[20] R.T. Morrison, R.N. Boyd, Chimica organica, Casa Editrice Ambrosiana,
1997, p. 243.
[21] ASTM D3359-02: Standard Test Methods for Measuring Adhesion by
Tape Test, ASTM International, 10 August, 2002.
[22] Commissione Beni Culturali e NORMAL: Materiali lapidei naturali
ed artificiali. Determinazione dellassorbimento dacqua per capillarita`
(2000) UNI 10859.
[23] Commissione Beni Culturali e NORMAL: Materiali lapidei naturali
ed artificiali. Determinazione dellassorbimento dacqua per immersione
totale e capacita` di imbibizione (1981) NORMAL 7/81.
Water Resources Management (2005) 19: 333347

DOI: 10.1007/s11269-005-3281-5 C _ Springer 2005
Delineation of Flood-Prone Areas Using Remote

Sensing Techniques
SANJAY K. JAIN, R. D. SINGH, M. K. JAIN and A. K. LOHANI
Remote Sensing Application Division, National Institute of Hydrology, Bhawan-Roorkee,

Jalvigyan 247 667, India
(author for correspondence, e-mail: sjain@nih.ernet.in)
(Received: 15 April 2004; in final form: 7 September 2004)
Abstract. Flood problems resulting due to heavy rainfall and drainage congestion are being regularly
experienced in plain areas of Bihar, India. Due to this problem, power plant located in Koa catchment,
Kahalgaon, Bihar, is faces huge loss at the time of flood. In this paper, the flood-affected areas
in Koa catchment have been mapped using remote sensing satellite data (IRS LISS III, 1999 and
Landsat TM, 1995). A range of image processing techniques has been used, including simple density
slicing, Tasseled Cap Transformation and water-specific index. The results obtained using different
approaches have been analysed. Result indicates that a Normalized Difference Water Index (NDWI)
based approach produced best results.
Key words: catchment, flood, rainfall, remote sensing, water
1. Introduction
Floods are the major disaster affecting many countries in the world year after year.
It is an inevitable natural phenomenon occurring from time to time in all rivers
and natural drainage systems that not only damages the lives, natural resources
and
environment, but also causes the loss of economy and health. In India also, floods
are
most frequent mainly in the eastern part (Orissa, Bengal, Andhra Pradesh, Bihar).
The annual precipitation in India, including snowfall is estimated at 4000 billion
m3 (BCM). Out of this, the seasonal rainfall in monsoon is of the order of 3000
BCM. The rainfall in India shows great temporal and spatial variations, unequal
seasonal distribution and geographical distribution and frequent departures from
the normal (Mohapatra and Singh, 2003).
Accurate information on the extent of water bodies is important for flood prediction,
monitoring, and relief (Smith, 1997; Baumann, 1999). Often, this information
is difficult to produce using traditional survey techniques because water bodies can
be fast moving as in floods, tides, and storm surges or may be inaccessible. The
synoptic, repetitive nature of satellite remotely sensed data have allowed
monitoring
of water bodies over large regions of land. In the optical range (visible-infrared)
water has a distinctively low spectral response. In many studies, Landsat data (TM
334 S. K. JAIN ET AL.
and MSS), IRS or SPOT have been used to determine the extent of water bodies
using simple classification procedures, usually with an infrared band. These studies
have relied on the water bodies having a unique spectral response in this range
of electromagnetic radiation when compared to the surrounding landscape (Wang,
2002; Frazier and Page, 2000). Many studies using density slicing of Landsat MSS
band 7 (Bennett, 1987), TM bands 4 and 7 (Wang, 2002; Frazier and Page, 2000)
have been reported in literature. Manavalan et al. (1993) used Landsat TM band 4
to map the extent of the Bhadra reservoir, India. Overton, 1997 used density slicing
of Landsat TM band 5 and a high flood spatial mask to map water bodies on the
Murray river between Blanchetown and Wentworth, South Australia. Shaikh et al.
(1997) used band 4 of Landsat TM to map flood extent of the Mississippi river
but experienced problems separating water from certain urban features without the
inclusion of an additional band.
The river Ganges as it flows down in the flatter plain of Bihar and further to
Bengal, its flood plain extended much beyond the defined riverbanks. In both banks
of the Ganges exists such land zones, which behaves not only as flood plain but
also as detention basin for up-land flows. Kahalgaon Super Thermal Power Project
(KhSTPP) of NTPC is located near Kahalgaon town, District Bhagalpur of Bihar
State, India. The plant area and its adjoining areas were subjected to severe flood
in August 1993, September 1995 and September 1999. The flood was brought by
Koa river coupled with back water effect due to high flood level of Ganges river
and inundated a vast areas upstream and adjoining the power station (situated in
Koa river basin) and also affected the drainage inside the plant area. The lower
part of Koa basin out falling to Ganges where the Kahalgaon Power plant and
other structures are situated is also in flood plain area. Topographically, the plant
is situated at the mouth of a catchment (Koa nala) near the confluence of Ganga.
Insufficient natural drainage aggravates the flood situation in the catchment of Koa
Nala. In this paper, IRS-1C LISS III and Landsat TM data for mapping of flooded
areas have been used. A digital elevation model (DEM) has also been used.
2. The Study Area and Data Used
The river Koa emerges from the hills at an elevation of about 400 m. The river
in general has very narrow width and traverses about 48 km and then joins river
Ganga. The general slope of the river is very mild except in some upper reaches
of the catchment. The catchment area of Koa river up to Eastern Railway Bridge
is 677.0 km2 and up to Marry Go Round (MGR), 584 km2. The catchment is
bounded between latitude 2500_00__N to 2515_10__N and Longitude 8713_51__E
to 8738_28__E. The catchment area has a shape of key, i.e. wide in the beginning
and
narrow at the mouth. The annual normal rainfall is about 122 cm. On an average, it
rains for about 5060 days in a year. Most of the rain occurs (about 7580%) during
the monsoon season (JuneSeptember). The PublicWorks Department road bridge
is the most downstream bridge in Kao river and is just upstream of confluence of
DELINEATION OF FLOOD PRONE AREAS 335
Koa with Ganges river. The deck level is of the order of elevation about 34.5 m.
The eastern railway bridge is approximately 100.0 upstream of the Public Works
Department road bridge and the clear water way is 100 m. The elevation of railway
track is about 35.50 (NHPC, 1995).
The satellite data used were as follows:
Satellite/sensor Path/row Date of pass
Landsat TM 13943 27 May 1995
18 October 1995
IRS-1C LISS III 10654 March 1999
December 1999
Topographical maps at a scale of 1:50,000 have been used. The area is covered
in topographical maps Nos. 72 O/4, 7, 8 and 12.
3. Methodology
3.1. PREPARATION OF DATABASE
In the present study for preparation of drainage map, contour map, etc., Survey
of India toposheets has been used. The Koa catchment is covered in two
topographical
maps at a scale of 1:50,000. The drainage and contour maps of the Koa
catchment were digitized directly from topographical maps. The drainage map of
the Koa catchment along with MGR and the plant is shown in Figure 1. The data
Figure 1. The study showing MGR and road.

in raster format were converted to vector form using the raster to vector software,
R2V. This software R2V for Windows is a powerful raster to vector conversion
software developed by Able Software Corp. (http://www.ablesoftware.com/r2v).
R2V combines the power of intelligent automatic vectorizing technology with an
easy-to-use, menu-driven, graphical user interface in the Microsoft Windows
environment.
The software converts scanned maps or images to vector formats for
mapping, geographic information systems (GIS), CAD, and scientific computing
applications. The data in vector format are required for providing elevation values
to the contours as depicted in the topographical maps. For generation of the
contour
map, each contour line is selected one by one and the elevation value is assigned.
For area estimation and creation of DEM, etc., the GIS software Integrated Land
andWater Information System (ILWIS)was used. ILWIS is a GIS that integrates image
processing capabilities, tabular databases and conventional GIS characteristics.
This software is developed at International Institute for Geo-Information Science
and Earth Observation (ITC) (www.itc.nl). The drainage network and catchment
boundary created in ILWIS was transferred to ERDAS 8.6 for further image
processing.
The satellite data of the years 1995 and 1999 received in digital format
from the National Remote Sensing Agency (NRSA), Hyderabad, India, was loaded
in ERDAS.
3.2. PROCESSING OF SATELLITE DATA
To study the extent of flood inundation, the analysis of Landsat TM scenes and
IRS-1C LISS III digital data have been performed. For flood mapping, two sets of
the remotely sensed data are required; one set consisting of data acquired during
the
occurrence of the flood. In reality, data availability may cause some compromise,
because it is difficult to obtain cloud free data during floods. Therefore, in this
study, efforts were made to take satellite data just before and after the flood event.
Selection of the dates of the satellite scenes is made based on observed flood
events.
Satellite data as mentioned earlier were geocoded with topographical maps using
ground control points. The Polyconic projection system is used for geocoding the
digital data. The geocoded scenes are masked by boundaries forKoa catchment
digitized
from topographical maps and imported in ERDAS IMAGINE. Some clearly
identifiable features like crossing of rivers, canals, sharp turns in the rivers, roads,
bridges, etc., were located on both map and image and were selected as control
points. About 10 such points were selected for geo-referencing in all the cases.
Now looking at the statistics, some points that generated big errors were deleted
and replaced by other points so as to obtain the better geo-referencing. After
completing
this process, the image was displayed over the map and the superimposition
was compared. The geo-referencing was found to be quite good. Then the
catchment
area was masked. The False Colour Composite (FCC) of Landsat TM for two
seasons, i.e. May and October 1995 are shown in Figures 2 and 3.
Figure 2. False Colour Composite (FCC) of Landsat TM for May 1995.
Figure 3. False Colour Composite (FCC) of Landsat TM for May 1995.

3.3. IDENTIFICATION OF WATER AREAS
Though spectral signatures of water are quite distinct from other land use like
vegetation, built-up area and soil surface, yet identification of water pixels at the
water/soil interface is very difficult and depends on the interpretative ability of
the analyst. Deep water bodies have quite distinct and clear representation in the
imagery. However, very shallow water/turbid water can be mistaken for soil while
saturated soil can be mistaken for water pixels. Secondly, it is also possible that a
pixel, only at the soil/water interface, may represent mixed conditions (some part
as water and other part as soil). Just by looking at the tone and colour of a pixel it is
difficult to tell the difference between suspended sediment and shallow water. After
all, it is basically the same material with the same reflection properties. Therefore,
very shallow water will have the same colour and brightness as very turbid (high
concentration of suspended sediment) water.
There are many possible methods for identifying water versus non-water areas
using satellite data. Initially, supervised and unsupervised classification approaches
were applied, but they did not produce very good results when compared with
original
satellite data. Therefore, the density-slicing approach using different band have
been attempted. Then two approaches, one based on the Tasseled Cap
Transformation
(wetness) and the other based on water index approaches were applied. The
results of the all the methods have been analysed and compared. These methods
are discussed in the following sections.
3.4. DENSITY SLICING OF SINGLE BAND
In many studies, Landsat data (TM and MSS), IRS or SPOT have been used to
determine the extent of water bodies using density-slicing approach, usually with
an infrared band. These studies have relied on the water bodies having a unique
spectral response in this range of electromagnetic radiation when compared to the
surrounding landscape. Baumann (1999) carried out a study using all the bands
of Landsat TM. He concluded that band 4 (NIR) represented the best spectral
band that identifies the flood-inundated area well. Wang et al. (2002) carried out
a study for identifying the water-affected area and he observed that bands 4 and 7
(2.082.35 m) are well suited for the purpose. In this study, initially, six bands of
Landsat TM (except thermal band) and three bands of IRS LISS III (except MIR
band) have been analysed. In Figure 4, six Landsat TM individual band histograms
for the October 1995 scene are presented. It is seen that in bands 13, there is
considerable spectral overlap between the digital values found on the water bodies
and those in the surrounding urban/vegetation areas. The infrared band gives a
much
better representation in the water-related features than the visible bands did. Each
of the visible bands displays a uni-model histogram with no indication of a separate
group of data for water pixel. The infrared bands show a bi-model histogram. The
density slicing ranges for infrared band starts at near zero, meaning thatwater
pixels
are the darkest in the image. In this study, the analysis of bands 4 and 7 of TM data
Figure 4. Individual Landsat TM band histograms October 1995.
and band 3 of LISS III have been made. It was found that the results obtained
using band 7 are not representing water-related features very well when compared
with raw data; therefore, only band 4 has been taken. Initially, near infrared band
(band 4 in TM data and band 3 in IRS LISS III) has been taken for mapping of
flood. As the reflectance of water in near infrared (NIR) band is minimum, the area
of low reflectance has been taken using this approach. Once the representation of
the reflectance values for water and non-water features was clear, a cut-off value of
digital numbers could be determined to separate the two categories. For the
October
TM image the cut-off value was 53, means if the digital number was less than 53,
that pixelwas assigned aswater category. For the May image ofTMdata, the cut-off
value was 60, i.e. if digital number was less than 60, that pixel was classified as
water, otherwise as non-water. In case of IRS LISS III data, the cut-off values for
December 1999 scene was 54 and for March 1999 scene it was 71. NIR band is
able to identify most of the major water features, but there is still some mixing with
the hill shadows, vegetation area, and therefore another method based on Tasseled
Cap Transformation (wetness) has been applied.
3.5. TASSELED CAP TRANSFORMATION
There are numerous methods available for enhancing spectral information content
of
satellite data. The Tasseled Cap Transformation compressed the total information
into three bands: greenness, brightness and wetness. Besides expressing a large
amount of image variability within three bands, tasseled cap bands could be directly
related to physical scene characteristics. A Tasseled Cap Transformation to data
from the Landsat TM has been given (Crist and Cicone, 1984). The coefficients for
wetness functions are:
1234567
0.1509 0.1793 0.3299 0.3406 0.306 0.7112 0.4572
One of the main reasons for supporting the use of the Tasseled Cap Transformation
method against, for example, the principal component technique is that the
coefficients of the transformation are defined a priori. This method was applied on
the Landsat TM data only. The coefficients for IRS LISS III data were not available.
The earlier wetness function was applied and threshold values of this index
52.0 to 34.0 for October and 35.0 to 19.0 for May have been chosen for water
features. This method was good for identification of water-related features when
compared with original satellite data. But it could not eliminate the shadow effects
caused mainly due to forest, etc. To overcome the problem of shadows, a band ratio
technique has been applied.
3.6. NDWI APPROACH
There are numerous vegetation indices developed to estimate vegetation cover with
the remotely sensed imagery. A vegetation index is a number that is generated by
some combination of remote sensing bands. The most common spectral index used
to evaluate vegetation cover is the Normalized DifferenceVegetation Index (NDVI).
McFeeters (1996) developed an index similar to the NDVI, which is called the
Normalized DifferenceWater Index (NDWI). Any sensor having a green band and a
NIR band can provide for this index. TheNDWIwas derived using principles similar
to those that were used to derive the NDVI. The NDWI is calculated as follows:
NDWI = (GREEN NIR)
(GREEN + NIR)
(1)
where GREEN is a band that encompasses reflected green light and NIR represents
reflected near-infrared radiation. The selection of these wavelengths was done to:
(1) maximise the typical reflectance of water features by using green light
wavelengths;
(2) minimise the low reflectance of NIR by water features; and (3) take
advantage of the high reflectance of NIR by terrestrial vegetation and soil features.
When Equation (1) is used to process a multi-spectral satellite image that contains
a reflected visible green band and NIR band, water features have positive values;
while soil and terrestrial vegetation features have zero or negative values, owing to
their typically higher reflectance of NIR than green light. Image processing software
can easily be configured to delete negative values. This effectively eliminates the
terrestrial vegetation and soil information and retains the openwater information for
analysis. The range of NDWI is then from zero to one. Multiplying Equation (1) by
a scale factor (e.g. 255) enhances the resultant image for visual interpretation. Now
in the output obtained water-related features were identified. The cut-off values of
NDWI for Landsat TM were chosen as 21 and above for May, 1995 and 39 and
above for October 1995. For IRS LISS III, the cut-off values for the March scene
was 55 and above while for the December scene it was chosen as 60 and above.
3.7. FLOOD INUNDATION USING DEM
The flood inundation map using DEM based on the river gauge readings before
the flood event was prepared. The river gauge station is near the meeting of Kao
river with Ganga River. The simple threshold used for this purpose was possible
due to the relatively flat terrain (no sink). In the present study, the area inundated
corresponding to gauge level of 33 m were obtained.
4. Results and Discusion
To quantify the flood-affected areas the whole catchment was subdivided in two
areas; one downstream of MGR experiencing maximum flood and other upstream
MGR where flood but most of the areas, which seems flood affected is actually
forest covered/shadow area, affect some area. In Figures 5 and 6 the results of
density slicing for TM band 4 are shown. This band was able to identify major
water bodies, but there is still some mixing with the urban areas, hill shadows,
etc. In fact, upstream of MGR a very little area is affected by flood; therefore,
the flood-affected areas downstream of MGR is only shown in all the figures. The
density slicing of the other infrared bands (5 and 7) of the TM data were also
tried but they did not produce good results. The density slicing of the visible bands
Figure 5. Flood-affected area using density-slicing approach, May 1995.
Figure 6. Flood-affected area using density-slicing approach, October 1995.
343
substantially overestimate the area ofwater on the image, the explanationwas given
earlier. The combination of mixed pixels and high turbidity limits the use of a single
NIR band to classify water bodies. The second approach applied was based on the
Tasseled Cap Transformation. In this method, the wetness index based on certain
coefficient has been computed. Using this approach also, mixing of land use and
overestimation of water area has been observed. It was observed that this method
is
good during post-monsoon season (after floods) when enough water is available but
it is not effective during pre-monsoon season. Also this method could not eliminate
shadow areas. The flood-inundated areas for May 1995 scene is shown in Figure 7
and for October 1995 shown in Figure 8.
In the NDWI approach, the results was better and the mixing of pixels have been
eliminated up to certain extent. Using this approach also, flood-affected areas could
not be delineated clearly because this area is not like water body. In case of IRS
LISSIII data, the water and sediment-laden water both are represented by positive
values of NDWI. While in case of TM data, the values of NDWI are positive for
clear water but the values for sediment-laden water is negative also. For TM data,
the flood-affected areas are shown in Figures 9 and 10 for May 1995 and October
1995, respectively.
Even applying different approaches of image processing, scattered water features
were observed. In case of river gauge/DEM method area submerged withwater
can be obtained clearly, but there are other limitations of the method. First, the
contour
interval available for the study area was at a interval of 20 m which is not quite
DELINEATION OF FLOOD PRONE AREAS
Figure 7. Flood-affected area using Tasseled Cap Transformation May, 1995.
Figure 8. Flood-affected area using Tasseled Cap Transformation October, 1995.

Figure 9. Flood-affected area using NDWI approach May, 1995.
345
Table I. Flood-affected areas using different techniques

Flood-affected area
Ratio of
D/S of MGR total flood
up to outlet Total flood- affected
Satellite/ of the catch- affected area/area
Methodology sensor Date of pass ment (km2) area (km2) D/S MGR
Density
slicing of
single band
Landsat TM May 1995 15.07 75.60 20
October 1995 23.04 105.00 22
IRS LISS III March 1999 8.56 81.30 10.5
December 1999 11.73 42.12 27.8
Tasseled Cap Landsat TM May 1995 8.05 76.32 10.5
Transformation October 1995 28.56 66.51 43
NDWI Landsat TM May 1995 14.41 50.76 28.5
approach October 1995 23.01 41.76 55
IRS-1C March 1999 7.42 19.98 37.1
LISS III December 1999 18.59 35.74 52
Figure 10. Flood-affected area using NDWI approach October, 1995.
good; therefore, the DEM prepared was not of the high quality. Second, the
gauge data available were not of the dates corresponding to the flood event.
After identifying water versus non-water areas on all the images using the
aforementioned
approaches, determination of flood-affected areas were made. The area
affected by flood in these two regions, i.e., upstream of MGR and downstream
of MGR is given in Table I. In the last column, the ratio of flood-affected area
in downstream of MGR with total flood-affected area is given. From this table, it
is observed that the flood-affected area downstream of MGR is maximum using
NDWI approach. Also in percentage it represents the maximum area (more than
50%) for post-monsoon season. It means that using NDWI approach, water-related
feature is best identified. The ratio in two cases, i.e. single band and Tasseled Cap
Transformation for pre-monsoon image is very lowi.e. about 10%. From the figures
obtained using these two methods, it is observed that the area falls in the upper
most
part of the catchment where shadow effect is prominent. In case of wetness-based
approach, the ratio is very low (about 10%) during pre-monsoon season, indicating
poor response.
5. Conclusions
In this study, methods for mapping flood extent through the use of IRS LISS III and
Landsat TM as well as DEM data have been applied. The density-slicing method
was based on a comparison of the reflectance features of the water versus nonwater
targets on a pair of satellite images (one acquired before and the other during the
flood event). All the visible bands proved to be inadequate for successful density
slicing for identification of water pixels. The combination of mixed pixels and high
turbidity limits the use of single band (NIR) to classify water. The other methods
that have been applied include wetness index from Tasseled Cap Transformation
and NDWI approaches.
NDWI-based approach is produces best results for mapping of flood-inundated
areas when verified with original satellite data. Also, this compares well with results
derived from DEM. In this study, there are certain limitations such as the data
availability during flood season due to cloud cover and quality of DEM.
References
Baumann, P., 1999, Flood analysis: 1993 Mississippi Flood, URL: http://www.research.umbc.edu/
tbenjal/baumann/mod2.html.
Bennett, M.W. A., 1987, Rapid monitoring of wetlandwater status using density slicing, in Proceedings
of the 4th Australian Remote Sensing Conference, 1418 September, Adelaide, pp. 682691.
Crist, E. P. and Ciconel, R. C., 1984, A physically based transformation of Thematic Mapper
dataThe TM Tasseled Cap, IEEE Trans. Geosci. Remote Sens. GE-22, 256263.
Frazier, P. and Page, K., 2000, Water body detection and delineation with Landsat TM data,
Photogrametric
Eng. Remote Sens. 66, 12.
McFeeters, S. K., 1996, The use of Normalised Difference Water Index (NDWI) in the delineation
of open water features, Int. J. Remote Sens. 17(7), 14251432.
Mohapatra, P. K. and Singh, R. D., 2003, Flood management in India, Nat. Hazards 28, 131143.
NHPC, 1995, Project Report on Kahalgaon Super Thermal Power Project, a Technical Report
submitted to National Thermal Power Corporation (NTPC) by Naitonal Hydro Power Corporation
(NHPC).
347
Kundu, B. S. and MohiKumar,K. E., 1995, Mapping and management of flood affected areas through
remote sensing. A case study of Sirsa district, Harayana, J. Indian Soc. Remote Sens. 23(3).
Smith, L. C., 1997. Satellite remote sensing of river inundation area, stage, and discharge: A review,
Hydrol. Process. 11, 14271439.
Wang, Y., Colby, J. D. and Mulcahy, K. A., 2002, An efficient method for mapping flood extent in a
coastal flood plain using Landsat TM and DEM data, Int. J. Remote Sens. 23(18), 36813696.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 1, JANUARY 2011 85
Satellite Remote Sensing and

Hydrologic Modeling
for Flood Inundation Mapping in Lake
Victoria
Basin: Implications for Hydrologic
Prediction in Ungauged Basins
Sadiq I. Khan, Yang Hong, Jiahu Wang, Koray K. Yilmaz, Jonathan J. Gourley, Robert F.
Adler,
G. Robert Brakenridge, Fritz Policelli, Shahid Habib, and Daniel Irwin
AbstractFloods are among the most catastrophic natural disasters
around the globe impacting human lives and infrastructure.
Implementation of a flood prediction system can potentially
help mitigate flood-induced hazards. Such a system typically requires
implementation and calibration of a hydrologicmodel using
in situ observations (i.e., rain and stream gauges). Recently, satellite
remote sensing data have emerged as a viable alternative or
supplement to in situ observations due to their availability over
vast ungauged regions. The focus of this study is to integrate the
best available satellite products within a distributed hydrologic
model to characterize the spatial extent of flooding and associated
hazards over sparsely gauged or ungauged basins. We present
a methodology based entirely on satellite remote sensing data
to set up and calibrate a hydrologic model, simulate the spatial
extent of flooding, and evaluate the probability of detecting inundated
areas. A raster-based distributed hydrologicmodel, Coupled

Routing and Excess STorage (CREST), was implemented for the
Nzoia basin, a subbasin of Lake Victoria in Africa. Moderate Resolution
Imaging Spectroradiometer Terra-based and Advanced
Spaceborne Thermal Emission and Reflection Radiometer-based
flood inundation maps were produced over the region and used to
benchmark the distributed hydrologic model simulations of inundation
areas. The analysis showed the value of integrating satellite
data such as precipitation, land cover type, topography, and other
products along with space-based flood inundation extents as inputs
to the distributed hydrologic model.We conclude that the quantification
of flooding spatial extent through optical sensors can help
to calibrate and evaluate hydrologicmodels and, hence, potentially
improve hydrologic prediction and flood management strategies in
ungauged catchments.
Manuscript received November 1, 2009; revised April 21, 2010 and
June 14, 2010; accepted June 24, 2010. Date of publication August 30,
2010; date of current version December 27, 2010. This work was supported
by the National Aeronautics and Space Administration (NASA) Headquarters
under the NASA Earth and Space Science Fellowship Program through
Grant NNX08AX63H and under the NASA Applied Sciences SERVIR Africa
Project.
S. I. Khan, Y. Hong, and J. Wang are with the School of Civil Engineering
and Environmental Sciences, The University of Oklahoma, Norman, OK
73019-1024 USA (e-mail: yanghong@ou.edu).
K. K. Yilmaz, R. F. Adler, F. Policelli, and S. Habib are with the NASA
Goddard Space Flight Center, Greenbelt, MD 20771 USA.
J. J. Gourley is with the NOAA National Severe Storms Laboratory, National
Weather Center, Norman, OK 73072 USA.
G. R. Brakenridge is with the Department of Geography and the Department
of Earth Sciences, Dartmouth College, Hanover, NH 03755 USA.
D. Irwin is with the NASA Marshall Space Flight Center, Huntsville, AL
35811 USA.
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TGRS.2010.2057513
Index TermsFloods, hydrology, optical sensors, rainfall

runoff modeling, satellite precipitation.
I. INTRODUCTION
LOODS are among the most recurring and devastating
natural hazards, impacting human lives and causing severe

economic damage throughout the world. It is understood that
flood risks will not subside in the future, and with the onset
of climate change, flood intensity and frequency will threaten
many regions of the world [1], [2]. The current trend and
future scenarios of flood risks demand accurate spatial and
temporal information on the potential hazards and risks of
floods. Techniques utilizing satellite remote sensing data can
provide objective information that help to detect floods and
to monitor their spatiotemporal evolution [3][5]. For example,
orbital sensors, such as National Aeronautics and Space
Administration (NASA)s Moderate Resolution Imaging Spectroradiometer
(MODIS), provide reliable data to help detect
floods in regions where no other means are available for flood
monitoring [6], [7]. Such data, after certain processing, provide
information on flooding areal extents with global coverage
and frequent (sometimes daily) observations over the region of
interest. To date, satellite observations have become practical
tools for development of cost-effective methods for hydrologic
prediction in poorly gauged or even ungauged basins around

the globe, regardless of the political boundaries. It has been
demonstrated that orbital remote sensing can be used for river
inundation mapping and has the potential to remotely estimate
runoff [7], [8].
The application of satellite imagery for flood mapping began
with the use of Landsat Thematic Mapper and Multispectral
Scanner, the Satellite Pour lObservation de la Terre [9][12],
the Advanced Very High Resolution Radiometer [13][16],
the Advanced Spaceborne Thermal Emission and Reflection
Radiometer (ASTER), MODIS, and Landsat-7 sensors [3],
[4], [17][19]. For a comprehensive review on space-based
observation of flood extent and surface water level from various
satellite sensors, please refer to [4], [5], [11], and [20][24].
Microwave measurements from space can be used for flood
monitoring even when cloud cover is present but the spatial
resolution is relatively coarse at approximately 10 km, such
0196-2892/$26.00 2010 IEEE
86 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 1, JANUARY 2011
Fig. 1. Map showing the Nzoia river basin in the Lake Victoria region, East Africa.
as with Advance Microwave Scanning Radiometer for the

Earth Observing System (AMSR-E) microwave data. Radar
imagery onboard satellites has proved invaluable in mapping
flood extent [25][27]. For example, flood extent maps derived
from Synthetic Aperture Radar (SAR) sensors have been used
to validate hydraulic models [28], [29]. However, limitations
of the SAR include inability to detect flooding in urban areas,
inaccurate image calibration that leads to geometric and
radiometric distortions, difficulties in data processing, and more
prohibitively, low temporal resolution of the current overpasses
with a revisit time of 35 days [27]. Other sensors such as the
Advanced SAR instrument onboard ENVISAT with a spatial
resolution of (1501000 m) and a revisit time of a few days
can be effective for flood detection [24], [27], [28]. Visible and
infrared sensors onboard the NASA MODIS Terra satellite can
detect floods with relatively high spatial (30 m with ASTER
and 250 m with MODIS) and temporal (daily if clear sky)
resolution around the globe. For the past decade, significant
efforts have been made in investigating the potential to use
flood inundation extent derived from optical sensors as a tool
to evaluate the performance of hydrologic models in sparsely
gauged or unguaged basins [6], [7].
This study presents a methodology based entirely on satellite
remote sensing data (including topography, land cover, precipitation,
and flood inundation extent) to calibrate a hydrologic
model, simulate the spatial extent of flooding, and evaluate
the probability of detecting inundated areas. MODIS- and
ASTER-based raster flood inundation maps were derived to
benchmark the distributed hydrologic model to simulate the
spatial extent of flooding. The objective of this paper is to
investigate the utility of flood spatial extent information obtained
from orbital sensors to calibrate and evaluate hydrologic
models in an effort to potentially improve hydrologic prediction
and flood management strategies in ungauged catchments.
This paper is organized as follows. Section II describes the
study area, data, and the hydrologic model. Section III outlines
the methods for space-based flood inundation mapping and

for hydrologic model-based flood extent mapping. Section IV
provides a comparison between the two methods and validates
the model performance using MODIS-/ASTER-derived flood
inundation maps, followed by final concluding remarks in
Section V.
II. STUDY AREA AND DATA
A. Nzoia Basin
The rainy season from October to early December brings
devastating floods in Uganda, Kenya, Tanzania, and other countries
in East Africa almost every year. This region, surrounding
Lake Victoria, is heavily populated with around 30 million
people (Fig. 1). During December 2006, the United Nations
Office for the Coordination of Humanitarian Affairs estimated
that 1.8 million people had been affected by the flooding in
Kenya, Ethiopia, and Somalia. Repeated flooding affects many
lives particularly in the Lake Victoria region. With an area
of 68 600 km2, Lake Victoria is the second largest freshwater
lake in the world [30]. Nzoia, a subbasin of the Lake Victoria
basin, was chosen as the study area because of its regional
importance as being susceptible to flooding and as a major
KHAN et al.: SATELLITE REMOTE SENSING AND HYDROLOGIC MODELING FOR FLOOD INUNDATION MAPPING 87
TABLE I
SELECTED FLOOD EVENTS, LOCATION, FLOODED AREAS/RIVER (VERIFIED WITH THE DFO FLOOD INVENTORY).
NUMBERS IN PARENTHESES ARE THE JULIAN DAYS OF THE CORRESPONDING YEAR
tributary to Lake Victoria. The Nzoia River basin covers approximately

12 900 km2 with elevations ranging between 1100
and 3000 m. Annual average rain within the region is 1500 mm.
Table I lists the recent flooding events investigated in this
study.
B. Satellite Remote Sensing Data for Flood
Inundation Mapping
In this paper, we used MODIS and ASTER for flood inundation
mapping. MODIS instruments onboard NASAs Terra and
Aqua satellites offer a unique combination of quasi-global daily
coverage with acceptable spatial resolution. These capabilities
are being utilized for flood monitoring at regional and global
scales. Brakenridge et al. [3], [4] demonstrated that MODIS
data can be used to distinguish between flooded and nonflooded
areas with suitable spatial resolution. This can be very crucial
in regions where no other means of flood monitoring are
available. The NASA Goddard Space Flight Center, through
the Rapid Response System, processes and displays images
in near real timewithin 24 h of retrieval. MODIS Rapid
Response data are available from Terra and Aqua in near real
time at http://rapidfire.sci.gsfc.nasa.gov/. This system, initially
developed for fire hazard detection and monitoring, can be
utilized for flood detection across the globe. Several spectral
bands at spatial resolutions of approximately 250 and 500 m
are appropriate for accurate discrimination of water from land.
Global coverage can be provided on a near-daily basis if sky
conditions are cloud free.
Another sensor used in this study is ASTER, an imaging instrument
onboard Terra satellite that was launched in December
1999 as part of NASAs Earth Observing System. ASTER is a
cooperative effort between NASA, Japans Ministry of Economy,
Trade and Industry, and Japans Earth Remote Sensing

Data Analysis Center. ASTER is an advanced multispectral
imager with high spatial, spectral, and radiometric resolution.
The ASTER instrument covers a wide spectral range, from
visible to thermal infrared with 14 spectral bands. It has a
total of 14 bands in visible to near-infrared (VNIR), short-wave
infrared (SWIR), and thermal-infrared (TIR) wavelengths. The
ground resolutions of the VNIR, SWIR, and TIR images are
15, 30, and 90 m, respectively [31], [32]. Data from this sensor
can be acquired on demand from Land Processes Distributed
Active Archive Center at the U.S. Geological Survey Earth Resources
Observation and Science Data Center, with the standard
Hierarchical Data Format (http://LPDAAC.usgs.gov). In this
study, the strategies for computing MODIS- and ASTER-based
inundation extent are described in Section III.
C. Data for Hydrologic Model Setup and Implementation
The key remote sensing data sets enabling the implementation
and testing of a distributed hydrologic model in the Nzoia
basin include the following.
1) Digital elevation data from the Shuttle Radar Topography
Mission (SRTM) [33] (http://www2.jpl.nasa.gov/srtm/)
and SRTM-derived hydrologic parameter files of
HydroSHEDS [34].
2) Rainfall data from the Tropical Rainfall Measuring Mission

(TRMM)-based Multisatellite Precipitation Analysis
3B42 Real-Time (TMPA 3B42RT) operating in near real
time [35]. The data are available on the TRMM Web site
(http://trmm.gsfc.nasa.gov) at 0.25 0.25 spatial and
3-h temporal scales within 50 northsouth latitude band.
3) Soil parameters provided by the Food and Agriculture
Organization (2003) (http://www.fao.org/AG/agl/agll/
dsmw.html).
4) MODIS land classification map used for land use/cover,
with 17 classes of land cover according to the International
GeosphereBiosphere Program classification [36].
5) Global daily Potential Evapotranspiration data from the
Famine Early Warning Systems Network (FEWS NET:
http://earlywarning.usgs.gov/Global/index.php).
III. METHODOLOGY
The methodology consists of three major steps. First, the data
fromMODIS and ASTER sensors were retrieved and processed
to derive flood inundation maps for the selected events (Table I).
Second, a grid-based distributed hydrologic model was implemented
and further calibrated using the satellite-derived flood
inundation maps in the study area. Finally, the performance
of hydrologic simulations in the Nzoia basin was evaluated
by comparing the simulated flood inundation extents with
those derived from MODIS and ASTER imageries. A similar
technique described below is used by the Dartmouth Flood
Observatory (http://www.dartmouth.edu/~floods/) to generate
flood maps.
A. Satellite Remote-Sensing-Based Flood
Inundation Mapping
There are several methods for identifying flooded versus
nonflooded areas using optical remote sensing imagery (e.g.,

[11] and [37]). The first step is to identify spectral classes within
the imagery. One of the widely used clustering algorithms
for this task is the Iterative Self-Organizing Data Analysis
Technique Algorithm (ISODATA), which uses the Euclidean
distance in the feature space to assign every pixel to a cluster
through a number of iterations [37]. ISODATA begins with
either arbitrary cluster means or means of an existing signature
set, and each time the clustering repeats, the means of these
clusters are shifted. The new cluster means are used for the next
iteration. To perform ISODATA, the analyst selects the number
of spectral classes, a convergence threshold, and number
of iterations for the algorithm, which introduces considerable
subjectivity into the classification process [38]. This process of
floodwater classification was performed using ENVI software.
The method for flood detection and mapping using satellite
imagery included the following steps.
1) Terra MODIS near-real-time subsets covering the region
of Lake Victoria were retrieved from the NASA Web site
http://rapidfire.sci.gsfc.nasa.gov/subsets.
2) Color composite images were downloaded for image
processing. The false-color composite of MODIS
bands 1, 2, and 7 (red, near-infrared, and SWIR) has a
resolution of 250 m. The true-color composite of MODIS
bands 1, 3, and 4 was used for visual interpretation.
3) False-color composite images were subset to the region
of interest, and ISODATA classification was performed
(20 classes and three iterations).
4) All the water classes were combined into one water class.
5) The raster-type images were exported in Geographical
Information System (GIS)-compatible format for further
processing.
6) The images obtained in step 5) were overlaid on the
true-color image to remove the cloud contamination and
shadows that were falsely classified as water.
7) The final product overlaid in GIS under a reference water
layer (SRTM-based water bodies) was used to identify the
flooded areas.
B. Hydrologic Modeling
A distributed hydrologic model, Coupled Routing and
Excess STorage (CREST), developed by Wang et al. [39] was
used to generate modeled flood areal extents for comparison
with the satellite-based flood inundation maps.
1) Hydrologic Model Description and Implementation: The
distributed CREST hydrologic model is a hybrid modeling
strategy that was recently developed by The University of
Oklahoma (www.hydro.ou.edu) and the NASA SERVIR
Project Team (www.servir.net). The main CREST components
are briefly described as follows. 1) Three connected layers
within the soil profile that control the maximum storage of
infiltrating water and thus yield surface runoff generation. The
representation of within-cell variability of soil moisture storage
capacity (via a spatial probability distribution) and within-cell
routing can be employed for simulations at different spatiotemporal
scales. 2) Cell-to-cell routing of surface water using a
kinematic wave assumption. 3) Coupling between the runoff

generation and routing components via feedback mechanisms.
This coupling allows for a realistic scalability of the hydrologic
variables, such as soil moisture, and is particularly important
for simulations at fine spatial resolution.
CREST simulates the spatiotemporal variation of water
fluxes and storages on a regular grid with the grid cell resolution
being user defined. The scalability of model simulations
is accomplished through subgrid scale representation of soil
moisture variability (through spatial probability distributions)
and physical process representation. CREST can also simulate
inundation extent in an effort to obtain spatial and temporal
variations of floodwater within the domain. For more information
of the CREST model, please refer to [39].
In CREST, parameters related to topography and soil properties
are directly estimated from the land surface data shown
in the framework in Fig. 2. To apply the CREST model over
the Nzoia basin at 1-km2 spatial resolution, local drainage
direction and accumulation maps were derived from the
30-arc-second resolution SRTM DEM from the HydroSHEDS
data set. The precipitation forcing data are TMPA 3B42RT
products [35]. The subscript RT refers to real time, which,
in reality, refers to a pseudo real time where data become available
to the user via the Internet with a 816-h latency. Before
Fig. 2. Schematic of the satellite remote-sensing- and hydrologic-modeling-based flood monitoring system.
Fig. 3. (In blue on top abscissa) Observed precipitation, (in black) observed
runoff, and (in red) simulated runoff for the Nzoia basin during the calibration
period (19851998).
forcing CREST with the TMPA 3B42RT rainfall, the data were
accumulated into daily amounts and linearly resampled onto the
30-arc-second resolution model grid.
2) Model Calibration and Validation: CREST was calibrated
using the available daily discharge observations for the
period between 1998 and 2004. A one-year period (1998) was
used for warming up the model states. The model utilizes a
global optimization approach to capture the parameter interactions.
An autocalibration technique based on the Adaptive
Random Search (ARS) method by Brooks [40] was used to
calibrate the CREST model. The ARS method is considered
adaptive in the sense that it uses information gathered during
previous iterations to guide the parameter search in the current
step. The two most commonly used indicators to assess model
skill in matching the model-simulated streamflow with observations
are the NashSutcliffe Coefficient of Efficiency (NSCE)
[41] and the relative bias ratio (Bias). These two criteria are
used as objective functions for the automatic calibration. The
best skill occurs with NSCE 1 and Bias 0%
NSCE =1 _(Qi,o Qi,c)2
_(Qi,o Qo)2
(1)
Bias = _Qi,o _Qi,c
_Qi,o
100% (2)
where Qi,o is the observed discharge at the ith time step, Qi,c
is the simulated discharge at the ith time step, and Qo is the
average of all the observed discharge values. The ARS search

method yielded an optimized parameter combination that had
an NSCE = 0.873 and Bias = 0.228% (Fig. 3).
3) Flood Inundation Module: The CREST flood inundation
model uses one of the model outputs, the grid-to-grid total free
water, to simulate the flood extents. A predefined total free
water depth threshold of approximately 70 mm is employed to
determine flood-inundated extents. This value is not fixed but
changes with the calibration of satellite-based flood inundation
images that are used during the autocalibration process. For
more information about CREST model inputs and outputs,
please refer to [39].
TABLE II
2 2 CONTINGENCY TABLE FOR CREST-SIMULATED AND
SATELLITE-BASED FLOOD EXTENT COMPARISONS
C. Flood Inundation Evaluation Indexes

Finally, the CREST-simulated inundation spatial extents
were compared with the satellite-derived flood inundation
maps. The following several categorical verification statistics,
which measure the correspondence between the estimated and
observed occurrence of events, were used in this study: probability
of detection (POD), false-alarm ratio (FAR), and critical
success index (CSI). POD measures the fraction of observed
events that were correctly diagnosed and is also called the hit
rate (Table II). FAR gives the fraction of diagnosed events that
were not observed. CSI gives the overall fraction of correctly
diagnosed events by CREST. Perfect values for these scores are
POD = 1, FAR = 0, and CSI = 1
POD =Hits/(Hits + Misses) (3)
FAR =False alarms/(Hits + False alarms) (4)
CSI =Hits/(Hits + Misses + False alarms). (5)
CREST was calibrated using MODIS-based flood extent
maps for the different events listed in Table I for which gauged
streamflow observations are not available. As pointed out earlier,
the purpose of this exercise is to investigate the possibility
of calibrating the hydrologic model through satellite remote
sensing data sets alone. The calibration period includes two
cloud-free MODIS images available in 2006 and 2007. The
objective function selected to guide the calibration process was
the CSI between satellite-based inundation maps and CRESTmodeled
flood extents. The calibration terminates when improvements
in CSI within the last three iterations are less than
0.001. The approach can have far-reaching implications for
hydrologic prediction in unguaged basins where no other means
are available to calibrate a hydrologic model.
IV. RESULTS AND ANALYSIS
In this section, we present the application of the two alternative
methods for inundation mapping, namely, CRESTsimulated
and satellite-based methods described in Section III,
to generate the flood inundation maps for three different flood
events in the study area (see Fig. 4 and Table I). The comparison
of CREST-simulated flood extent with satellite-based
observations will provide an evaluation of the CREST model
performance in simulating the spatiotemporal evolution of the
flood inundation extent. Unsupervised classification is an automatic
and objective process that generates precise flooding

maps. However, flood inundation mapping from the binary
flood classification using optical sensors can be influenced by
cloud and vegetation cover. These lead to underdetection or
overdetection due to the influence of riparian vegetation and the
natural variability of the water surface, respectively. Comparisons
between the CREST- and satellite-based flood inundation
extent for the three events listed in Table I are discussed next.
A. Evaluation of Inundation Extent for Event 1
Fig. 4(B1) and (C1) show true- and false-color composite
(bands 7, 2, 1) MODIS scenes, respectively. After the MODIS
data were acquired for December 4, the flood extent was derived
using the ISODATA classification [Fig. 4(A1)]. The December
event was also simulated using the distributed hydrologic model
CREST. Intercomparison was made between the satellite-based
flood detection and CREST-simulated flood inundation map
[Fig. 5(A1)]. The river channel and water bodies are shown
as light blue, MODIS detections are in black, and CREST
inundations are in blue. The overlapping flooded areas from
MODIS and CREST are shown in red. Further examination of
flood extents from CREST and MODIS indicates that the spatial
patterns of the flood extent are similar, as shown in Fig. 5(A1).
To quantify this similarity, a spatial correlation is introduced
and analyzed on a pixel-by-pixel basis. If a pixel is classified in
the same category (river channel and water bodies and flooded
area), on both inundation maps, the pixel is recoded as 1 (a
hit); otherwise, it is recorded as a nonflooded area denoted as 0
(a miss).
Fig. 6(A1) shows the statistical comparison between the
flood extents derived from MODIS and CREST for the December
2006 event, evaluated as a function of search radius. POD
shows an increase from 0.23 to 0.75 to 0.98 for radii of 250,
1000, and 2000 m, respectively. Fig. 6(A1) also shows that,
within a 250-m radius, FAR can be as high as 0.7. However, by
increasing the radius to 1000 m, FAR is reduced to 0.18. The
CSI is improved from 0.14 to 0.64 for radii of 250 and 1000 m,
respectively. By further increasing the radius to 2000 m, the
CSI is improved to 0.92. Thus, the two maps show a spatial
agreement of 92% at a radius of 2000 m in Fig. 6(A1).
B. Evaluation of Inundation Extent for Event 2
A well-documented flood event that occurred during August
2007, with an estimated return period of ten years, was used
to validate CREST model performance. Fig. 4(B2)(B4) and
(C2)(C4) shows the true- and false-color composite (bands 7,
2, and 1) MODIS scenes, respectively. MODIS-based flood
extent maps shown in Fig. 6(A1)(A4) are derived from the
false-color composite scenes for August 15, 22 and 24, 2007,
respectively. The statistical comparison between the CREST
and MODIS flood inundation extents for these events is shown
in Fig. 6(A2)(A4). Fig. 6(A2) shows that, on August 15, 2007,
the POD is increased from 0.37 to 0.93 with an increase in
radius from 250 to 1000 m. Similarly, FAR and CSI have shown
improvements by increasing the radius for the other days of this
event [Fig. 6(A3) and (A4)].
Fig. 4. (A1A4) MODIS-based flood inundation maps for December 4, 2006 and August 15, 22, and 24, 2007. (A5)
ASTER inundation map for November 12,
2008. (B1B5) MODIS true-color composite of bands 1, 3, and 4. (C1C5) MODIS false-color composite of bands 7, 2,
and 1.
C. Evaluation of Inundation Extent for Event 3

For the November 2008 event, the ASTER image with higher
spatial resolution is shown in Fig. 4(A5). The POD of CREST
shows an increase from 0.46 at a radius of 30 m to 0.88 at a
radius of 600 m. Fig. 6(A5) also shows that, within a 30-m
radius, FAR is as high as 0.75. However, with the increase of
the search radius to 600 m, FAR substantially decreases to 0.15.
The CSI is improved from 0.19 at 30 m to more than 0.76 at a
radius of 600 m.
Fig. 5. Comparison of satellite-based and CREST-simulated flood inundation extents. First legend entry is the year
and the Julian day of the flood event, followed
by the event identification number (refer to Table I).
V. CONCLUSION AND FUTURE WORK
We have studied the feasibility of setting up and calibrating

a distributed hydrologic model using satellite-based forcing,
parameters, and observations of flood inundation extents. This
approach is in contrast to the conventional method of floodmodeling
techniques, which are often not available in ungauged
basins or data sparse regions. The proposed approach implements
a distributed hydrologic model with remote sensing data
and further calibrates the model parameters through satellitebased
flood inundation maps. Utilizing satellite-based data sets
Fig. 6. Evaluation of hydrologic model-based spatial flood extents using inundated areas derived from (A1A4)
MODIS and (A5) ASTER. Evaluation is
performed as a function of search radius.
that are freely available in the public domain and integrating

them with distributed hydrologic models has the potential to
improve simulation and prediction of the spatial extent of
floods, even in ungauged basins. The broader impact of such
a demonstrated technique is to provide a cost-effective tool
to progressively build capacity for flood disaster prediction
and risk reduction in poorly or ungauged basins located in
many lesser developed or developing countries in Africa or
South Asia. Operationally implementing this strategy in those
areas will provide flood managers and international aid organizations
a realistic decision-support tool in order to better
assess imminent flood impacts. This paper has demonstrated
the applicability of distributed hydrologic model calibration
using satellite-derived flood inundation maps from MODIS and
ASTER images in gauged basins.
The recent release of the ASTER Global Digital Elevation
Model with a resolution of 30 m and satellite-based precipitation
products at high temporal resolution (e.g., TRMM at
3-h resolution) can be used to implement, calibrate, and force
distributed hydrologic models for flood prediction purposes.
Thus, further research on how to utilize spatially distributed
observations, such as high-resolution imagery and other microwave
sensors, should be carried out in various geographical
locations for the calibration and evaluation of distributed
hydrologic models. The case studies on the Nzoia basin that we

have presented illustrate that, for regions with scarce groundbased
observations, remote sensing data can be used to implement

hydrologic models with sufficient accuracy in predicting
spatial flood extents. It is noted that further improvements in
flood monitoring will be made possible with the integration of
satellite remote sensing products with ground-based observations
and details of catchment properties.
ACKNOWLEDGMENT
The authors would like to thank the three anonymous reviewers
for their critical comments and constructive suggestions.
REFERENCES
[1] S. N. Jonkman, Global perspectives on loss of human life caused by
floods, Nat. Hazards, vol. 34, no. 2, pp. 151175, Feb. 2005.
[2] J. McCarthy, Climate Change 2001: Impacts, Adaptation, and Vulnerability:
Contribution of Working Group II to the Third Assessment Report
of the Intergovernmental Panel on Climate Change. Cambridge, U.K.:
Cambridge Univ. Press, 2001.
[3] G. Brakenridge, E. Anderson, S. Nghiem, S. Caquard, and T. B. Shabaneh,
Flood warnings, flood disaster assessments, and flood hazard reduction:
The roles of orbital remote sensing, in Proc. 30th Int. RSE, Honolulu, HI,
Nov. 1014, 2003.
[4] L. C. Smith, Satellite remote sensing of river inundation area, stage, and
discharge: A review, Hydrol. Process., vol. 11, no. 10, pp. 14271439,
1997.
[5] G. R. Brakenridge, S. V. Nghiem, E. Anderson, and R. Mic, Orbital
microwave measurement of river discharge and ice status, Water Resour.
Res., vol. 43, no. 4, p. W04 405, 2007.
[6] R. Brakenridge, MODIS-based flood detection, mapping and measurement:
The potential for operational hydrological applications, in
Transboundary Floods: Reducing Risks Through Flood Management.
New York: Springer-Verlag, 2006, pp. 112.
[7] C. M. Birkett, L. A. K. Mertes, T. Dunne, M. H. Costa, and M. J. Jasinski,
Surface water dynamics in the Amazon Basin: Application of satellite
radar altimetry, J. Geophys. Res.Atmos., vol. 107, no. D20, p. 8059,
2002.
[8] F. Blasco, M. F. Bellan, and M. U. Chaudhury, Estimating the extent of
floods in Bangladesh using SPOT data, Remote Sens. Environ., vol. 39,
no. 3, pp. 167178, Mar. 1992.
[9] M. France and P. Hedges, A hydrological comparison of Landsat TM,
Landsat MSS, and black and white aerial photography, in Proc. 7th Int.
Symp. ISPRS, Enschede, The Netherlands, Aug. 1986, pp. 717720.
[10] J. R. Jensen, M. E. Hodgson, E. Christensen, H. E. Mackey, L. R. Tinney,
and R. R. Sharitz, Remote-sensing inland wetlandsA multispectral
approach, Photogramm. Eng. Remote Sens., vol. 52, no. 1, pp. 87100,
1986.
[11] J. P. Watson, A visual interpretation of a LANDSAT mosaic of the
Okavango-delta and surrounding area, Remote Sens. Environ., vol. 35,
no. 1, pp. 19, Jan. 1991.
[12] I. J. Barton and J. M. Bathols, Monitoring floods with AVHRR, Remote
Sens. Environ., vol. 30, no. 1, pp. 8994, Oct. 1989.
[13] S. J. Gale and S. Bainbridge, The floods in eastern Australia, Nature,
vol. 345, no. 6278, p. 767, 1990.
[14] H. Rasid and M. Pramanik, Areal extent of the 1988 flood in Bangladesh:
How much did the satellite imagery show? Nat. Hazards, vol. 8, no. 2,
pp. 189200, Sep. 1993.
[15] I. Sandholt, L. Nyborg, B. Fog, M. L, O. Boucum, and K. Rasmussen,
Remote sensing techniques for flood monitoring in the Senegal River
Valley, Geogr. Tidsskr., vol. 103, no. 1, pp. 7181, 2003.
[16] G. Brakenridge and E. Anderson, Satellite gaging reaches: A strategy for
MODIS-based river monitoring, in Proc. 9th Int. Symp. Remote Sens.,
Crete, Greece, 2003, vol. 4886, pp. 479485.
[17] G. Stancalie, A. Diamandi, C. Corbus, and S. Catana, Application of
EO data in flood fore-casting for the Crisuri Basin, Romania, in Flood
Risk Management: Hazards, Vulnerability and Mitigation Measures.
New York: Springer-Verlag, 2004, p. 101.
[18] Y. Wang, Using Landsat 7 TM data acquired days after a flood event
to delineate the maximum flood extent on a coastal floodplain, Int. J.
Remote Sens., vol. 25, no. 5, pp. 959974, Mar. 2004.

[19] Y. Wang, J. D. Colby, and K. A. Mulcahy, An efficient method for
mapping flood extent in a coastal floodplain using Landsat TM and
DEM data, Int. J. Remote Sens., vol. 23, no. 18, pp. 36813696,
Sep. 2002.
[20] C. Puech and D. Raclot, Using geographical information systems and
aerial photographs to determine water levels during floods, Hydrol.
Process., vol. 16, no. 8, pp. 15931602, Jun. 2002.
[21] D. Alsdorf, D. Lettenmaier, and C. Vrsmarty, The need for global,
satellite-based observations of terrestrial surface waters, EOS Trans.
AGU, vol. 84, no. 29, pp. 275276, 2003.
[22] D. Alsdorf, E. Rodrguez, and D. Lettenmaier, Measuring surface water
from space, Rev. Geophys., vol. 45, no. 2, p. RG2 002, 2007.
[23] W. Marcus and M. Fonstad, Optical remote mapping of rivers at submeter
resolutions and watershed extents, Earth Surf. Process. Landforms,
vol. 33, no. 1, pp. 424, Jan. 2008.
[24] G. Schumann, P. Bates, M. Horritt, P. Matgen, and F. Pappenberger,
Progress in integration of remote sensing-derived flood extent and stage
data and hydraulic models, Rev. Geophys., vol. 47, no. 4, p. RG4 001,
2009.
[25] M. S. Horritt, Calibration of a two-dimensional finite element flood flow
model using satellite radar imagery, Water Resour. Res., vol. 36, no. 11,
pp. 32793291, 2000.
[26] M. S. Horritt and P. D. Bates, Evaluation of 1D and 2D numerical
models for predicting river flood inundation, J. Hydrol., vol. 268, no. 14,
pp. 8799, Nov. 2002.
[27] G. Schumann, R. Hostache, C. Puech, L. Hoffmann, P. Matgen,
F. Pappenberger, and L. Pfister, High-resolution 3-D flood information
from radar imagery for flood hazard management, IEEE Trans. Geosci.
Remote Sens., vol. 45, no. 6, pp. 17151725, Jun. 2007.
[28] G. Di Baldassarre, G. Schumann, and P. D. Bates, A technique for the
calibration of hydraulic models using uncertain satellite observations of
flood extent, J. Hydrol., vol. 367, no. 3/4, pp. 276282, Apr. 2009.
[29] M. S. Horritt, G. Di Baldassarre, P. D. Bates, and A. Brath, Comparing
the performance of a 2-D finite element and a 2-D finite volume model
of floodplain inundation using airborne SAR imagery, Hydrol. Process.,
vol. 21, no. 20, pp. 27452759, Sep. 2007.
[30] S. Swenson and J. Wahr, Monitoring the water balance of Lake Victoria,
East Africa, from space, J. Hydrol., vol. 370, no. 14, pp. 163176,
May 2009.
[31] H. Fujisada, F. Sakuma, A. Ono, andM. Kudoh, Design and preflight performance
of ASTER instrument protoflight model, IEEE Trans. Geosci.
Remote Sens., vol. 36, no. 4, pp. 11521160, Jul. 1998.
[32] Y. Yamaguchi, A. B. Kahle, H. Tsu, T. Kawakami, and M. Pniel,
Overview of Advanced Spaceborne Thermal Emission and Reflection
Radiometer (ASTER), IEEE Trans. Geosci. Remote Sens., vol. 36, no. 4,
pp. 10621071, Jul. 1998.
[33] B. Rabus, M. Eineder, A. Roth, and R. Bamler, The Shuttle Radar
Topography MissionA new class of digital elevation models acquired
by spaceborne radar, ISPRS J. Photogramm., vol. 57, no. 4, pp. 241262,
Feb. 2003.
[34] B. Lehner, K. Verdin, and A. Jarvis, New global hydrography derived
from spaceborne elevation data, EOS, vol. 89, no. 10, pp. 9394, 2008.
[35] G. J. Huffman, R. F. Adler, D. T. Bolvin, E. J. Nelkin, D. B.Wolff, G. Gu,
Y. Hong, K. P. Bowman, and E. F. Stocker, The TRMM multisatellite
precipitation analysis (TMPA): Quasi-global, multiyear, combined-sensor
precipitation estimates at fine scales, J. Hydrometeorol., vol. 8, no. 1,
pp. 3855, Feb. 2007.
[36] M. A. Friedl, D. K. McIver, J. C. F. Hodges, X. Y. Zhang, D. Muchoney,
A. H. Strahler, C. E. Woodcock, S. Gopal, A. Schneider, A. Cooper,
A. Baccini, F. Gao, and C. Schaaf, Global land cover mapping from
MODIS: Algorithms and early results, Remote Sens. Environ., vol. 83,
no. 1/2, pp. 287302, Nov. 2002.
[37] J. Jensen, Introductory Digital Image Processing: A Remote Sensing
Perspective. Upper Saddle River, NJ: Prentice-Hall PTR, 2005.
[38] R. L. Lang, G. F. Shao, B. C. Pijanowski, and R. L. Farnsworth, Optimizing
unsupervised classifications of remotely sensed imagery with a
data-assisted labeling approach, Comput. Geosci., vol. 34, no. 12,
pp. 18771885, Dec. 2008.
[39] J. Wang, Y. Hong, L. Li, J. J. Gourley, K. Yilmaz, S. Khan, F. S. Policelli,

R. F. Adler, S. Habib, D. Irwn, T. Korme, and L. Okello, The Coupled
Routing and Excess STorage (CREST) distributed hydrological model,
Hydrol. Sci. J., 2010, to be published.
[40] S. H. Brooks, A discussion of random methods for seeking maxima,
Oper. Res., vol. 6, no. 2, pp. 244251, Mar./Apr. 1958.
[41] J. Nash and J. Sutcliffe, River flow forecasting through conceptual
models. Part IA discussion of principles, J. Hydrol., vol. 10, no. 3,
pp. 282290, Apr. 1970.
Sadiq I. Khan is currently working toward the

Ph.D. degree in the College of Atmospheric and
Geographic Sciences, The University of Oklahoma,
Norman.
He is currently also a Graduate Research Assistant
with the School of Civil Engineering and Environmental
Science, The University of Oklahoma.
His fields of interest include flood hydrology and
remote sensing in hydrology, particularly optical and
microwave remote sensing for flood hydrology and
management.
Mr. Khan is a member of the American Geophysical Union, Association
of American Geographers, and American Society for Photogrammetry and
Remote Sensing. He is the recipient of the NASA Earth and Space Science
Fellowship and was selected for an outstanding student paper award for the
research work he presented at the American Geophysical Union Fall 2009
Meeting in San Francisco, CA.
Yang Hong received the B.S. and M.S. degrees in
geosciences and environmental sciences from Peking
University, Beijing, China, and the Ph.D. degree,
major in hydrology and water resources and minor
in remote sensing and spatial analysis, from
The University of Arizona, Tucson.
Following a postdoctoral appointment at the
Center for Hydrometeorology and Remote Sensing,
University of California, Irvine, he joined the NASA
Goddard Space Flight Center, Greenbelt, MD, in
2005. He is currently an Associate Professor with
the School of Civil Engineering and Environmental Sciences, The University
of Oklahoma, Norman, where he is also directing the Remote Sensing Water
Hydrology research group. He also serves as the Associate Director of the
Center for Natural Hazards and Disaster Research and an affiliated Faculty
Member with the Atmospheric Radar Research Center, Center for Spatial
Analysis, andWater Technology for Emerging Regions Center. He has served in
the editorial boards of the International Journal of Remote Sensing, the Natural
Hazards journal, and the Landslides journal. His primary research interests
are in remote-sensing retrieval and validation, hydrology and water resources,
natural hazard prediction, land surface modeling, and data assimilation systems
for water resource planning under changing climate.
Dr. Hong is currently the American Geophysical Union Precipitation
Committee Chair.
JiahuWang, photograph and biography not available at the time of publication.
Koray K. Yilmaz, photograph and biography not available at the time of
publication.
Jonathan J. Gourley received the B.S. and M.S.
degrees in meteorology with a minor in hydrology
and the Ph.D. degree in civil engineering and environmental
science from the University of Oklahoma,
Norman.
He is currently a Research Hydrometeorologist
with NOAAs National Severe Storms Laboratory, is
an affiliate Associate Professor with the School of
Meteorology, University of Oklahoma, and Director
of the National Weather Centers seminar series.
His research focuses on rainfall observations from
remote sensing platforms with an emphasis on ground-based radars and
implementing these high-resolution observations into hydrologic models. He
completed a postdoctoral study with researchers in Paris, France to demonstrate
the capabilities of dual-polarimetric radar in improving data quality, microphysical
retrievals, and precipitation estimation. MeteoFrance has subsequently

upgraded several of their operational radars with polarimetric technology.
Dr. Gourley received the Department of Commerce Silver Medal Award
in 1999 For developing an important prototype Warning Decision Support
System for weather forecasters and making significant enhancements to the
NEXRAD system, resulting in more timely and reliable warnings. He also
received an Honorable Mention in 2004 from the Universities Council onWater
Resources Dissertation Awards Committee.
Robert F. Adler received the B.S. and M.S. degrees
from The Pennsylvania State University, University
Park, in 1965 and 1967, respectively, and the Ph.D.
degree from Colorado State University, Fort Collins,
in 1974.
His research focuses on the analysis of precipitation
observations from space on global and regional
scales using Tropical Rainfall Measuring Mission
(TRMM) data along with data from other satellites.
He studies precipitation variations in relation
to phenomena such as El Nio/Southern Oscillation,
volcanoes, and tropical cyclones, as well as longer interdecadal changes or
variations. He also leads the group that produces the global monthly and
daily precipitation analyses for the World Climate Research Program Global
Precipitation Climatology Project. He has published 80 papers in scientific
journals on the aforementioned topics. He is currently the TRMM Project
Scientist of the NASA Goddard Space Flight Center, Greenbelt, MD.
Dr. Adler is a Fellow of the American Meteorological Society. He was the
recipient of the Goddard Space Flight Center Exceptional Performance Award
in 1980, the NASA Exceptional Scientific Achievement Medal in 1989, the
NASA Outstanding Leadership Medal in 2002, and the Goddard Laboratory
for Atmospheres Scientific Leadership Award in 2002.
G. Robert Brakenridge, photograph and biography not available at the time
of publication.
Fritz Policelli, photograph and biography not available at the time of
publication.
Shahid Habib, photograph and biography not available at the time of
publication.
Daniel Irwin, photograph and biography not available at the time of
publication.
The use of optical remote sensing for mapping

flooded areas
ARTICLE OCTOBER 2013
CITATIONS
2
READS
67
3 AUTHORS:
Igor Ogashawara
Indiana University-Purdue University India
SEE PROFILE
Marcelo Curtarelli
National Institute for Space Research, Brazil
SEE PROFILE
Celso M. Ferreira
George Mason University
SEE PROFILE

Available from: Igor Ogashawara
Igor Ogashawara et al Int. Journal of Engineering Research and Application www.ijera.com

ISSN : 2248-9622, Vol. 3, Issue 5, Sep-Oct 2013, pp.1956-1960
www.ijera.com 1956 |
Page
The Use of Optical Remote Sensing For Mapping Flooded Areas

Igor Ogashawara*, Marcelo Pedroso Curtarelli* and Celso M. Ferreira**
*(Remote Sensing Division, National Institute for Space Research, Brazil)
** (Department of Civil, Environmental and Infrastructure Engineering, George Mason University, USA)
ABSTRACT
Flood maps are a crucial tool to support emergency management, disaster recovery and risk reduction planning.
Traditional flood mapping methods are time-consuming, labor intensive, and costly. Our goal in this paper is to
introduce a novel technique to aggregate knowledge and information to map coastal flooded areas. We proposed
a Difference of Normalized Difference Water Indices (DNDWI) derived from two LANDSAT-5/TM surface
reflectance product acquired before and after the passage of Hurricane Ike, for Upper Texas in September of
2008. The reference flooded area was delineated interpolating the maximum surge in each location using a spline
with barriers method with high tension and a 30 meter Digital Elevation Model (DEM). It was noticed that
NDWI values decreased after the hurricane landfall on average from 0.226 to 0.122 for flooded area. However
for the non-flooded areas it increased from 0.292 to 0.300. Results from the Monte Carlo simulation showed that
mapping flooded areas with DNDWI got an accuracy of 85.68% while the non-flooded areas got an accuracy of
92.13%. Thus, DNDWI is promising tool for mapping flooded areas since it is a cheaper and simple technique
which can be applied rapidly for several areas of the planet.
Keywords - Flood mapping, Hurricane, Monte Carlo Simulation, NDWI, Optical Remote Sensing
I. INTRODUCTION
Hurricanes are one of the most costly natural
disasters in the United States [1] and recent storms
such as Hurricane Sandy (2012), Ike (2008) and
Katrina (2005) have caused major infrastructure
damage and losses of lives along the East and Gulf
Coasts [2]. Tropical cyclones are also a major cause of
flooding and damage in several regions worldwide
including recent events in the Bay of Bengal, Typhoon
Phailin (2013), and historically deadly storms such as
Tropical cyclone Thelma (1991), in the North Pacific
Ocean, and Cyclone Zoe (2002), in the South Pacific
Ocean, among others. Coastal flooding is one of the
major hazards to accompany a tropical cyclone
landfall [3] and can be aggravated by the combination
of the storm tidal surge and rainfall-runoff from the
heavy precipitations.
Traditionally, coastal flooding due to tropical
cyclones has been estimated by measured water levels
on buoys and coastal gages (e.g., [4-5]). Although
these monitoring networks provide good historical
data for coastal flooding, it lacks spatial information
due to the limited number of stations over large areas.
Recent developments of physics based numerical
models using High Performance Computing (HPC)
(e.g., [6]) in addition to an increasingly large volume
of high resolution data (e.g., topo/bathymetry, land
use, wind fields) has led to an unprecedented
improvement in accuracy of tropical cyclones flood
prediction and mapping. A combination of numerical
modeling and measured data for model validation and

calibration is currently the state of the art in predicting
coastal flooding [7]. Recently the United States
Geological Survey (USGS) launched the Inland
Storm-Tide Monitoring Program [8] that provides
unprecedented detail in coastal flooding monitoring
during hurricane events on the United States coastal
areas. Although recent advances in technology and
methods, the forecasting and mapping of coastal
flooding spatial extent is still a challenge especially in
large areas without extensive instrumentation.
In this way, remote sensing data can provide
useful information to mapping coastal flooding over
large area. The main advantages in the use of remote
sensing data are the synoptic view of large areas,
spatial variability of data and repetitive acquisition.
Moreover, the cost of data acquired by sensors
onboard orbital platforms can be lower than the data
acquired by conventional methods (discounting the
cost of the satellites) [9]. Recently, remote sensing
data have been applied to identify flood areas using
the Environmental Satellite (ENVISAT) advanced
synthetic aperture radar (ASAR) and Landsat
Thematic Mapper (TM) optical imagery to document
the flooded extent of Hurricane Ike (2008) by [10].
Subsequently, [11] presented a summary of the
limitations and potentials of satellite imagery to
monitor and map coastal flooding for Hurricanes
Gustav and Ike (2008) demonstrating that the
correspondence between ground data and ASARbased
flood mapping ranged from 86 to 96% for water
levels higher than 0.8 m.
Although the good correspondence between
ground data and RADAR-based flood mapping, such
as demonstrated by [11], RADAR (Radio Detection
and Ranging) sensors (e.g., ASAR, TerraSAR,
RADARSAT) are not always imaging freely and
generally their products have high costs. Moreover,
RESEARCH ARTICLE OPEN ACCESS

Page
the data acquisition is sparse and over small areas.
Therefore, it is necessary to develop new techniques
based only on optical remote sensing data, which
generally is available freely, have more options of
spatial and temporal resolutions and has a historical
time series with more than 10 years records (e.g.,
MODIS and TM sensors), which allows mapping the
flooding extent of past events.
The objective of this study is to introduce a
novel technique based on optical remote sensing
techniques in order to aggregate knowledge and
information to map coastal flood areas. We present a
case study for the coastal flooding caused by

Hurricane Ike on the Texas coast and develop our
method using the Normalized Difference Water Index
(NDWI) which was derived from two the medium
resolution LANDSAT-5/TM surface reflectance
product from the LANDSAT Climate Data Record
(CDR).
II. STUDY AREA AND EVENT

The study area is the upper Texas coast,
USA, which was directly impacted by Hurricane Ike
in 2008 more specifically between the coordinates
3046'23"N 9541'48"W and 2919'53"N 941'57"W
(Fig. 1). According to [12], the frequency of
hurricanes in this region is one about every six years
and the annual probability of hurricanes occurrence is
around 31%. However, more than 40 events were
registered in the past century (1 at each three year),
which indicates an increase in the frequency of events.
Hurricane Ike was a Category 2 hurricane
when it made landfall on the Texas coast [2] causing
severe damage to the State of Texas, Bahamas and
Cuba as well. The storm winds reached 230 km h-1
with the lowest central pressure at 935 mbar leaving
an estimated death toll of 103 people and around 40
billion dollars in damages.
Figure 1. Study area in the Upper Texas, near Hilton
and reference flooded area. LANDSAT-5/TM image
acquired on 4 September true color composition.
III. DATASET
The dataset used in this study include in situ
measurements and remote sensing images. The in situ
data were collected by the USGS mobile storm surge
network [8], which provide atmospheric pressure and
water level data at each 6 minutes. The data were
measured using a pressure transducer (HOBO
Onset) and stored in the USGS database [8].
Topography was obtained from a Digital Elevation
Model (DEM) dataset extracted from the National
Elevation Dataset (NED) [13] representing the entire
region topography at a resolution of 1 arc-second.
The remote sensing data comprise images
collected by the TM sensor, onboard LANDSAT-5
satellite. This images are provided with 7 spectral
bands (from the visible to thermal spectral regions),
quantized in 8 bits and with 30 meters spatial
resolution, except the thermal band which has 120 m
spatial resolution. The images are acquired at each 16
days. In this study we used the surface reflectance
product from the LANDSAT CDR. This product is
generated from specialized software called Landsat
Ecosystem Disturbance Adaptive Processing System
(LEDAPS) which provide images converted for
reflectance values and corrected for the atmospheric
effects. More information about this product is
available at [14]
IV. METHODS
4.1. REFERENCE FLOODED AREA

The reference flooded area was determined
based on the methodology proposed by [15] and
consists of spatially interpolating the maximum flood
height at each monitoring station and calculating the
water depths based on the DEM. For this study we
considered 59 recording stations along the study area.
The maximum water heights were extracted from the
recorded time series to represent the maximum flood
level at each location. A spatial interpolation using a
spline with higher tension (weight of 20 as suggested
by [15] was used to develop the maximum water level
coverage for the region. The water depths were
calculated by subtracting the maximum flood surface
from the DEM using the vertical NAVD88 Datum for
the region. The resulting coverage was re-classified to
remove the dry areas defining the estimated flood
extent.
4.2. NORMALIZED DIFFERENCE WATER INDEX
As vegetation is one of the most affected
targets by flood in coastal areas, changes in its water
content could be used to identify the affected areas.
The NDWI, which is also called the leaf area waterabsent
index, could be an alternative for optical
remote sensing to map flooded areas. It is possible
since this index (1) estimates the water content within
vegetation [16]:
mm
mm
NDWI
0.86 1.24
0.86 1.24

Page
where, mis the reflectance at 0.86 m and
mis the reflectance at 1.24 m.
According to [16], it measures the liquid
water molecules in vegetation that interact with solar
radiation. The NDWI has been widely used because it
is less sensitive to atmospheric scattering effects
compared with the NDVI. However, similar to the
NDVI, the NDWI did not completely remove the soil
background reflectance effects. Using Landsat-5/TM
bands, the NDWI was estimated as follows (2):
4 5
( 4) ( 5)
TMB TMB
TMB TMB
NDWI
where, is the reflectance at LANDSAT-5/TM

band 4 and is the reflectance at LANDSAT5/TM band 5. We used band 5 (1.65 m) as an
approximation of 1.24 m [16]. This procedure was
previously described in several studies [17-19].
4.3. METHOD DEVELOPMENT
The use of NDWI to classify flooded areas
from a hurricane event consists in analyzing NDWI
values from two different dates (before and after the
hurricane). The difference among the NDWI values
from the two dates is here called Difference of NDWI
(DNDWI). Two thousand values of DNDWI were
used to classify flooded and non-flooded areas
according to the established limits for each of these
two classes. These limits were computed from the
univariate statistics of DNDWI values from each of
the reference class. Average and standard deviations
for each class was used to determine the lower and
upper limit to categorize the DNDWI values. The
DNDWI threshold for each class was chosen as
plus/minus one standard deviation from the mean
value plus a constant (see Table 1 in the results and
discussion section).
4.4. VALIDATION
It was used to validate the NDNWI
classification a Monte Carlo Method (MCM). The
MCM, also known as statistical simulation, is defined
as any method that uses sequences of random numbers
to perform a simulation. This process was based on
the outcome of many simulations using random
numbers to obtain a probable solution. This technique
provides an approximate, quick answer and a high
level of accuracy; however, it is possible that accuracy
is increased with additional simulations [20]. We
collected 2000 stratified points of the DNDWI image,
each class (flooded and non-flooded) got 1000 random
values each. From these values an average from 20 of
them were calculated 10,000 times. These 10,000
average values were used to evaluate the proposed
classification using the thresholds analysis. The
analysis were conducted by fitting all 10,000 values in
the thresholds and them analyzing the number that got
in the correct class.
V. RESULTS AND DISCUSSIONS
5.1. DNDWI CLASSIFICATION
For the DNDWI classification we collected
1000 points for each class based on the reference and

calculate their univariate statistics and a histogram
(Fig. 2). For the 1000 points from the flooded area,
the average value was 0.107 while for the non-flooded
are the average was -0.011.The median for the flooded
area was 0.140 and the non-flooded got a median
value of 0.003. It showed that the DNDWI is higher
for flood areas and lower for non-flooded ones. It
happened due to the fact that in both areas there were
precipitation, however, as the NDWI measures the
water content in the vegetation, at flooded areas the
index was saturated while in non-flooded area it
worked well.
Figure 2. Histograms of DNDWI from the two classes
according to the reference classification.
The values of DNDWI can be explained by
the mean values of NDWI before and after the
Hurricane Ike. The mean value of the NDWI before
the hurricane was 0.292 for the non-flooded areas and
0.2264 for the flooded areas. They changed to 0.300 in
the non-flooded areas and to 0.122 for flooded areas.
It showed an increase of 2.74% in the NDWI values
for the non-flooded area and a decrease of 45.82% of
NDWI values for the flooded areas. Thus as the index
values increases from the first to the second image in

Page
the non-flooded areas, the DNDWI tends to decrease
while the opposite is also true, in the flooded areas
DNDWI values tend to increase due to the decrease of
NDWI in the second image. Due to these
characteristics the thresholds were calculated from the
average plus/minus one standard deviation from the
mean value plus a constant (Table 1).
Table 1. Thresholds for the DNDWI classification.
Classes Threshold
Non-flooded <0.05
Flooded >0.05
5.2. VALIDATION USING MCM
To validate the choice of thresholds the
10,000 values from the simulated average of DNDWI
values were analyzed. For flooded areas the values
ranged from -0.099 to 0.285, while the values for nonflooded
areas ranged from -0.193 to 0.143. The mean
values of these two series were 0.106 and -0.010 for
the flooded and non-flooded areas respectively. A
histogram of all 10,000 DNDWI values and the
threshold limit are shown on Fig. 3.
Figure 3. Histogram of DNDWI simulated average by
MCM.
MCM results also showed that for flooded
areas 8568 values were classified as flooded by the
threshold analysis. For non-flooded the number of
values in the range proposed in the threshold was
9213. It showed that DNDWI got an accuracy of

85.68% for mapping flooded areas, showing that it is a
potential tool to help managers and policy makers to
analyze the impacted areas.
We also calculate the results of the Wilcoxon
signed rank test for the 10,000 values of DNDWI
through the MCM. It showed that the null hypothesis
was refuted with 5% significance (p-value < 0.001);
thus, there was undoubtedly a difference in the
DNDWI values from flooded and non-flooded areas.
VI. CONCLUSION
A methodology for mapping spatial
variations of flood inundation caused by hurricanes
events using optical remotely sensed image series was
developed in this study. Since flood prevention,
management and emergency response is an important
issue for policy makers, flood detection using remote
sensing can improve the number of monitored areas in
remotely accessed places or in places without any
monitoring program. As optical orbital sensors, like
Landsat TM, ETM+ and OLI family, are collecting
non-stop data from the entire planet, the use of an
optical sensor could enhance the knowledge of
flooding spatial mapping in areas with lack of data.
Our results showed that the DNDWI and a threshold
analysis method for distinguish flooded areas could be
a potential tool to enhance the knowledge of tropical
cyclones flooding mapping. It was observed, from a
MCM technique, that 85.68% of 10000 mean values
from an interaction of 20 to 1000 DNDWI values were
accurately classified as flooded areas. For the nonflooded
area an accuracy of 92.13% was found.
However, these results were based on only
one event (Hurricane Ike) and should be extended to
other study areas. We also observed that additional
spectral behavior studies are needed to explain the
relationship between water content and vegetation
spectral response in flooded areas. Nevertheless, we
proposed a methodology which could be an useful tool
for countries without any flooding monitoring
program and RADAR imagery cover, since the
mapping of flooded areas is an important issue for the
economy and the rebuilding of the affected region.
Although the use of Synthetic Aperture Radar has
been previously applied, the cost for buying the
images is high as well as it is labor intensity. Using
optical remote sensing is not only cheaper but also
much easier to manage the data. Through the advance
of orbital hyperspectral sensors like the Hyperspectral
Imager for the Coastal Ocean (HICO), it will be
possible to identify the key spectral ranges to identify
the elevated degree of water content in vegetation.
Thus the hyperspectral studies will enhance the use of
optical remote sensing to map flooded areas.
VII. ACKNOWLEDGEMENTS
The first author is grateful to the Brazilian
Federal Agency for Support and Evaluation of

Graduate Education (CAPES) for the masters
scholarship. The second author wishes to thanks the
National Counsel of Technological and Scientific
Development (CNPQ) (grants 161233/2013-9) for the
PhD scholarship.
REFERENCES
[1] N. Lott and T. Ross, Tracking and
Evaluating U.S. Billion Dollar Weather
Disasters, 1980-2005, National Climatic
Data Center (NCDC), 2006. Available at:
http://www1.ncdc.noaa.gov/pub/data/papers/
200686ams1.2nlfree.pdf. Accessed on: 12
Oct 2013.
[2] National Oceanic and Atmospheric
Administration (NOAA), Tropical Cyclone
Report: Hurricane Ike, 2013a. Available at:

Page
http://www.nhc.noaa.gov/2008atlan.shtml.
Accessed on:10 Oct. 2013.
[3] D. Resio and J. Westerink, Modeling the
physics of storm surges, Physics Today,
61(9), 2008, 33-38.
[4] National Oceanic and Atmospheric
Administration (NOAA), Tides and currents,
2013b. Available at:
http://tidesandcurrents.noaa.gov/. Accessed
on: 15 Oct. 2013.
[5] United Nations Education, Scientific and
Cultural Organization (UNESCO), The
Global Ocean Observing System (GOOS),
2013. Available at: http://www.iocgoos.
org/Accessed on: 11 Oct. 2013.
[6] J. Dietrich, M. Zijlema, J. Westerink, L.
Holthuijsen, C. Dawson, R. Luettich, R.
Jensen, J. Smith, G. Stelling and G. Stone,
Modeling hurricane waves and storm surge
using integrally-coupled, scalable
computations, Coastal Engineering, 58(1),
2011, 45-65.
[7] C. Dawson, E.J. Kubatko, J.J. Westerink, C.
Trahan, C. Mirabito, C. Michoski, N. Panda,
Discontinuous Galerkin methods for
modeling hurricane storm surge, Advances in
Water Resources, 34(9), 2011, 1165-1176.
[8] United States Geological Survey (USGS),
The storm-tide monitoring program, 2013a.
Available at:
http://water.usgs.gov/osw/programs/storm_su
rge1.html.Accessed on: 15 Oct. 2013.
[9] J.R. Jensen, Remote sensing of the
environment: An earth resource perspective
(Boca Raton, FL: Prentice Hall, 2006).
[10] E. Ramsey, D. Werle, Z. Lu, A.

Rangoonwala and Y. Suzuoki, A case of
timely satellite acquisition in support of
coastal emergency and environmental
response management, Journal of Coastal
Research, 25(5), 2009, 1168-1172.
[11] E. Ramsey, D. Werle, Y. Suzuoki, A.
Rangoonwala and Z. Lu, Limitations and
potential of satellite imagery to monitor
environmental response to coastal flooding,
Journal of Coastal Research, 28(2), 2012,
457-476.
[12] D. Roth, Texas Hurricane History, National
Weather Service, 2010. Available at:
http://www.wpc.ncep.noaa.gov/research/txhu
r.pdf.Accessed on: 11 Oct. 2013.
[13] United States Geological Survey (USGS),
National Elevation Dataset, 2013b. Available
at: http://ned.usgs.gov/.Accessed on: 04 Aug.
2013.
[14] J.G. Masek, E.F. Vermote, N. Saleous, R.
Wolfe, F.G. Hall, F. Huemmrich, F. Gao, J.
Kutler, and T.K. Lim, A Landsat surface
reflectance data set for North America, 19902000, Geoscience and Remote Sensing
Letters, 3(1), 2006, 68-72.
[15] C. Berenbrok, R.R. Mason and S.F.
Blanchard, Mapping Hurricane Rita inland
storm tide, Journal of Food Risk
Management, 2(1), 2009, 76-82
[16] B.C. Gao, NDWI - A Normalized Difference
Water Index for remote sensing of vegetation
liquid water from space, Remote Sensing of
Environment, 58(3), 1996, 257266.
[17] X.L. Chen, H.M. Zhao, P.X. Li and Z.Y. Yin,
Remote sensing image-based analysis of the
relationship between urban heat island and
land use/cover changes, Remote Sensing of
Environment, 104(2), 2006, 133146.
[18] D.D. Bosch and M.H. Cosh, SMEX03
Landsat Thematic Mapper NDVI and NDWI
(Boulder, CO: National Snow and Ice Data
Center, 2008).
[19] T.J. Jackson and M.H. Cosh, SMEX03
Landsat Thematic Mapper NDVI and NDWI
(Boulder, CO: National Snow and Ice Data
Center, 2007).
[20] Y. Hong, K.-L. Hsu, H. Moradkhani and S.
Sorooshian, Uncertainty quantification of
satellite precipitation estimation and Monte
Carlo assessment of the error propagation
into hydrologic response, Water Resource
Research, 42(8), 2006, 1-15.
Modeling Floodplain Inundation on a

Regulated River: Integrating GIS, Remote
Sensing, and Hydrological Models
ARTICLE in RIVER RESEARCH AND APPLICATIONS NOVEMBER 2005
Impact Factor: 2.03 DOI: 10.1002/rra.867
CITATIONS
55
READS
248
1 AUTHOR:
Ian Overton
The Commonwealth Scientific and Indus
SEE PROFILE
Available from: Ian Overton
RIVER RESEARCH AND APPLICATIONS
River Res. Applic. 21: 9911001 (2005)

Published online in Wiley InterScience
(www.interscience.wiley.com). DOI: 10.1002/rra.867
MODELLING FLOODPLAIN INUNDATION ON A REGULATED RIVER:

INTEGRATING GIS, REMOTE SENSING AND HYDROLOGICAL MODELS
I. C. OVERTON*
CSIRO Division of Land and Water, Urrbrae, SA, 5064, Australia
ABSTRACT
The lower River Murray in South Australia is highly regulated through weirs and water extraction for
irrigation. Management of
the river for environmental purposes requires an understanding of the extent of floodplain inundation
from various flows and
weir manipulations. This study aimed to produce a floodplain inundation model for the 600 km long
and 15 km wide portion of
the River Murray in South Australia from the New SouthWales border to Lake Alexandrina. The model
was developed using a
Geographical Information System (GIS), remote sensing and hydrological modelling. Flood inundation
extents were monitored
from Landsat satellite imagery for a range of flows, interpolated to model flood growth patterns and
linked to a hydrological
model of the river. The resulting model can be analysed for flows ranging from minimum flow to a 1-in13-year flood event for
any month and weir configuration and has been independently tested using aerial photography to an
accuracy of approximately
15% underestimate. The results have proven the approach for determining flood inundation over a
large area at approximately
one-tenth of the cost of detailed elevation and hydrodynamic modelling. The GIS model allows
prediction of impacts on infrastructure,
wetlands and floodplain vegetation, allowing quantitative analysis of flood extent to be used as an
input into the
management decision process. Copyright # 2005 John Wiley & Sons, Ltd.
key words: flood inundation; modelling; hydrology; GIS; remote sensing; River Murray
INTRODUCTION
The River Murray is Australias largest river and plays a vital role in Australias economic and
environmental
resources. The MurrayDarling Basin catchment occupies one-seventh of Australia and
contributes 70% of
Australias irrigated crops and pastures. The River Murray provides 100% of Adelaides water
supply in dry
months and is the major water supply source to the towns in the basin. The study region
receives almost all of
the catchment run-off, minus the water extracted due to irrigation and evaporation, as all
tributaries including
the Darling River enter the River Murray further upstream.
Management of the River Murray in southeastern Australia has been implemented to
mitigate large floods and to
protect infrastructure, while maintaining storages for regular water supply to irrigators.
Concerns over river health
have increased the attention on environmental flow strategies over the past decade. The
focus is on releasing and
managing flows to provide environmental benefits to the floodplain, wetlands and in-stream
water quality. No
quantitative tool for measuring the extent and volume of flood inundation existed for the
lower part of the River
Murray prior to the original work (Overton et al., 1999). This project was commissioned to
create such a model to
provide input into the management decision process. Previous studies on modelling river
flow and inundation have
been undertaken for wetlands on the Darling River in New South Wales (Shaikh et al., 2001)
and floodplains on
Roanoke River in North Carolina (Townsend and Walsh, 1998) and the River Murray in
Australia (Overton et al.,
1999; Frazier et al., 2003).
Spatial decision support is one of the main roles of Geographical Information Systems (GIS)
and provides an
excellent framework for the integration of multi-criterion evaluation results (Taylor et al.,
1999; Jankowski et al.,
2001). The flood inundation model was developed using a spatial information system
framework that allowed the
integration of non-spatial river flow models with the extent of flood inundation. Riverine
ecosystems benefit from
spatial analysis studies because they encompass three important temporally dynamic
spatial dimensions (along the
Received 12 October 2004
Revised 13 December 2004
Copyright # 2005 John Wiley & Sons, Ltd. Accepted 19 January 2005
*Correspondence to: I. C. Overton, CSIRO Land and Water, Urrbrae, SA, 5064, Australia; School of Earth and
Environmental Sciences,
University of Adelaide, SA, 5005, Australia. E-mail: ian.overton@csiro.au
channel, river height and floodplain width) and the physical and ecological processes taking
place are complex.
Much work has been done on the longitudinal (upstream and downstream) dimension
examining the ecological
impacts of river regulation on native flora, fauna and the physical changes occurring in the
littoral zone (Stanford
et al., 1996). However, long-term modification of flow rates, changes to the frequency of
flooding events and alteration
to the timing of flows have now been identified as causing degradation beyond the littoral
zone into both the
lateral and vertical dimensions (Young, 2001). This is particularly prevalent in the River
Murray ecosystem where
rising saline groundwater has contributed to land degradation throughout the region and
increased accession of
saline groundwater to the river (Jolly, 1996). Furthermore the semi-arid nature of the Murray
floodplain means
that the variability in the size and timing of flows is extreme and unpredictable (Walker and
Thoms, 1993). Consequently,
river regulation in the Murray has severely impacted the balance of physical and ecological
processes
that maintain this unique riverine environment. Managing the river system to balance social,
economic and environmental
requirements, when resources are scarce, requires quantitative analytical tools (Young et al.,
2000). This
paper describes the development of a predictive model to assess the effect of flow
management options on floodplain
inundation.
Mapping flood inundation
The degradation of the River Murray has increased interest and the need for information and
predictive modelling
of river health and the impacts from management and land use options. An Environmental
Flows Decision
Support System (Young et al., 2000) has been developed to look at a number of river
environmental factors such as
area of wetland and aquatic and terrestrial fauna. This model has been advanced into the
Murray Flow Assessment
Tool (Young et al., 2003). Neither model, however, looks at the extent of varying flood sizes
and considers floodplains
as one unit with a commence-to-fill flow and a rate of fill.
Remote sensing is particularly useful for monitoring flood extents because it provides basic
data more cheaply
and efficiently than ground-based methods (Whitehouse, 1989). Previous studies to
determine the aerial extent
of flooding have commonly involved optical satellite image analysis (Usachev, 1985; Walker
et al., 1986;
Townsend and Walsh, 1998; Overton et al., 1999; Shaikh et al., 2001; Sheng et al., 2001;
Frazier et al., 2003),
radar remote sensing (Townsend and Walsh, 1998) or an integration of remote sensing and
GIS (Brivio et al.,
2002). These methods have proved to be very useful and economical for large-area flood
analysis. More detailed
studies have used digital elevation models to create a floodplain surface that can be
inundated at certain river
heights (Townsend and Walsh, 1998). Elevation methods are particularly useful for predictive
studies of changing
flow paths through the floodplain by manipulating flow barriers. However, surface modelling
may not give the best
representation of flood inundation, as there are numerous impediments and small channels
across a predominantly
flat floodplain. The modelling of flood inundation from surface elevation also requires
detailed information on
stage heights, backwater curves, flow impedances and roughness coefficients as simple
height levels are insufficient
in this dynamic environment.
Due to the high level of control over flows in this region, the river exists as a series of pools
under most conditions.
Floods rise and fall in days or weeks and last for several weeks to months, meaning that the
extent of flood
inundation for a given flow has usually extended to its full potential extent before the flood
peak passes. This type
of regulated river is therefore modelled sufficiently from images of various events. Upstream
in the unregulated
reaches, stream flooding is very rapid and sporadic and two flood peaks of equal magnitude
are unlikely to create
the same extent of inundation (Frazier et al., 2003).
Study area
The study area was chosen to capture the floodplain environments of the River Murray in
South Australia from
the New SouthWales border toWellington (Figure 1). This is generally referred to as the lower
River Murray and is
highly regulated by six weirs (which are situated alongside locks for river navigationthe
total structure is usually
referred to as a lock) and the barrages at the mouth of the river. It has large extents of
floodplain and is a discharge
area for the regional saline groundwater system of the MurrayDarling Basin. The ecological
significance of the
Chowilla floodplain, the largest floodplain in this region, is highlighted by its listing under the
UNESCO Ramsar
Convention as a Wetland of International Importance (National Environmental Consultancy,
1988) and its
992 I. C. OVERTON
Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 9911001 (2005)
inclusion in the list of Significant Ecological Assets of the River Murray under the Murray
Darling Basin
Ministerial Councils The Living Murray Initiative (Murray Darling Basin Commission, 2003).
The regulation
of this section of the river allows flows to be manipulated for environmental management to
mitigate increasing
salinization on the floodplain and enhance wetland environments.
Figure 1. Study area and detail showing locks and localities
MODELLING FLOODPLAIN INUNDATION ON A REGULATED RIVER 993
The floodplains along this stretch of the river range from gorge sections below Overland
Corner (Figure 1),
23 km wide and 3040m deep, to valleys 510 km wide flanked by broad floodplains
(Walker and Thoms,
1993). Figure 2 shows a schematic cross-section of the floodplain. GIS data are available for
the study area including
the floodplain boundary and the permanent water in the river channel and wetlands;
however, no data were
available for the flood extent under different flows other than the extent of the 1956 flood
(the largest on record)
at approximately 290 000 Ml/day which covered what we now define as the floodplain
boundary.
METHODOLOGY
Satellite imagery was obtained for this region at a range of magnitudes of flood events and
therefore correct interpretation
of these data provides information on the spatial extents of flood inundation. The process of
developing
the flood inundation model is described in detail below. Figure 3 shows a schematic flow
diagram of the process
involved.
Figure 2. Cross-section of the local geomorphology
Figure 3. Flow diagram of the process of building the Flood Inundation Model
994
I. C. OVERTON
Production of flood inundation maps from satellite imagery

Most studies on identifying water bodies have identified the landscape through image
classification procedures
on multi-band imagery (Munyati, 2000). To reduce costs, most of the information required for
flood mapping can
be obtained from satellite image bands that detect reflected light from the Earths surface in
the mid-infrared wavelength.
Radiation in this region is almost completely absorbed by water and hence images show
sharp contrast
between the high reflectance in soil and vegetation on dry areas and the very low
reflectance of water (Whitehouse,
1989). This study used Landsat Thematic Mapper (TM) satellite data to provide a complete
coverage of the study
site at a cell (pixel) resolution of 30m by 30 m. Landsat TM Band 5 (which detects light
waves in the mid-infrared
region, 1.551.75 mm) was therefore chosen as the single band to determine the presence
of surface water.
Water is detected by isolating the pixels within the image that have a very low reflectance
value, indicating that
the light was absorbed and not reflected, for this band. However, other features, especially
shadow, also have very
low reflectance in this region of the light spectrum. Detecting water from dark shadow is not
possible using a singleband image. This problem is compounded by the slightly higher reflectance of turbid or
shallow water leading
to these features being detected as non-water. The assessment on the cut-off value of
percentage reflectance for
determining surface water in a single-band image (referred to as a density slice) is a
judgement made by the analyst
based on the histogram (Figure 4) and image characteristics. Despite the problems involved
in individual pixel
misclassification, the method of density slicing a single mid-infrared band to detect surface
water has been successfully
used in previous studies and represents an economical method for determining flooding over
large areas
(Sheng et al., 2001). Frazier and Page (2000) found that a simple density slice on Band 5
achieved an overall accuracy
of 96.9% when compared to aerial photographic interpretation, and proved as successful in
delineating water
as a six-band maximum likelihood classification (a multi-band statistical analysis commonly
used in remote sensing
for image classification). Figure 4 shows the cut-off made for one of the images in the study
area.
Twenty-one satellite images (Table I) were chosen to correspond to a range of flow values as
recorded at the
gauging station near the New SouthWales border (Gauging Station 426 200). The flows
range from 3460 Ml/day to
101 900 Ml/day with dates ranging from October 1986 to November 1996. The images were
chosen to occur in
periods of rising hydrographs (increasing flow in the river); this reduced the potential for
remnant water from
higher flows to still be present on the floodplain, and flood edges are harder to delineate
with a receding flood
boundary. Rainfall events were not taken into consideration in choosing the image dates as
water not connected
to the river was able in most cases to be filtered out at a later stage.
Figure 4. Histogram of a 102 Gl/day flow with the central line indicating the point at which the reflectance value is
cut (density slice) to
indicate water or not water. Two low-reflectance-value peaks represent Lake Victoria (with deep water) and the
River Murray (with shallow
turbid water)
Registration and spatial accuracy of the imagery

The successful comparison of images to model flood growth with increasing flow relied on
accurate image registration.
The satellite images had to be registered to map grid co-ordinates so that flood masks could
be used within
the GIS. Registration was performed on the first image using both map co-ordinates and
digital data. All subsequent
images were registered to this image.
The accuracy of the registration was visually assessed as being plus or minus one pixel.
Problems with overlaying
registered images must be considered as errors get compounded with additional overlays.
This problem is of
special concern in this study as individual pixels were being monitored over time for flooding
or not flooding.
Errors in the registration caused a shadowing effect on the edges of the flood extent in some
cases.
Flood map editing and coding of the flood masks
The flood inundation images were used to predict the flood extent for a given flow once any
anomalies had been
removed. Flood extent anomalies can occur from the identification of water that is not
hydrologically connected
to the river, such as remnants from previously larger flows, rainfall events, irrigation
practices or misclassified
pixels. These areas were removed in order to determine the exact extent of flooding that
would occur at a particular
flow. The process involved removing any identified water outside the 1956 flood boundary;
this was the largest
flood event in recorded history. The process also involved investigating the relationship
between images of increasing
flow rate. It was necessary to remove water that was identified to be present at lower flows
but not at higher
flows. The choice of imagery on rising hydrographs assisted with the reduction in such
anomalies. This will be an
ongoing process of editing the flood masks as new information becomes available from
satellite imagery, aerial
photography and ground recordings. Satellite imagery does not detect every pixel of water
within the scene as
some areas are covered by vegetation or have high turbidity or shallow depth. For this
reason, the areas that lie
within the river channel itself and areas classified as permanent wetlands (Pressey, 1986)
were assigned a unique
code.
Interpolation of flood masks
Satellite images were obtained to provide a range of flows, but it was necessary to consider
flows between these
events to provide a more continuous predictive model. Interpolation between the discrete
flow intervals was performed
to produce finer intervals of flood extent.
The extents of the flood masks of each flow from the satellite imagery provided a boundary
line of equal flood
extent. These pixels were interpolated to obtain the flow at all other pixels in the image.
There are many ways to
Table I. Image path and dates for the 21 satellite Landsat TM images used in the study. All images were
from Row 84
Flow Path 97/Row 8485 Path 96/Row 84 Path 95/Row 84
(Gl/day) Lakes to Overland Corner Overland Corner to Lake Victoria Lake Victoria to Wentworth
Date (Gl/day) Date (Gl/day) Date (Gl/day)
<20 9-Nov-94 6.45 30-Sep-94 3.5 9-Nov-94 6.1
2030 20-May-89 26.7 20-Nov-89 25.8 8-May-90 27.6
3040 10-Aug-96 37.8 11-Oct-92 38.2 12-Aug-96 38
4045 25-Jul-90 40.6 27-Oct-86 43.8
4550 3-Aug-90 46.7 21-Sep-93 47.4
5060 20-Sep-96 57 2-Sep-95 55.1
6070 23-Nov-96 68.3 23-Nov-96 68.3 7-Oct-93 66.4
7075 31-Jul-89 70.5
7580 16-Aug-89 78.1 16-Aug-89 77.4
80100 20-Sep-90 93.4 3-Oct-89 82.4 26-Sep-89 80.2
>100 22-Nov-93 109.9 26-Oct-90 101.6 29-Sep-90 101.9
996
I. C. OVERTON
perform this kind of interpolation. The true situation is defined by the local topography of the
area which is
unknown in this case. Therefore interpolation of the flow level is similar to the problem of
interpolating the landscape
height at each point. Kriging is the most common surface interpolation method (Burrough,
1986) but is influenced
by areas outside the adjacent known boundaries being the two closest satellite masks. It
was decided to use
an image morphological process called a marker-based watershed segmentation algorithm
rather than traditional
interpolation methods to ensure that the information from each satellite image was
preserved. The watershed algorithm
is commonly used in mathematical morphological problems and is often used in relation to
topographic
analysis of digital elevation models (Vincent and Soille, 1991). The advantage of this method
over other
approaches to contour interpolation is that it can be applied to very noisy data with broken
contours. This was
the case here since the data are quantized so coarsely.
Flood interpolation was undertaken by dividing the combined satellite mask image into
regions of constant
minimum flow. Using this method, interpolated values were ensured of lying within the
minimum and maximum
flow bounds by interpolating each region independently. In each region, contours of equal
flow are interpolated
from the boundary points, at which the minimum flow is known. This contour interpolation is
repeated iteratively,
each time based on the contours that have already been estimated. Each contour is
interpolated using a flooding
simulation extending from the next higher and lower boundaries. Regions having the
minimum flow value or representing
land that did not flood at the largest flow were not included and were used as the lower and
upper limit of
interpolation accordingly.
The difficulty in applying the watershed algorithm to flood interpolation is in choosing the
source and sink
points. The sink points are points adjacent to the region that have the next highest
quantized flow level. The source
points are those adjacent points that have the highest flow that is lower than the sink points
for this region. The
locations of these sink and source points were chosen to model the behaviour of flood
growth across the floodplain
but also to represent the filling of wetlands from a single inflow channel. A set of colour
aerial photographs of a
70 000 Ml/day flood were used to validate the growth behaviour of floods, and a small
number of edits were made
to the final flood inundation grid.
GIS development
The result of the interpolation stage was a raster grid of cell values that represent the
commence-to-flow based
on the flows on the day of the satellite images. This grid was then filtered to remove noise
and converted to a
polygon coverage.
Filtering is a technique used to enhance the quality of digital imagery by changing the
values of cells in raster
images, and can be used to ease the computation burden of raster to vector conversion by
removing isolated pixels
(Trotter, 1991). Filtering uses neighbouring cells to determine the value of the cell in
question and can sharpen or
smooth images to emphasize features or reduce the effects of erroneous cells (Wilkinson,
1996). To remove
anomalies in the interpolated flood masks, such as higher or lower value pixels in the middle
of lakes, a majority
filter was used. This filter replaced the value of each pixel with the value of the eight nearest
neighbours if there is a
majority of neighbouring cells with a different value from the cell.
At this stage of the processing the imagery was still in raster format (pixel-based data). The
raster data were
converted to vector data (area- or shape-based data) to reduce the data redundancy and
allow for easy retrieval,
updating and generalization of graphics and attributes that are especially important if
equations are to be used to
generate models.
Hydrological modelling
Output from the hydrological models was provided as input data to the GIS. The GIS model
then used table
look-ups to query the required hydrological parameters that then link to the spatial layers in
the GIS. This type
of loose coupling allows existing hydrological models to be run without reprogramming
within the GIS (Karami
and Houston, 1996; Sui and Maggio, 1999). As the hydrological parameters do not change,
this proved an effective
and economical approach.
River flow and height model. To create a model that would predict flood inundation from a
regulated river, the
river height at every kilometre is required. Actual recorded data of river height at certain
kilometres from the
mouth of the river, and flow at the gauging station during the flooding events, were obtained
from the South
Australian Water Authority, the organization which manages the river in this region. The
limited number of
gauging stations meant that a modelled series of backwater curves were also used. These
were a series of water
surface elevation curves between weirs for a steady uniform flow computed by the Murray
River Level Module
(MURLEV), part of the River Murray Flow and Salt Transport (RMFST) computer model (Water
Studies, 1992).
Figure 5 shows the flow curves derived from these methods for a section of the river from
the border to weir 2. The
flow rate of 3000 Ml/day equates to a river height referred to as weir pool level.
The river is held higher than pre-regulation by the weirs, which causes the water levels to
have a backwater curve
rather than a straight flow down hill. River heights near the mouth of the river are much
lower than near the border
due to downfall flow of the river and attenuation of the flow. Once flow levels have reached
60 000 Ml/day the river
is above the maximum height of the weirs and backwater curves become straight.
Flood units. As river heights can be manipulated by the six weirs, predominantly
independent of each other, the
floodplain needed to be divided up into regions that could be related to the local conditions
in the river. Areas where
the floodplain inundation responded to the same point in the river were identified using flood
behaviour and river
morphology as a guide. Regions identified that respond to a particular point in the river were
termed Flood Units.
Flood Units were assigned the closest kilometre marker from the hydrological model (trigger)
or the trigger that
was assumed would most likely cause that area to flood. As a height difference of several
metres can occur between
the upper and lower pool levels of the weirs, behaviour of the flow path around the weir is
complex. Aweir manipulation
trial at Lock 5 demonstrated that the area around the weir responded to changes in the river
height below the
weir. There is very little difference between the river height within 2-km intervals in the
modelled backwater
curves, which implies that there is probably an error margin of approximately two triggers
up- or downstream
in the assignment of a Flood Unit to a trigger. Each polygon for each particular reach now
contained attributes
for Reach, Flood Unit, Flow and Trigger.
Once the data had been converted to polygons in the GIS it was clear that each Flood Unit
contained small areas
of floodplain that were inundated at the same particular flow. These areas were termed
unique Ecological Units as
they will have similar ecology due to their similar elevation and the fact that distribution of
floodplain vegetation
communities is strongly related to flooding frequency (Van Der Sommen, 1987).
Hydrological inputs to the model. The same flow rate at the border gauging station can
produce different river
elevations along the river depending on the weir elevations, the antecedent conditions and
the time of year. Local
river height predominantly determines whether the river will break the banks and flood an
adjacent Flood Unit.
River height at kilometre markers from state border to lock 2
0.00
5.00
10.00
15.00
20.00
25.00
362.1
368
378
388
403
409
418
429
439
454
480
489
500
515
520
531
542
555
566
575
584
594
603
623
633
642
River km markers
River height (m)
3
26
47
55
82
102
Flow rate at border
(Gl/day)
Figure 5. River levels for kilometres of the river from weir 2 to the SANSW border for different flows
998 I. C. OVERTON
Plate 1. Example of seasonal attenuation curves
Backwater Profile for Lock Pool 3,
at 5cm above normal pool level
9.2
9.4
9.6
9.8
10.0
10.2
10.4
10.6
10.8
11.0
11.2
11.4
11.6
11.8
12.0
12.2
12.4
12.6
12.8
13.0
13.2
13.4
13.6
13.8
14.0
430 440 450 460 470 480 490 500 510 520
AMTD (km)
AHD (m)
56295
50000
40000
30000
20000
15000
10000
5000
pool level
Flow (ML/day)
Plate 2. Example of a backwater profile for Lock 3

Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: (2005)
Plate 3. Example output of the GIS Flood Inundation Model showing an area around Lock 6 at Chowilla with two flood
predictions of
60 000 Ml/day (dark) and 100 000 Ml/day (light)
Plate 4. Area of flood inundation versus flow size for the South Australian Border (Chowilla floodplain) to Wellington.
The graph shows the
area of inundation compared to the area of permanent water and the extent of the 1956 flood (_258 000 Ml/day)
Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: (2005)
Other factors such as antecedent conditions were not included in the model. In the hot dry
summer period a flow
rate at the border will be further reduced downstream than in wetter months as water is
evaporated and extracted
from the river and less is replaced by rainfall. Curves were generated which show the
attenuation of flow at each
weir for a given month. Plate 1 shows an example of the change in flow rate during the year
in each lock reach for a
flow of 60 000 Ml/day at the border.
The RMFST model was used to generate backwater curves for a range of weir manipulations
and flow values.
The model estimates the stagedischarge matrices for a nominated water level and
discharge at the weir, which
defines the upstream water levels to the next weir. In this way relationships can be derived
between the water levels
at the downstream and upstream ends of the river segment and discharge. The river level
and discharge relationships
for each weir reach were established using a linear relationship. Plate 2 shows the backwater
curve for the
weir pool of Lock 3 at a level of 5 cm above normal pool level. The precision of the model is
5 cm in the river
height predictions.
Decision support system development
The height of the river at triggers had to be related to the management areas, the Ecological
Units derived from
the original flow values and the interpolation process. With area inundated extremely
sensitive to slight alterations
in river height, it was decided that a relationship between river height and area of flood
inundation should be used
in each Flood Unit. This relationship has the potential to be improved with further image
acquisition and statistical
analysis. The model will always be limited to the spatial resolution at which the images are
acquired, which for
Landsat TM is 30m, regardless of the number of images used. If a change in the height of
the river results in less
than a 30m change at the flood boundary, the area of inundation will not increase. Other
satellite systems may be
useful in reducing the pixel size and therefore identifying finer changes in flood inundation.
In the region downstream of Lock 3, the height of the river continues to increase without
breaking its banks until
it reaches approximately 35 000 Ml/day. As the water flows onto the floodplain there is a
gradual increase in area
with increasing flow until the river level rises to reach the edges of the trench (Figure 2). In
the region upstream of
Lock 3 the river height increases until it breaks its banks and continues to increase as it fills
up wetlands and then
gradually tapers out as the water spreads over the floodplain. The relationships at each
floodplain unit were used to
code the flood masks with a river height which would cause the Ecological Unit to be
inundated. The height of the
river was assigned to the Ecological Unit for all Flood Units using the Flood Unit trigger as
the point that would
cause this Ecological Unit to flood. A dialogue screen was then designed with input fields to
ask the user the values
for the flow at the border, the month of the year and the height of each lock. These values
are then used to perform a
query which selects all the river height codings in the map layer where the local river height
is less than that predicted
for the given flow (attenuated for the time of the year).
The flood inundation model is a steady-state model that predicts the extent of flooding from
a given flow on the
first day of the flood. It does not consider the effect of antecedent conditions or the effect of
flood duration. Further
research on the wetting and drying behaviour of the floodplain and its wetlands needs to be
incorporated into the
model to be able to predict time sequences for management scenarios. The model is limited
by its ability to predict
areas of inundation given the same floodplain topography. Changes in elevations due to
construction of levee banks
and regulators will cause changes in the area of inundation not modelled by the Flood
Inundation Model. The
model can provide wetting and drying cycles for wetlands, days since last flooding, average
return period, volume
of water, extent of inundation and other hydrological indicators that could be linked to
floodplain ecological processes.
Better understanding is needed to determine, or at least demonstrate in most cases, these
linkages between
the hydrological indicators and the benefits to the environment.
RESULTS
The Flood Inundation Model produced has been used as the basis for an environmental flow
strategy in South
Australia (SKM and Mapping and Beyond, 2000) and a number of other studies on the
hydrology and ecology
of the floodplain. The Flood Inundation Model was used to predict the area of inundation
expected to occur as
a result of the raising of Lock 5 near Renmark in 2000 (DWLBC, 2002). Aerial photography
was obtained during
the raised weir and was interpreted for area of inundation. The spatial accuracy of the model
in predicting the area
inundated was estimated to be 15% underestimate of the actual inundated area from the
photography. A database of
the surveyed heights of 126 bores on the floodplain was used to independently test the
accuracy of the height data
derived by the model. The mean difference between the surveyed height and modelled
height was 0.07m with a
standard deviation of 0.88 m. Plate 3 shows an example of the GIS Flood Inundation Model.
The initial stages of the project provided us with Ecological Units and the height of the river
at a particular
trigger that would cause these areas to flood. The hydrological modelling allows the monthly
simulation of a flow
from the border under different weir configurations, with the GIS providing the spatial view
of the area inundated.
The inputs to the decision support system are the flow at the border, the weir configurations
of all six weirs and the
month. The outputs of the model will be the river heights at each trigger kilometre and the
area of inundation. The
flood inundation model created is in a GIS framework and has incorporated layers such as
riparian vegetation,
transport and water infrastructure and major wetlands. The science of incorporating spatial
patterns in decision
support systems through the integration of GIS and process models is suggested as a critical
tool for the future
of information integration in the environmental sciences (Taylor et al., 1999; Hendriks and
Dirk, 2000).
The total area of inundation of the floodplain in South Australia, including the whole of the
Chowilla floodplain,
which is partly in New South Wales, down to Wellington, is 117 827 hectares (downstream of
the gauging station
on Figure 1). The floodplain is defined as the extent of the 1956 flood which was
approximately 258 000 Ml/day.
Plate 4 shows the increasing area of inundation of the River Murray floodplain from
5000Ml/day to 100 000Ml/
day. The area of permanent water is approximately 24 000 hectares and is made up of the
river channel, wetlands
and anabranch creeks.
CONCLUSION
Satellite imagery of known flood events has been interpolated to create a model of flood
growth. This flood growth
has been linked to a hydrological model of the river, making predictions of the extent of
flood inundation from
given river flows and manipulation possible. The GIS framework could be utilized to
determine flow patterns and
losses across the floodplain, incorporating wetlands as sources and sinks to create a more
dynamic model of flooding
(Costelloe et al., 2003).
The integration of the hydrological model with the GIS has enabled simulation for both
scientific research and
policy management. The visualization, quantitative analysis and spatial correlation of
environmental and infrastructure
data of the GIS has improved the usefulness of the hydrological modelling.
The model has been used successfully to predict flood inundation in a weir manipulation trial
(DWLBC, 2002)
and has been used as the basis for the development of a flow management strategy for the
River Murray in South
Australia (SKM and Mapping and Beyond, 2002). The model has also provided a useful
surface elevation model in
further floodplain modelling work. Current work on modelling floodplain vegetation health
affected by soil salinization
processes will be linked to the flood inundation model to improve its usefulness as a decision
support
system. Further research on the behaviour of floods on the floodplain and in wetlands is
required before it can
be used as a temporal model.
ACKNOWLEDGEMENTS
The author wishes to acknowledge funding through the Murray Darling Basin Commission,
the River Murray
Catchment Management Board and the South Australian Water Corporation (MDBC Natural
Resource Management
Strategy R6045 project). Assistance on the initial project design was provided by Brenton
Erdmann and Bob
Newman of the South Australian Water Corporation. Jamie Sherrah performed the
mathematical image interpolation,
while Suzanne Slegers and Jane Lawley, of Mapping and Beyond Pty Ltd, assisted with the
data manipulation.
REFERENCES
Brivio PA, Colombo R, Maggi M, Tomasoni R. 2002. Integration of remote sensing data and GIS for accurate mapping
of flooded areas. International
Journal of Remote Sensing 23(3): 429441.
Burrough AP. 1986. Principles of Geographical Information Systems for Land Resources Assessment. Oxford
University Press.
1000 I. C. OVERTON
Costelloe JF, Grayson RB, Argent RM, McMahon TA. 2003. Modelling the flow regime of an arid zone floodplain river,
Diamantina River,
Australia. Environmental Modelling and Software 18(89): 693703.
DWLBC. 2002. Lock 5 Flow Enhancement Trial. Murray-Darling Division, Department of Water, Land and Biodiversity
Conservation, South
Australia. Report No. 1.
Frazier PS, Page KJ. 2000. Water body detection and delineation with Landsat TM data. Photogrammetric
Engineering and Remote Sensing
66(12): 14611467.
Frazier PS, Page KJ, Louis J, Briggs S, Robertson A. 2003. Relating wetland inundation to river flow using Landsat TM
data. International
Hendriks P, Dirk V. 2000. From geographical information systems to spatial group decision support systems: a
complex itinerary. Geographical
and Environmental Modelling 4: 1.
Jankowski P, Andrienko N, Andrienko G. 2001. Map-centred exploratory approach to multiple criteria spatial decision
making. International
Journal of Geographical Information Science 15(2): 101127.
Jolly ID. 1996. The effects of river management on the hydrology and hydroecology of arid and semi-arid
floodplains. In Floodplain Processes,
Anderson MG, Walling DE, Bates PD (eds). John Wiley and Sons: New York; 577609.
Karami HA, Houston BH. 1996. Evaluating strategies for integrating environmental models with GIS: current trends
and future needs. Computers,
Environment and Urban Systems 20(6): 413425.
Munyati C. 2000. Wetland change detection on the Kafue Flats, Zambia, by classification of a multitemporal remote
sensing image dataset.
International Journal of Remote Sensing 21(9): 17871806.
Murray Darling Basin Commission. 2003. The Murray Darling Basin Commission website. http://www.mdbc.gov.au
[November 2003].
National Environmental Consultancy. 1988. Chowilla Salinity Mitigation SchemeDraft Environmental Impact
Statement. Report prepared by
National Environmental Consultancy for the Engineering and Water Supply Department of South Australia: Adelaide.
Overton IC, Newman B, Erdmann B, Sykora N, Slegers S. 1999. Modelling floodplain inundation under natural and
regulated flows in the Lower
River Murray. Second Australia Stream Management Conference Proceedings, 811 February 1999, Adelaide,
Australia: 493498.
Pressey RL. 1986. Wetlands of the River Murray below Lake Hume. Environmental Report 86/1. Murray-Darling Basin
Commission: Canberra.
Shaikh M, Green D, Cross H. 2001. A remote sensing approach to determine the environmental flows for wetlands of
the Lower Darling River,
New South Wales, Australia. International Journal of Remote Sensing 22(9): 17371751.
Sheng Y, Gong P, Xiao Q. 2001. Quantitative dynamic flood monitoring with NOAA AVHRR. International Journal of
Remote Sensing 22(9):
17091724.
SKM and Mapping and Beyond. 2002. Decision Support System for Managing Flows in the River Murray, South
Australia. Report prepared for
the River Murray Catchment Management Board, the South Australian Department of Water, Land and Biodiversity
Conservation, and the
Natural Heritage Trust.
Stanford JA,Ward JV, Liss WJ, Frissell CA,Williams RN, Lichatowich JA, Countant CC. 1996. A general protocol for
restoration of regulated
rivers. Regulated Rivers: Research and Management 12: 391413.
Sui DZ, Maggio RC. 1999. Integrating GIS with hydrological modelling: practices, problems and prospects.
Computers, Environment and
Urban Systems 23: 3351.
Taylor K, Walker G, Abel D. 1999. A framework for model integration in spatial decision support. International Journal
of Geographical
Information Science 13(6): 533555.
Townsend PA,Walsh SJ. 1998. Modelling floodplain inundation using an integrated GIS with radar and optical remote
sensing. Geomorphology
21: 295312.
Trotter C. 1991. Remotely-sensed data as an information source for geographical information systems in natural
resource management: a
review. International Journal of Geographical Information Systems 5(2): 225239.
Usachev VF. 1985. Evaluation of flood plain inundations by remote sensing methods. Hydrological applications of
remote sensing and remote
data transmission. Proceedings of the Hamburg Symposium, August 1983, Goodison BE (ed.). Publication No. 145.
International Association
Hydrological Sciences: Washington; 235243.
Van Der Sommen FJ. 1987. A Review of Components and Processes Influencing the Ecological Status of the
Vegetation Component of the
Riparian Ecosystems of the River Murray Floodplain in South Australia. South Australian Department of Environment
and Planning Report.
Vincent L, Soille P. 1991. Watersheds in digital spaces: an efficient algorithm based on immersion simulations. IEEE
Transactions on Pattern
Analysis and Machine Intelligence 13(6): 583598.
Walker KF, Thoms MC. 1993. Environmental effects of Flow Regulation on the Lower River Murray, Australia.
Regulated Rivers: Research and
Management 8: 103119.
Walker RD, Aubrey MC, Fraser SJ. 1986. Remote Sensing in Australia. Canbec Press: Melbourne.
Water Studies. 1992. River Murray Flow Model, Vol. 3. User Manual. Department of Engineering and Water Supply.
Whitehouse G. 1989. Flood monitoring and floodplain studies. In Remote Sensing in Hydrological and
Agrometeorological Applications, de
Vries D (ed.). COSSA Publications: Canberra.
Wilkinson GG. 1996. A review of the current issues in the integration of GIS and Remote Sensing data. International
Journal of Geographic
Information Systems 10(1): 85101.
Young WJ (ed.). 2001. Rivers as Ecological Systems: The Murray-Darling Basin. Murray Darling Basin Commission:
Canberra, Australia.
Young WJ, Lam DC, Ressel V, Wong IW. 2000. Development of an Environmental Flows Decision Support System.
Environmental Modelling
and Software 15: 257265.
Young WJ, Scott AC, Cuddy SM, Rennie BA. 2003. Murray Flow Assessment Tool: A Technical Description. CSIRO
Client Report for the CRC
for Freshwater Ecology.
HYDROLOGICAL PROCESSES
Hydrol. Process. 19, 36993716 (2005)
Published online 1 August 2005 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/hyp.5852
Remote sensing and GIS-based flood

vulnerability
assessment of human settlements: a case
study
of Gangetic West Bengal, India

Joy Sanyal* and X. X. Lu
Department of Geography, National University of Singapore, Singapore 117570, Singapore
Abstract:
Flooding due to excessive rainfall in a short period of time is a frequent hazard in the flood plains of
monsoon Asia. In
late September 2000, a devastating flood stuck Gangetic West Bengal, India. This particular event has
been selected for
this study. Instead of following the conventional approach of flooded area delineation and overall
damage estimation,
this paper seeks to identify the rural settlements that are vulnerable to floods of a given magnitude.
Vulnerability
of a rural settlement is perceived as a function of two factors: the presence of deep flood water in and
around the
settlement and its proximity to an elevated area for temporary shelter during an extreme hydrological
event. Landsat
ETMC images acquired on 30 September 2000 have been used to identify the non-flooded areas within
the flooded
zone. Particular effort has been made to differentiate land from water under cloud shadow. ASTER
digital elevation
data have been used to assess accuracy and rectify the classified image. The presence of large
numbers of trees around
rural settlements made it particularly difficult to extract the flooded areas from their spectral
signatures in the visible
and infrared bands. ERS-1 synthetic aperture radar data are found particularly useful for extracting the
settlement
areas surrounded by trees. Finally, all information extracted from satellite imageries are imported into
ArcGIS, and
spatial analysis is carried out to identify the settlements vulnerable to river inundation. Copyright
2005 John Wiley
& Sons, Ltd.
KEY WORDS flood; remote sensing; settlement vulnerability; GIS
INTRODUCTION
Flood is a perpetual natural hazard in the flood plains of monsoon Asia, where over 80% of
annual precipitation
is received in the four wet months from June to September. The problem of river flooding is
of great concern
in the Indian state of West Bengal. The Irrigation and Waterways Department of West Bengal
Government
reports that, since its independence in 1947, there has only been 5 years in which it has
been spared from the
effects of monsoon inundation. One of the most devastating of its kind was experienced in
SeptemberOctober
2000. A total of 23 756 km2 land area was inundated and 221 million people were affected
(Rudra, 2001).
This particular flood event caught everybody unaware. The unprepared condition of the
administration to
cope with this kind of natural calamity was thoroughly exposed. In recent years, a number of
studies have
recognized the importance of estimating peoples vulnerability to natural hazards, rather
than retaining a
narrow focus on the physical processes of the hazard itself (Hewitt, 1997; Varley, 1994;
Mitchell, 1999).
Cannon (2000) argued that natural disaster is a function of both natural hazard and
vulnerable people. He
emphasized the need to understand the interaction between hazard and peoples
vulnerability.
Although most of the developed countries are well equipped with detailed flood hazard
maps, up-to-date
flood insurance maps and post-disaster hazard mitigation technical support (FEMA, 2003),
there is hardly any
* Correspondence to: Joy Sanyal, Department of Geography, 1 Arts Link, National University of Singapore, Singapore
117570, Singapore.
E-mail: g0202381@nus.edu.sg
Received 23 January 2004

Copyright 2005 John Wiley & Sons, Ltd.
3700 J. SANYAL AND X. X. LU
Accepted 24 September 2004
detailed spatial database for flood prevention and mitigation in the developing countries. In
recent years, efforts
have been made to use remote sensing and geographic information systems (GISs) for
creating national-level
flood hazard maps for Bangladesh (Islam and Sado, 2000a). Population density and other
socio-economic
data have been integrated with hydrologic information to identify priority zones for
implementing anti-flood
measures (Islam and Sado, 2002). These studies were undertaken on a regional scale using
coarse-resolution
AVHRR imageries from NOAA satellites. The results of such investigations would only be
useful for nationallevel
macro planning.
The focus of this paper is on individual settlements. Its purpose is to analyse how the
location of an
individual settlement vis-à-vis the flood-prone zone and its socio-economic characteristics
make it vulnerable
to monsoon floods. The theoretical framework of this study is based upon the hypothesis
that settlements
are vulnerable to flood from three aspects: (1) whether the people have access to relatively
higher ground to
take shelter during an extreme hydrological event; (2) whether a settlement falls in a zone
that is expected
to experience a high flood discharge causing extraordinary damage of life and property; (3)
whether the
population density of the area is high enough to result in huge loss of property even in a
moderate flood.
The example used for this study is Gangetic West Bengal, which is a natural flood-prone area
whose fluvial
characteristics have made it very suitable for rice cultivation. The population density in this
region is one of
the highest in the world. Although human settlement is abundant in this region, the local
people traditionally
settled only on relatively higher ground, known locally as danga. Rapidly increasing
population density, due
to both natural growth and the influx of millions of refugees in the post-independence era
from the then East
Pakistan, left very little choice for the people to settle on higher ground selectively. Severe
shortage of land
has forced people to settle indiscriminately over the highly flood-prone zone.
Landsat ETMC and ERS synthetic aperture radar (SAR) imageries are used in this study to
classify nonflooded
areas and flood depth within flooded zones, and to delineate human settlements at village
level. The
high spatial resolution of satellite imageries enables us to obtain detailed classification
results that are suitable
for formulating planning measures on a small scale. An added advantage is that the high
resolution hydrologic
information can be conveniently integrated with demographic data collected from smaller
administrative units.
This would greatly enhance the capability of the spatial database to estimate vulnerability of
individual
settlements to an extreme flood event.
STUDY AREA
The study area extends over three major river basins of southern West Bengal, namely
Bhagirathi-Hoogly,
Jalangi and Churni. All these three rivers are distributaries of the main branch of the River
Ganga. Although
we have tried to cover the natural region of the three river basins, the extent of the
investigation is marginally
compromised due to limited availability of digital terrain data. The eastern parts of the
districts of Bardhaman,
Murshidabad, and most of Nadia form the administrative entity for the area (Figure 1). The
three river
basins are overwhelmingly rural, with agriculture as the main source of livelihood. According
to Bagchis
(1945) subregional classification of the Bengal Delta, the study area is identified as a
moribund delta. In this
section of the delta, the rivers are decaying and the land-building process has entirely
ceased. Owing to its
comparatively higher elevation and high levees, this area is traditionally less flood prone
than the area that
lies further south. The area falling between the Rivers Bhagirathi and Jalangi is an elongated
depression,
and the Churni basin area is almost entirely low lying in comparison with rest of Gangetic
West Bengal.
Therefore, this zone is liable to flooding. The study states that, in the Nadia and Hoogly
districts, this belt is
bounded by 10 m contour lines. Interfluves of the numerous distributaries are ill drained
(Spate, 1965) and
very often cause waterlogging during the monsoon season. This situation ultimately led to
stagnation of water
and development of cut-off channels known as bills. An abundance of oxbow lakes and misfit
river channels
also characterize this part of Gangetic West Bengal. There is a marked distinction in the
channel pattern of
the streams lying east and west of the River Hoogly. Sinuosity indices of the rivers on the
eastern side of the
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
REMOTE SENSING AND GIS-BASED FLOOD VULNERABILITY ASSESSMENT 3701
Bhatar
Chapra
Pandua
Kaliganj
Kandi
Mangolkot
Nakshipara
Domkal
Sagardighi
Chakdaha
Noada
Jamalpur
Lalgola
Berhampur
Balagar
Nabagram Jalangi
Manteshwar
Tehatta-I
Bardhaman-I Santipur
Memari-I
Hanskhali
Kalna-I
Ranaghat-II
Hariharpara
Katwa-I
Memari-II
Karimpur-II
Kalna-II
Ketugram-I
Ranaghat-I
Krishnanagar-I
Beldanga-II
Karimpur-I
Tehatta-II
Katwa-II
Bharatpur-I
Beldanga-I
Raninagar-I
Purbasthali-II
Krishaganj
Raninagar-II
Ketugram-II
Bharatpur-II
Bhagawangola-I
Nabadwip
Murshidabad-Jiaganj
Purbasthali-I
Bhagawangola-II
Krishnanagar-II
Bardhaman-II
Legend
Rivers
Development Blocks
Study Area
Study Area
Location within India
50 25 0 50 Kilometers
N
Figure 1. Administrative boundary of the study area
River Hoogly are very high compared with the western side (Goswami, 1983). The overall
geomorphology
of the study area depicts a degenerating fluvial system.
Owing to its geographical location, i.e. at the tail end of the extensive Ganga basin, West
Bengal has a
very limited capacity to control extreme hydrological events ensuing from the upper
catchment of the River
Ganga and its tributaries. Very high precipitation over a short period of time is cited as the
most important
factor responsible for triggering devastating floods in Gangetic West Bengal (The Times of
India, 2000). After
the Independence of India, 1956, 1959, 1978, 1995, 1999 and 2000 are identified as years
that received
abnormally high precipitation and, hence, severe floods (Basu, 2001). The 2000 flood in
SeptemberOctober
was the worst in terms of its scale and damage caused. The West Bengal Government
estimated that a total of
171 blocks of the state _23 756 km2_ area was affected. The total loss was estimated to be
56 600 million rupees
(US$1132 million) (Ganashakti, 2000). Abnormally high rainfall for 4 days in the upper
catchment areas of
the western tributaries of the River Bhagirathi were responsible for this natural calamity. The
severity of the
event was so high that many low-lying areas of the Nadia district remained waterlogged for
over 3 weeks,
with the depth of water estimated as high as 3 m (Rudra, 2001).
DATA AND METHODS

Work flow of the current investigation has four components: (1) delineating extent of nonflooded surface
in and around the major flood zone; (2) extracting the high flood-depth zone from the
flooded areas;
(3) demarcating the human settlements of the area; (4) importing the above three layers
into a vector GIS
environment and performing spatial analysis to obtain relevant results.
Data sources for the first three components are satellite images. Semi-automatic digital
image processing
techniques and manual digitization have been employed to extract the relevant information
from remotely
sensed data. The fourth component involves incorporation of demographic parameters with
the information
extracted from satellite imageries to identify effectively the settlements that are highly
vulnerable to flood
hazard. In this section, each component of the total work flow has been dealt with
separately. Special attention
has been given to illustrating how each of these components contributes to the ultimate
objective of this study.
Delineating non-flooded areas from flooded areas
The current study is concerned more with dry/land area than flooded area. Delineation of the
non-flooded
area is particularly important because these areas can serve as a temporary shelter for the
nearby settlements.
This information is necessary for identifying the settlements that are highly vulnerable to
flooding. Settlements
having no immediate access to dry areas would be considered highly vulnerable to flooding.
From the early era of passive remote sensing, special attention has been given to
distinguishing water
from dry surface. MSS band 7 _0811 m_ has been found to be particularly suitable for
distinguishing
water or moist soil from dry surface due to its strong absorption of water in the near-infrared
(NIR) range of
the spectrum (Smith, 1997). MSS data were used to deal with the flood-affected areas in
Iowa (Rango and
Solomonson, 1974), Arizona (Morrison and Cooley, 1973), and the River Mississippi basin
(Deutsch et al.,
1973; Deutsch and Ruggles, 1974; Rango and Anderson, 1974; McGinnis and Rango, 1975).
From the early 1980s, Landsat TM data with an improved spatial resolution of 30 m have
become one of
the major sources of remotely sensed data for flood management research. Landsat TM
band 4 is spectrally
a near equivalent of MSS band 7. Water yields very low reflectance in the NIR region of the
spectrum and,
therefore, can be effectively used to discriminate water from land surface. This property of
Landsat band 4 has
been extensively used to delineate flooded areas in West Africa (Berg and Gregiore, 1983),
India (Bhavsar,
1984) and Thailand (Raungsiri et al., 1984).
For the current research, two Landsat ETMC scenes of the study area acquired on 30
September 2000 have
been obtained. The imageries were acquired nearly at the peak of the flood. The scenes
were geometrically
and radiometrically corrected (level 1G product from US Geological Survey (USGS)). The two
scenes have
been accurately georeferenced using GPS control points collected during a field visit to the
study area. All
bands of the two scenes have been mosaicked. TM bands 4, 3 and 2 have been projected in
RGB to generate
a standard false colour composition (FCC) of the study area. Although TM band 4 is useful in
delineating
land and water boundaries, asphalt road surfaces and rooftops also yield very low
reflectance in this band.
Reflectance from water varies sufficiently from roads and dark rooftops in Landsat band 7
(228235 m,
mid infrared). Therefore, water and non-water can be effectively discriminated by adding TM
band 4 and
band 7 (Wang et al., 2002).
The predominance of cloud cover during the flooding season is one of the major obstacles to
using optical
remote sensing in flood management. Although SAR can penetrate cloud cover, scientific
communities in
the developing countries still prefer optical data over radar because of two reasons.
Purchasing radar data
is beyond the means of most of the government agencies in the developing countries
because of its high
price. Radar satellite needs prior programming to capture data for a particular area. Owing
to its very high
rate of data capture it requires a real-time data transfer to a receiving station near the
coverage area. Lack of
ground receiving stations and proper coordination has resulted in limited coverage of SAR
data for developing
countries. Major satellites that operate in the optical portion of the spectrum maintain a
year-round archive
of global coverage and, therefore, are still more acceptable in the field of remote sensing.
After adding TM bands 4 and 7, an effort was made to mask out the cloud-contaminated
pixels, as their
existence can interfere with any classification effort. There are a number of algorithms for
screening clouds
in optical imageries (England and Hunt, 1985; Saunders, 1986). Cloud appears very bright in
a standard FCC
image. After a detailed comparison of the band _4 C 7_ image with the FCC, it has been
found empirically
that any pixel having a digital number (DN) of over 124 in band _4 C 7_ is cloud. The pixels
with a (DN)
range >124 have been masked. It has been observed that it is very easy to extract cloudcovered pixels over a
dark background (e.g. flooded area in band _4 C 7_), but over a bright background, like
settlements or healthy
vegetation, pixels at the periphery of a cloud cluster have a very similar reflectance to the
background.
Therefore, the above-mentioned threshold is not able to remove small numbers of cloudcovered pixels
over bright and dry surfaces. Since this study focuses on flooded areas, the existence of
small numbers of
cloud-contaminated pixels in the predominantly dry areas is not of great concern.
Band _4 C 7_ is quite efficient in discriminating flood water from dry land. Major boundaries
of flood water
can be delineated without much effort. However, there are still a few water pixels that are
not classified
as flooded within the rural settlements (Figure 2). These pixels can be readily identified in a
standard FCC.
The reflectance of these pixels is very close to the nearby wet soil surface. During floods, the
albedo from
the water body increases significantly because of a high concentration of debris and silt in
the water. Thus,
the reflectance peak moves toward the red band. On the other hand, increasing soil
moisture decreases soil
albedo, making reflectance from some non-flooded pixels very similar to flood pixels (Sheng
et al., 1998).
From a close comparison of band _4 C 7_ with the FCC, a binary classification has been done
as follows for
band _4 C 7_:
Pixel value > 78 D dry land
Pixel value _ 78 D water
This classification effectively extracts water pixels within the settlement area, but the main
disadvantage of
this classification is that it cannot distinguish between dry surface and water under cloud
shadow.
Areas under cloud shadow receive only scattered sunlight, but the low illumination results in
a suppressed
reflectance from all land-cover categories. Therefore, it is very difficult to discriminate
between land and
water for the areas under cloud shadow (Sheng et al., 1998). Water has a significantly low
reflectance in
the NIR region of the electromagnetic spectrum compared with dry land surface. Owing to
low reflectance
of non-flooded areas under cloud shadow, the above-mentioned threshold spuriously
classifies it as water
(Figure 3). As a consequence, the classified image appears as an underestimation of nonflooded area and an
overestimation of water. Actually, the difference in the reflectance between flooded and nonflooded regions
becomes so low that it is not possible to separate them by means of a threshold value. Land
surface reflects
Flood affected area within
a rural settlement (water
pixel)
Figure 2. Landsat ETMC FCC (zoomed eight times from optimum resolution) showing flooded area within a
settlement
Land under cloud shadow
Water under cloud shadow
Masked out cloud
Figure 3. FCC showing flooded and non-flooded areas under cloud shadow
higher energy in the red band (Landsat band 3) compared with water. On the other hand,
reflectance from
water is significantly lower in the NIR band compared with land. Therefore, the ratio of NIR
and red (band
4/band 3) increases the difference in the reflectance of flooded and non-flooded pixels. This
difference in the
ratio image would be sufficient to distinguish water from land under cloud shadow. After
comparison with
the FCC it has been found that, although the ratio image is effective in differentiating water
and land under
cloud shadow, it is not as sensitive as band _4 C 7_ to water surface. The problem arises
from situations
where the water pixels mixed with non-flooded pixels have a higher value than the open
water pixels in the
ratio image. Sometimes, values of such pixels overlap with the non-flooded pixels
surrounded by water under
cloud shadow. Hence, the ratio image is not able to identify a cluster of water pixels
surrounded by land.
In this paper, an attempt has been made to synthesize the advantage of band _4 C 7_ and
the ratio image
for extracting the non-flooded area from the flood scene. After thorough observation of the
DN of the ratio
image and Landsat FCC, it has been found empirically that pixel values ranging from 052
to 1 in the ratio
image represent non-flooded areas under cloud shadow. Pixels that fall within the specified
range in the
ratio image have been selected and merged with non-flooded pixels derived from band _4 C
7_. It should be
pointed out that these two regions derived from different manipulation of Landsat bands
have not been found
mutually exclusive in their spatial coverage. Overlapped areas are dissolved to create one
classified layer
depicting the non-flooded area. In the synthetic image, the advantages of band _4 C 7_ and
band (4/3) have
been incorporated to extract the non-flooded area in the flood scene accurately.
There is no doubt that the above classification scheme enhances the accuracy. However, it
still suffers from
limitations. The roofs of submerged houses appear as dry surface, and in areas of compact
settlement this
error becomes significant. The tree canopy also creates confusion in delineating the
boundary between water
and land. Flooded areas under canopy appear as non-flooded. Although forest constitutes an
insignificant
land-cover category in Gangetic West Bengal, rural settlements are often surrounded by
trees. This factor
constitutes a major source of error in the classification scheme.
Site visits to some selected settlements were made during the fieldwork in 2003 and local
people were
interviewed to locate specific sites for validation. However, after incorporating that
information it was found
that it is not scientifically viable to do an accuracy assessment based on peoples
recollection of a flood
event that occurred 2 years ago. In the absence of any aerial photograph during the flood
we embarked
upon using a digital elevation model (DEM) to assess the accuracy of our classification
indirectly. To assess
the accuracy of the classified image, pixels classified as non-flooded have been
superimposed on a DEM
derived from the advanced space-bome thermal emission and reflection radiometer (ASTER)
images of the
study area. The ASTER DEM has a spatial resolution of 30 m with a relative accuracy of more
than 10 m.
These digital data are suitable to meet the 1 : 50 000 to 1 : 250 000 map accuracy standard
(USGS, 2003).
The elevation distribution of the non-flooded pixels is shown in Figure 4. The graph shows a
sudden upward
2.5105
2.0105
1.5105
Frequency
1.0105
5.0104
8m
-10 0
0
10
Elevation (m)
20 30
Figure 4. Elevation distribution of the non-flooded area extracted from the ASTER DEM
trend approximately at an elevation of 8 to 9 m. The rest of the non-flooded area has

elevation ranging from
10 to 35 m. It is assumed that non-flooded pixels having an elevation of below 8 m are most
probably tree
canopy or other confusing land cover in the flooded area that has been incorrectly classified
as non-flooded.
To rectify this error, and to reduce overestimation of dry area, all non-flooded pixels having
an elevation
of less than 8 m have been eliminated. Since we have used a relative ASTER DEM, the
absolute heights of
the pixels are not very accurate from sea level, but the relative difference of elevation
between two pixels
meets our requirement. An 8 m elevation has been used to illustrate the point of inflection in
the elevation
distribution curve of the non-flooded pixels. The absolute height is not very relevant in this
discussion. The
elevation distribution of the non-flooded pixels is the purpose of the graph.
After the non-flooded pixels had been imported into ArcGIS, it was realized that any nonflooded surface
occupying less than nine Landsat TM pixels is potentially a tree canopy, or part of some road
surface that
was not submerged by flood water. After considering the general topography of Gangetic
West Bengal it
was realized that there is the rare possibility of the existence of such a small abruptly higher
area. Nine
pixels were chosen as a threshold because this presents a block of 90 m 90 m. Thus, any
apparently dry
land surface that is less than a 3 3 block is not likely to be a suitable place for flood
shelter. Anything
smaller than three pixels may also contain some inundated area in the sub-pixel level and,
therefore, cannot
be relied upon as a potential shelter. Considering the population density of rural West
Bengal, even if some
of these isolated pixels represent actual non-flooded surfaces, then they are too small to
provide any effective
shelter to the adjacent population. Therefore, patches with less than nine pixels have been
eliminated from
the non-flooded layer. Table I summarizes the extent of area eliminated at different levels of
correction of the
non-flooded area.
Anything other than the non-flooded area depicts the actual water area. This classified
water layer includes
permanent water bodies of the area. For calculating actual flood-affected area, permanent
water bodies must be
eliminated from the water layer (Yang et al., 1999). Two Landsat ETMC scenes of 16 April
2003, depicting
normal hydrological conditions, have been used to extract the permanent water bodies.
These pixels were
subtracted from the water layer to obtain the actual flooded area. The classified image is
shown in Figure 5.
Delineating the high flood-depth zone
Flood depth is considered as the most important indicator of flood hazard (Wadge et al.,
1993; Townsend
and Walsh, 1998; Islam and Sado, 2002). A higher depth of flood is associated with a high
discharge, which
is a determining factor in flood-induced destruction of life and property. During the 2000
September flood
the depth of water was as high as 25 to 3 m in some areas. The roofs of the majority of the
houses in rural
West Bengal are not concrete; they are commonly built with hay or thatch and corrugated
sheets of iron. This
fact excludes the local people from climbing over their roofs to escape inundation. Flood
depth determination
from remotely sensed imagery is very difficult, but an indirect method exists to classify a
flooded area into
different flood-depth zones. The amount of radiant energy reflected by water in the visible
light portion of
the electromagnetic spectrum is essentially determined by the colour of the water and its
turbidity. Except
for the blue band, all optical bands have very high correlation with the turbidity and
sediment concentration
of the water. Deeper water has more turbidity than shallower waters because of its high
velocity (Islam and
Sado, 2000b).
Interband correlation is a major impediment in analysing multispectral data. Principal
component (PC)
transformation is performed to overcome this problem. This image-processing technique
makes the bands less
correlated and reduces the dimensionality of the original dataset (Lillesand and Kiefer,
2000). To enhance
contrast and facilitate classification, a PC transformation has been applied over bands 2, 3,
4, 5 and 7 of the
Landsat ETMC data acquired during the flood. The first three components, explaining about
9965% of the
total variation, were selected for further analysis. The other components were excluded from
further analysis
as their noise-to-signal ratio is expected to be very high. Kunte and Wagle (2003) attempted
to classify depth
of water in the Gulf of Kutch and reported that the PC2 band is particularly sensitive to the
concentration
of suspended sediments and, therefore, can be effectively used for broad classification of
water depth. To
enhance the amount of information, different combinations of the three PC bands into RGB
were tried to
create an FCC, and it was found that PC2, PC1 and PC3 (RGB) generate the best FCC. Figure
6 shows the
turbidity/sediment concentration in the flooded zone.
Figure 6 clearly represents at least two turbidity zones in the shades of yellow and violet.
The general trend
of tonal variation reveals that highly turbid water exists at the core of the flooded zone and
the sediment
concentration of the water gradually decreases towards the boundary of water and land.
After consulting Islam
and Sados (2000b), we have assumed that highly turbid water is associated with high
velocity. During a
flood, high water flows at high velocity through the highly inundated zone. Highly
turbid/deep water appears
in yellow and shallow water with less turbidity is presumed to appear in violet tint. The major
rivers of
the region, namely the Bhagirathi and Jalangi, fall in the yellow zone, which further confirms
our visual
interpretation of flood depth.
Table I. Correction of non-flooded area under different levels of processing
Processing level Total area _km2_ Area reduced _km2_
Digital image processing of Landsat flood scene 351298
After eliminating area under 8 m 332498 741
After eliminating polygons less than nine Landsat pixels 328708 3790
Classification
category
Area
(Km2)
Flood water
Land
Permanent water body
Cloud cover
2745.79
3287.08
331.39
1327.53
Legend
Flood water
Land
Permanent Water Body
Cloud Cover
Classification of the Study Area

into Flooded and Non-flooded Regions
N
Figure 5. Classified image showing the flooded area on 30 September 2000
It is interesting to note that along the River Bhagirathi the majority of the flooded area falls
in the high
depth zone and there is little evidence of a shallow depth zone along the margin of the
flooded area. This
phenomenon is attributed to the presence of extensive embankments along the river. Land
use is very intensive
in the immediate flood plain of the River Bhagirathi. To protect the land, the West Bengal
Government has built
hundreds of kilometres of embankment. These embankments, along with other minor floodcontrol measures,
put an abrupt stop to advancing flood water during the flood of SeptemberOctober 2000.
The low-lying active
flood plain, located between the embankment and the river, records very high water
discharge during floods
and, consequently, the area suffers from high water depth. Owing to the existence of the
embankments, there
Masked out cloud
cover
Non-flooded area
Shallow flood depth
Deep flood water
Figure 6. Different flood depth/turbidity zones identified over an FCC (PC2, PC1 and PC3 as RGB)
was hardly any zone of transition from deep to shallow flood water along the bank. The River
Bhagirathi
represents the most dominant distributary of the River Ganga that flows through the Indian
state of West
Bengal. The discharge and carrying capacity of this river is much higher than the other
smaller rivers in this
region. Owing to a higher velocity and discharge, the sediment concentration and
consequent turbidity in the
River Bhagirathi is much higher than the other smaller rivers, like the Jalangi. This factor also
contributes to
deep inundation along the Bhagirathi river bank.
After comparing the PC2 band with the FCC, a threshold value has been selected empirically
to extract
highly turbid/deep water from the PC2 band. Although river flooding is not only a function of
terrain
configuration, the topography of the flood plain does play an important role in the course of
an advancing flood.
Therefore, the ASTER DEM has been used to verify the accuracy of the high flood-depth
zone. Elevation
distribution of the high flood-depth zone has been plotted in Figure 7. This plot ratifies that
classification of
high flood depth is, on average, accurate and realistic. The area having an elevation of more
than 16 m was
not likely to experience high flood depth. The fact that the wave drops sharply after an
elevation of more
than 16 m attests to the high accuracy of the classified high flood-depth zone.
Delineating human settlements
As mentioned in the Introduction, the main focus of this study revolves around identifying
the floodvulnerable
settlements. Human settlement is the main input layer of the present study, and all other
layers of
information are used to estimate its vulnerability to floods. Land-use maps of 1 : 250 000
have been used as
the basic source of information for delineating human settlements in this area. These maps
were published
in 1991 and, therefore, are considered relatively outdated. Satellite imageries have been
used to update these
maps. Application of remotely sensed data for identifying human settlements is not new.
Radar imageries
1.5105
1.0105
Frequency
5.0104
0
-10 -5 0 5 10 15
Elevation (m)
20 25
Figure 7. Elevation distribution of the area affected by high flood depth
are more widely used than optical data because settlements are particularly accentuated in
radar imageries
owing to their geometric shapes and dielectric constants. These parameters of SAR data are
different from
optical sensors (Henderson and Xia, 1997). Studies in Germany (Henderson, 1995), China
(Lo, 1986), and
the Ganges Plain (Imhoff et al., 1987) reported that settlements having a population of more
than 1000 are
generally recognizable from SAR imageries.
The difficulty associated with identifying human settlement from radar data is that most SAR
sensors obtain
data at a single wavelength with fixed polarization. The best band combination for any
classification should
cover major portions of the electromagnetic spectrum (visible, infrared and microwave) that
are normally
used for remote sensing (Haack et al., 2000). Haack and Slonecker (1994) reported that
neither Landsat TM
nor SAR data can independently locate villages in Sudan. For the present study, Landsat
ETMC data of 13
April 2003 and ERS-1 SAR data of 9 October 1995 have been used to interpret the land cover
of the study
area visually and to update the settlement layer previously digitized from the land-use map.
The Landsat ETMC data obtained are geometrically corrected (level-1G) from the USGS. It has
been
further registered to the 1 : 250 000 land-use maps. The radar data have been obtained as a
precision image
(PRI). ERS SAR PRI products are projected to ground range and resampled in 125 m pixel
size. The PRIs
have been coregistered with the land-use map with a root-mean-square error (RMSE) of
106 pixels. After
georeferencing, the SAR imageries have been mosaicked. A low-pass filter of a 5 5 pixel
window has been
applied to it to reduce speckle and improve visual interpretability in identifying settlements.
After stacking
TM bands 4, 3 and 2 in RGB, a coloured image was generated by fusing it with the processed
radar image.
HSV sharpening tools have been used to perform this operation. This function transforms an
RGB image to
HSV colour space, replaces the value band with the high-resolution radar image and
automatically resamples
the hue and saturation bands to the high-resolution pixel size (125 m) of the SAR PRI using
the nearest
neighbour method. Finally, it transforms the image back to the RGB colour space. The output
was a coloured
image of 125 m spatial resolution. It is necessary to point out that a coregistration RMSE of
more than one
pixel is acceptable for SAR scenes because the 125 m 125 m pixels of radar imageries
are fused with
30 m 30 m ETMC pixels.
Figure 8 represents part of the study area. The small rural settlements are easily
distinguishable by their
bright appearance over a reddish background of vast cropland. The vectorized settlement
layer, already
extracted from the land-use maps, has been superimposed on this image to update the
boundaries of settlements
manually. Automated classification methods have not been considered, as a majority of
studies reported low
accuracy for smaller settlements (Dowman and Morris, 1982; Lo, 1984; Liu et al., 1986). It
has also been
found that the reflectance of fallow land and bare soil are quite similar to the rural
settlements in the TM
bands. Such substantial error in classifying human settlements would jeopardize the whole
gamut of results.
This factor induced us to rule out the option of automated classification.
Processing different data layers in a GIS environment
In order to reach the objective of obtaining a flood vulnerability of settlements, it is
necessary to make the
other information layers compatible with the settlement layer to allow for the analysis of
their vulnerability
to river inundation. Apart from the hydrological information extracted from the Landsat ETM C
flood scenes,
some socio-economic data, like population density, have also been incorporated in the
current framework to
facilitate vulnerability analysis of the settlements. Considering the scale of this investigation,
development
blocks have been selected as the appropriate administrative unit for reporting population
density. Development
blocks are the smallest administrative unit in West Bengal, as well as India, for collecting
majority of the
data. Boundaries of the development blocks over the study area are shown in Figure 1.
Rural
settlements
Figure 8. Landsat ETMC bands 4, 3 and 2 merged with ERS SAR image to identify the rural settlements in Gangetic
West Bengal visually
Administrative boundaries of the development blocks have been digitized from a 1 : 500 000
map.
Demographic data, collected from the 2001 Indian Census, have been integrated in the
attribute table of the
development blocks. It has been observed over the study area that human settlements,
especially rural hamlets,
develop across administrative boundaries of the development blocks. The frequency of flood
occurrence in
each development block for the decade 1991 to 2000 has been considered to identify those
areas that have
been suffering from chronic flooding (Sanyal and Lu, 2005). This information was obtained
from the Annual
Flood Reports of the Irrigation and Waterways Department, Government of West Bengal,
India.
To facilitate spatial analysis, the shape of the file containing settlements has been clipped
by the
administrative boundaries of the development blocks. In the output, individual settlements
have been
subdivided where an administrative boundary cuts across them. This overlay operation
proved very useful in
subsequent analysis, as in the output layer each settlement contains certain information
about the development
block it is located in. This information includes name, population density and frequency of
flood occurrence
over the last decade. It should be noted that the above-mentioned attributes are assumed to
be constant for all
individual settlements falling within one development block. It is evident that the accuracy
of this information
decreases with increasing size of the development blocks. Owing to the monotonous flat
terrain of the study
area, there is not much scope for large variation of topography within a development block.
RESULTS AND DISCUSSION
The main purpose of this investigation is to analyse the interaction between different flood
hazard indicators
that contribute to vulnerability of human settlements in the study area. The previous section
illustrated a
systematic approach to extracting the spatial extent of these hazard factors. A GIS
environment is used to
evaluate the interaction of these factors in a spatial dimension and locate the settlements
that should be given
priority in implementing remedial measures. It should be pointed out that this methodology
is likely to yield
better perception of flood hazard if a very high magnitude flood is taken into consideration.
Using a moderate
magnitude event would increase the risk of leaving some vulnerable settlements
unidentified that have chances
of be affected should a very high magnitude event hit the area. We found that an
overwhelming majority of
the settlements contain some pixels classified as non-flooded. The reason behind this
phenomenon is that the
local people traditionally build their houses over relatively higher ground, leaving the floodprone lowlands
for paddy cultivation. This settlement pattern evolved as an outcome of local inhabitants
adaptation to living
in a flood-prone area. In Gangetic West Bengal, owing to the vast expanse of cropland, trees
are only found in
and around rural settlements and along roads. Misclassification of some tree canopy as nonflooded pixels also
contributes to overestimation of non-flooded areas. It has been decided that any individual
settlement having
very little intersection with the non-flooded pixels should be practically considered as highly
vulnerable to
flood. For computational ease we have calculated the hazard indicator as
HI1 D _A/B_ 100 _1_
where A is the area of intersection between non-flooded pixels and individual settlements
and B is the total
area of an individual settlement.
The intersection function in ArcGIS is used to calculate the common area between the
settlement and nonflooded
layer. Part of the attribute table of the output shape file is shown in Table II. This table shows
that,
very often, each settlement polygon intersects with more than one polygon representing
non-flooded areas.
Therefore, the area of intersection is distributed in more than one row, which makes it
difficult to calculate HI1.
It is noted in Table II that settlement ID 23 is distributed in 13 rows. In other words,
settlement 23 contains 13
patches of non-flooded land area. Table II is not the complete table. It only shows a typical
portion of it that
exemplifies the difficulty in calculating HI1. The attribute table of the output shape file has
been summarized
by dissolving the polygons on the basis of their settlement ID number. Part of the processed
table is illustrated
in Table III. After calculating HI1 it has been found that out of 921 settlement polygons in the
study area 390
Table II. Part of the attribute table illustrating how the intersection of a non-flooded layer with
individual settlements is distributed in different polygons. Note that 13 polygons represent the
area of intersection between settlement 23 and the non-flooded layer
Settlement
polygon ID
Intersection
area _m2_
Block
name
District Settlement
ID
56796 21 1919 Balagar Hugly 13
56791 35 1324 Balagar Hugly 13
56720 1 100 000 0000 Balagar Hugly 20
56720 1 100 000 0000 Pandua Hugly 22
56720 1 100 000 0000 Pandua Hugly 22
56793 27 8300 Chakdaha Nadia 23
56782 93430 Chakdaha Nadia 23
56769 99209 Chakdaha Nadia 23
56720 1 100 000 0000 Pandua Hugly 24
56525 49 3324 Pandua Hugly 26
56525 49 3324 Pandua Hugly 26
56720 1 100 000 0000 Pandua Hugly 26
56720 1 100 000 0000 Pandua Hugly 28
56720 1 100 000 0000 Pandua Hugly 28
56720 1 100 000 0000 Pandua Hugly 28
Table III. Area of intersection between settlement layers and non-flooded
area summarized on the basis of individual settlements
Development
block
District Settlement ID Intersection
area _m2_
Balagar Hugly 13 45 79000
Balagar Hugly 20 813684
Chakdaha Nadia 21 000
Pandua Hugly 22 580 05279
Chakdaha Nadia 23 685 68778
Pandua Hugly 24 1 200 00000
Chakdaha Nadia 25 000
Balagar Hugly 27 000
settlements have less than 50% of their area classified as non-flooded. In other words, more
than 50% of the
area of these settlements was affected by inundation. For more than 75% inundation, the
number is still high
at 206 (2236%). There are 124 settlements having more than 90% of their area submerged
under flood water
(Figure 9). These settlements are located in a high flood-hazard zone and, therefore, are
extremely vulnerable
Location of vulnerable settlements N
Legand
Vulnerable Settlements
Development Blocks
Studyarea
River
Figure 9. Location of the settlement that does not have access to higher ground as shelter during the flood on 30
September 2000
to inundation. In addition, a buffer operation reveals that 13 of these 124 settlements have
no non-flooded
areas within a buffer zone of 500 m. For the sake of simplicity, a straight-line distance has
been used to
calculate the buffer. Since Gangetic West Bengal is remarkably flat there is no conspicuous
physical barrier
such as, mountain to travel for short distance. It is been also realized that 500 m is the
maximum distance
that people can travel even at a moderate flood depth. Thus, these settlements have no
immediate access to a
potential flood shelter. Out of these 13 settlements, five are located in the Ranaghat-II block,
four are in the
Ranaghat-I block, two are in the Balagar block, one is in the Chakdah block and one is in the
Chapra block
(see Figure 1 for block names).
The location of a particular settlement with respect to the high flood-depth zone forms the
basis for
computing the second flood hazard indicator: any settlement having the majority of its area
under deep flood
water has been considered as vulnerable. The hazard indicator has been calculated as
HI2 D _C/D_ 100 _2_
where C is the area of intersection between high flood depth and an individual settlements
and D is the total
area of an individual settlement.
The vector layer of the high depth zone, extracted from Landsat imageries, has been
overlaid with the
settlement layer and the intersection between the two layers was calculated using ArcGIS.
Attribute tables
are compiled in a similar way to HI1. It is evident from previous discussion that a high HI 2
value for any
settlement indicates its vulnerability to flood-induced disaster. More than 50% of the area of
19 settlements
was affected by high flood depth. Ten of these 19 settlements are situated in Nadia District
and nine are in
the Bardhaman District of West Bengal.
Vulnerability analysis for each settlement can be done more meaningfully by synthesizing
HI1 and HI2
with population density. It was mentioned before that each settlement polygon already
contains information
regarding average population density as the number of people per square kilometre. An SQL
query was built
to identify those settlements that have a high population density and which are vulnerable
with respect to
both hazard factors.
A total of 18 settlements meet the criteria where HI 1 < 25, HI2 > 50 and Pop density >750.
These
settlements have been classified as extremely vulnerable. The distribution of these
settlements in different
development blocks is shown in Table IV. Table V shows the location coordinates of the
centroids of
the polygons. The coordinates will help local planners and administrators easily locate highly
vulnerable
settlements over the administrative boundaries of revenue villages. During any highmagnitude floods, the
local administration should place high priority on providing relief to the population of these
settlements.
Table IV. Development block-wise distribution of extremely flood-vulnerable
settlements
Development
block
District Flood occurrences
(19912000)
Vulnerable
settlements
Chapra Nadia 3 2
Kalna-I Bardhaman 2 3
Krishnanagar Nadia 3 1
Nabadwip Nadia 3 2
Purbasthali-I Bardhaman 4 3
Purbasthali-II Bardhaman 3 2
Ranaghat-II Nadia 2 1
Shantipur Nadia 2 4
Table V. Precise location of centroids of the settlements that are highly vulnerable
to flooding
Settlement ID Block Name Latitude (N) Longitude (E)
1 Kalna-I 23296 88361
2 Kalna-I 23317 88291
3 Purbasthali-I 23336 88348
5 Kalna-I 23354 88290
7 Purbasthali-II 23500 88340
8 Purbasthali-II 23509 88283
9 Santipur 23218 88412
10 Santipur 23296 88364
11 Ranaghat-II 23292 88650
12 Santipur 23310 88369
13 Nabadwip 23329 88380
14 Santipur 23332 88349
15 Nabadwip 23335 88349
16 Krishnanagar-I 23456 88545
17 Chapra 23501 88615
18 Chapra 23530 88606
Table V shows that these settlements are not only vulnerable to extreme flood events like in
2000, but most
of them are also located in very frequently flooded development blocks because of a rapidly
increasing
population density and hence rapid expansion of settlements to highly flood-prone zones.
To verify rural settlement increases, an old 1 : 63 360 map of Nakashipara development

block, produced
between 1917 and 1921, was digitized. This map depicts the pre-independence settlement
pattern in the region.
A query was built in the current settlement layer to identify the polygons in Nakashipara
development block
that have more than 50% of their area under flood water. A total of 33 polygons from the
current settlement
layer met this criterion. The results show that 17 of these 33 settlements have no area of
intersection with
the settlements present during 1921. Hence, it can be concluded that the majority of
vulnerable settlements
are of new origin and that people chose to build settlements in a flood-prone zone most
probably out of
socio-economic compulsion.
CONCLUSIONS
This study demonstrates a cost-effective and efficient way to create a moderate-resolution
spatial database for
identifying human settlements that are highly vulnerable to monsoon flooding. The results of
this study are,
as yet, not conclusive. It is an effort to build the architecture of a flood hazard database that
can be analysed
from a spatial dimension. Any other combination of the attributes based on in-depth local
knowledge might
prove more rational for implementing mitigation measures. The accuracy of classifying
flooded areas from
non-flooded areas is limited by partial cloud cover over the area and the predominance of
tree canopy in the
rural settlements. Unavailability of updated large-scale maps also restricts our ability to
undertake a detailed
mapping of settlements in the study area. The lack of high-resolution digital terrain data for
developing
countries also leads to difficulty in assessing the accuracy of the classification results
(Sanyal and Lu, 2004).
In spite of these constraints, the study has resulted in a reasonably accurate spatial
information database that
is suitable for generating 1 : 250 000 hazard maps. Since a very high magnitude and lowfrequency flood has
been studied, there is very little possibility of leaving any potentially flood-vulnerable
settlement unidentified.
We propose that efforts should be made to generate cost intensive, high-resolution terrain
data, at least for
areas of high flood-depth zone. Such an effort will immensely enhance our capability to
estimate flood hazard
and assess the vulnerability of people.
ACKNOWLEDGEMENTS
This research has been funded by the National University of Singapore (NUS) research grant
R-109-000-049112. NUSs support of this research is gratefully acknowledged by the authors.
REFERENCES
Bagchi K. 1945. The Ganges Delta. University of Calcutta: Kolkata, India.
Basu PK. 2001. Flood 2000: a comprehensive study with major floods in the state Sechpatra 3(1): 612.
Berg A, Gregiore JM. 1983. Use of remote sensing techniques for rice production forecasting in West Africa (Mali and
Guinea: NigerBani
project). In ESA Satellite Remote Sensing for Developing Countries, Ispra, Italy; 161168.
Bhavsar PD. 1984. Review of remote sensing applications in hydrology and water resource management in India.
Advances in Space Research
4(11): 193200.
Cannon T. 2000. Vulnerability analysis and disasters. In Floods Vol I, Parker DJ (ed.). Routledge: New York; 4555.
Deutsch M, Ruggles F. 1974. Optical data processing and projected application of the ERTS1 imagery covering the
1973 Mississippi river
valley floods. Water Resource Bulletin 10(5): 10231039.
Deutsch M, Ruggles F, Guss P, Yost E. 1973. Mapping the 1973 Mississippi floods from the Earth Resource
Technology satellites. In
Proceedings of International Symposium on Remote Sensing and Water Resource Management. American Water
Resources Association,
No. 17. AWRA: Burlington, Ontario; 3955.
Dowman IJ, Morris AH. 1982. The use of synthetic aperture radar for mapping. Photogrammetric Record 10(60):
687696.
England CF, Hunt GE. 1985. A bispectral method for the automatic determination of parameters for use in imaging
satellites cloud retrievals.
International Journal of Remote Sensing 6(8): 15451553.
FEMA. 2003. Federal Emergency Management Agency. http://www.fema.gov/nfip [20 September 2000] (in Bengali).
Ganashakti. 2000. November 3 (In Bengali).
Goswami AK. 1983. Studies on the nature of floods in the depressed belt of southern West Bengal. Geographical
Review of India 41(2):
189201.
Haack BN, Slonecker ET. 1994. Merged spaceborne radar and thematic mapper digital data for locating villages in
Sudan. Photogrammetric
Engineering and Remote Sensing 60(10): 12531257.
Haack BN, Herold ND, Bechdol MA. 2000. Photogrammetric Engineering and Remote Sensing 66(6): 709716.
Henderson FM. 1995. An analysis of settlement characterization in central Europe using SIR-B radar imagery.
Remote Sensing of Environment
54: 6170.
Henderson FM, Xia ZG. 1997. SAR applications in human settlement detection, population estimation and urban
land use pattern analysis:
a status report. IEEE Transactions on Geoscience and Remote Sensing 35(1): 7985.
Hewitt K. 1997. Regions of risk: A Geographical Introduction to Disasters. Addison-Wesley Longman: Harlow, UK.
Imhoff ML, Vermillion C, Story MH, Choudhury AM, Gafoor A, Polcyn P. 1987. Monsoon flood boundary delineation
and damage
assessment using space borne imaging radar and Landsat data. Photogrammetric Engineering and Remote Sensing
53(4): 405413.
Islam MM, Sado K. 2000a. Flood hazard assessment in Bangladesh using NOAA AVHRR data with geographical
information system.
Hydrological Processes 14(3): 605620.
Islam MM, Sado K. 2000b. 1997. Development of flood hazard maps of Bangladesh using NOAA-AVHRR images with
GIS. Hydrological
Sciences Journal 45(3): 337355.
Islam MM, Sado K. 2002. Development of priority map flood countermeasures by remote sensing data with
geographic information system.
Journal of Hydrologic Engineering 7(5): 346355.
Kunte PD, Wagle BG. 2003. Sediment transportation and depth variation study of the Gulf of Kutch using remote
sensing. International
Lillesand TM, Kiefer RW. 2000. Remote Sensing and Image Interpretation, 4th edn. John Wiley: Chichester.
Liu J, Teng X, Xiao J. 1986. Application of Shuttle imaging radar data for land use investigation. Remote Sensing of
Environment 19:
291301.
Liu Y, Islam MN, Gao J. 2003. Quantification of shallow water quality parameters by means of remote sensing.
Progress in Physical
Geography 27(1): 2443.
Lo CP. 1986. Settlement, population and land use analysis of the North China Plain using Shuttle imaging radar-A
data. Professional
Geographer 38(2): 141149.
McGinnis DF, Rango A. 1975. Earth Resource Satellite System for flood monitoring. Geophysical Research Letters
2(4): 132135.
Mitchell JK (ed.). 1999. Crucibles of Hazard: Mega-Cities and Disaster in Transition. United Nations Publication: New
York.
Morrison RB, Cooley ME. 1973. Assessment of flood damage in Arizona by means of ERTS-1 imagery. In Proceedings
of Symposium on
Significant Result Obtained from the Earth Resource Satellite 1, New Carrollton, Maryland, Vol. 1; 755760.
Rango A, Anderson AT. 1974. Flood hazard studies in the Mississippi river basin using remote sensing. Water
Resources Bulletin 10(5):
10601081.
Rango A, Solomonson VV. 1974. Regional flood mapping from space. Water Resources Research 10(3): 473484.
Ruangsiri P, Sripumin R, Polongam S, Kanjanasuntorn P, Wongparn S. 1984. State of flooding in the Mun-Chi river
basin area, N. E.
Thailand by digital Landsat data analysis. Report: Remote Sensing Division, National Resource Council of Thailand,
Bangkok, Thailand.
Rudra K. 2001. The Flood in West Bengal: September 2000 . Jayasree Press: Kolkata, India.
Sanyal J, Lu XX. 2004. Application of remote sensing in flood management with special reference to monsoon Asia:
a review. Natural
Hazards 33: 283301.
Sanyal J, Lu XX. 2005b. GIS based flood hazard mapping in Gangetic West Bengal. Singapore Journal of Tropical
Geography (in press).
Saunders RW. 1986. An automated scheme for the removal of cloud contamination from AVHRR radiance over
western Europe. International
Sheng Y, Su Y, Xiao Q. 1998. Challenging the cloud-contamination problem in flood monitoring with NOAA/AVHRR
imagery. Photogrammetric
Engineering and Remote Sensing 64(3): 191198.
Spate OHK, Learmonth ATA, Learmonth AM. 1965. India and Pakistan: A General and Regional Geography. Methuen:
Bungay, Suffolk.
Smith LC. 1997. Satellite remote sensing of river inundation area, stage and discharge: a review. Hydrological
Processes 11: 14271439.
The Times of India. 2000. Rain reality mired in recording controversy. 18 November.
Townsend PA, Walsh SJ. 1998. Modelling flood plain inundation using integrated GIS with radar and optical remote
sensing. Geomorphology
21(98): 295312.
USGS. 2003. http://edcdaac.usgs.gov/aster/ast14dem.html [6 December 2003].
Varley P (ed.). 1994. Disaster, Development and the Environment. Wiley: Chichester, UK.
Wadge G, Wisloki AP, Pearson J, Whittow JB. 1993. Mapping natural hazard with spatial modelling system. In
Geographic Information
Handling Research and Application, Mather PM (ed.). Wiley: New York; 312324.
Wang Y, Colby JD, Mulcahy KA. 2002. An efficient method for mapping flood extent in a coastal flood plain using
Landsat TM and DEM
data. International Journal of Remote Sensing 23(18): 36813696.
Yang C, Zhou C, Wan Q. 1999. Deciding the flood extent with Radarsat SAR data and image fusion. In Proceedings of
20th Asian Conference
of Remote Sensing, Hong Kong, 2225 November.
Application of Remote Sensing in Flood

Management with Special Reference to
Monsoon Asia: A Review
ARTICLE in NATURAL HAZARDS SEPTEMBER 2004
Impact Factor: 1.72 DOI: 10.1023/B:NHAZ.0000037035.65105.95
CITATIONS
130
READS
748
2 AUTHORS:
Joy Sanyal
Presidency University

SEE PROFILE
"X.X." Lu
National University of Singapore
SEE PROFILE
Available from: "X.X." Lu
Natural Hazards 33: 283301, 2004.

2004 Kluwer Academic Publishers. Printed in the Netherlands.
283

Management with Special Reference to Monsoon
Asia: A Review
JOY SANYAL_ and X. X. LU
Department of Geography, Faculty of Arts and Social Science, National University of Singapore, 1
Arts Link, Singapore 117570
(Received: 20 June 2003; accepted: 4 November 2003)
Abstract. The conventional means to record hydrological parameters of a flood often fail to record
an extreme event. Remote sensing technology along with geographic information system (GIS) has
become the key tool for flood monitoring in recent years. Development in this field has evolved
from optical to radar remote sensing, which has provided all weather capability compared to the
optical sensors for the purpose of flood mapping. The central focus in this field revolves around
delineation of flood zones and preparation of flood hazard maps for the vulnerable areas. In this
exercise flood depth is considered crucial for flood hazard mapping and a digital elevation model
(DEM) is considered to be the most effective means to estimate flood depth from remotely sensed or
hydrological data. In a flat terrain accuracy of flood estimation depends primarily on the resolution
of the DEM. River flooding in the developing countries of monsoon Asia is very acute because of
their heavy dependence on agriculture but any flood estimation or hazard mapping attempt in this
region is handicapped by poor availability of high resolution DEMs. This paper presents a review of
application of remote sensing and GIS in flood management with particular focus on the developing
countries of Asia.
Key words: flood, remote sensing, geographic information system, digital elevation model, hazard
mapping, monsoon Asia
1. Introduction
Among all kinds of natural hazards of the world flood is probably most devastating,
wide spread and frequent. In the humid tropics and subtropical climates, especially
in the realms of monsoon, river flooding is a recurrent natural phenomenon. Floods
resulting from excessive rainfall within a short duration of time and consequent
high river discharge damage crops and infrastructures. They also result in siltation
of the reservoirs and hence limit the capacity of existing dams to control floods.
For formulating any flood management strategy the first step is to identify the
area most vulnerable to flooding. This step is even more critical for the developing
countries in the monsoon Asia as the funding available for developmental activities
is very limited. Thus these funds need to be utilized optimally for the areas that
_
Author for correspondence. E-mail: g0202381@nus.edu.sg
284 JOY SANYAL AND X. X. LU

suffer from river flooding most frequently. With the equipment currently installed
at river gauging stations it is sometimes difficult to record an extreme flood event
having a very high return period. In the developing counties, the density of gauging
stations is very low and thus any flood prediction or risk assessment model tested
in the developed countries faces acute shortage of ground data when applied.
Remote
sensing is a reliable way of providing synoptic coverage over a wide area
in a very cost effective manner. It also overcomes the limitation of the ground
stations to register data in an extreme hydrological event. In addition multi-date
imageries equip the investigators with an additional tool of monitoring the change
or reconstruct progress of a past flood.
For the last two decades advancement in the field of remote sensing and
geographic
information system (GIS) have greatly facilitated the operation of flood

mapping and flood risk assessment. It is evident that GIS has a great role to play in
natural hazard management because natural hazards are multi dimensional and the
spatial component is inherent (Coppock, 1995). The main advantage of using GIS
for flood management is that it not only generates a visualization of flooding but
also creates potential to further analyze this product to estimate probable damage
due to flood (Hausmann et al., 1998; Clark, 1998).
Smith (1997) reviews the application of remote sensing for detecting river
inundation,
stage and discharge. Since then, the focus in this direction is shifting
from flood boundary delineation to risk and damage assessment. Therefore, there
is a need to review the current literature with a holistic view of dealing with various
prospects and constraints of using the technology of remote sensing and GIS
in flood management. This paper presents recent developments on delineation of
flooded areas and flood hazard mapping using remote sensing and GIS. In particular
this paper draws attention to some of the issues associated with application of
remote sensing in combating floods in extremely flat flood plains of monsoon Asia.
Our review includes three aspects. First, we focus on the development of remote
sensing as a tool of flood delineation. Second, we emphasize the assessment of the
intensity of flood hazards and damage. Third, we highlight some of the issues in
the application of the technology with particular reference in the monsoon Asia, an
agricultural area with lack of high resolution spatial data bases.
2. Remote sensing as a tool of flooded area delineation
2.1. APPLICATION OF OPTICAL REMOTE SENSING
In the initial stages of satellite remote sensing the data available was from Landsat
Multi Spectral Scanner (MSS) with 80 m resolution. The pioneering investigations
in the field of application of remote sensing in flood mitigation were predominantly
concentrated on the flood prone regions of USA. MSS data were used to deal with
the flood affected areas in Iowa (Hallberg et al., 1973; Rango et al.,1974), Arizona
(Morrison et al., 1973), and Mississippi River basin (Deutsch et al., 1973; Deutsch
et al., 1974; Rango et al., 1974; McGinnis et al., 1975; Morrison et al., 1976). MSS
APPLICATION OF REMOTE SENSING IN FLOOD MANAGEMENT 285
band 7 (0.81.1 m) has been found particularly suitable for distinguishing water
or moist soil from dry surface due to strong absorption of water in the near infrared
range of the spectrum (Smith, 1997).
From the early 1980s, Landsat Thematic Mapper (TM) imageries with 30 m
resolution became the prime source of data for monitoring floods and delineating
the boundary of inundation. Special attention was given to dealing with monsoon
flooding in the developing countries like West Africa (Berg et al., 1983), India
(Bhavsar, 1983) and Thailand (Ruangsiri et al., 1984). Landsat TM band 4 proves
to be very useful in discriminating water from the dry land surface because it is a
near equivalent of MSS band 7. But Landsat TM NIR band cannot be used optimally
in developed land use areas such as downtown commercial or industrial areas.
The main reason is that the NIR band reflects very little energy from asphalt areas,
appearing black in the imageries. Therefore makes it easy to confuse developed
areas with water. Wang et al. (2002) successfully solved this problem by adding
Landsat TM band 7 to the NIR (band 4) band to delineate the inundated areas.
In TM band 7 (2.082.35 micrometre) images the reflectance from water, paved
road surfaces, and roof tops differ significantly and therefore in the Band4+Band7
image, it becomes easier to choose the density slice for extracting the flood water.
But in some cases a simple density slice or supervised classification is not enough
to identify the inundated area accurately.
During later stages SPOT multi spectral imageries, were also used for flood
delineation with the similar assumption that water has very low reflectance in
the near infrared portion of the spectra. SPOT imageries, for example, were used
along with a DEMfor delineation of monsoon flood in Bangladesh (Brouder, 1994;
Oberstadler et al., 1997; Profeiti et al., 1997; Sado et al., 1997).
Apart from these medium resolution imageries, coarse resolution imageries like
Advanced Very High Resolution Radiometer Radiometer (AVHRR) data have been
also found useful for floods of a regional dimension ( Wiesnet et al., 1974; Huh et
al., 1985ac; Ali et al., 1987; Islam et al., 2000ac, 2001, 2002). Although AVHRR
imageries are coarse in resolution and frequently contaminated by cloud cover their
merit lies in their high temporal resolution. This advantage enables us to monitor
the progress of a flood in near real-time.
To use the capability of the near infra red band more effectively to detect water
Normalized Difference Vegetation Index (NDVI) can be used to monitor river
inundation
from AVHRR images. It is well known that water has a unique spectral
signature in the near infrared which is very different from other surface features.
Therefore, when a surface feature is inundated its NDVI value changes considerably
from the normal situation. Wang et al. (2002) observed that in the lower
reaches of the Yangtze River, the NDVI value for inundated surface features remains
negative while the value for non inundated surface is commonly greater than
0. But choice of this threshold is critical because natural condition of river flooding
varies greatly from place to place. The main difficulties of selecting an appropriate
threshold arise from two facts. Firstly the albedo of water bodies increases signi286
JOY SANYAL AND X. X. LU
ficantly due to high concentration of sediment in the flooded water and secondly,
albedo of bare soil decreases considerably due to its high moisture content during
the monsoon season. These two factors collectively reduce the difference in NDVI
value between inundated and dry surface. In some studies, NDVI values of flood
water were found to be significantly positive (Barton et al., 1989) Thus, a straight
forward approach of using simple NDVI values might not be universally effective
in delineation of inundated area. Moreover, many other factors such as atmospheric
condition, cloud cover and satellite viewing angle also influence NDVI values and
attempts should be made to minimize these effects before calculating the NDVI.
2.2. APPLICATION OF MICROWAVE REMOTE SENSING
The existence of cloud cover appears as the single most important impediment to
capture the progress of floods in bad weather condition (Rango et al., 1977; Lowry
et al., 1981; Imhoff et al., 1987; Rashid et al., 1993; Melack et al., 1994). The
development of microwave remote sensing, particularly radar imageries, solve the
problem because the radar pulse can penetrate cloud cover. Currently the most
common approach to flood management is to use synthetic aperture radar (SAR)
imagery and optical remote sensing imagery simultaneously in one project (Honda
et al., 1997; Liu et al., 1999; Chen et al., 1999). Apart from its all weather capability
the most important advantage of using SAR imagery lies in its ability to sharply
distinguish between land and water.
Thresholding is one of the most frequently used techniques in active remote
sensing to segregate flooded areas from non flooded areas in a radar image (Liu
et al., 1999; Townsend et al., 1998; Brivio et al., 2002). Commonly, a threshold
value of radar back scatter is set in decibel (dB) and a binary algorithm is followed
to determine whether a given raster cell is flooded or not. Radar back scatter is
computed as a function of the incidence angle of the sensor and digital number
(DN) (Chen et al., 1999). The threshold values are determined by a number of
processes depending on the study area and overall spectral signature of the
imagery.
Change detection can be used as a powerful tool to detect flooded area in SAR
imagery. It is generally performed by acquiring two imageries taken before and
after the flood. Coherence and amplitude change detection techniques are widely
applied in SAR domain. In the amplitude approach, areas are estimated as flooded
where the radar back scatter is observed to be in considerable decline from before
flood to after flood imagery. In the coherence approach areas are generally
identified as flooded where the coherence or correlation of radar backscatters from
before and after flood imagery are very low (Nico et al., 2000). Multi-date SAR
scenes for the same area can be projected to red, green and blue channels to create
a colour composite. Long et al. (2001) used three ERS SAR scenes to produce
this kind of composite image. The composite image effectively depicts progress of
a flood during a specific time period. This methodology is simple to execute and
provides an opportunity to readily identify the area that remains water logged for a
maximum period of time.
The existing studies pointed out some common problems encountered in accurately
extracting the flood affected area from SAR imageries. A major problem is
associated with the relation between radar wave length and roughness of the
terrain
and water body. Normally pure and calm water acts as a specular reflector to the
radar signals. Thus the radar antennae receive no backscatter and the water
appears
in dark tone in the SAR imageries. Rough water surface appears in brighter tone
in the SAR imageries than the calm water (Yang et al., 1999). During floods, bad
and windy condition usually prevails over the affected area. Wind induced ripples
in the water surface frequently creates problems for the interpreter to determine
the
threshold value to delineate the flooded area.
Forest cover also poses an obstacle to accurately identify inundated areas from a
SAR image (Hess et al., 1990). The key to identify the inundated areas under forest
cover lies in the fact that flooded forests produces a bright radar back scatter in
contrast to non flooded forests due to a double bounce effect (Kundus et al., 2001),
whereas the flooded areas without a forest canopy appear dark in SAR imageries.
The flooded forests reflect 2dB as radar back scatter in the L-band (Pope et al.,
1997). This particular property has been utilized to separate the inundated areas
from the non inundated ones by applying the thresholding technique (Townsend
et al., 1998; Ali et al., 2001; Rosenqvist et al., 2002). A correct separation of
the flooded and non-flooded settlements is also problematic. Normally the high
back scatter of the buildings overlay the back scatter of flood water within the
settlements. Rural settlements, especially in monsoon Asia are often surrounded
by trees. Therefore, due to the effect of trees, inundation within the settlements is
very difficult to detect (Oberstadler et al., 1997).
Establishing a universal threshold value for detecting flood is not justified. A
particular algorithm of deriving the threshold for differentiating inundated areas
from the dry land may not work with the same efficiency in varied natural settings.
Jin (1999) used DMPS SSM/I (Specific Sensor Microwave/Imager) data for this
purpose and came to the conclusion that a particular algorithm developed on the
basis of a wet land environment produces spurious results for inundated areas
under
a forest canopy. He emphasized the importance of knowledge in the regional
geography of the area under investigation in setting any threshold value.
Combination of wave length, incidence angle and polarization play a major role
in influencing the interpreters ability to segregate flood areas from the non-flooded
ones under a forest canopy. Given the same wave length and incidence angle the
ratio of back scatter from flooded forests to the non-flooded ones is higher in
horizontal
polarization than at vertical polarization. It can be further demonstrated
that while wave length and polarization parameters are held constant, the aforesaid
ratio is larger under a small incidence angle than large (Wang et al., 1995). Radar
back scatter also depends on the orientation of the rough surface (e.g., furrows
of a ploughed field). As radar signals are directional in nature, the same surface
may produce different tonal signatures depending on the relative orientation of the
rough surface to the radar antenna.
Radar incidence angle and the consequent variation in back scatter also pose
difficulties in the delineation of inundated areas from SAR imageries. Generally,
it is found that dealing with forested area imageries taken at a lower incidence
angle prove more useful compared to imageries taken at a higher incidence angle.
Townsend et al. (1998) recommended the use of JERS SAR imageries to ERS for
identification of flooded areas under forest canopy due to the smaller incidence
angle of the former as compared to the latter. This satellite has some advantages
over other earth observation satellite systems to detect flooding under forest as the
L-Band signal of JERS1 is particularly sensitive to standing water below the forest
canopy (Rosenqvist et al., 2002). Radarsat SAR imageries is also preferable to ERS
for flood mapping as the former can rotate its sensor disseminating the radar signal
at different incidence angles. This manoeuvring capability proves very useful in
locating the flood affected area in different types of terrain and land cover (Andre
et al., 2002).
2.3. A COMBINED APPROACH
In recent years, flood mapping efforts synthesize the advantages of both optical
and microwave remote sensing technologies for better results. In some occasions
this approach also leads to the formulation of better flood management strategy. In
mountains, slopes positioned perpendicular to the radar beam only appear bright
and all other areas appear as dark or shaded. This poses an obstacle to effectively
identify the flooded areas in the mountains. Due to its shaded appearance it is very
common to erroneously identify the mountainous areas as inundated. Yang et al.
(1999) came out with a solution to this problem by fusing Landsat TM imageries
with SAR imageries. After extracting the mountain regions from a Landsat TM
data it was overlaid on the SAR imagery. To get rid of the spuriously demarcated
flooded area in the SAR imagery, the portion identified as mountain in the
TM image was eliminated from it. This process led to accurate delineation of the
inundated area. However, the authors realized that adoption of this method may
result in removal of some existing water bodies from the actual hydrological layer
of the mountainous area. To rectify this error, the TM data was consulted and the
water bodies in the mountain region have been restored in the final output. This
paper illustrates that the method of using data captured by different sensors is not
straight forward and often requires iterative experiments to arrive at an optimal
result.
In this context, the utility of generating land cover maps is realized as it provides
information about permanent water bodies in normal hydrological conditions and
rescues the interpreter from possible confusion of including permanent water bodies
as inundated area (Tholey et al., 1997). But while dealing with the monsoon
flood one should consult the imagery of a wet season to carve out the natural
drainage. Use of an imagery captured in dry season would lead to underestimation
of the natural drainage, which in turn would lead to an overestimation of the flooded
area (Islam et al., 2002).
The efforts towards delineation of inundated area are ultimately aimed at assessing
the impact of flood on our economy and livelihood. Land use maps obtained
primarily from optical remote sensing are overlaid on the flood maps for assessing
the degree of damage on different kind of land uses. Varieties of land use
classification
strategies have been adopted by researchers to estimate flood risk. An
analysis of most of the papers show predominance of classifying land use pattern of
the flood prone areas into conventional categories like cropland, urban area, barren
land etc (Islam et al., 2001). However, these conventional land cover classifications
do not serve the purpose optimally. Since damage inflicted over areas of high value
land use is of greater concern a more detailed classification of the economically
intensive land use is recommended. It can be done meaningfully by clubbing low
value land use areas like barren land, forest, marshy land into one category and
applying a rigorous classification of arable land and urban areas (Honda et al.,
1997). It is also realized that a data archive for selected satellites should be built for
the developing countries. This kind of archive would depict the background land
cover condition for the study area over different seasons and facilitate the process
of change detection. Only with such a data set can remote sensing of flooded areas
become a useful tool in flood management and mitigation.
3. Flood Hazard & Risk Mapping with GIS and Remote Sensing
Another primary issue for flood management is to identify the area having higher
hazard potential. Hazard can be defined as some threat, natural, technological, or
civil to people, property and environment and risk is viewed as the probability that
a hazard will occur during a particular time period. Flood is a natural hazard and
flood risk is defined in terms of hundred-year flood (Godschalk, 1991). The issue
of preparing a reliable hazard map is one of the latest concerns within the subject
of flood management.
Different approaches have been taken to map the potential hazard. Rejesk
(1993) introduces three different methods for hazard zoning. His first method
describes
a binary model which evaluates whether the hazard is present or not in a
particular raster cell. The second method involves ranking different locations of an
area depending upon the intensity of the hazard present. In the last approach some
hazard values have been assigned to each of the raster cells based on the results
of a multivariate model which were built up on a host of variables related to river
flooding and associated hazards. But the quantification of the severity of the hazard
in some units or numbers as proposed by Rejesk (1993) in his last approach was
criticized by some authors (Wadge et al., 1993).

Flood depth is considered as the most important indicator of the intensity of
the hazard (Islam et al., 2001, 2000; Townsend et al., 1998; Wadge et al., 1993).
Flood risk maps are prepared depending on the estimated depth of inundation. The
estimation is commonly derived from various hydrological and remotely sensed
data. For identifying flood depth it is important to classify the phenomenon of river
flooding into two categories, namely non-source flood and source flood.
Nonsource
flood is defined as inundation caused by well distributed rain storm over
a large area while source flood describes inundation caused primarily by over
bank flow, predominantly affecting areas near the river channel (Liu et al., 2002).
These two basic characteristics of overland flow have significant implications for
formulating the GIS model. In the case of non-source floods, all the raster cells or
vector points having an elevation below the water level are considered as
inundated
while for source floods, it is necessary to simulate the path of over bank flow
from the main channel to the adjacent flood plain to accurately estimate the flood
affected area.
The concept of topographical convergence or wetness index (Beven et al., 1979;
Moore et al., 1991; Wolock et al., 1995) was used to measure the depth of
inundation
(Townsend et al., 1998). The concept of wetness index is based on the
assumption that the accumulation of water in a particular cell of a raster depends
upon the area of the upslope region contributing water to that particular cell. The
main problem of using this index is that when a slope tends to zero the wetness
index becomes undefined. Thus this index is not very useful for modelling in
extremely flat flood plains. Townsend et al. (1998) also used another model for
simulating source flooding. The model assumes that the potential for any site to
be inundated is directly related to the difference in elevation between that site and
the river at its nearest hydrological link.
Perhaps the most innovative, simple and cost effective study regarding flood
hazard management has been conducted by Islam et al. (2001). He assessed the
flood depth from NOAA AVHRR imageries simply by the tonal difference of the
flood water. In this study, the flood affected area was subdivided into different flood
depth zones using supervised classification. To accurately identify the training sets,
the AVHRR data was superimposed over a DEM. Flood hazard has been assessed
by calculating a weighted score for each land use, physiographic and geologic
division of the country. The highlight of this methodology is that it assigns greater
weight to the categories of deeper flood depth in an exponential manner. In other
words, for the shallow depth, the weight increases in a progressive manner but
beyond a certain depth the increment of weight is much higher than the previous
depth categories. This process ensures that areas having higher depth of inundation
will be assigned a high hazard denomination. The rationale is that after a certain
depth, the flood water becomes very destructive and identifying this critical depth
is crucial for mapping hazard zone. This depth of water, or more precisely the
critical river stage, is likely to vary from region to region depending on the local
topography, building materials, settlement pattern etc.
Population density and hazard ranking have been multiplied to produce land
development
priority maps. Zones of this type of maps indicate development priority

for flood counter measures (Islam et al., 2002). This method is heavily dependent
upon the skill of the analyst who chooses the training sites for different depth of
inundation. The process of choosing the training sets is likely to be hampered by
forest cover and varying sediment load of the flood water which may alter the
natural
relationship of depth of the water and its tonal appearance. The main limitation
of this approach is that AVHRR imageries are too coarse to be used for analyzing a
local flood and cloud free images are very rare, especially in the monsoon season.
Boyle et al. (1998) used an even more diverse dataset to estimate flood risk.
Land use, hydraulic characteristics and human activities like demography, property
values and land ownership have been taken into consideration to determine the
type, location and severity of the hazard produced by the floods. Maps have been
prepared showing the properties under potential risk in the event of floods having
500, 100 to 52 years return period.
Flood hazard maps derived from coarse to moderate resolution imageries is
misfit in the developed countries. This kind of flood hazard maps can hardly make
any improvement to the existing flood insurance infrastructure in North America
or Western Europe. But in contrast, these maps are extremely useful for the
developing
countries of monsoon Asia, countries like China, India and Bangladesh etc.
These countries often suffer from devastating monsoon floods and a large
proportion
of the population in these countries live in the flood prone areas. In this part
of the world flood insurance maps are often unheard of. Here flood hazard maps
are primarily required for saving the lives and livelihood of millions of marginal
farmers. Thus, these flood hazard maps can be used meaningfully by planners to
formulate strategies to combat this natural hazard.
Developed countries also need to evaluate the accuracy of flood risk mapping.
A very accurate flood risk map will exclude those individuals, found to be located
at a very high flood risk area, from a flood insurance scheme. On the other hand,
the insurance sector would lower the premium for those individuals who would be
found at minimum flood risk zone. Thus, a very high resolution flood hazard map
is likely to reverse the basic principle of insurance that loss of few should fall upon
many (Clark, 1998). Therefore, generalizing the risk zones to a certain extent is
likely to be more beneficial to the community as a whole.
Hazard maps should have a good visual effect so that the end user can have an
overview of the entire situation at a glance. In recent times hazard maps have
become
so technical in nature that they convey very little information to the planners
and policy makers (Rocha et al., 1994). Unless the experts of hydrology and GIS
overcome this handicap, hazard maps cannot be used optimally by all the relevant
users.
4. Some Issues of Remote Sensing Applications with Special Reference to
Monsoon Asia
Having briefly given an overview of the development and methods of the
application
of remote sensing in flood management, this section tries to highlight the
constraints in applying this technique. Some of the problems, for example,

agricultural
damage assessment, are typical for monsoon Asia which is heavily dependent
on agriculture
4.1. DEPENDENCY OF DIGITAL ELEVATION MODELS IN FLOOD MANAGEMENT
In a majority of the studies dealing with the application of remote sensing in
inundated
area delineation and flood risk assessment, digital elevation models (DEM)
are used to visualize the interface of flood water with the terrain. Flood depth is
normally calculated by subtracting the elevation of each cell in a raster from its
water level (Brouder, 1994; Townsend et al., 1998; Ali et al., 2001; Islam et al.,
2002). DEMs are also used to simulate the flood depth from discharge data and
very often the result is compared with actual flooded area derived from satellite
imageries. As the spatial extent of inundation is subjected to a method of cross
checking this methodology is likely to yield a more accurate flood map as compared
to one derived from pure hydrological modelling. The main drawback of this
approach is that it is fully dependent on the accuracy of the DEM (Jones et al.,
1998). In the largely featureless plains of monsoon Asia, the accuracy required of
the DEMs is almost unattainable.
In a flat flood plain, where a vertical error of 1 m in the DEM may lead to an
error of 100s of square kilometres in flood estimation, recognition of the magnitude
of errors in the DEM is important in hydrological modelling. This issue has been
addressed from the view point of its significance in flood plain mapping (Lee
et al., 1992; Hunter et al., 1995). In the near absence of high resolution DEMs
required for flood mapping in a very flat terrain, multi-date SAR imageries can
serve as a very potent alternative data set in monsoon Asia. For example, multidate
Radarsat imageries were used in monsoon Asia to depict progress of a flood from
its inception to the peak (Liu, 1999; Chen, 1999). This particular operation can
create a visualization of the course of inundation from the river channel to the
adjacent low lying areas of the flood plain and complement the method of flow
direction simulation. Although multi-date imageries can serve as an alternative to
flow direction simulation, it has severe limitation of determining the flood depth.
A very accurate terrain data base regarding the local flood plain morphology is an
essential prerequisite for such an operation. The flood water intersecting with the
slope is taken as the primary indicator of determining flood depth. Thus for the
gently sloping topography, the resolution of the terrain data actually controls the
accuracy of the estimated flood depth (Brakenridge et al., 1998). High resolution
satellite imageries or aerial photographs are needed for preparing an accurate DEM
which can meet the precision level of a flood depth investigation.
One of the recent developments in the application of remote sensing to flood
related problems is the use of LIDAR (Light Detecting and Ranging) sensor. In the
developed countries, especially in USA, this technology has become very popular
for creating DEMs for flood prone areas. LIDAR sensors can readily identify the
vertical differences in the landforms and can be exploited as powerful instruments
to create DEMs of exceptional accuracy. This sensor can also detect the flood depth.
Although LIDAR sensor can attain the vertical accuracy of 5 cm or better, it is
difficult to map it in that resolution. Due to the limitation of GPS systems to locate
an aircraft/sensor functionally LIDAR generated DEMs are released at an accuracy
of 15 to 25 cm RMSE. Moreover the accuracy decreases gradually with increase
in the density of vegetation cover of the ground (Hodgson et al., 2003). Although
LIDAR data is even more expensive than the SAR imageries, sometimes it provides
the only appropriate option to do flood mapping in the extremely flat flood plains.
Fowler (2002) maintained that the resolution of the LIDAR data depends upon the
intensity of the laser pulses and any attempt to make the survey more intensive by
increasing the laser pulses would increase the cost of the data exponentially.
4.2. AGRICULTURAL DAMAGE ASSESSMENT
Boyle et al. (1998) classified flood damages into two categories; tangible and
intangible.
Tangible damage occurs due to direct contact with the flood water whereas
intangible damage is exemplified by the loss of historical monuments, heritage sites
etc. A disease assuming a form of an epidemic due to flood is also categorized as
an intangible damage. In the agricultural landscape of monsoon Asia, estimation
of agricultural damages requires special attention. Asias population is
predominantly
rural and its economy is heavily reliant on agriculture. Monsoon flood often
creates havoc on the economy by damaging the standing paddy. This problem is
unique for monsoon Asia because in the developed world damage to urban area
and infrastructure facilities is of concern to government authorities. Hence, in this
section we shall emphasize studies dealing with application of remote sensing in
assessment of agricultural damage.
Erosion of top soil due to a flash flood and deposition of flood borne coarse
sand reduce the fertility of soil very severely and thus have a negative impact on
agricultural economy. The process of change detection is found useful to monitor
this kind of damage to agricultural land. The most widely used procedure is to
monitor the change in brightness value (VB) at a particular wave length or different
bands to identify the erosion caused by a flood. Several change detection
techniques
like Spectral Image Differencing (SID), Tasseled Cap Brightness Image Differencing
(TCBID), Principal Component Analysis (PCA), and Spectral Change Vector
Analysis (SCVA) are employed for the purpose of detecting the erosions due to
flooding, but for Landsat TM data, SCVA is found to yield most accurate results
(Dhakal et al., 2002).
In the flood plains of monsoon Asia rice is the dominant crop. Landsat TM4 band (Near Infrared) has been extensively used to estimate the damage to rice
crops due to flood (Shibayama et al., 1989; Okamoto et al., 1996) while the use of
band 3 is also considered to be very effective in assessment of paddy field damage
(Yamagata et al., 1988) due to its very high reflectance in turbid or muddy water
typically associated with flooded paddy field in the riverine plains of monsoon Asia
(Miller et al., 1983; Patel et al., 1985). The studies in this direction even attempted
to estimate the quantum of crop damage from satellite data. For example, a DN
value of 84 (in TM-4) indicates a yield of 3.0 metric tons/ha, whereas a value
around 55 corresponds to no crop in North Korea (Okamoto et al., 1998). The
main shortcoming of formulating a universal range of DN number is that it might
fail to extract accurate information for different varieties of rice crops.
Traditional supervised classification by maximum likelihood method generally
perform well in flood affected agricultural area where the instantaneous field of
view (IFOV) is less than the inundated segment of land and the land use is
dominated
by crop lands (Jensen et al., 1995; Hugenin et al., 1997). The mean value
of Normalized Difference Vegetation Index (NDVI) derived from JERS-1 OPS
data was used to differentiate damaged crops from undamaged one in North Korea
(Choen et al., 1998) because vegetation condition or biomass over a portion of land
is highly correlated with NDVI (Singh, 1989; Tappen et al., 1992; Gamon et al.,
1995; Michener et al., 1997; Lyon et al., 1998).
4.3. PROBLEM OF TEMPORAL RESOLUTION IN FLOOD MANAGEMENT
Devastating floods are generally low frequency, high magnitude natural
phenomena.
Flash floods occur within a very short interval of time and the peak stage
remains only for a couple of hours, but the most extensive and severe damage
takes
place during that time. With the current Radarsat resources it is very difficult to
capture the spatial extent of a flood at its peak. Thus, attempts have been made to
extrapolate the extent of inundation at the peak of a flood from an image acquired
at a later stage of the event. Some GIS algorithms in ARC/INFO are promising to
perform this extrapolation from an image that captures some standing water only at
a time when the flood peak had already passed. The method of least accumulation
cost distance can provide a viable solution to this problem (Brivio et al., 2002).
This principle simulates flow direction from the river channel to the flood plain
based on the assumption that water flows through the path where the work done in
doing so is least.
This methodology yielded a remarkable accuracy of 96.7 percent when compared
with the aerial photographs of the peak of the particular flood event which
occurred in November 1994 in Italy. The main weakness of this approach is that
the least accumulation cost distance operates on certain values of the raster cells
which represent the roughness of the terrain causing frictional drag to the
overflowing
flood water. Since roughness is a function of a host of other geomorphic and
lithological factors, it is very difficult to control the parameters of the experiments
in an area having diverse lithology and land use. Apart from the DEMs and remote
sensing data, field work conducted in recently flood affected area can prove useful.
Wang et al. (2002) used high-flood level marks of a recent flood on houses as a
supplementary data set to the DEMs for estimating flood depth. This kind of data
serve very well to reconstruct dimension of a past flood peak.
General studies of flood geomorphology also provide some insights to resolve
this problem. Many very high magnitude floods may leave permanent imprints in
the flood plain morphology rather than creating some transient features (Brunsden
et al., 1979).When the morphology of the basin perfectly suites with the magnitude
and frequency of the flood event, the flood created features attain a state of
permanency.
Since the frequency of such an event is high in the humid tropics, the interval
between two high magnitude floods gets reduced here. It indirectly contributes to
the permanency of the flood created characteristic morphological features in the
flood plain (Gupta, 1983). The permanency of this kind of topographic features
would be more prominent in areas of high relief and coarser valley sediments
(Gupta, 1988). Hence in the humid tropics there is enough scope to use relatively
inexpensive high resolution optical remote sensing not necessarily at the time of
flooding to demarcate the areas vulnerable to river flooding. But the analyst has to
be trained enough to identify the morphological features in the flood plain typically
associated with high magnitude floods.
5. Conclusion and Prospective
The preceding discussion regarding various facades of the prospects and constraints
of using remote sensing and GIS for flood management unveils some
significant facts. A majority of the researchers favoured multi date radar imageries
to observe a particular flood event and considered different image processing
techniques
to overcome the limitations of remotely sensed data in flood delineation.
The main weaknesses of this type of approach are manifold. Most of the
investigations
mentioned in this paper are heavily dependent on the availability of satellite
data, which is not always guaranteed for the time of peak flood. Rather than moving
towards a comprehensive flood management strategy, these papers concentrated
on
some specific issues like delineation of flooded area. Most of these projects are on
a very high budget and no attempts were made try to keep the cost low. Monsoon
floods affect the developing countries more acutely than the developed ones.
Therefore,
while framing the methodology one should be aware of its feasibility in the
operational area. Most of the studies in this field cannot achieve its desired level
of accuracy without a very high resolution DEM, therefore, we need to develop
alternative methodology to shed our dependence on high resolution terrain data.
A detailed hydro-geomorphological mapping depicting the trace of past floods
may help us in this direction. The hazard of monsoon flood and the destitution
of people associated with it are very different from the industrially developed
countries. Hence, we should adopt an improvised but effective methodology of
risk and damage assessment to come out with meaningful flood hazard maps for
this region.
Application of remote sensing and GIS is convincingly a very efficient and cost
effective way of flood management. Use of very high resolution imageries like
IKONS or SPOT 5 have not been very popular yet in the field of flood management
due to its high price, but it is likely that with these imageries would be available
at a reasonable price and would be widely used for flood mapping. In the age
of internet, GIS has assumed new dimensions, especially for coping with natural
disasters like river flooding. The most important advantage of using internet based
GIS is that it has opened the door of GIS technology to the end users who would not
like to install expensive GIS software. One of the numerous examples is Arc IMS
technology. This technology has been used to develop a web enabled application
named Map Action Processing Digital Interactive Geo Resource (MAPDIGR) for
providing very recent information regarding flood risk to an analyst via internet
(Smith et al., 2002). This technology is at present at an embryonic stage of
development
but has great potential for expand the user base of GIS technology for flood
management by substantially reducing the cost of operation.
Since the problem of flood is very acute in the developing countries of monsoon
Asia, special attention should be given to deal with this problem in the regional
context. GIS models having low cost and simple data requirement are likely to
attract the local authorities in the developing countries to adopt this technology
as an essential input towards a comprehensive flood management system. In the

age of all embracing flood plain management, these sophisticated technologies can
be very useful for the planners to formulate effective strategy for combating the
perpetual natural disaster of river flooding.
Acknowledgement
This study has been done as part of an ongoing project titled Application
of Remote Sensing and GIS in Flood Hazard Management: A Case Study of
West Bengal, India, funded by National University of Singapore research grant
R-109-000-049-112.
References
Ali, A. and Quadir, D. A.: 1987, Agricultural hydrologic and oceanographic studies in Bangladesh
with NOAA AVHRR data, International Journal of Remote Sensing 8(6), 917925.
Ali, S., Hassan, A.,Martin, T. C., and Hassan Q. K.: 2001, Geospatial tools for monitoring flood plain
water dynamics, In: M. Owe, K. Brubaker, J. Ritchie, and A. Rango, A. (eds), Remote Sensing
and Hydrology 2000, IAHS Publication, Oxford, pp. 465471.
Andre, G., Guillande, R., and Bahoken, F.: 2002, Flood mapping using spatial radar and optical
imagery and digital elevation model: Limits and capacities, Houille Blanche-Revue Int. De LEAU
1, 4954.
APPLICATION OF REMOTE SENSING IN FLOOD MANAGEMENT
297
Barton, I. and Bathols, J.: 1989, Monitoring floods with AVHRR, International Journal of Remote
Sensing 10(12), 18731892.
Berg, A. and Gregiore, J. M.: 1983, Use of remote sensing techniques for rice production forecasting
in West Africa, (Mali and Guinea: Niger-Bani project), ESA Satellite Remote Sensing for
Developing Countries, Ispra, Italy, pp. 161168
Beven, K. J. and Kirkby, M. J.: 1979, A physically based, variable contributing area model of basin
hydrology, Hydrol. Sci. Bull. 24, 4369.
Bhavsar, P. D.: 1984, Review of remote sensing applications in hydrology and water resource
management in India, Advances in Space Research 4(11), 193200.
Blyth, K.: 1995, A telenetwork for acquisition, processing and dissemination of Earth Observation
data for monitoring and emergency management of floods, In: Proc. First ERS Thematic Working
Group Meeting on Flood Monitoring. European Space Agency/ESRIN, Frascati, Italy, 2627
June.
Boyle, S. J., Tsams, I. K., Manber, A. S. C. E., and Kanaroglu, P. S.: 1998, Developing geographical
information system for land use impact assessment in flooding conditions, Journal of Water
Resource Planning and Management 124(2), 8998.
Brakenridge, G. R., Tracy, R. T., and Know, J. C.: 1998, Orbital SAR remote sensing of the river
flood waves, International Journal of Remote Sensing 19(7), 14391445
Brimicombe, J., and Bratlet, J. M.: 1996, Linking geographical information system with hydraulic
simulation modeling for flood risk assessment: The Honk Kong Approach, In: M. F. Goodchild
(eds), GIS and Environmental Modeling, Oxford University Press, New York, pp. 165168.
Brivio, P. A., Colombo, R., Maggi, M., and Tomasoni, R.: 2002, Integration of remote sensing data
and GIS for accurate mapping of flooded areas, International Journal of Remote Sensing 23(3),
429441.
Brouder, J. A. M.: 1994, Flood study in the Meghna-Dhonagoda Polder, Bangladesh, In: Proc. Asian
Institute of Remote Sensing, Bangalore, India, 1723 November.
Brunsden, D. and Thorns, J. B.: 1979, Landscape sensitivity and change, Trans. Inst. Br. Geogr., New
Series 4, 463484.
Chen, P., Liew, S. C., and Lim, H.: 1999, Flood detection using multitemporal Radarsat and ERS
SAR data, In: Proc. 20th Asian Conference of Remote Sensing, Hong Kong, 2225 November.
Choen, K. and Kwang-Hoon, C.: 1998, Flood damage mapping in North Korea using multi-sensor
data, In: Proc. 19th Asian Conference of Remote Sensing, Manila, Philippines, 1620 November.
Clark, M. J.: 1998, Putting water in its place: A perspective on GIS in hydrology and water
management, Hydrological Processes 12, 823834.
Collins, S. H. and Moon, G. C.: 1981, Algorithms for dense digital terrain models, Photogrammetric
Engineering and Remote Sensing 47, 7176.
Coppock, J. T.: 1995, GIS and natural hazard: an overview from a GIS perspective, In: A. Carrara
and F. Guzzetti (eds), Geographical Information System in Assessing Natural Hazard, Kluwer
Academic, Netherlands, pp. 2134.
Deutsch, M.: 1976, Optical processing of ERTS data for determining extent of 1973 Mississippi River
flood, USGS Professional Paper 929, ERTS-1, A New Window on Our Planet, pp. 209222.
Deutsch, M. and Ruggles, F.: 1974, Optical data processing and projected application of the ERTS1
imagery covering the 1973 Mississippi River Valley floods,Water Resource Bulletin 10(5), 1023
1039.
Deutsch, M., Ruggles, F., Guss, P., and Yost, E.: 1973, Mapping the 1973 Mississippi floods from
the Earth Resource Technology satellites, In: Proc. International Symposium on Remote Rensing
and Water Resource Management, American Water Resource Association, No. 17, Burlington,
Ontario, pp. 3955.
Dhakal, A. S., Amda, T., Aniya, M., and Sharma, R. R.: 2002, Detection of areas associated
with flood and erosion caused by a heavy rainfall using multi temporal Landsat TM data,
Photogrammetric Engineering and Remote Sensing 68(3), 233239.
298
Fowler, R. A.: 2002, LIDAR for flood mapping, Earth Observation Magazine 9(7)
(http://www.eomonline.com/Common/Archives/July00/robert.htm, true as par 30th May, 2003)
Gamon, J. A., Field, C. B., Goulden, M. L., Griffin, K. L., Hartley, A. E., Joel, G., Penuela, J., and
Valentini, R.: 1995, Relationship between NDVI, canopy structure and photosynthesis in their
Californian vegetation types, Ecological Applications 5, 2841.
Godschalk, D. R.: 1991, Disaster mitigation and hazard management, In: T. E. Drabek and G. J. Hoetmer
(eds), Emergency Management: Principles and Practice for Local Government, International
City Management Association, Washington, D.C. pp. 131160.
Gumley, L. and King, M. D.: 1995, Remote sensing of flooding in the US Upper Midwest during the
summer of 1993, Bulletin of American Meteorological Society 76(6), 933943.
Gupta, A.: 1983, High magnitude floods and stream channel response, Special Publication of
International Association of Sedimentology 6, 219227.
Gupta, A.: 1988, Large floods as geomorphic events in the humid tropics, In: Victor R. Barker, Craig
R. Kochel, and Peter C. Paton (eds), Flood Geomorphology, John Wiley and Sons, pp. 301321.
Hallberg, G. R., Hoyer, B. E., and Rango, A.: 1973, Application of ERTS1 imagery to flood inundation
mapping, NASA Special Publication No. 327, Symposium on Significant Results Obtained
from the Earth Resources Satellite 1, Vol. 1, Technical presentations, section A, pp. 745753.
Hausmann, P. and Weber, M.: 1988, Possible contributions of hydroinformatics to risk analysis in
insurance, In: Proc. 2nd International Conference on Hydroinformatics, Zurich, Switzerland,
913 September, Balkema, Rotterdam.
Hess, L. L., Melack, J. M., and Simonett, D. S.: 1990, Radar detection of flooding beneath the forest
canopy a review, International Journal of Remote Sensing 11(7), 13131325.
Hodgson, M. E., Jensen, J. R., Schmidt, L., Schill, S., and Davis, B.: 2003, An evaluation of LIDAR
and IFSAR derived digital elevation models in leaf-on conditions with USGS level-1 and level-2
DEMs, Remote Sensing of Environment 84, 295308.
Honda, K. C., Francis, X.J., and Sah V. P.: 1997, Flood monitoring in central plain of Thailand using
JERS-1 SAR data, In: Proc. 18th Asian Conference of Remote Sensing,Malaysia, 2024 October.
Horritt, M. S. and Bates, P. D.: 2001, Predicting flood plain inundation: Raster based modeling versus
the finite element approach, Hydrological Processes 15(5), 825842.
Huguenin, R. L., Karaska, M. A., Blaricom, D. V., and Jensen, J. R.: 1997, Subpixel classification of
bald cypress and tupelo gum trees in Thematic Mapper imagery, Photogrammetric Engineering
& Remote Sensing 63, 717725.
Huh, O. K., Ali, A., and Quadir, D. A.: 1985a, Mapping of green leaf biomass over Bangladesh
with NOAA satellite AVHRR data, Manuscript Report prepared for UN Flood and Agriculture
Organization, Coastal Studies Institute, Louisiana State University., Baton Rouge, LA.
Huh, O. K., Ali, A., and Quadir, D. A.: 1985b, Observations on the surface waters of the Bay
of Bengal with NOAA satellite AVHRR data, Manuscript Report prepared for UN Flood and
Agriculture Organization, Coastal Studies Institute , Louisiana State University, Baton Rouge,
LA.
Huh, O. K., Ali, A., and Quadir, D. A.: 1985c, Use of NOAA satellite AVHRR data to monitor
river flood hydrology in Bangladesh, Manuscript Report prepared for UN Flood and Agriculture
Organization, Coastal Studies Institute, Louisiana State University, Baton Rouge, LA.
Hunter, G. J. and Goodchild, M. F: 1995, Dealing with error in spatial databases a simple casestudy,
Photogrammetric Engineering & Remote Sensing 61(5), 529537.
Imhoff, M. L., Vermillion, C., Story, M. H., Choudhury, A. M., Gafoor, A., and Polcyn, P.: 1987,
Monsoon flood boundary delineation and damage assessment using space borne imaging radar
and Landsat data, Photogrammetric Engineering and Remote Sensing 53(4), 405413.
Islam, M. M. and Sadu, K.: 2000a, Development of flood hazard maps of Bangladesh using NOAAAVHRR
images with GIS, Hydrological Science Journal 45(3), 337355.
Islam, M. M., and Sadu, K.: 2000b, Flood hazard assessment in Bangladesh using NOAA-AVHRR
data with geographical information system, Hydrological Processes 14(3), 605620.
299
Islam, M. M. and Sadu, K.: 2000c, Satellite remote sensing data analysis for flood damaged zoning
with GIS for flood management, Journal of Hydraulic Engineering 44, 301306.
Islam, M. M. and Sadu, K.: 2001, Flood damage and modelling using satellite remote sensing data
with GIS: Case study of Bangladesh; In: Jerry Ritchie et al. (eds), Remote Sensing and Hydrology
2000, IAHS Publication, Oxford, pp. 455458.
Islam, M. M. and Sadu, K.: 2002, Development of priority map remote sensing data for flood counter
measures by geographical information system, Journal of Hydrological Engineering 7(5), 346
355.
Jensen, J. R., Rutchey, K., Koch, M., and S. Narumalani: 1995, Inland Wetland Change Detection
in the Everglades Water Conservation Area 2A Using A Time Series of Normalized Remotely
Sensed Data, Photogrammetric Engineering & Remote Sensing 61(2), 199209.
Jin, Y. Q.: 1999, A flooding index and its regional threshold values for monitoring in china from
SSM/I data, International Journal of Remote Sensing 20(5), 10251030.
Jones, K. H.: 1998, A comparison of algorithms used to compute hill slope as a property of the DEM,
Comput. Geosci. 24(4), 315323.
Kundus, P., Karszenbaum, H., Pultz, T., Paramuchi, G., and Bava, J.: 2001, Influence of flood conditions
and vegetation status on the radar back scatter of wetland ecosystem, Canadian Journal of
Remote Sensing 27(6), 651662.
Lee, J., Snyder, P. K., and Fisher, P. F.: 1992, Modeling the effect of data errors on feature extraction
from digital elevation models, Photogrammetric Engineering & Remote Sensing 58(10), 1461
1467.
Liu, R. and Liu, N.: 2002, Flood area and damage estimation in Zhejiang, China, Journal of
Environmental Management 66, 18.
Liu, Z. L., Huang, F., Li, L.Y., and Wan, E. P.: 1999, Dynamic monitoring and damage evaluation of
flood in northwest Jilin with remote sensing, In: Proc. 20th Asian Conference of Remote Sensing,
Hong Kong, 2225 November.
Long, N. T. and Trong, B. D.: 2001, Flood monitoring of Mekong River Delta, Vietnam using ERS
SAR Data, In: Proc. 22nd Asian Conference of Remote Sensing, Singapore, 59 November.
Lowry, R. T., Langham, E. J., and Murdy, N.: 1981, A preliminary analysis of SAR mapping of
Manitoba flood, May 1979, In: Proc. Satellite Hydrology, Fifth Anniversary. William T. Pecora
Memorial Symposium on Remote Sensing 1979, AmericanWater Resource Association Technical
Publication No. TPS81-1, pp. 316323.
Lyon, J. D., Yuan, G., Lunetta, R. S., and Elvidge, C. D.: 1998, A change detection experiment using
vegetation indices, Photogrammetric Engineering and Remote Sensing, 64, 143150.
McGinnis, D. F. and Rango, A.: 1975, Earth Resource Satellite System for flood monitoring,
Geophys. Res. Lett. 2(4), 132135.
Melack, J. M., Hess, L. L., and Sippel, S.: 1994, Remote sensing of lakes and floodplains in the
Amazon Basin, Remote Sensing Review 10, 127142.
Michener, W. K. and Houhoulis, P. F.: 1997, Detection of vegetation changes associated with extensive
flooding in a forested ecosystem, Photogrammetric Engineering & Remote Sensing 63(12),
13631374.
Miller, L. D., Yang, Y., Matthews, M., and Irons, R. T.: 1983, Correlations of rice yields to radiometric
estimates of canopy biomass as a junction of growth stage, In: Proc. 4th Asian Conference of
Remote Sensing, 1015 November, Colombo, Srilanka, A.6.1A 6.21.
Moore, I. D., Grayson, R. B., and Ladson, A. R.: 1991, Digital terrain modelling: A review of
hydrological, geomorphological, and biological applications, Hydrological Processes 5, 330.
Morrison, R. B. and Cooley, M. E.: 1973, Assessment of flood damage in Arizona by means of
ERTS-1 imagery, In: Proc. Symposium on Significant Result Obtained from the Earth Resource
Satellite 1, Vol. 1, New Carrollton, Maryland, pp. 755760.
Morrison, R. B. and White, P. G.: 1976, Monitoring flood inundation, USGS Professional Papers No.
929, ERTS-1, A New Window on Our Planet, pp. 196208.
300
Nico, G., Pappalepore, M., Pasquariello, G., Refice, A., and Samarelli, S.: 2000, Comparison of
SAR amplitude vs. coherence flood detection methods- a GIS application, International Journal
of Remote Sensing 21(8), 16191631.
Oberstadler, R., Honsch, H., and Huth, D.: 1997, Assessment of the mapping capabilities of ERS-1
SAR data for flood mapping: A case study of Germany, Hydrological Processes 10(10), 1415
1425.
Okamoto, K. and Fukuhara,M.: 1996, Estimation of paddy field area using the area ratio of categories
in each pixel of Landsat TM, International Journal of Remote Sensing 17, 17351749.
Okamoto, K., Yamakawa, S., and Kawashima, H.: 1998, Estimation of flood damage of North Korea
in 1995, International Journal of Remote Sensing 19(2), 365371.
Patel, N. K., Singh, T. P., Sahai, B., and Patel, M.: 1985, Spectral response of rice crops and its
relation to yield and yield attributes, International Journal of Remote Sensing 6, 657664.
Pope, K. O., Rejmankova, E., Paris, J.F. et al.: 1997, Detecting seasonal flooding cycles in marshes
of the Yucatan Peninsula with SIR-C polarimetric radar imagery, Remote Sensing of Environment
59(2), 157166.
Profeiti, G. and MacIntosh, H.: 1997, Flood management through Landsat TM and ERS SAR data:
a case study, Hydrological Processes 11(10), 13971408.
Quan,W.,Watanabe, M., Hayashi, S., and Murakami, S.: 2003, Using NOAA AVHRR data to assess
flood damage in China, Environmental Monotoring and Assessment 82, 119148.
Rango, A. and Anderson, A. T.: 1974, Flood hazard studies in the Mississippi River Basin using
remote sensing, Water Resource Bulletin 10(5), 10601081.
Rango, A. and Solomonson, V. V.: 1974, Regional flood mapping from space, Water Resource
Research 10(3), 473484.
Rango, A. and Solomonson, V. V.: 1977, The utility of short wavelength (<1 mm) remote sensing
techniques for the monitoring and assessment of hydrological parameters, In: Proc. 11th Int.
Symp. on Remote Sensing of Environ., Ann Arbor, MI, 2529 April 1977, pp. 5564.
Rashid, H. and Pramanik, M. A. H.: 1993, Areal extent of the 1988 flood in Bangladesh: How much
did the satellite imagery show? Natural Hazards 8, 189200.
Reed, S., Johnson, D., and Sweeney, T.: 2002, Application of national geographic information system
data base to support two-year flood and the threshold run-off estimate, Journal of Hydrological
Engineering 7(3), 209219.
Rejesk, D.: 1993, GIS and risk: A three cultural problem, In: M. F. Goodchild, B. O. Parks, and L.
T. Steyaerts (eds), Environmental Modeling With GIS, Oxford University Press, New York, pp.
49267.
Rocha, J. S.,Marques, Z., Ramos, I., and Ameida, R.: 1994, Simulation of risk flood areas on GIS, In:
Tsakiris, G. and Santos, M. A. (eds), Advances in Water Resource Technology and Management,
Balkema, Rotterdam, pp. 375383.
Rosenqvist, A., Shimada, M., Chapman, B., Freeman, A., Grande, G. De, and Saatchi, S.: 2000,
The global rainforest mapping a review, International journal of Remote Sensing 21(6 & 7),
13751387.
Rosenqvist, A., Forsberg, B. R., Pimentel, T., Rauste, Y. A., and Riche, J. E.: 2002, The use of space
borne radar data to model inundation patterns and trace gas emission in the central Amazon flood
plain, International Journal of Remote Sensing 23(7), 13031328.
Ruangsiri, P., Sripumin, R., Polongam, S., Kanjanasuntorn, P., and Wongparn, S.: 1984, State of
flooding in the Mun-Chi River Basin area, N. E. Thailand by digital Landsat data analysis,
Report: Remote Sensing Division, National Resource Council of Thailand, Bangkok, Thailand.
Sado, K., and Islam, M. M.: 1997, Satellite remote sensing data analysis for flooded area and weather
study: Case study of Dhaka city, Bangladesh, Journal of Hydraulic Engineering 41, 945950.
Saikh, M., Green, D., and Cross, H.: 2001, A remote sensing approach to determine environmental
flows for wet lands of the Lower Darling River, New South Wales, Australia, International
Journal of Remote Sensing 22(9), 17371751.
301
Sharma, P.: 1999, Flood risk mapping of Dikrong subbasin of Assam, Proc. Map Asia 1999. Source
: http://www.gisdevelopment.net/proceedings/mapindia/1999/index.htm
Shibayama, M. and Akiyama, T.: 1989, Seasonal visible, near-infrared and multi infrared spectra
of rice canopies in relation to LAI and above- ground dry phytomass, Remote Sensing of the
Singh, A.: 1989, Digital change detection techniques using remotely-sensed data, International
Smith, L. C.: 1997, Satellite remote sensing of river inundation area, stage and discharge: A review,
Hydrological Processes 11, 14271439.
Smith, C. and Price, J.: 2002, A new generation of flood management, Arc User 5(4), 2022.
Tappen, G. D., Tyler, D.,Wehde,M., andMoor, D.: 1992, Monitoring rangeland dynamics in Senegal
with AVHRR data, Geocarto International 1, 8798.
Tholey, N., Clandillon, S., and De Fraipont, P.: 1997, The contribution of space borne SAR and
optical data in monitoring flood events; examples in Northern and Southern France, Hydrological
Processes 11(10), 14271439.
Toutin, T.: 2002, DEMfrom stereo Landsat 7 ETM+ data over high relief areas, International Journal
Townsend, P. A. and Walsh, S. J.: 1998, Modelling flood plain inundation using integrated GIS with
radar and optical remote sensing, Geomorphology 21(98), 295312.
Wadge, G., Wislocki, A. P., Pearson, J., and Whittow, J. B.: 1993, Mapping natural hazards with
spatial modeling System, In: P. M. Mather (eds), Geographic Information Handling Research
and Applications, John Wiley and Sons Inc., New York.
Wang, W., Hess, L. L., Filoso, S., and Melack, T. M.: 1995, Understanding the radar back scattering
from flooded and non-flooded Amazonian Forests: result from canopy back scatter modelling,
Remote Sensing of Environment 54(3), 324332.
Wang, Y., Colby, J. D., and Mulcahy, K. A.: 2002, An efficient method for mapping flood extent in
a coastal flood plain using Landsat TM and DEM data, International Journal of Remote Sensing
23(18), 36813696.
Wiesnet, D. R., McGinnis, D. F., and Pritchard, J. A.: 1994, Mapping of the 1973 Mississippi River
floods by the NOAA-2 satellite, Water Resource Bulletin 10(5), 10401049.
Wolock, D. M.: 1995, Effects of subbasin size on topographic characteristics and simulated flow
paths in sleepers river watershed, Vermont, Water Resour. Res. 31(8), 19891997.
Yamagata, Y. and Akiyama, T.: 1988, Flood damage analysis using multi temporal Landsat TM data,
International Journal of Remote Sensing 9(3), 503514.
Yang, C., Zhou, C., and Wan, Q.: 1999, Deciding the flood extent with Radarsat SAR data and Image
Fusion, In: Proc. 20th Asian Conference of Remote Sensing, Hong Kong, 2225 November.

Management with Special Reference to
Monsoon Asia: A Review
ARTICLE in NATURAL HAZARDS SEPTEMBER 2004
Impact Factor: 1.72 DOI: 10.1023/B:NHAZ.0000037035.65105.95
CITATIONS
127
READS
702
2 AUTHORS:
Joy Sanyal
Presidency University

SEE PROFILE
"X.X." Lu
National University of Singapore
SEE PROFILE
Available from: "X.X." Lu
Natural Hazards 33: 283301, 2004.

2004 Kluwer Academic Publishers. Printed in the Netherlands.
283

Management with Special Reference to Monsoon
Asia: A Review
JOY SANYAL_ and X. X. LU
Department of Geography, Faculty of Arts and Social Science, National University of Singapore, 1
Arts Link, Singapore 117570
(Received: 20 June 2003; accepted: 4 November 2003)
Abstract. The conventional means to record hydrological parameters of a flood often fail to record
an extreme event. Remote sensing technology along with geographic information system (GIS) has
become the key tool for flood monitoring in recent years. Development in this field has evolved
from optical to radar remote sensing, which has provided all weather capability compared to the
optical sensors for the purpose of flood mapping. The central focus in this field revolves around
delineation of flood zones and preparation of flood hazard maps for the vulnerable areas. In this
exercise flood depth is considered crucial for flood hazard mapping and a digital elevation model
(DEM) is considered to be the most effective means to estimate flood depth from remotely sensed or
hydrological data. In a flat terrain accuracy of flood estimation depends primarily on the resolution
of the DEM. River flooding in the developing countries of monsoon Asia is very acute because of
their heavy dependence on agriculture but any flood estimation or hazard mapping attempt in this
region is handicapped by poor availability of high resolution DEMs. This paper presents a review of
application of remote sensing and GIS in flood management with particular focus on the developing
countries of Asia.
Key words: flood, remote sensing, geographic information system, digital elevation model, hazard
mapping, monsoon Asia
1. Introduction
Among all kinds of natural hazards of the world flood is probably most devastating,
wide spread and frequent. In the humid tropics and subtropical climates, especially
in the realms of monsoon, river flooding is a recurrent natural phenomenon. Floods
resulting from excessive rainfall within a short duration of time and consequent
high river discharge damage crops and infrastructures. They also result in siltation
of the reservoirs and hence limit the capacity of existing dams to control floods.
For formulating any flood management strategy the first step is to identify the
area most vulnerable to flooding. This step is even more critical for the developing
countries in the monsoon Asia as the funding available for developmental activities
is very limited. Thus these funds need to be utilized optimally for the areas that
_
Author for correspondence. E-mail: g0202381@nus.edu.sg

suffer from river flooding most frequently. With the equipment currently installed
at river gauging stations it is sometimes difficult to record an extreme flood event
having a very high return period. In the developing counties, the density of gauging
stations is very low and thus any flood prediction or risk assessment model tested
in the developed countries faces acute shortage of ground data when applied.
Remote
sensing is a reliable way of providing synoptic coverage over a wide area
in a very cost effective manner. It also overcomes the limitation of the ground
stations to register data in an extreme hydrological event. In addition multi-date
imageries equip the investigators with an additional tool of monitoring the change
or reconstruct progress of a past flood.
For the last two decades advancement in the field of remote sensing and
geographic
information system (GIS) have greatly facilitated the operation of flood
mapping and flood risk assessment. It is evident that GIS has a great role to play in
natural hazard management because natural hazards are multi dimensional and the
spatial component is inherent (Coppock, 1995). The main advantage of using GIS
for flood management is that it not only generates a visualization of flooding but
also creates potential to further analyze this product to estimate probable damage
due to flood (Hausmann et al., 1998; Clark, 1998).
Smith (1997) reviews the application of remote sensing for detecting river
inundation,
stage and discharge. Since then, the focus in this direction is shifting
from flood boundary delineation to risk and damage assessment. Therefore, there
is a need to review the current literature with a holistic view of dealing with various
prospects and constraints of using the technology of remote sensing and GIS
in flood management. This paper presents recent developments on delineation of
flooded areas and flood hazard mapping using remote sensing and GIS. In particular
this paper draws attention to some of the issues associated with application of
remote sensing in combating floods in extremely flat flood plains of monsoon Asia.
Our review includes three aspects. First, we focus on the development of remote
sensing as a tool of flood delineation. Second, we emphasize the assessment of the
intensity of flood hazards and damage. Third, we highlight some of the issues in
the application of the technology with particular reference in the monsoon Asia, an
agricultural area with lack of high resolution spatial data bases.
2. Remote sensing as a tool of flooded area delineation
2.1. APPLICATION OF OPTICAL REMOTE SENSING
In the initial stages of satellite remote sensing the data available was from Landsat
Multi Spectral Scanner (MSS) with 80 m resolution. The pioneering investigations
in the field of application of remote sensing in flood mitigation were predominantly
concentrated on the flood prone regions of USA. MSS data were used to deal with
the flood affected areas in Iowa (Hallberg et al., 1973; Rango et al.,1974), Arizona
(Morrison et al., 1973), and Mississippi River basin (Deutsch et al., 1973; Deutsch
et al., 1974; Rango et al., 1974; McGinnis et al., 1975; Morrison et al., 1976). MSS
band 7 (0.81.1 m) has been found particularly suitable for distinguishing water
or moist soil from dry surface due to strong absorption of water in the near infrared
range of the spectrum (Smith, 1997).
From the early 1980s, Landsat Thematic Mapper (TM) imageries with 30 m
resolution became the prime source of data for monitoring floods and delineating
the boundary of inundation. Special attention was given to dealing with monsoon
flooding in the developing countries like West Africa (Berg et al., 1983), India
(Bhavsar, 1983) and Thailand (Ruangsiri et al., 1984). Landsat TM band 4 proves
to be very useful in discriminating water from the dry land surface because it is a
near equivalent of MSS band 7. But Landsat TM NIR band cannot be used optimally
in developed land use areas such as downtown commercial or industrial areas.
The main reason is that the NIR band reflects very little energy from asphalt areas,
appearing black in the imageries. Therefore makes it easy to confuse developed
areas with water. Wang et al. (2002) successfully solved this problem by adding
Landsat TM band 7 to the NIR (band 4) band to delineate the inundated areas.
In TM band 7 (2.082.35 micrometre) images the reflectance from water, paved
road surfaces, and roof tops differ significantly and therefore in the Band4+Band7
image, it becomes easier to choose the density slice for extracting the flood water.
But in some cases a simple density slice or supervised classification is not enough
to identify the inundated area accurately.
During later stages SPOT multi spectral imageries, were also used for flood
delineation with the similar assumption that water has very low reflectance in
the near infrared portion of the spectra. SPOT imageries, for example, were used
along with a DEMfor delineation of monsoon flood in Bangladesh (Brouder, 1994;
Oberstadler et al., 1997; Profeiti et al., 1997; Sado et al., 1997).
Apart from these medium resolution imageries, coarse resolution imageries like
Advanced Very High Resolution Radiometer Radiometer (AVHRR) data have been
also found useful for floods of a regional dimension ( Wiesnet et al., 1974; Huh et
al., 1985ac; Ali et al., 1987; Islam et al., 2000ac, 2001, 2002). Although AVHRR
imageries are coarse in resolution and frequently contaminated by cloud cover their
merit lies in their high temporal resolution. This advantage enables us to monitor
the progress of a flood in near real-time.
To use the capability of the near infra red band more effectively to detect water
Normalized Difference Vegetation Index (NDVI) can be used to monitor river
inundation
from AVHRR images. It is well known that water has a unique spectral
signature in the near infrared which is very different from other surface features.
Therefore, when a surface feature is inundated its NDVI value changes considerably
from the normal situation. Wang et al. (2002) observed that in the lower
reaches of the Yangtze River, the NDVI value for inundated surface features remains
negative while the value for non inundated surface is commonly greater than
0. But choice of this threshold is critical because natural condition of river flooding
varies greatly from place to place. The main difficulties of selecting an appropriate
threshold arise from two facts. Firstly the albedo of water bodies increases signi286
ficantly due to high concentration of sediment in the flooded water and secondly,
albedo of bare soil decreases considerably due to its high moisture content during
the monsoon season. These two factors collectively reduce the difference in NDVI
value between inundated and dry surface. In some studies, NDVI values of flood
water were found to be significantly positive (Barton et al., 1989) Thus, a straight
forward approach of using simple NDVI values might not be universally effective
in delineation of inundated area. Moreover, many other factors such as atmospheric
condition, cloud cover and satellite viewing angle also influence NDVI values and
attempts should be made to minimize these effects before calculating the NDVI.
2.2. APPLICATION OF MICROWAVE REMOTE SENSING
The existence of cloud cover appears as the single most important impediment to
capture the progress of floods in bad weather condition (Rango et al., 1977; Lowry
et al., 1981; Imhoff et al., 1987; Rashid et al., 1993; Melack et al., 1994). The
development of microwave remote sensing, particularly radar imageries, solve the
problem because the radar pulse can penetrate cloud cover. Currently the most
common approach to flood management is to use synthetic aperture radar (SAR)
imagery and optical remote sensing imagery simultaneously in one project (Honda
et al., 1997; Liu et al., 1999; Chen et al., 1999). Apart from its all weather capability
the most important advantage of using SAR imagery lies in its ability to sharply
distinguish between land and water.
Thresholding is one of the most frequently used techniques in active remote
sensing to segregate flooded areas from non flooded areas in a radar image (Liu
et al., 1999; Townsend et al., 1998; Brivio et al., 2002). Commonly, a threshold
value of radar back scatter is set in decibel (dB) and a binary algorithm is followed
to determine whether a given raster cell is flooded or not. Radar back scatter is
computed as a function of the incidence angle of the sensor and digital number
(DN) (Chen et al., 1999). The threshold values are determined by a number of
processes depending on the study area and overall spectral signature of the
imagery.
Change detection can be used as a powerful tool to detect flooded area in SAR
imagery. It is generally performed by acquiring two imageries taken before and
after the flood. Coherence and amplitude change detection techniques are widely
applied in SAR domain. In the amplitude approach, areas are estimated as flooded
where the radar back scatter is observed to be in considerable decline from before
flood to after flood imagery. In the coherence approach areas are generally
identified as flooded where the coherence or correlation of radar backscatters from

before and after flood imagery are very low (Nico et al., 2000). Multi-date SAR
scenes for the same area can be projected to red, green and blue channels to create
a colour composite. Long et al. (2001) used three ERS SAR scenes to produce
this kind of composite image. The composite image effectively depicts progress of
a flood during a specific time period. This methodology is simple to execute and
provides an opportunity to readily identify the area that remains water logged for a
maximum period of time.
The existing studies pointed out some common problems encountered in accurately
extracting the flood affected area from SAR imageries. A major problem is
associated with the relation between radar wave length and roughness of the
terrain
and water body. Normally pure and calm water acts as a specular reflector to the
radar signals. Thus the radar antennae receive no backscatter and the water
appears
in dark tone in the SAR imageries. Rough water surface appears in brighter tone
in the SAR imageries than the calm water (Yang et al., 1999). During floods, bad
and windy condition usually prevails over the affected area. Wind induced ripples
in the water surface frequently creates problems for the interpreter to determine
the
threshold value to delineate the flooded area.
Forest cover also poses an obstacle to accurately identify inundated areas from a
SAR image (Hess et al., 1990). The key to identify the inundated areas under forest
cover lies in the fact that flooded forests produces a bright radar back scatter in
contrast to non flooded forests due to a double bounce effect (Kundus et al., 2001),
whereas the flooded areas without a forest canopy appear dark in SAR imageries.
The flooded forests reflect 2dB as radar back scatter in the L-band (Pope et al.,
1997). This particular property has been utilized to separate the inundated areas
from the non inundated ones by applying the thresholding technique (Townsend
et al., 1998; Ali et al., 2001; Rosenqvist et al., 2002). A correct separation of
the flooded and non-flooded settlements is also problematic. Normally the high
back scatter of the buildings overlay the back scatter of flood water within the
settlements. Rural settlements, especially in monsoon Asia are often surrounded
by trees. Therefore, due to the effect of trees, inundation within the settlements is
very difficult to detect (Oberstadler et al., 1997).
Establishing a universal threshold value for detecting flood is not justified. A
particular algorithm of deriving the threshold for differentiating inundated areas
from the dry land may not work with the same efficiency in varied natural settings.
Jin (1999) used DMPS SSM/I (Specific Sensor Microwave/Imager) data for this
purpose and came to the conclusion that a particular algorithm developed on the
basis of a wet land environment produces spurious results for inundated areas
under
a forest canopy. He emphasized the importance of knowledge in the regional
geography of the area under investigation in setting any threshold value.
Combination of wave length, incidence angle and polarization play a major role
in influencing the interpreters ability to segregate flood areas from the non-flooded
ones under a forest canopy. Given the same wave length and incidence angle the
ratio of back scatter from flooded forests to the non-flooded ones is higher in
horizontal
polarization than at vertical polarization. It can be further demonstrated

that while wave length and polarization parameters are held constant, the aforesaid
ratio is larger under a small incidence angle than large (Wang et al., 1995). Radar
back scatter also depends on the orientation of the rough surface (e.g., furrows
of a ploughed field). As radar signals are directional in nature, the same surface
may produce different tonal signatures depending on the relative orientation of the
rough surface to the radar antenna.
Radar incidence angle and the consequent variation in back scatter also pose
difficulties in the delineation of inundated areas from SAR imageries. Generally,
it is found that dealing with forested area imageries taken at a lower incidence
angle prove more useful compared to imageries taken at a higher incidence angle.
Townsend et al. (1998) recommended the use of JERS SAR imageries to ERS for
identification of flooded areas under forest canopy due to the smaller incidence
angle of the former as compared to the latter. This satellite has some advantages
over other earth observation satellite systems to detect flooding under forest as the
L-Band signal of JERS1 is particularly sensitive to standing water below the forest
canopy (Rosenqvist et al., 2002). Radarsat SAR imageries is also preferable to ERS
for flood mapping as the former can rotate its sensor disseminating the radar signal
at different incidence angles. This manoeuvring capability proves very useful in
locating the flood affected area in different types of terrain and land cover (Andre
et al., 2002).
2.3. A COMBINED APPROACH
In recent years, flood mapping efforts synthesize the advantages of both optical
and microwave remote sensing technologies for better results. In some occasions
this approach also leads to the formulation of better flood management strategy. In
mountains, slopes positioned perpendicular to the radar beam only appear bright
and all other areas appear as dark or shaded. This poses an obstacle to effectively
identify the flooded areas in the mountains. Due to its shaded appearance it is very
common to erroneously identify the mountainous areas as inundated. Yang et al.
(1999) came out with a solution to this problem by fusing Landsat TM imageries
with SAR imageries. After extracting the mountain regions from a Landsat TM
data it was overlaid on the SAR imagery. To get rid of the spuriously demarcated
flooded area in the SAR imagery, the portion identified as mountain in the
TM image was eliminated from it. This process led to accurate delineation of the
inundated area. However, the authors realized that adoption of this method may
result in removal of some existing water bodies from the actual hydrological layer
of the mountainous area. To rectify this error, the TM data was consulted and the
water bodies in the mountain region have been restored in the final output. This
paper illustrates that the method of using data captured by different sensors is not
straight forward and often requires iterative experiments to arrive at an optimal
result.
In this context, the utility of generating land cover maps is realized as it provides
information about permanent water bodies in normal hydrological conditions and
rescues the interpreter from possible confusion of including permanent water bodies
as inundated area (Tholey et al., 1997). But while dealing with the monsoon
flood one should consult the imagery of a wet season to carve out the natural
drainage. Use of an imagery captured in dry season would lead to underestimation
of the natural drainage, which in turn would lead to an overestimation of the flooded
area (Islam et al., 2002).

The efforts towards delineation of inundated area are ultimately aimed at assessing
the impact of flood on our economy and livelihood. Land use maps obtained
primarily from optical remote sensing are overlaid on the flood maps for assessing
the degree of damage on different kind of land uses. Varieties of land use
classification
strategies have been adopted by researchers to estimate flood risk. An
analysis of most of the papers show predominance of classifying land use pattern of
the flood prone areas into conventional categories like cropland, urban area, barren
land etc (Islam et al., 2001). However, these conventional land cover classifications
do not serve the purpose optimally. Since damage inflicted over areas of high value
land use is of greater concern a more detailed classification of the economically
intensive land use is recommended. It can be done meaningfully by clubbing low
value land use areas like barren land, forest, marshy land into one category and
applying a rigorous classification of arable land and urban areas (Honda et al.,
1997). It is also realized that a data archive for selected satellites should be built for
the developing countries. This kind of archive would depict the background land
cover condition for the study area over different seasons and facilitate the process
of change detection. Only with such a data set can remote sensing of flooded areas
become a useful tool in flood management and mitigation.
3. Flood Hazard & Risk Mapping with GIS and Remote Sensing
Another primary issue for flood management is to identify the area having higher
hazard potential. Hazard can be defined as some threat, natural, technological, or
civil to people, property and environment and risk is viewed as the probability that
a hazard will occur during a particular time period. Flood is a natural hazard and
flood risk is defined in terms of hundred-year flood (Godschalk, 1991). The issue
of preparing a reliable hazard map is one of the latest concerns within the subject
of flood management.
Different approaches have been taken to map the potential hazard. Rejesk
(1993) introduces three different methods for hazard zoning. His first method
describes
a binary model which evaluates whether the hazard is present or not in a
particular raster cell. The second method involves ranking different locations of an
area depending upon the intensity of the hazard present. In the last approach some
hazard values have been assigned to each of the raster cells based on the results
of a multivariate model which were built up on a host of variables related to river
flooding and associated hazards. But the quantification of the severity of the hazard
in some units or numbers as proposed by Rejesk (1993) in his last approach was
criticized by some authors (Wadge et al., 1993).
Flood depth is considered as the most important indicator of the intensity of
the hazard (Islam et al., 2001, 2000; Townsend et al., 1998; Wadge et al., 1993).
Flood risk maps are prepared depending on the estimated depth of inundation. The
estimation is commonly derived from various hydrological and remotely sensed
data. For identifying flood depth it is important to classify the phenomenon of river
flooding into two categories, namely non-source flood and source flood.
Nonsource
flood is defined as inundation caused by well distributed rain storm over
a large area while source flood describes inundation caused primarily by over
bank flow, predominantly affecting areas near the river channel (Liu et al., 2002).
These two basic characteristics of overland flow have significant implications for
formulating the GIS model. In the case of non-source floods, all the raster cells or
vector points having an elevation below the water level are considered as
inundated
while for source floods, it is necessary to simulate the path of over bank flow
from the main channel to the adjacent flood plain to accurately estimate the flood
affected area.
The concept of topographical convergence or wetness index (Beven et al., 1979;
Moore et al., 1991; Wolock et al., 1995) was used to measure the depth of
inundation
(Townsend et al., 1998). The concept of wetness index is based on the
assumption that the accumulation of water in a particular cell of a raster depends
upon the area of the upslope region contributing water to that particular cell. The
main problem of using this index is that when a slope tends to zero the wetness
index becomes undefined. Thus this index is not very useful for modelling in
extremely flat flood plains. Townsend et al. (1998) also used another model for
simulating source flooding. The model assumes that the potential for any site to
be inundated is directly related to the difference in elevation between that site and
the river at its nearest hydrological link.
Perhaps the most innovative, simple and cost effective study regarding flood
hazard management has been conducted by Islam et al. (2001). He assessed the
flood depth from NOAA AVHRR imageries simply by the tonal difference of the
flood water. In this study, the flood affected area was subdivided into different flood
depth zones using supervised classification. To accurately identify the training sets,
the AVHRR data was superimposed over a DEM. Flood hazard has been assessed
by calculating a weighted score for each land use, physiographic and geologic
division of the country. The highlight of this methodology is that it assigns greater
weight to the categories of deeper flood depth in an exponential manner. In other
words, for the shallow depth, the weight increases in a progressive manner but
beyond a certain depth the increment of weight is much higher than the previous
depth categories. This process ensures that areas having higher depth of inundation
will be assigned a high hazard denomination. The rationale is that after a certain
depth, the flood water becomes very destructive and identifying this critical depth
is crucial for mapping hazard zone. This depth of water, or more precisely the
critical river stage, is likely to vary from region to region depending on the local
topography, building materials, settlement pattern etc.
Population density and hazard ranking have been multiplied to produce land
development
priority maps. Zones of this type of maps indicate development priority
for flood counter measures (Islam et al., 2002). This method is heavily dependent
upon the skill of the analyst who chooses the training sites for different depth of
inundation. The process of choosing the training sets is likely to be hampered by
forest cover and varying sediment load of the flood water which may alter the
natural
relationship of depth of the water and its tonal appearance. The main limitation
of this approach is that AVHRR imageries are too coarse to be used for analyzing a
local flood and cloud free images are very rare, especially in the monsoon season.
Boyle et al. (1998) used an even more diverse dataset to estimate flood risk.
Land use, hydraulic characteristics and human activities like demography, property
values and land ownership have been taken into consideration to determine the
type, location and severity of the hazard produced by the floods. Maps have been
prepared showing the properties under potential risk in the event of floods having
500, 100 to 52 years return period.
Flood hazard maps derived from coarse to moderate resolution imageries is
misfit in the developed countries. This kind of flood hazard maps can hardly make
any improvement to the existing flood insurance infrastructure in North America
or Western Europe. But in contrast, these maps are extremely useful for the
developing
countries of monsoon Asia, countries like China, India and Bangladesh etc.
These countries often suffer from devastating monsoon floods and a large
proportion
of the population in these countries live in the flood prone areas. In this part
of the world flood insurance maps are often unheard of. Here flood hazard maps
are primarily required for saving the lives and livelihood of millions of marginal
farmers. Thus, these flood hazard maps can be used meaningfully by planners to
formulate strategies to combat this natural hazard.
Developed countries also need to evaluate the accuracy of flood risk mapping.
A very accurate flood risk map will exclude those individuals, found to be located
at a very high flood risk area, from a flood insurance scheme. On the other hand,
the insurance sector would lower the premium for those individuals who would be
found at minimum flood risk zone. Thus, a very high resolution flood hazard map
is likely to reverse the basic principle of insurance that loss of few should fall upon
many (Clark, 1998). Therefore, generalizing the risk zones to a certain extent is
likely to be more beneficial to the community as a whole.
Hazard maps should have a good visual effect so that the end user can have an
overview of the entire situation at a glance. In recent times hazard maps have
become
so technical in nature that they convey very little information to the planners
and policy makers (Rocha et al., 1994). Unless the experts of hydrology and GIS
overcome this handicap, hazard maps cannot be used optimally by all the relevant
users.
4. Some Issues of Remote Sensing Applications with Special Reference to
Monsoon Asia
Having briefly given an overview of the development and methods of the
application
of remote sensing in flood management, this section tries to highlight the
constraints in applying this technique. Some of the problems, for example,
agricultural
damage assessment, are typical for monsoon Asia which is heavily dependent
on agriculture
4.1. DEPENDENCY OF DIGITAL ELEVATION MODELS IN FLOOD MANAGEMENT
In a majority of the studies dealing with the application of remote sensing in
inundated
area delineation and flood risk assessment, digital elevation models (DEM)
are used to visualize the interface of flood water with the terrain. Flood depth is
normally calculated by subtracting the elevation of each cell in a raster from its
water level (Brouder, 1994; Townsend et al., 1998; Ali et al., 2001; Islam et al.,
2002). DEMs are also used to simulate the flood depth from discharge data and
very often the result is compared with actual flooded area derived from satellite
imageries. As the spatial extent of inundation is subjected to a method of cross
checking this methodology is likely to yield a more accurate flood map as compared
to one derived from pure hydrological modelling. The main drawback of this
approach is that it is fully dependent on the accuracy of the DEM (Jones et al.,
1998). In the largely featureless plains of monsoon Asia, the accuracy required of
the DEMs is almost unattainable.
In a flat flood plain, where a vertical error of 1 m in the DEM may lead to an
error of 100s of square kilometres in flood estimation, recognition of the magnitude
of errors in the DEM is important in hydrological modelling. This issue has been
addressed from the view point of its significance in flood plain mapping (Lee
et al., 1992; Hunter et al., 1995). In the near absence of high resolution DEMs
required for flood mapping in a very flat terrain, multi-date SAR imageries can
serve as a very potent alternative data set in monsoon Asia. For example, multidate
Radarsat imageries were used in monsoon Asia to depict progress of a flood from
its inception to the peak (Liu, 1999; Chen, 1999). This particular operation can
create a visualization of the course of inundation from the river channel to the
adjacent low lying areas of the flood plain and complement the method of flow
direction simulation. Although multi-date imageries can serve as an alternative to
flow direction simulation, it has severe limitation of determining the flood depth.
A very accurate terrain data base regarding the local flood plain morphology is an
essential prerequisite for such an operation. The flood water intersecting with the
slope is taken as the primary indicator of determining flood depth. Thus for the
gently sloping topography, the resolution of the terrain data actually controls the
accuracy of the estimated flood depth (Brakenridge et al., 1998). High resolution
satellite imageries or aerial photographs are needed for preparing an accurate DEM
which can meet the precision level of a flood depth investigation.
One of the recent developments in the application of remote sensing to flood
related problems is the use of LIDAR (Light Detecting and Ranging) sensor. In the
developed countries, especially in USA, this technology has become very popular
for creating DEMs for flood prone areas. LIDAR sensors can readily identify the
vertical differences in the landforms and can be exploited as powerful instruments
to create DEMs of exceptional accuracy. This sensor can also detect the flood depth.
Although LIDAR sensor can attain the vertical accuracy of 5 cm or better, it is
difficult to map it in that resolution. Due to the limitation of GPS systems to locate
an aircraft/sensor functionally LIDAR generated DEMs are released at an accuracy
of 15 to 25 cm RMSE. Moreover the accuracy decreases gradually with increase
in the density of vegetation cover of the ground (Hodgson et al., 2003). Although
LIDAR data is even more expensive than the SAR imageries, sometimes it provides
the only appropriate option to do flood mapping in the extremely flat flood plains.
Fowler (2002) maintained that the resolution of the LIDAR data depends upon the
intensity of the laser pulses and any attempt to make the survey more intensive by
increasing the laser pulses would increase the cost of the data exponentially.
4.2. AGRICULTURAL DAMAGE ASSESSMENT
Boyle et al. (1998) classified flood damages into two categories; tangible and
intangible.
Tangible damage occurs due to direct contact with the flood water whereas
intangible damage is exemplified by the loss of historical monuments, heritage sites
etc. A disease assuming a form of an epidemic due to flood is also categorized as
an intangible damage. In the agricultural landscape of monsoon Asia, estimation

of agricultural damages requires special attention. Asias population is
predominantly
rural and its economy is heavily reliant on agriculture. Monsoon flood often
creates havoc on the economy by damaging the standing paddy. This problem is
unique for monsoon Asia because in the developed world damage to urban area
and infrastructure facilities is of concern to government authorities. Hence, in this
section we shall emphasize studies dealing with application of remote sensing in
assessment of agricultural damage.
Erosion of top soil due to a flash flood and deposition of flood borne coarse
sand reduce the fertility of soil very severely and thus have a negative impact on
agricultural economy. The process of change detection is found useful to monitor
this kind of damage to agricultural land. The most widely used procedure is to
monitor the change in brightness value (VB) at a particular wave length or different
bands to identify the erosion caused by a flood. Several change detection
techniques
like Spectral Image Differencing (SID), Tasseled Cap Brightness Image Differencing
(TCBID), Principal Component Analysis (PCA), and Spectral Change Vector
Analysis (SCVA) are employed for the purpose of detecting the erosions due to
flooding, but for Landsat TM data, SCVA is found to yield most accurate results
(Dhakal et al., 2002).
In the flood plains of monsoon Asia rice is the dominant crop. Landsat TM4 band (Near Infrared) has been extensively used to estimate the damage to rice
crops due to flood (Shibayama et al., 1989; Okamoto et al., 1996) while the use of
band 3 is also considered to be very effective in assessment of paddy field damage
(Yamagata et al., 1988) due to its very high reflectance in turbid or muddy water
typically associated with flooded paddy field in the riverine plains of monsoon Asia
(Miller et al., 1983; Patel et al., 1985). The studies in this direction even attempted
to estimate the quantum of crop damage from satellite data. For example, a DN
value of 84 (in TM-4) indicates a yield of 3.0 metric tons/ha, whereas a value
around 55 corresponds to no crop in North Korea (Okamoto et al., 1998). The
main shortcoming of formulating a universal range of DN number is that it might
fail to extract accurate information for different varieties of rice crops.
Traditional supervised classification by maximum likelihood method generally
perform well in flood affected agricultural area where the instantaneous field of
view (IFOV) is less than the inundated segment of land and the land use is
dominated
by crop lands (Jensen et al., 1995; Hugenin et al., 1997). The mean value
of Normalized Difference Vegetation Index (NDVI) derived from JERS-1 OPS
data was used to differentiate damaged crops from undamaged one in North Korea
(Choen et al., 1998) because vegetation condition or biomass over a portion of land
is highly correlated with NDVI (Singh, 1989; Tappen et al., 1992; Gamon et al.,
1995; Michener et al., 1997; Lyon et al., 1998).
4.3. PROBLEM OF TEMPORAL RESOLUTION IN FLOOD MANAGEMENT
Devastating floods are generally low frequency, high magnitude natural
phenomena.
Flash floods occur within a very short interval of time and the peak stage
remains only for a couple of hours, but the most extensive and severe damage
takes
place during that time. With the current Radarsat resources it is very difficult to
capture the spatial extent of a flood at its peak. Thus, attempts have been made to
extrapolate the extent of inundation at the peak of a flood from an image acquired
at a later stage of the event. Some GIS algorithms in ARC/INFO are promising to
perform this extrapolation from an image that captures some standing water only at
a time when the flood peak had already passed. The method of least accumulation
cost distance can provide a viable solution to this problem (Brivio et al., 2002).
This principle simulates flow direction from the river channel to the flood plain
based on the assumption that water flows through the path where the work done in
doing so is least.
This methodology yielded a remarkable accuracy of 96.7 percent when compared
with the aerial photographs of the peak of the particular flood event which
occurred in November 1994 in Italy. The main weakness of this approach is that
the least accumulation cost distance operates on certain values of the raster cells
which represent the roughness of the terrain causing frictional drag to the
overflowing
flood water. Since roughness is a function of a host of other geomorphic and
lithological factors, it is very difficult to control the parameters of the experiments
in an area having diverse lithology and land use. Apart from the DEMs and remote
sensing data, field work conducted in recently flood affected area can prove useful.
Wang et al. (2002) used high-flood level marks of a recent flood on houses as a
supplementary data set to the DEMs for estimating flood depth. This kind of data
serve very well to reconstruct dimension of a past flood peak.
General studies of flood geomorphology also provide some insights to resolve
this problem. Many very high magnitude floods may leave permanent imprints in
the flood plain morphology rather than creating some transient features (Brunsden
et al., 1979).When the morphology of the basin perfectly suites with the magnitude
and frequency of the flood event, the flood created features attain a state of
permanency.
Since the frequency of such an event is high in the humid tropics, the interval
between two high magnitude floods gets reduced here. It indirectly contributes to
the permanency of the flood created characteristic morphological features in the
flood plain (Gupta, 1983). The permanency of this kind of topographic features
would be more prominent in areas of high relief and coarser valley sediments
(Gupta, 1988). Hence in the humid tropics there is enough scope to use relatively
inexpensive high resolution optical remote sensing not necessarily at the time of
flooding to demarcate the areas vulnerable to river flooding. But the analyst has to
be trained enough to identify the morphological features in the flood plain typically
associated with high magnitude floods.
5. Conclusion and Prospective
The preceding discussion regarding various facades of the prospects and constraints
of using remote sensing and GIS for flood management unveils some
significant facts. A majority of the researchers favoured multi date radar imageries
to observe a particular flood event and considered different image processing
techniques
to overcome the limitations of remotely sensed data in flood delineation.
The main weaknesses of this type of approach are manifold. Most of the
investigations
mentioned in this paper are heavily dependent on the availability of satellite
data, which is not always guaranteed for the time of peak flood. Rather than moving
towards a comprehensive flood management strategy, these papers concentrated
on
some specific issues like delineation of flooded area. Most of these projects are on
a very high budget and no attempts were made try to keep the cost low. Monsoon
floods affect the developing countries more acutely than the developed ones.
Therefore,
while framing the methodology one should be aware of its feasibility in the
operational area. Most of the studies in this field cannot achieve its desired level
of accuracy without a very high resolution DEM, therefore, we need to develop
alternative methodology to shed our dependence on high resolution terrain data.
A detailed hydro-geomorphological mapping depicting the trace of past floods
may help us in this direction. The hazard of monsoon flood and the destitution
of people associated with it are very different from the industrially developed
countries. Hence, we should adopt an improvised but effective methodology of
risk and damage assessment to come out with meaningful flood hazard maps for
this region.
Application of remote sensing and GIS is convincingly a very efficient and cost
effective way of flood management. Use of very high resolution imageries like
IKONS or SPOT 5 have not been very popular yet in the field of flood management
due to its high price, but it is likely that with these imageries would be available
at a reasonable price and would be widely used for flood mapping. In the age
of internet, GIS has assumed new dimensions, especially for coping with natural
disasters like river flooding. The most important advantage of using internet based
GIS is that it has opened the door of GIS technology to the end users who would not
like to install expensive GIS software. One of the numerous examples is Arc IMS
technology. This technology has been used to develop a web enabled application
named Map Action Processing Digital Interactive Geo Resource (MAPDIGR) for
providing very recent information regarding flood risk to an analyst via internet
(Smith et al., 2002). This technology is at present at an embryonic stage of
development
but has great potential for expand the user base of GIS technology for flood
management by substantially reducing the cost of operation.
Since the problem of flood is very acute in the developing countries of monsoon
Asia, special attention should be given to deal with this problem in the regional
context. GIS models having low cost and simple data requirement are likely to
attract the local authorities in the developing countries to adopt this technology
as an essential input towards a comprehensive flood management system. In the
age of all embracing flood plain management, these sophisticated technologies can
be very useful for the planners to formulate effective strategy for combating the
perpetual natural disaster of river flooding.
Acknowledgement
This study has been done as part of an ongoing project titled Application
of Remote Sensing and GIS in Flood Hazard Management: A Case Study of
West Bengal, India, funded by National University of Singapore research grant
R-109-000-049-112.
References
Ali, A. and Quadir, D. A.: 1987, Agricultural hydrologic and oceanographic studies in Bangladesh
with NOAA AVHRR data, International Journal of Remote Sensing 8(6), 917925.
Ali, S., Hassan, A.,Martin, T. C., and Hassan Q. K.: 2001, Geospatial tools for monitoring flood plain
water dynamics, In: M. Owe, K. Brubaker, J. Ritchie, and A. Rango, A. (eds), Remote Sensing
and Hydrology 2000, IAHS Publication, Oxford, pp. 465471.
Andre, G., Guillande, R., and Bahoken, F.: 2002, Flood mapping using spatial radar and optical
imagery and digital elevation model: Limits and capacities, Houille Blanche-Revue Int. De LEAU
1, 4954.
297
Barton, I. and Bathols, J.: 1989, Monitoring floods with AVHRR, International Journal of Remote
Sensing 10(12), 18731892.
Berg, A. and Gregiore, J. M.: 1983, Use of remote sensing techniques for rice production forecasting
in West Africa, (Mali and Guinea: Niger-Bani project), ESA Satellite Remote Sensing for
Developing Countries, Ispra, Italy, pp. 161168
Beven, K. J. and Kirkby, M. J.: 1979, A physically based, variable contributing area model of basin
hydrology, Hydrol. Sci. Bull. 24, 4369.
Bhavsar, P. D.: 1984, Review of remote sensing applications in hydrology and water resource
management in India, Advances in Space Research 4(11), 193200.
Blyth, K.: 1995, A telenetwork for acquisition, processing and dissemination of Earth Observation
data for monitoring and emergency management of floods, In: Proc. First ERS Thematic Working
Group Meeting on Flood Monitoring. European Space Agency/ESRIN, Frascati, Italy, 2627
June.
Boyle, S. J., Tsams, I. K., Manber, A. S. C. E., and Kanaroglu, P. S.: 1998, Developing geographical
information system for land use impact assessment in flooding conditions, Journal of Water
Resource Planning and Management 124(2), 8998.
Brakenridge, G. R., Tracy, R. T., and Know, J. C.: 1998, Orbital SAR remote sensing of the river
flood waves, International Journal of Remote Sensing 19(7), 14391445
Brimicombe, J., and Bratlet, J. M.: 1996, Linking geographical information system with hydraulic
simulation modeling for flood risk assessment: The Honk Kong Approach, In: M. F. Goodchild
(eds), GIS and Environmental Modeling, Oxford University Press, New York, pp. 165168.
Brivio, P. A., Colombo, R., Maggi, M., and Tomasoni, R.: 2002, Integration of remote sensing data
and GIS for accurate mapping of flooded areas, International Journal of Remote Sensing 23(3),
429441.
Brouder, J. A. M.: 1994, Flood study in the Meghna-Dhonagoda Polder, Bangladesh, In: Proc. Asian
Institute of Remote Sensing, Bangalore, India, 1723 November.
Brunsden, D. and Thorns, J. B.: 1979, Landscape sensitivity and change, Trans. Inst. Br. Geogr., New
Series 4, 463484.
Chen, P., Liew, S. C., and Lim, H.: 1999, Flood detection using multitemporal Radarsat and ERS
SAR data, In: Proc. 20th Asian Conference of Remote Sensing, Hong Kong, 2225 November.
Choen, K. and Kwang-Hoon, C.: 1998, Flood damage mapping in North Korea using multi-sensor
data, In: Proc. 19th Asian Conference of Remote Sensing, Manila, Philippines, 1620 November.
Clark, M. J.: 1998, Putting water in its place: A perspective on GIS in hydrology and water
management, Hydrological Processes 12, 823834.
Collins, S. H. and Moon, G. C.: 1981, Algorithms for dense digital terrain models, Photogrammetric
Engineering and Remote Sensing 47, 7176.
Coppock, J. T.: 1995, GIS and natural hazard: an overview from a GIS perspective, In: A. Carrara
and F. Guzzetti (eds), Geographical Information System in Assessing Natural Hazard, Kluwer
Academic, Netherlands, pp. 2134.
Deutsch, M.: 1976, Optical processing of ERTS data for determining extent of 1973 Mississippi River
flood, USGS Professional Paper 929, ERTS-1, A New Window on Our Planet, pp. 209222.
Deutsch, M. and Ruggles, F.: 1974, Optical data processing and projected application of the ERTS1
imagery covering the 1973 Mississippi River Valley floods,Water Resource Bulletin 10(5), 1023
1039.
Deutsch, M., Ruggles, F., Guss, P., and Yost, E.: 1973, Mapping the 1973 Mississippi floods from
the Earth Resource Technology satellites, In: Proc. International Symposium on Remote Rensing
and Water Resource Management, American Water Resource Association, No. 17, Burlington,
Ontario, pp. 3955.
Dhakal, A. S., Amda, T., Aniya, M., and Sharma, R. R.: 2002, Detection of areas associated
with flood and erosion caused by a heavy rainfall using multi temporal Landsat TM data,
Photogrammetric Engineering and Remote Sensing 68(3), 233239.
298
Fowler, R. A.: 2002, LIDAR for flood mapping, Earth Observation Magazine 9(7)
(http://www.eomonline.com/Common/Archives/July00/robert.htm, true as par 30th May, 2003)
Gamon, J. A., Field, C. B., Goulden, M. L., Griffin, K. L., Hartley, A. E., Joel, G., Penuela, J., and
Valentini, R.: 1995, Relationship between NDVI, canopy structure and photosynthesis in their
Californian vegetation types, Ecological Applications 5, 2841.
Godschalk, D. R.: 1991, Disaster mitigation and hazard management, In: T. E. Drabek and G. J. Hoetmer
(eds), Emergency Management: Principles and Practice for Local Government, International
City Management Association, Washington, D.C. pp. 131160.
Gumley, L. and King, M. D.: 1995, Remote sensing of flooding in the US Upper Midwest during the
summer of 1993, Bulletin of American Meteorological Society 76(6), 933943.
Gupta, A.: 1983, High magnitude floods and stream channel response, Special Publication of
International Association of Sedimentology 6, 219227.
Gupta, A.: 1988, Large floods as geomorphic events in the humid tropics, In: Victor R. Barker, Craig
R. Kochel, and Peter C. Paton (eds), Flood Geomorphology, John Wiley and Sons, pp. 301321.
Hallberg, G. R., Hoyer, B. E., and Rango, A.: 1973, Application of ERTS1 imagery to flood inundation
mapping, NASA Special Publication No. 327, Symposium on Significant Results Obtained
from the Earth Resources Satellite 1, Vol. 1, Technical presentations, section A, pp. 745753.
Hausmann, P. and Weber, M.: 1988, Possible contributions of hydroinformatics to risk analysis in
insurance, In: Proc. 2nd International Conference on Hydroinformatics, Zurich, Switzerland,
913 September, Balkema, Rotterdam.
Hess, L. L., Melack, J. M., and Simonett, D. S.: 1990, Radar detection of flooding beneath the forest
canopy a review, International Journal of Remote Sensing 11(7), 13131325.
Hodgson, M. E., Jensen, J. R., Schmidt, L., Schill, S., and Davis, B.: 2003, An evaluation of LIDAR
and IFSAR derived digital elevation models in leaf-on conditions with USGS level-1 and level-2
DEMs, Remote Sensing of Environment 84, 295308.
Honda, K. C., Francis, X.J., and Sah V. P.: 1997, Flood monitoring in central plain of Thailand using
JERS-1 SAR data, In: Proc. 18th Asian Conference of Remote Sensing,Malaysia, 2024 October.
Horritt, M. S. and Bates, P. D.: 2001, Predicting flood plain inundation: Raster based modeling versus
the finite element approach, Hydrological Processes 15(5), 825842.
Huguenin, R. L., Karaska, M. A., Blaricom, D. V., and Jensen, J. R.: 1997, Subpixel classification of
bald cypress and tupelo gum trees in Thematic Mapper imagery, Photogrammetric Engineering
& Remote Sensing 63, 717725.
Huh, O. K., Ali, A., and Quadir, D. A.: 1985a, Mapping of green leaf biomass over Bangladesh
with NOAA satellite AVHRR data, Manuscript Report prepared for UN Flood and Agriculture
Organization, Coastal Studies Institute, Louisiana State University., Baton Rouge, LA.
Huh, O. K., Ali, A., and Quadir, D. A.: 1985b, Observations on the surface waters of the Bay
of Bengal with NOAA satellite AVHRR data, Manuscript Report prepared for UN Flood and
Agriculture Organization, Coastal Studies Institute , Louisiana State University, Baton Rouge,
LA.
Huh, O. K., Ali, A., and Quadir, D. A.: 1985c, Use of NOAA satellite AVHRR data to monitor
river flood hydrology in Bangladesh, Manuscript Report prepared for UN Flood and Agriculture
Organization, Coastal Studies Institute, Louisiana State University, Baton Rouge, LA.
Hunter, G. J. and Goodchild, M. F: 1995, Dealing with error in spatial databases a simple casestudy,
Photogrammetric Engineering & Remote Sensing 61(5), 529537.
Imhoff, M. L., Vermillion, C., Story, M. H., Choudhury, A. M., Gafoor, A., and Polcyn, P.: 1987,
Monsoon flood boundary delineation and damage assessment using space borne imaging radar
and Landsat data, Photogrammetric Engineering and Remote Sensing 53(4), 405413.
Islam, M. M. and Sadu, K.: 2000a, Development of flood hazard maps of Bangladesh using NOAAAVHRR
images with GIS, Hydrological Science Journal 45(3), 337355.
Islam, M. M., and Sadu, K.: 2000b, Flood hazard assessment in Bangladesh using NOAA-AVHRR
data with geographical information system, Hydrological Processes 14(3), 605620.
299
Islam, M. M. and Sadu, K.: 2000c, Satellite remote sensing data analysis for flood damaged zoning
with GIS for flood management, Journal of Hydraulic Engineering 44, 301306.
Islam, M. M. and Sadu, K.: 2001, Flood damage and modelling using satellite remote sensing data
with GIS: Case study of Bangladesh; In: Jerry Ritchie et al. (eds), Remote Sensing and Hydrology
2000, IAHS Publication, Oxford, pp. 455458.
Islam, M. M. and Sadu, K.: 2002, Development of priority map remote sensing data for flood counter
measures by geographical information system, Journal of Hydrological Engineering 7(5), 346
355.
Jensen, J. R., Rutchey, K., Koch, M., and S. Narumalani: 1995, Inland Wetland Change Detection
in the Everglades Water Conservation Area 2A Using A Time Series of Normalized Remotely
Sensed Data, Photogrammetric Engineering & Remote Sensing 61(2), 199209.
Jin, Y. Q.: 1999, A flooding index and its regional threshold values for monitoring in china from
SSM/I data, International Journal of Remote Sensing 20(5), 10251030.

Jones, K. H.: 1998, A comparison of algorithms used to compute hill slope as a property of the DEM,
Comput. Geosci. 24(4), 315323.
Kundus, P., Karszenbaum, H., Pultz, T., Paramuchi, G., and Bava, J.: 2001, Influence of flood conditions
and vegetation status on the radar back scatter of wetland ecosystem, Canadian Journal of
Remote Sensing 27(6), 651662.
Lee, J., Snyder, P. K., and Fisher, P. F.: 1992, Modeling the effect of data errors on feature extraction
from digital elevation models, Photogrammetric Engineering & Remote Sensing 58(10), 1461
1467.
Liu, R. and Liu, N.: 2002, Flood area and damage estimation in Zhejiang, China, Journal of
Environmental Management 66, 18.
Liu, Z. L., Huang, F., Li, L.Y., and Wan, E. P.: 1999, Dynamic monitoring and damage evaluation of
flood in northwest Jilin with remote sensing, In: Proc. 20th Asian Conference of Remote Sensing,
Hong Kong, 2225 November.
Long, N. T. and Trong, B. D.: 2001, Flood monitoring of Mekong River Delta, Vietnam using ERS
SAR Data, In: Proc. 22nd Asian Conference of Remote Sensing, Singapore, 59 November.
Lowry, R. T., Langham, E. J., and Murdy, N.: 1981, A preliminary analysis of SAR mapping of
Manitoba flood, May 1979, In: Proc. Satellite Hydrology, Fifth Anniversary. William T. Pecora
Memorial Symposium on Remote Sensing 1979, AmericanWater Resource Association Technical
Publication No. TPS81-1, pp. 316323.
Lyon, J. D., Yuan, G., Lunetta, R. S., and Elvidge, C. D.: 1998, A change detection experiment using
vegetation indices, Photogrammetric Engineering and Remote Sensing, 64, 143150.
McGinnis, D. F. and Rango, A.: 1975, Earth Resource Satellite System for flood monitoring,
Geophys. Res. Lett. 2(4), 132135.
Melack, J. M., Hess, L. L., and Sippel, S.: 1994, Remote sensing of lakes and floodplains in the
Amazon Basin, Remote Sensing Review 10, 127142.
Michener, W. K. and Houhoulis, P. F.: 1997, Detection of vegetation changes associated with extensive
flooding in a forested ecosystem, Photogrammetric Engineering & Remote Sensing 63(12),
13631374.
Miller, L. D., Yang, Y., Matthews, M., and Irons, R. T.: 1983, Correlations of rice yields to radiometric
estimates of canopy biomass as a junction of growth stage, In: Proc. 4th Asian Conference of
Remote Sensing, 1015 November, Colombo, Srilanka, A.6.1A 6.21.
Moore, I. D., Grayson, R. B., and Ladson, A. R.: 1991, Digital terrain modelling: A review of
hydrological, geomorphological, and biological applications, Hydrological Processes 5, 330.
Morrison, R. B. and Cooley, M. E.: 1973, Assessment of flood damage in Arizona by means of
ERTS-1 imagery, In: Proc. Symposium on Significant Result Obtained from the Earth Resource
Satellite 1, Vol. 1, New Carrollton, Maryland, pp. 755760.
Morrison, R. B. and White, P. G.: 1976, Monitoring flood inundation, USGS Professional Papers No.
929, ERTS-1, A New Window on Our Planet, pp. 196208.
300
Nico, G., Pappalepore, M., Pasquariello, G., Refice, A., and Samarelli, S.: 2000, Comparison of
SAR amplitude vs. coherence flood detection methods- a GIS application, International Journal
Oberstadler, R., Honsch, H., and Huth, D.: 1997, Assessment of the mapping capabilities of ERS-1
SAR data for flood mapping: A case study of Germany, Hydrological Processes 10(10), 1415
1425.
Okamoto, K. and Fukuhara,M.: 1996, Estimation of paddy field area using the area ratio of categories
in each pixel of Landsat TM, International Journal of Remote Sensing 17, 17351749.
Okamoto, K., Yamakawa, S., and Kawashima, H.: 1998, Estimation of flood damage of North Korea
in 1995, International Journal of Remote Sensing 19(2), 365371.
Patel, N. K., Singh, T. P., Sahai, B., and Patel, M.: 1985, Spectral response of rice crops and its
relation to yield and yield attributes, International Journal of Remote Sensing 6, 657664.
Pope, K. O., Rejmankova, E., Paris, J.F. et al.: 1997, Detecting seasonal flooding cycles in marshes
of the Yucatan Peninsula with SIR-C polarimetric radar imagery, Remote Sensing of Environment
59(2), 157166.
Profeiti, G. and MacIntosh, H.: 1997, Flood management through Landsat TM and ERS SAR data:
a case study, Hydrological Processes 11(10), 13971408.
Quan,W.,Watanabe, M., Hayashi, S., and Murakami, S.: 2003, Using NOAA AVHRR data to assess
flood damage in China, Environmental Monotoring and Assessment 82, 119148.
Rango, A. and Anderson, A. T.: 1974, Flood hazard studies in the Mississippi River Basin using
remote sensing, Water Resource Bulletin 10(5), 10601081.
Rango, A. and Solomonson, V. V.: 1974, Regional flood mapping from space, Water Resource
Research 10(3), 473484.
Rango, A. and Solomonson, V. V.: 1977, The utility of short wavelength (<1 mm) remote sensing
techniques for the monitoring and assessment of hydrological parameters, In: Proc. 11th Int.
Symp. on Remote Sensing of Environ., Ann Arbor, MI, 2529 April 1977, pp. 5564.
Rashid, H. and Pramanik, M. A. H.: 1993, Areal extent of the 1988 flood in Bangladesh: How much
did the satellite imagery show? Natural Hazards 8, 189200.
Reed, S., Johnson, D., and Sweeney, T.: 2002, Application of national geographic information system
data base to support two-year flood and the threshold run-off estimate, Journal of Hydrological
Engineering 7(3), 209219.
Rejesk, D.: 1993, GIS and risk: A three cultural problem, In: M. F. Goodchild, B. O. Parks, and L.
T. Steyaerts (eds), Environmental Modeling With GIS, Oxford University Press, New York, pp.
49267.
Rocha, J. S.,Marques, Z., Ramos, I., and Ameida, R.: 1994, Simulation of risk flood areas on GIS, In:
Tsakiris, G. and Santos, M. A. (eds), Advances in Water Resource Technology and Management,
Balkema, Rotterdam, pp. 375383.
Rosenqvist, A., Shimada, M., Chapman, B., Freeman, A., Grande, G. De, and Saatchi, S.: 2000,
The global rainforest mapping a review, International journal of Remote Sensing 21(6 & 7),
13751387.
Rosenqvist, A., Forsberg, B. R., Pimentel, T., Rauste, Y. A., and Riche, J. E.: 2002, The use of space
borne radar data to model inundation patterns and trace gas emission in the central Amazon flood
plain, International Journal of Remote Sensing 23(7), 13031328.
Ruangsiri, P., Sripumin, R., Polongam, S., Kanjanasuntorn, P., and Wongparn, S.: 1984, State of
flooding in the Mun-Chi River Basin area, N. E. Thailand by digital Landsat data analysis,
Report: Remote Sensing Division, National Resource Council of Thailand, Bangkok, Thailand.
Sado, K., and Islam, M. M.: 1997, Satellite remote sensing data analysis for flooded area and weather
study: Case study of Dhaka city, Bangladesh, Journal of Hydraulic Engineering 41, 945950.
Saikh, M., Green, D., and Cross, H.: 2001, A remote sensing approach to determine environmental
flows for wet lands of the Lower Darling River, New South Wales, Australia, International
Journal of Remote Sensing 22(9), 17371751.
301
Sharma, P.: 1999, Flood risk mapping of Dikrong subbasin of Assam, Proc. Map Asia 1999. Source
: http://www.gisdevelopment.net/proceedings/mapindia/1999/index.htm
Shibayama, M. and Akiyama, T.: 1989, Seasonal visible, near-infrared and multi infrared spectra
of rice canopies in relation to LAI and above- ground dry phytomass, Remote Sensing of the
Singh, A.: 1989, Digital change detection techniques using remotely-sensed data, International
Smith, L. C.: 1997, Satellite remote sensing of river inundation area, stage and discharge: A review,
Hydrological Processes 11, 14271439.
Smith, C. and Price, J.: 2002, A new generation of flood management, Arc User 5(4), 2022.
Tappen, G. D., Tyler, D.,Wehde,M., andMoor, D.: 1992, Monitoring rangeland dynamics in Senegal
with AVHRR data, Geocarto International 1, 8798.
Tholey, N., Clandillon, S., and De Fraipont, P.: 1997, The contribution of space borne SAR and
optical data in monitoring flood events; examples in Northern and Southern France, Hydrological
Processes 11(10), 14271439.
Toutin, T.: 2002, DEMfrom stereo Landsat 7 ETM+ data over high relief areas, International Journal
Townsend, P. A. and Walsh, S. J.: 1998, Modelling flood plain inundation using integrated GIS with
radar and optical remote sensing, Geomorphology 21(98), 295312.
Wadge, G., Wislocki, A. P., Pearson, J., and Whittow, J. B.: 1993, Mapping natural hazards with
spatial modeling System, In: P. M. Mather (eds), Geographic Information Handling Research
and Applications, John Wiley and Sons Inc., New York.
Wang, W., Hess, L. L., Filoso, S., and Melack, T. M.: 1995, Understanding the radar back scattering
from flooded and non-flooded Amazonian Forests: result from canopy back scatter modelling,
Remote Sensing of Environment 54(3), 324332.
Wang, Y., Colby, J. D., and Mulcahy, K. A.: 2002, An efficient method for mapping flood extent in
a coastal flood plain using Landsat TM and DEM data, International Journal of Remote Sensing
23(18), 36813696.
Wiesnet, D. R., McGinnis, D. F., and Pritchard, J. A.: 1994, Mapping of the 1973 Mississippi River
floods by the NOAA-2 satellite, Water Resource Bulletin 10(5), 10401049.
Wolock, D. M.: 1995, Effects of subbasin size on topographic characteristics and simulated flow
paths in sleepers river watershed, Vermont, Water Resour. Res. 31(8), 19891997.
Yamagata, Y. and Akiyama, T.: 1988, Flood damage analysis using multi temporal Landsat TM data,
International Journal of Remote Sensing 9(3), 503514.
Yang, C., Zhou, C., and Wan, Q.: 1999, Deciding the flood extent with Radarsat SAR data and Image
Fusion, In: Proc. 20th Asian Conference of Remote Sensing, Hong Kong, 2225 November.
SATELLITE REMOTE SENSING OF RIVER INUNDATION AREA,

STAGE, AND DISCHARGE: A REVIEW
LAURENCE C. SMITH1
1University
of California, Los Angeles, Department of Geography, Los Angeles, CA 90095-1524, U.S.A.
ABSTRACT
The growing availability of multi-temporal satellite data has increased opportunities for monitoring
large rivers from
space. A variety of passive and active sensors operating in the visible and microwave range are
currently operating, or
planned, which can estimate inundation area and delineate ood boundaries. Radar altimeters show
great promise for
directly measuring stage variation in large rivers. It also appears to be possible to obtain estimates of
river discharge
from space, using ground measurements and satellite data to construct empirical curves that relate
water surface area to
discharge. Extrapolation of these curves to ungauged sites may be possible for the special case of
braided rivers.
Where clouds, trees and oating vegetation do not obscure the water surface, high-resolution
visible/infrared sensors
provide good delineation of inundated areas. Synthetic aperture radar (SAR) sensors can penetrate
clouds and can
also detect standing water through emergent aquatic plants and forest canopies. However, multiple
frequencies and
polarizations are required for optimal discrimination of various inundated vegetation cover types.
Existing singlepolarization, xed-frequency SARs are not sucient for mapping inundation area in all riverine
environments. In the
absence of a space-borne multi-parameter SAR, a synergistic approach using single-frequency, xedpolarization SAR
and visible/infrared data will provide the best results over densely vegetated river oodplains. #1997
John Wiley &
Sons, Ltd.
Hydrol. Process., Vol. 11, 14271439 (1997)
(No. of Figures: 3 No. of Tables: 0 No. of Refs: 103)
KEY WORDS
remote sensing; river inundation; review; synthetic aperture radar, Landsat, oods
INTRODUCTION
Satellite-derived ood inundation maps produced in near-real time are invaluable to state or
national
agencies for disaster monitoring and relief eorts. New facilities are being developed that
will utilize
Internet/World Wide Web technology to disseminate satellite data rapidly during ood events
(Biasutti and
Lombardi, 1995; Blyth 1995). Precise mapping of the maximum ood extent is also required
for detecting
deciencies in existing ood control measures and for arbitrating damage claims later.
Remote sensing has proved useful in ecological, hydrological and geomorphological river
studies. Its
value in remote regions has been demonstrated in the Amazon Basin (Sipple et al., 1992;
Koblinsky et al.,
1993; Hess et al., 1995), where seasonal to interannual variations in stage and oodplain
inundation area are
needed for assessing biogeochemical processes such as methane ux and main

stemoodplain exchange
(Asselmann and Crutzen, 1989; Richey et al., 1989; Richey et al., 1991). Smith et al.
(1995a,b) describe a
method for using ERS (European Space Agency) high-resolution SAR (synthetic aperture
radar) satellite
imagery to estimate discharge in remote, braided, glacial rivers that may be sensitive to
changing regional or
global climate. ERSSAR data have also been proposed as a source for validation of
numerical hydraulic
ow models, which predict oodwave surface proles and inundation patterns (Bates and
Anderson, 1995).
Swamp and other habitat types in the large and ecologically diverse Okavango Delta of
Botswana have
been mapped using Landsat MSS (Ringrose et al., 1988) and TM (Watson, 1991) imagery. A
similar study
established the morphometry and connectivity of the thousands of channels, lakes and
ponds that constitute
the Mackenzie Delta in Canada's Northwest Territories (Mouchot et al., 1991); the derived
information is
useful for watercraft navigation, sheries and wildlife habitat monitoring. Intermittently
ooded areas that
CCC 08856087/97/10142713$17_50 Received 30 October 1995
#1997 John Wiley & Sons, Ltd. Accepted 30 April 1996
HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997)
are potential breeding grounds for mosquitoes that carry the dangerous Rift Valley Fever
virus have been
mapped in Kenya using Landsat TM and airborne polarimeter data (Pope et al., 1992).
Floodplain boundaries and land surface types can be delineated with Landsat (Rango et al., 1975; Hollyday,
1976; Sollers et al.,
1978); Nagarajan et al. (1993) used Landsat images and aerial photographs over the Rapti
River in India to
identify areas vulnerable to channel migration and oods.
The sensors used in these and other river studies may be classied into two types: (1)
passive, in which the
sensor receives energy naturally reected by or emitted from the earth's surface; and (2)
active, in which the
sensor provides its own illumination and records the amount of incident energy returned
from the imaged
surface. Passive sensors include all of the visible and infrared instruments such as the
Landsat Thematic
Mapper (TM) and Multi-Spectral Scanner (MSS), the Advanced Very High Resolution
Radiometer
(AVHRR), the Satellite Pour l'Observation de la Terre (SPOT) and the anticipated Advanced
Spaceborne
Thermal Emission and Reection Radiometer (ASTER), Moderate-Resolution Imaging
Spectroradiometer
(MODIS) and Landsat-7 sensors. Passive microwave radiometers such as the Special Sensor
Microwave/
Imager (SSM/I) measure the amount of microwave energy naturally emitted from the Earth's
surface.
However, the poor spatial resolution of spaceborne microwave radiometers (ca. 27 km at 37
GHz) limits
their use to very large areas. This problem has been mitigated through use of spectral
mixing models, which
extract subpixel proportions of spectrally distinct end members (Sipple et al., 1992;
Hamilton et al., 1996).
The active sensors described in this review consist of imaging radars and radar altimeters.
Radars can
penetrate clouds, darkness and, at the longer wavelengths, tree canopies. Cloud penetration
is particularly
important for monitoring ood events, as they commonly occur during periods of extended
rainfall.
However, interpretation of synthetic aperture radar (SAR) imagery is less straightforward
than for the
visible/infrared range. In addition, the presence of wind-induced waves or emergent
vegetation can roughen
the surface of open water bodies, making them dicult to discriminate from other, nonooded land surface
types.
The aim of this paper is to review briey eorts to use active and passive remote sensing to
estimate water
surface area, stage and discharge. Methods for mapping surface area are by far the best
developed. More
recently, improvements in satellite orbital precision and the increasing availability of multitemporal satellite
data have enabled the estimation of river stage and discharge from space. While these
techniques are largely
in their infancy and not yet used operationally, three general approaches have emerged: (1)
direct measurement of water surface level from radar altimeter waveform data; (2) determination of water
surface
elevations at their point of contact with the land surface using high-resolution satellite
imagery and
topographic data; and (3) correlation of satellite-derived water surface areas with ground
measurements of
stage or discharge. It should be noted that river ow velocity cannot be directly measured
from space.
Satellite estimates of discharge therefore require the use of ground-based empirical
relationships between
discharge and inundation area or stage.
Studies that have used passive (visible, infrared and microwave range) and active
(microwave) sensors to
delineate inundation area are reviewed in the rst section, and eorts to estimate river
stage and discharge
from space are reviewed in the second section.
REMOTE SENSING OF INUNDATION AREA WITH PASSIVE AND ACTIVE SENSORS
Visible/infrared Remote Sensing of Inundation Area
Much of the pioneering work on the remote sensing of oods was accomplished using the
Multi-Spectral
Scanner (MSS) sensor on ERTS-1 (the rst Earth Resources Technology Satellite, later
renamed Landsat-1),
launched on 23 July 1972. With a spatial resolution of about 80 m, MSS data were used to
map the extent of
ooding in Iowa (Hallberg et al., 1973; Rango and Salomonson, 1974), Arizona (Morrison and
Cooley,
1973), Virginia (Rango and Salomonson, 1974) and along the Mississippi River (Deutsch et
al., 1973;
Deutsch and Ruggles, 1974; Rango and Anderson, 1974; McGinnis and Rango, 1975;
Deutsch, 1976;
Morrison and White, 1976). Maximum ood boundaries derived from the MSS imagery were
shown to
1428 L. C. SMITH
HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997) #1997 John Wiley & Sons, Ltd.
agree well with those derived from aerial photography (Hallberg et al., 1973; Morrison and
Cooley, 1973)
and US Army Corps of Engineers and US Geological Survey ood hazard maps (Rango and
Anderson,
1974). In one case, the MSS-derived ood hazard map performed best, near a major
tributary where
backwater ooding occurred (Rango and Anderson, 1974). In all studies, MSS band 7
(0.81.1 mm) was
most useful for discriminating water from dry soil or vegetated surfaces owing to the strong
absorption of
water in the near-infrared range. This was further conrmed by comparing panchromatic
and colour
infrared aerial photographs over inundated areas (Hallberg et al., 1973; Moore and North,
1974), and by
analysing MSS band 5 (0.60.7 mm), band 7 and eld spectral radiometer data along
shoreline waterwet
soildry soil transitions (Gupta and Banerji, 1985). Errors in MSS-derived ooded areas have
been
estimated to be less than 5% (Rango and Salomonson, 1974); a similar error estimate has
been reported for
lakes (Gupta and Banerji, 1985). Chidley and Drayton (1986) suggested that the SPOT
system can detect
water bodies smaller than 0.5 ha. France and Hedges (1986) found minimum detectable lake
areas of 0.6 ha
for TM and 2.4 ha for MSS. However, only a 64% accuracy was reported for MSS
classications of ooded
land in Bangladesh, where the 80 m spatial resolution failed to resolve adequately the raised
dike system that
separates rice paddies and serves as infrastructure for transport and dwellings (Imho et al.,
1987). Even
larger errors can occur in ooded forests, because trees are highly reective in the visible
and near-infrared
range (Hallberg et al., 1973; Moore and North, 1974). Floating emergent macrophytes are
also a problem in
many riverine environments, particularly in tropical systems such as the Amazon (Melack et
al., 1994; Hess,
1995). Vila da Silva and Kux (1992) noted a signicant underestimation of inundation area
over such
aquatic habitats when using Landsat TM.
Perhaps the greatest problem with visible/infrared sensors is their inability to image the
Earth's surface
during cloudy conditions (Rango and Salomonson, 1977; Lowry et al., 1981; Van den Brink,
1986; Blyth
and Biggin, 1993; Rasid and Pramanik, 1993; Melack et al., 1994). For the purpose of
determining
maximum ood extent, this diculty is somewhat mitigated by the fact that residually wet
soils and stressed
vegetation can be mapped even after ood stage recession (Rango and Anderson, 1974;
Deutsch, 1976). This
eect can last from one to two weeks (Hallberg et al., 1973; Rango and Salomonson, 1974;
Morrison and
White, 1976; Salomonson, 1983).
Other studies have continued the methodology rst developed with MSS, using TM and
SPOT data
(France and Hedges, 1986; Jensen et al., 1986; Watson, 1991; Blasco et al., 1992; Pope et
al., 1992; Vila
da Silva and Kux, 1992). There has also been some success studying very large rivers or
oods with the
coarser resolution (ca. 1 km) US National Ocean and Atmospheric Administration Very High
Resolution
Radiometer (VHRR) (McGinnis and Rango, 1975; Dey et al., 1977), the Advanced Very High
Resolution
Radiometer (AVHRR) (Ali et al., 1989; Barton and Bathols, 1989; Gale and Bainbridge, 1990;
Rasid and
Pramanik, 1993) and the Nimbus-7 Coastal Zone Color Scanner (CZCS) (Wiesnet and
Deutsch, 1987).
However, since the 1970s there has been little change in the way visible/infrared satellite
data are used to
map open water bodies. The approach is relatively straightforward and now considered
operational where
trees and vegetation do not obscure the water surface and clear skies prevail to permit data
acquisition.
Passive Microwave Remote Sensing of Inundation Area
Natural thermal emission of microwaves from the Earth's surface may be measured by
passive microwave
radiometers such as the Scanning Multichannel Microwave Radiometer (SMMR) which ew on
board the
Nimbus-7 satellite from 1978 to 1987, and the Special Sensor Microwave/Imager (SSM/I)
currently
operating on a DMSP (Defense Meteorological Satellite Program) platform. Brightness
temperatures (K)
measured at the sensor are proportional to the product of the eective surface temperature
and the
emissivity of the emitting medium (Choudhury, 1989). At the 37 GHz frequency, atmospheric
water vapour
content and temperature inuence the received brightness temperatures. To mitigate these
eects, the
dierence between the vertically and horizontally polarized brightness temperatures (DT) is
often used
instead of the actual brightness temperatures. Large values of DT are found over surfaces
that emit a
strongly polarized microwave signal (such as open water). Depolarization by scattering from
vegetation or
REMOTE SENSING OF RIVER INUNDATION 1429
#1997 John Wiley & Sons, Ltd. HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997)
surface roughness causes values of DT to decrease (Schmugge et al., 1986; Choudhury,

1989). Scattering
from precipitating clouds can also exert a depolarizing eect, reducing DT (Hamilton et al.,
1995).
A theoretical basis for the use of DT over land surfaces is provided by Choudhury (1989). In
general, the
polarization dierence DT decreases with increasing vegetation density and surface
roughness, and increases
over wet or inundated soils.
In South America, monthly SMMR-derived values of DT were used to determine seasonal

inundation
patterns for 15 rivers and wetlands in the Amazon, La Plata, Orinoco and Sao Francisco
river systems
(Giddings and Choudhury, 1989; Choudhury, 1991). Distinct dierences in the timing of
seasonal ow
cycles can be seen in the DT time-series from a selection of basins, with more subtle
dierences found among
rivers of the same system. These studies were limited to the continental scale by the ca. 27
km resolution of
SMMR at 37 GHz. An improved 37 GHz resolution of 14 km will be provided by the MultiFrequency
Imaging Microwave Radiometer (MIMR), scheduled for launch on the EOS-PM platform in the
year 2000.
At present, subpixel estimates of inundation area may be extracted from 37 GHz SMMR data,
using
empirically derived values of DT for open water, seasonally ooded land and non-ooded
land as endmember inputs to a linear spectral mixing model (Sipple et al., 1994; Hamilton et al., 1996).
In Figure 1,
SMMR 37 GHz polarization dierences for a seasonally inundated oodplain from the
Pantanal wetland of
South America are presented with monthly mean ground measurements of discharge in the
Paraguay River.
Polarization dierences for a nearby, non-inundated upland area are also shown. A screening
procedure was
used to eliminate SMMR images that were aected by heavy rains. Values of DT over the
inundated
oodplain correspond closely with ground measurements of stage in the Paraguay River, the
major outlet of
the Pantanal. Polarization dierences over non-inundated upland show little or no change
over time,
demonstrating that uctuations in inundation area are responsible for the corresponding
changes in
microwave emission from the ooded land surface.
Radar (active microwave) Remote Sensing of Inundation Area
Spaceborne radars can image the Earth's surface in all weather conditions, day or night. As
described
earlier, the inability of visible/infrared sensors to penetrate clouds poses a severe problem
for regular
monitoring of river conditions. For example, in the Amazon basin, a search of Landsat data
archives from
19721985 found just 13 useful images per year, most occurring between July and
October (Melack et al.,
Figure 1. Time trends from 1979 to 1987 of the SMMR 37 GHz polarization dierence observed at the satellite (DTobs) over a
seasonally inundated Pantanal oodplain (bold solid line) and a nearby upland area (solid line). Polarization
dierence values are
averaged from four SMMR grid cells (oodplain) and sixteen grid cells (upland). Monthly mean stage from the
Paraguay River
(dashed line) is also shown. The strong correlation between the 37 GHz polarization dierence and observed river
stage indicate that
seasonal dierences in DT are driven largely by inundation cycles. (From Hamilton et al., 1996)
1430
L. C. SMITH
1994). Rasid and Pramanik (1993) cited pervasive cloud cover as the single greatest
limitation to their study of
ooding in Bangladesh using AVHRR imagery. Van den Brink (1986) reported the same
diculty in Kenya.
An early attempt to map ood inundation extent with active microwaves used a side-looking
airborne
radar (SLAR) at the X- and L-bands (Lowry et al., 1981). X-band SLAR imagery, acquired in
the late 1970s
was also used to map oodplain lakes along the Amazon River, Brazil (Sipple et al., 1992). At
the First ERS
Thematic Working Group Meeting on Flood Monitoring (ESA/ESRIN, 1995), numerous
investigators
presented ood inundation maps derived from C-band ERS-1 synthetic aperture radar (SAR)
images. A
general observation was that for smooth open water bodies without vegetation, especially
trees, radar
returns are normally low due to specular reection from the water surface. This
characteristic permits ood
boundaries to be determined with a good level of accuracy under many conditions. However,
turbulence,
wind, emergent vegetation and trees can all cause signicant increases in radar backscatter, making
inundation extent dicult or impossible to determine.
When a micro-scale element (at the scale of the radar wavelength) of an imaged surface is
subjected to an
incident microwave, the ratio of incident energy reected away from the surface to the
energy refracted
downwards into the media is proportional to the dielectric constant and the incidence angle
at which the
microwave intersects the imaged surface. The complex dielectric constant of a material is
proportional to the
strength of its dipole moment in the presence of a time-varying external electric eld, while
the incidence
angle (at the micro-scale) is determined by the surface roughness. Over a larger area (such
as a pixel), the
angular radiation pattern reected by the surface is also aected by the local slope, which
exerts a net
directionality to the radiation pattern. For smooth water bodies, surface roughness and the
local slope are
both nearly at. Consequently, most incident microwave energy impacts the surface
obliquely and is
specularly reected away from the satellite, yielding low radar returns. Wind reduces this
eect by roughening the water surface, increasing radar back-scattering to the satellite. Since water bodies
are perfectly
saturated, dielectric contrast eects are less important than surface roughness eects. This
is not the case for
unsaturated soils, where the presence of liquid water causes increased radar back-scatter
owing to an
increased complex dielectric constant of the soilwater mixture.When the soil becomes
saturated and water
is ponded, specular reection is enabled (in the absence of wind or emergent vegetation)
and back-scatter to
the satellite is dramatically reduced.
Where vegetation or trees are present, a wavelength-dependent increase in radar backscatter is commonly
observed. Surface roughness elements of the order of the radar wavelength exert maximum
scattering eect.
For example, X-band (2.43.8 cm) and C-band (3.87.5 cm) signals are eectively scattered
by the leaves,
twigs and branches within a forest canopy. L-band (15.030.0 cm) and P-band (30.0100.0
cm) display
greater canopy penetration but interact strongly with trunks and large branches. Radar
impulses are
transmitted and received in plane-polarized form by the radar antenna. Polarizations may be
transmitted
and received horizontally (HH), vertically (VV) or cross-polarized (HV or VH). Radars with
multiple
frequencies and polarizations provide much more information about the imaged surface than
singlepolarization, xed-frequency radars. L-band radar back-scatter is commonly increased over
ooded forests
through a double-bounce mechanism between the water surface and inundated tree trunks
or branches
(Hoer et al., 1985; Harris and Digby-Argus, 1986; Richards et al., 1987; Hess et al., 1990;
Hess andMelack,
1994). An HV phase dierence of 1538 was found in airborne polarimeter data over a
forested swamp
(Durden et al., 1989), agreeing closely with a phase dierence of 1528 modelled for the site
(a pure doublebounce return is dened as having a 1808 phase dierence between the vertical and
horizontal polarizations).
The magnitude of the back-scatter increase is signicantly aected by vegetation type
(Krohn et al., 1983;
Evans et al., 1986; Harris and Digby-Argus, 1986; Pope et al., 1994) and may be suppressed
altogether by a
dense undergrowth between the tree canopy and water surface (Waite et al., 1981).
Airborne multipolarization radar studies have generally indicated that ooding beneath forest canopies is
best detected
with co-polarizations rather than cross-polarizations (Wedler and Kessler, 1981; Hoer et al.,
1985; Evans
et al., 1986; Pope et al., 1992; Hess et al., 1995), although Wu and Sader (1987) favour use
of a co- and crosspolarization ratio.
REMOTE SENSING OF RIVER INUNDATION 1431
From L-HH Shuttle Imaging Radar B (SIR-B) SAR data collected over the Altamaha River in
Georgia,
Hess and Melack (1994) conrmed the observation of Hoer et al. (1985) that L-band
enhancement is
aected by radar incidence angle. Flooded pine forest could be discriminated from dry forest
at incidence
angles of 188 and 458 from vertical, but not 588. Horizontally polarized (HH) C-, L-, and Pband NASA/JPL
AIRSAR data were also acquired over this site, and illustrate superbly the features that can
be observed in
wetlands and ooded forests with dierent frequencies. Increased returns at C-band over
marshes were
interpreted as double-bounce reections between emergent plant stalks and a smooth water
surface. These
stalk dimensions were too small to increase returns at L- and P-band, so marshes appeared
dark at these
longer wavelengths. The converse was true in ooded bottomland forests, where L- and Pbands penetrated
the upper canopy but experienced double-bounce reections between the water surface and
inundated tree
trunks. C-band was attenuated by the leaves and twigs of the forest canopy, except in areas
defoliated by
caterpillar damage and permanent inundation by beaver ponds (Hess and Melack, 1994).
Comprehensive
reviews of the interaction of various radars with a wide range of ooded vegetation types are
provided by
Hess et al. (1990) and Melack et al. (1994).
Recent eorts have used fully polarimetric SAR data to classify various inundated land
surface types.
Amplitudes at multiple frequencies and polarizations vary with vegetation structure and
species. Multiparameter SAR can thus be used to detect ooding in vegetated riverine environments to an
extent not
possible with single-frequency, xed-polarization radars. Pope et al. (1994) presented four
biophysical
indices that can be used to discriminate thickets and areas of forest regrowth, marshes,
swamp forests and
upland forests. Hess et al. (1994, 1995) used a decision-tree model to classify polarimetric
SIR-C (Shuttle
Imaging Radar C) acquired over the Negro and Amazon rivers near Manaus, Brazil in April
and October,
1994. Both C- and L-bands were used to distinguish open water, oating macrophytes and
ooded forest
from upland forest and clearings. While the total area of open water did not change
signicantly between
data acquisitions, the extent of ooded forest was nearly 50% less in October. It is clear that
multiparameter SAR oers a distinct advantage over single-frequency, xed-polarization SAR
where diverse
vegetation cover is found.
REMOTE SENSING OF RIVER STAGE AND DISCHARGE
In the previous section studies that used passive or active sensors to measure river
inundation area or ood
extent were reviewed. In this section, a small group of studies that attempted to use satellite
data to
estimate river stage or discharge are discussed. Such eorts will likely become more
common in the future
as satellite orbital errors decrease and high-resolution multi-temporal data become more
plentiful. Data
availability has already increased with the launch of the ERS-1, ERS-2, JERS-1 and RADARSAT
spaceborne SARs, which are not limited by weather conditions or darkness. Even more data
are anticipated from future high-resolution SARs (JERS-2, ENVISAT-1, RADARSAT-2) and visible/infrared
sensors (Landsat-7, ASTER, SPOT 5 and 6).
Radar Altimetry of Water Surface Elevation
Radar altimetry has emerged as a promising method for directly measuring stage variations
in large rivers.
Radar altimeters emit a short, nadir-directed radar pulse to the Earth's surface. The two-way
return time is
used to calculate the range between the satellite and the target. Because the pulse is in the
form of a curved
wave front, a time-varying waveform is returned to the satellite. An on-board tracker, which
assumes a
gradually changing surface (i.e. ice sheets and water bodies), is used to estimate when the
return echo will
arrive. Interactions with the surface cause the waveform to be distorted. If the tracker
receives a rapidly
changing echo, the tracker may not `lock' properly and range estimates are lost. For this
reason, radar
altimeters generally do not work well over land. However, where a lock on the land power
return is
maintained, it is possible to identify peaks in the waveform data that result from specular
returns over large
rivers (Guzkowska et al., 1990). These spikes progress through the waveform as the satellite
passes over the
1432 L. C. SMITH
river. Koblinsky et al. (1993) used this approach with Geosat waveform data to estimate river
stage at four
sites along the Amazon. The derived water surface levels are presented with the
corresponding ground
measurements in Figure 2. Altimeter ranges were calculated by manually selecting specular
returns from the
waveform data. An average root mean square error of 0.7 m was found between the ground
and altimeter
estimates of water surface level. Uncertainties in the ground stage measurements (ca. 10
cm or more) and the
radial component of the satellite orbit (estimated at 50 cm) were major contributors to this
error.
Birkett (1994) obtained lake level changes using a similar analysis of Geosat altimeter
waveforms. Her
results suggest that radar altimetry can be used to measure relative water level changes in
lakes within 10 cm.
For very large lakes, waveform analysis is not required to obtain good estimates of water
level: Morris and
Gill (1994) used average height estimates from the Geosat altimeter to estimate levels in
the ve Great Lakes
of the United States, with an average root mean square error of 11.1 cm. Average heights
are a standard
altimeter product and represent the average water level over the altimeter footprint, which
typically has a
radius of several kilometres.
Figure 2. Comparisons in river stage at four sites in the Amazon basin, as estimated from river gauges (solid line)
and Geosat altimeter
waveforms (pluses). Standard deviations of the dierences between the altimeter and river stage data are also
shown. Period of
observation is from November 1986 to November 1988. (Adapted from Koblinsky et al., 1993)
REMOTE SENSING OF RIVER INUNDATION
1433
Errors in orbital precision pose the greatest problem to using satellite altimetry to measure
stage in rivers
or lakes. The Geosat orbital error has been estimated at about 50 cm (Koblinsky et al.,
1993), although it
has been subsequently reduced through recomputation of some of the 17-day exact repeat
mission (ERM)
orbits (Morris and Gill, 1994). However, this problem is steadily diminishing as satellite
tracking systems
improve. The radial component of the ERS-1 orbit is determined to within 15 cm; the
equivalent error is
only 3 cm for TOPEX/Poseidon (Le Traon et al., 1995). ERS-1 orbital errors have been reduced
even
further by using TOPEX/Poseidon data to minimize the variation in measured sea surface
heights at points
where the two satellite tracks cross (Le Traon et al., 1995). As altimeters with small orbital
errors (such as
TOPEX/Poseidon) become more common, radar altimetry over inland water surfaces may
become routine
for those sites where a lock on the land surface can be maintained.
Watersurface Elevations from Satellite Imagery and Topographic Data
River stage can be estimated at the landwater contact, using high-resolution satellite
imagery in
combination with topographic maps or a digital elevation model (DEM). Requirements for
this approach
are high image and topographic resolution and an unobscured water edge boundary along
an area of gently
sloping relief. Gupta and Banerji (1985) used topographic maps and Landsat MSS data to
obtain water
surface elevations for a large reservoir in India. Landwater contact elevations were chosen
along two lowgradient, inundated stream channels near their conuences with the reservoir. The derived
water surface
elevations agreed closely with values estimated from measurements surveyed on the
ground.
Miller (1986) used Landsat MSS and 1 : 50 000 topographic maps to obtain spot water
surface elevations
over the Belize River in Central America. Flood volumes subsequently estimated by
assuming mean
inundation depths of 1.53.0 m were in general agreement with volumes calculated from
hydrograph
analysis. However, signicant parts of the landwater boundary were obscured by
vegetation, limiting the
spot elevations to areas near permanent lagoons. Brakenridge et al. (1994) used ERS-1 SAR
images and
1 : 24 000 US Geological Survey topographic maps to pick water edge elevations at gently
sloping alluvial
fans along the Mississippi River during the `Great Flood' of 1993. Instantaneous ood stages
diered by as
much as 2.4 m at opposite sides of the Mississippi River, illustrating the value of this
approach for
estimating ood proles and monitoring ood wave dynamics.
Correlation of Inundated Area with Ground Measurements of Stage or Discharge
A third method for estimating river stage or discharge from space is by correlating ground
measurements
of these variables with satellite-derived estimates of inundation area. Until recently, the
major obstacle to
this approach has been the requirement of a large number of satellite images to construct
empirical rating
curves relating inundation area to stage or discharge. Kruus et al. (1981) compared total
inundation areas
from seven Landsat MSS scenes over the St John River in New Brunswick, Canada with
simultaneous
measurements of river stage. They found a general increase of stage with inundation area,
but at some times
an increase in stage was associated with a decrease in inundation area. Usachev (1983)
used a similar
approach to determine the relationship between inundation area and ground measurements
of stage and
discharge in the Ob River of Siberia. In all of the Ob River analyses, increasing stage or
discharge was
associated with increasing inundation area. The correlation was quasi-linear and quite
similar for three
study sections at Mogochin, Kolpashevo and Kargasok. Xia et al. (1983) presented a strongly
linear
correlation between water surface area and ground measurements of water level for
Dongting Lake, China's
second largest freshwater body.
Until additional empirical rating curves relating inundation area to ground measurements of
stage or
discharge are made, it is dicult to assess their potential for extrapolation to other rivers of
similar
morphology. However, it seems likely that such curves will vary signicantly between rivers
and therefore
must be constructed for each site. An exception may be the special case of braided rivers,
which display an
extreme spatial sensitivity of water surface area to changing discharge, and also share some
common
morphological properties. Smith et al. (1995) used multi-temporal ERS-1 SAR data and
simultaneous
1434 L. C. SMITH
ground measurements of discharge to derive a stable, power-law correlation between

satellite-derived water
surface area and discharge for a large braided river in Canada (Figure 3). Subsequent work
(Smith et al.,
1996) found similar curves for two other braided rivers in Alaska and Canada. A theoretical
power-law
correlation between water surface area and discharge was also found (Smith et al., 1996),
using a cellular
model of stream braiding (Murray and Paola, 1994). Moderate dierences in the satellitederived rating
curves relating water surface area to discharge are explained primarily by the intensity of
braiding, with
braid channel morphology and valley slope exerting smaller eects. In the absence of
ground data, it appears
that this method can be used to estimate relative discharge in order to determine the shape
and timing of
seasonal hydrographs in braided rivers. Absolute discharge can be estimated within a factor
of two, with
more accurate estimates requiring one or more ground measurements of discharge acquired
simultaneously
with a satellite image acquisition. More work is needed to determine if known morphological
controls, such
as total sinuosity, valley slope, bank material and stability, and braid channel hydraulic
geometry, can be
used to parameterize empirically derived rating curves for extrapolation to other ungauged
braided rivers.
CONCLUSIONS
High-resolution visible/infrared sensors such as Landsat provide good delineation of ood
extent where
clouds, trees or oating vegetation do not obscure the water surface. Passive microwave
radiometers show
promise for acquiring estimates of inundation area over very large rivers or wetlands, but
sensor resolution
is coarse and the landwater boundary cannot be located accurately. Spaceborne synthetic
aperture radars
(SARs) are not limited by weather conditions or darkness. Single polarization, C-band SARs
are eective
for mapping smooth, open water bodies. However, emergent vegetation, trees, wind or ow
turbulence can
all increase radar back-scatter returns, making delineation of inundated areas problematic.
L-band SARs
can penetrate forest canopies and often display increased returns over ooded forest
produced by a doublebounce mechanism between inundated tree trunks and the water surface. Multi-frequency
polarimetric SAR
provides much more information and can be used to classify dierent types of inundated
terrain to an extent
not possible with single-frequency SARs and visible/infrared sensors. However, there are no
current plans to
launch a multi-frequency, fully polarimetric SAR on a satellite platform. As a consequence, a
synergistic
Figure 3. Relationship between satellite-derived eective width and river discharge for a large braided river in
Canada. Eective width
is the total water surface area contained within a 10 km _ 3 km river reach, divided by the reach length. Each point
is determined from
a single ERS-1 SAR image and a ground measurement of discharge. The outlier is an extreme ood event; maximum
annual ows do
not normally exceed 2000 m3/s. (From Smith et al., 1995)
1435
remote sensing approach utilizing both visible/infrared and SAR data may be the only
eective way
to monitor inundation area in vegetated riverine environments. Visible/infrared data will
provide good
discrimination of ood extent in open areas, identify places where vegetation is a problem
and help with
interpretation of SAR data. SARs will provide excellent temporal coverage and, in certain
situations, will be
able to determine the ood extent through emergent plants and forest canopies.
As satellite orbital precision and data availability improve, promising methods for estimating
river
stage and discharge should continue to develop. Radar altimetry shows potential for
measuring stage in
large rivers to within 10 cm. Orbital laser altimeters such as GLAS (Geoscience Laser
Altimeter System),
planned for launch in 2003, should also permit monitoring of stage variation in large rivers.
Landsat and
ERS-1 SAR data have been combined with topography to obtain water surface elevations at
the point of
contact between water and land. This approach can also be used to determine lateral
asymmetries in river
stage and estimate the ood prole. Satellite-derived estimates of water surface area have
been correlated
with ground measurements of stage or discharge to obtain rating curves that can
subsequently be used to
estimate these ow variables from satellite data alone. Transfer of such empirical rating
curves to estimate
discharge in ungauged basins appears feasible in the special case of braided rivers, which
are spatially
sensitive to changing discharge and commonly share some morphological similarities
between sites.
ACKNOWLEDGEMENTS
This work was carried out at the Department of Geological Sciences, Cornell University, with
funding from
NASA through Graduate Student Researchers Program Fellowship NGT-51223, Mission to
Planet Earth/
Earth Observing System (EOS) grant NAGW-2638 and Spaceborne Imaging Radar (SIR-C)
grant 958745.
Helpful comments and assistance were provided by A. Bloom, B. Isacks, J. Liebeskind and R.
Forster of the
Department of Geological Sciences, Cornell University. S. Hamilton of the W.K. Biological
Station,
Michigan State University, and C. Koblinsky of NASA/Goddard Space Flight Center provided
Figures 1
and 2.
REFERENCES
Ali, A., Quadir, D. A., and Huh, O. K. 1989. `Study of river ood hydrology in Bangladesh with AVHRR data', Int. J.
Remote Sens.,
10(12), 18731891.
Asselmann, I. and Crutzen, P. J. 1989. `Global distribution of natural freshwater wetlands and rice paddies, their net
primary
productivity, seasonality, and possible methane emissions', J. Atmos. Chem., 8, 307358.
Astaras, T. 1985. `Drainage network analysis of Landsat images of the Olympus-Pieria mountain area, Northern
Greece', Int. J.
Remote Sens., 6, 673686.
Astaras, T., Lambrinos, N., and Soulakellis, N. 1990. À drainage system analysis evaluation of, and comparison
between,
Landsat-3 RBV, Landsat-5 TM and SPOT PA imageries covering the Central Macedonia district, Greece', Int. J.
Remote
Sens., 11(9), 15491559.
Barrett, E. C., Herschy, R. W., and Stewart, J. B. 1988. `Satellite remote sensing requirements for hydrology and
water management
from the mid 1990's, in relation to the Columbus Programme of the European Space Agency', Hydrol. Sci. J., 33(1),
117.
Barton, I. J., and Bathols, J. M. 1989. `Monitoring oods with AVHRR', Remote Sens. Environ., 30, 8994.
Bates, P. D. and Anderson, M. G. 1995. Ìssues of hydraulic model validation and design using remotely sensed
data', Proceedings,
First ERS Thematic Working Group Meeting on Flood Monitoring, European Space Agency/ESRIN, Frascati, Italy,
2627 June
1995.
Bhavsar, P. 1984. `Review of remote sensing applications in hydrology and water resources management in India',
Adv. Space Res.,
4(11), 193200.
Biasutti, R. and Lombardi, A. 1995. ÈARTHWATCH: a service provided by EURIMAGE the ood case', Proceedings,
First ERS
Thematic Working Group Meeting on Flood Monitoring, European Space Agency/ESRIN, Frascati, Italy, 2627 June
1995.
Birkett, C. M. 1994. `Radar altimetry: a new concept in monitoring lake level changes', EOS Trans., AGU, 75(24),
273275.
Blasco, F., Bellan M. F., and Chaudhury, M. U. 1992. Èstimating the extent of oods in Bangladesh using SPOT data',
Remote Sens.
Environ., 39, 167178.
Blyth, K. 1995. `FLOODNET a telenetwork for acquisition, processing, and dissemination of Earth Observation data
for
monitoring and emergency management of oods', Proceedings, First ERS Thematic Working Group Meeting on
Flood Monitoring,
European Space Agency/ESRIN, Frascati, Italy, 2627 June 1995.
Blyth, K. and Biggin, D. S. 1993. `Monitoring oodwater inundation with ERS-1 SAR', Earth Obs. Quart., 42, 68.
1436
L. C. SMITH
Brakenridge, G. R., Knox, J. C., Paylor, E. D., and Magilligan, F. J. 1994. `Radar remote sensing aids study of the great
ood of 1993',
EOS Trans., AGU, 75(45), 521527.
Brooner, W. G. and Binaghi, C. M. V. 1983. `Landsat monitoring of temporal hydrological variations on the Pilcomayo
River,
19721981', Proceedings, 17th International Symposium on Remote Sensing of the Environment., Ann Arbor, MI,
pp. 399407.
Chidley, T. R. E. and Drayton, R. S. 1986. `The use of SPOT-simulated imagery in hydrological mapping', Int. J.
Remote Sens., 7(6),
791799.
Choudhury, B. J. 1989. `Monitoring global land surface using Nimbus-7 37 GHz data: theory and examples', Int. J.
Remote Sens.,
10(10), 15791605.
Choudhury, B. J. 1991. `Passive microwave remote sensing contribution to hydrological variables', Surv. Geophys.,
12, 6384.
Cihlar, J., Pultz, T. J., and Gray, A. L. 1992. `Change detection with synthetic aperture radar', Int. J. Remote Sens.,
13(3), 401414.
Deutsch, M. 1976. Òptical processing of ERTS data for determining extent of the 1973 Mississippi River ood', U.S.
Geol. Surv. Prof.
Pap. 929, ERTS-1, a New Window on Our Planet, pp. 209222.
Deutsch, M. and Ruggles, F. 1974. Òptical data processing and projected applications of the ERTS-1 imagery
covering the 1973
Mississippi River Valley oods', Wat. Resour. Bull., 10(5), 10231039.
Deutsch, M., Ruggles, F. H., Guss, P., and Yost, E. 1973. `Mapping of the 1973 Mississippi river oods from the Earth
Resources
Technology Satellite', Proceedings, International Symposium on Remote Sensing and Water Resources Management,
American Water
Resources Association, Proceedings no. 17, Burlington, Ontario, pp. 3955.
Dey, B., Moore, H., and Gregory, A. F. 1977. `The use of satellite imagery for monitoring ice break-up along the
Mackenzie River,
N.W.T', Arctic, 30(4), 234242.
Durden, S. L., Van Zyl, J. J., and Zebker, H. A. 1989. `Modeling and observation of the radar polarization signature of
forested areas',
IEEE Trans. Geosci. Remote Sens., 27(3), 290301.
ESA/ESRIN, 1995. First ERS thematic Working Group Meeting on Flood Monitoring, Frascati, Italy, 2627 June 1995.
Evans, D. L., Farr, T. G., Ford, J. P., Thompson, T. W., and Werner, C. L. 1986. `Multipolarization radar images for
geologic
mapping and vegetation discrimination', IEEE Trans. Geosci. Remote Sens., GE-24(2), 246257.
Ferguson, H. L., Deutsch, M., and Kruus, J. 1980. Àpplications to oods of remote sensing from satellites', Adv.
Space Res., 9,
195206.
France, M. J. and Hedges, P. D. 1986. À hydrological comparison of Landsat TM, Landsat MSS, and black and white
aerial
photography', in Damen, Smit, and Verstappen (Eds), Proceedings, 7th International Symposium, ISPRS (Internat.
Soc. Photogram.
and Remote Sens.) Commission VII, Enschede, 2529 August 1986, pp. 717720.
Gale, S. J. and Bainbridge, S. 1990. `The oods in eastern Australia', Nature, 345, 767.
Giddings, L. and Choudhury, B. J. 1989. Òbservation of hydrological features with Nimbus-7 37 GHz data, applied to
South
America', Int. J. Remote Sens., 10(10), 16731686.
Gupta, R. P. and Banerji, S. 1985. `Monitoring of reservoir volume using Landsat data', J. Hydrol., 77, 159170.
Gupta, R. P. and Bodechtel, J. 1982. `Geotechnical applications of Landsat image analysis of Bhakra dam reservoir,
India', Remote
Sens. Environ., 12, 313.
Guzkowska, M. A. J., Rapley, C. G., Ridley, J. K., Cudlip, W., Birkett, C. M., and Scott, R. F. 1990. `Developments in
inland water
and land altimetry', European Space Agency Contract Report 7839/88/F/FL.
Hallberg, G. R., Hoyer B. E., and Rango, A. 1973. Àpplication of ERTS-1 imagery to ood inundation mapping', NASA
Special Pub.
No. 327, Symposium on signicant results obtained from the Earth Resources Satellite-1, Vol. 1, Technical
presentations, section A,
pp. 745753.
Hamilton, S. K., Sipple, S. J., and Melack, J. M. 1996. Ìnundation patterns in the Pantanal wetland of South America
determined
from passive microwave remote sensing.' Arch. Hydrobiol., 137, 115141.
Harris, J. and Digby-Arbus, S. 1986. `The detection of wetlands on radar imagery', Proceedings, 10th Canadian
Remote Sensing
Symposiums Edmonton, Alberta, 58 May, 1986. Canadian Aeronautics and Space Institute, Ottawa, pp. 529543.
Hess, L. L. and Melack, J. M. 1994. `Mapping wetland hydrology and vegetation with synthetic aperture radar', Int. J.
Ecol. Environ.
Sci., 20, 197205.
Hess, L. L., Melack, J. M., and Simonett, D. S. 1990. `Radar detection of ooding beneath the forest canopy: a
review', Int. J. Remote
Sens., 11(7), 13131325.
Hess, L. L., Melack, J. M., and Davis, F. W. 1995. `Mapping of oodplain inundation with multi-frequency polarimetric
SAR: use of a
tree-based model', Proceedings of the 1994 Int. Geosci. and Remote Sens. Symp (IGARSS '94), pp. 10721073.
Hess, L. L., Melack, J. M., Filoso, S., and Wang, Y. 1995. `Delineation of inundated area and vegetation along the
Amazon oodplain
with the SIR-C synthetic aperture radar', IEEE Trans. Geosci. Remote Sens., 33(4), 895903.
Hockey, B., Richards, T., and Osmaston, H. 1990. `Prospects for operational remote sensing of surface water',
Remote Sens. Rev., 4(2),
265283.
Hoer, R. M., Mueller, P. W., and Lozano-Garcia, D. F. 1985. `Multiple incidence angle shuttle imaging radar data for
discriminating
forest cover types', Proceedings, ASCM-ASPRS Fall Technical Meeting, Indianapolis, 813 September 1985, pp.
476485.
Hollyday, E. F. 1976. Ìmproving estimates of streamow characteristics by using Landsat-1 imagery', J. Res. US
Geol. Surv., 4(5),
517531.
Imho, M. and Gesch, D. 1988. `The derivation of sub-canopy surface terrain models of coastal forests using
synthetic aperture radar',
Proceedings, 1988 Int. Geosci. and Remote Sens. Symp (IGARSS '88), pp. 613617.
Imho, M. L., Vermillion, C., Story, M. H., Choudhury, A. M., Gafoor, A., and Polcyn, F. 1987. `Monsoon ood boundary
delineation and damage assessment using space borne imaging radar and Landsat data', Photogramm. Eng.
Remote Sens., 53(4),
405413.
Jensen, J. R., Hodgson, M. E., Christensen, E., Mackey, H. E., Jr., Tinney, L. R., and Sharitz, R. 1986. `Remote sensing
inland
wetlands: a multispectral approach', Photogramm. Eng. Remote Sens., 52, 87100.
1437
Koblinsky, C. J., Clarke, R. T., Brenner, A. C., and Frey, H. 1993. `Measurement of river level variations with satellite
altimetry',
Wat. Resour. Res., 29(6), 18391848.
Krohn, M. D., Milton, N. M., and Segal, D. B. 1983. `Seasat synthetic aperture radar (SAR) response to lowland
vegetation types in
eastern Maryland and Virginia', J. Geophys. Res., 88(C3), 19371952.
Kruus, J., Deutsch, M., Hansen, P. L., and Ferguson, H. L. 1981. `Flood applications of satellite imagery' in Satellite
Hydrology,
Proceedings, Fifth Ann. William T. Pecora Memorial Symp. on Remote Sens., 1979, Amer. Wat. Resour. Assoc. Tech.
Pub. No. TPS
811, pp. 292301.
Le Traon, P. Y., Gaspar, P., Ogor, F., and Dorandeu, J. 1995. `Satellites work in tandem to improve accuracy of data',
EOS Trans.,
AGU, 76(39), 385389.
Lowry, R. T., Langham, E. J., and Mudry, N. 1981. À preliminary analysis of SAR mapping of the Manitoba Flood,
May 1979',
in Satellite Hydrology, Proceedings, Fifth Ann. William T. Pecora Memorial Symp. on Remote Sens., 1979, Amer. Wat.
Resour.
Assoc. Tech. Pub. No. TPS81-1, pp. 316323.
Macdonald, H. C., Waite, W. P., and Demarcke, J. S. 1980. Ùse of Seasat satellite radar imagery for the detection of
standing
water beneath forest vegetation', Tech. Papers of the Amer. Soc. of Photogramm. Fall Tech. Meeting, Niagara Falls,
New York,
710 October, 1980, RS3B1RS3Bl3.
McGinnis, D. F. and Rango, A. 1975. Èarth Resources Satellite systems for ood monitoring', Geophys. Res. Let.,
2(4), 132135.
Melack, J. M., Hess, L. L., and Sippel, S. 1994. `Remote sensing of Lakes and Floodplains in the Amazon Basin',
Remote Sens. Rev.,
10, 127142.
Mertes, L. A. K., Smith, M. O., and Adams, J. B. 1993. Èstimating suspended sediment concentrations in surface
waters of the
Amazon River wetlands from Landsat images', Remote Sens. Environ., 43, 281301.
Miller, S. T. 1986. `The quantication of oodplain inundation by the use of Landsat and metric camera information,
Belize, Central
America', Proceedings, 7th International Symposium, ISPRS (Internat. Society Photogrammetry and Remote
Sensing) Commission
VII, Enschede August 1986, pp. 733738.
Moore, G. K. and North, G. W. 1974. `Flood inundation in the southeastern United States from aircraft and satellite
imagery',
Wat. Resour. Bull., 10(5), 10821096.
Morris, C. S. and Gill, S. K. 1994. `Variation of Great Lakes water levels derived from Geosat altimetry', Wat. Resour.
Res., 30(4),
10091017.
Morrison, R. B. and Cooley, M. E. 1973. Àssessment of ood damage in Arizona by means of ERTS-1 imagery',
Proceedings,
Symposium on Signicant Results Obtained from the Earth Resources Satellite-1, Vol. 1, New Carrollton, Maryland,
pp. 755760.
Morrison, R. B. and White, P. G. 1976. `Monitoring ood inundation', U.S. Geol. Surv. Prof. Pap. 929, ERTS-1, A New
Window on
Our Planet, pp. 196208.
Mouchot, M. C., Alfoldi, T., DeLisle, D., and McCullough, G. 1991. `Monitoring the water bodies of the Mackenzie
Delta by remote
sensing methods', Arctic, 44 (suppl. 1), 2128.
Muller, E., Decamps, H., and Dobson, M. K. 1993. `Contribution of space remote sensing to river studies',
Freshwater Biol., 29,
301312.
Murray, A. B. and Paola, C. 1994 À cellular model of braided rivers', Nature, 371, 5457.
Nagarajan, R., Marathe, G. T., and Collins, W. G. 1993. Ìdentication of ood prone regions of Rapti river using
temporal remotelysensed data', Int. J. Remote Sens., 14(7), 12971303.
Ormsby, J. P., Blanchard, B. J., and Blanchard, A. J. 1985. `Detection of lowland ooding using active microwave
systems',
Photogramm. Eng. Remote Sens., 51(3), 317328.
Philipson, W. R. and Hafker, W. R. 1981. `Manual versus digital Landsat analysis for delineating river ooding',
Photogramm. Eng.
Remote Sens., 47(9), 13511356.
Pope, K. O., Shener, E. J., Linthicum, K. J., Bailey, C. L., Logan, T. M., Kasichke, E. S., Birney, K., Njogu, A. R., and
Roberts, C. R.
1992. Ìdentication of Central Kenyan Rift Valley Fever virus vector habitats with Landsat TM and evaluation of
their ooding
status with airbourne imaging radar', Remote Sens. Environ., 40, 185196.
Pope, K. O., Rey-Benayas, J. M., and Paris, J. F. 1994. `Radar remote sensing of forest and wetland ecosystems in the
Central
American tropics', Remote Sen. Envion., 48, 205219.
Rango, A. and Anderson, A. T. 1974. `Flood hazard studies in the Mississippi River basin using remote sensing', Wat.
Resour. Bull.,
10(5), 10601081.
Rango, A. and Salomonson, V. V. 1973. `Repetitive ERTS-1 observations of surface water variability along rivers and
low-lying areas',
Amer. Wat. Resour. Assoc. Proc. no. 17, Remote Sensing and Water Resources Management, pp. 191199.
Rango, A. and Salomonson, V. V. 1974. `Regional ood mapping from space', Wat. Resour. Res., 10(3), 473484.
Rango, A. and Salomonson, V. V. 1977. `The utility of short wavelength (51 mm) remote sensing techniques for the
monitoring and
assessment of hydrological parameters' Proceedings, 11th Int. Symp. on Remote Sens. of Environ., Ann Arbor, MI,
2529 April 1977,
pp. 5564.
Rango, A., Foster, J., and Salomonson, V. V. 1975. Èxtraction and utilization of space acquired physiographic data
for water
resources development', Wat. Resour. Bull, 11(6), 12451255.
Rapley, C. G., Baker, S. G., Birkett, C. M., Laxon, S. W., Ridley, J. K., Scott, R. F., and Strawbridge, F. 1992. ÈRS-1
altimetry of inland
water and land', Proceedings, First ERS-1 Symp., ESA SP-359 (March, 1993), Cannes, France, 46 November, 1992,
pp. 539542.
Rasid, H. and Pramanik, M. A. H. 1993. Àreal extent of the 1988 ood in Bangladesh: how much did the satellite
imagery show?',
Natural Hazards, 8, 189200.
Rast, M., Jaskolla, F., and Arnason, K. 1991. `Comparative digital analysis of Seasat-SAR and Landsat-TM data for
Iceland', Int. J.
Remote Sens., 12(3), 527544.

Richards, J. A., Sun, G.-Q., and Simonett, D. S. 1987a. `L-band radar backscatter modeling of forest stands', IEEE
Trans. Geosci.
Remote Sens., 25, 487498.
1438
L. C. SMITH
Richards, J. A., Woodgate, P. W., and Skidmore, A. K. 1987b. Àn explanation of enhanced radar backscattering from
ooded
forests', Int. J. Remote Sens., 8(7), 10931100.
Richey, J. E., Mertes, L. A. K., Dunne, T., Victoria, R. L., Forsberg, B. R., Tancredi, A., and Oliveira, E. 1989. ` Sources
and routing
of the Amazon River ood wave', Global Biogeochem. Cycles, 3, 1991204.
Richey, J. E., Victoria, R. L., Salati, E., and Forsberg, B. R. 1991. `The biogeochemistry of a major river system: the
Amazon case
study', in Degens, E. T., Kempe, S., and Richey, J. E. (eds), Biogeochemistry of Major World Rivers, John Wiley &
Sons, Ltd,
Chichester. pp. 5774.
Ringrose, S., Matheson, W., and Boyle, T. 1988. `Dierentiation of ecological zones in the Okavango Delta,
Botswana, by classication and contextural analysis of Landsat MSS data', Photogramm. Eng. Remote Sens., 54(5), 601608.
Salomonson, V. V. (Ed.), 1983. `Water resources assessment', in Manual of Remote Sensing, Chap. 29, Amer. Soc. of
Photogramm.
pp. 14971570.
Salomonson, V. V. and Rango, A. 1974. ÈRTS-1 applications in hydrology and water resources', J. Amer. Wat. Works
Assoc., 66(3),
168172.
Schmugge, T. 1987. `Remote sensing applications in hydrology', Rev. Geophys., 25(2), 148152.
Schmugge, T. J., O'Neill, P. E., and Wang, J. R. 1986. `Passive microwave soil moisture research', IEEE Trans. Geosci.
Remote Sens.,
24, 1222.
Schultz, G. A. 1988. `Remote sensing in hydrology', J. Hydrol., 100, 239265.
Sipple, S. J., Hamilton, S. K., and Melack, J. M. 1992. Ìnundation area and morphometry of lakes on the Amazon
River oodplain,
Brazil', Arch. Hydrobiol., 123(4), 385400.
Sipple, S. J., Hamilton, S. K., Melack, J. M., and Choudhury, B. J. 1994. `Determination of inundation area in the
Amazon River
oodplain using the SMMR 37 GHz polarization dierence', Remote Sens. Environ., 48, 7076.
Smith, L., Isacks, B. L., Forster, R. R., Bloom, A. L., and Preuss, I. 1995. Èstimation of discharge from braided glacial
rivers using
ERS-1 SAR: rst results', Wat. Resour. Res., 31, 13251329.
Smith, L. C., Isacks, B. L., Bloom, A. L., and Murray, A. B. 1996. Èstimation of discharge from three braided rivers
using synthetic
aperture radar (SAR) satellite imagery: potential for application to ungaged basins', Water Resour. Res., 32(7),
20212034.
Sollers, S. C., Rango, A., and Henninger, D. L. 1978. `Selecting reconnaissance strategies for oodplain surveys',
Wat. Resour. Bull.,
14(2), 359373.
Solomon, S. I. 1992. `Methodological considerations for use of ERS-1 imagery for the delineation of river networks in
tropical
forest areas', Proceedings, First ERS-1 Symp., ESA SP-359 (March, 1993), Cannes, France, 46 November, 1992, pp.
595600.
Usachev, V. F. 1983. Èvaluation of ood plain inundations by remote sensing methods', Proceedings of the Hamburg
Symposium,
IAHS Publ., 145, 475482.
Van den Brink, J. W. 1986. À study with NOAA-7 AVHRR imagery in monitoring ephemeral streams in the lower
catchment area of
the Tana River, Kenya', Damen, Smit, and Verstappen (Eds.), Proceedings, 7th International Symposium, ISPRS
(Internat. Soc.
Photogram. and Remote Sens.) Commission VII, Enschede, 2529 August 1986, pp. 783785.
Van Es, E., Bomez, H., and Soeters, R. 1975. Àn inundation study of the lower Magdalena-Cauca River Basin', NASA
Tech. Memo
X-58168, Vol. 1-D, pp. 22952297.
Vila da Silva, J. S. and Kux, H. J. H. 1992. `Thematic mapper and GIS data integration to evaluate the ooding
dynamics within the
Panatal, Mato Grosso do Sul State, Brazil', Proceedings, 1992 Int. Geosci. Remote Sens. Symp. (IGARSS '92), pp.
14781480.
Waite, W. P., MacDonald, H. C., Kaupp, V. H., and Demarcke, J. S. 1981. `Wetland mapping with imaging radar',
Proceedings, 1981
Int. Geosci. and Remote Sens. Symp. (IGARSS '81), pp. 794799.
Watson, J. P. 1991. À visual interpretation of a Landsat mosaic of the Okavango Delta and surrounding area',
Remote Sens. Environ.,
35, 19.
Wedler, E. and Kessler, R. 1981. Ìnterpretation of vegetative cover in wetlands using four-channel SAR imagery',
Tech. Papers of the
Amer. Soc. of Photogramm., ASP-ASCM Convention, Washington, D.C., 2227 February, 1981, pp. 111124.
Whitelaw, A., Howes, S., Fletcher, P. and Rast, M. 1995. `Remote sensing applications in hydrological modelling',
SPIE (Soc. of
Photo-Optical Instrumentation Engineers) vol. 2314, Multispectral and Microwave Sensing of Forestry, Hydrology,
and Natural
Resources, Rome, Italy, 2630 September, 1994, pp. 618627.
Wiesnet, D. R. and Deutsch, M. 1987. `Flood monitoring in South America from the Landsat NOAA and NIMBUS
Satellites',
Adv. Space Res., 7(3), 7784.
Wu, S. T. and Sader, S. A. 1987. `Multipolarization SAR data for surface feature delineation and forest vegetation
characterization',
IEEE Trans. Geosci. Remote Sens., GE-25(1), 6776.
Xia, Liu, Shulin, Z., and Xianglian, L. 1983. `The application of Landsat imagery in the surveying of water resources
of Dongting
Lake', Proceedings of the Hamburg Symposium, IAHS Publ., 145, 483489.
1439
Satellite remote sensing of earthquake, volcano, flood,

landslide and coastal inundation hazards
David M. TralliT, Ronald G. Blom, Victor Zlotnicki, Andrea Donnellan,
Diane L. Evans
Earth Science and Technology Directorate, Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak
Grove Drive,
Pasadena, CA 91109, USA
Received 7 June 2004; accepted 2 February 2005
Available online 1 April 2005
Abstract
Satellite remote sensing is providing a systematic, synoptic framework for advancing scientific
knowledge of the Earth as a
complex system of geophysical phenomena that, directly and through interacting processes, often lead
to natural hazards.
Improved and integrated measurements along with numerical modeling are enabling a greater
understanding of where and when
a particular hazard event is most likely to occur and result in significant socioeconomic impact.
Geospatial information products
derived from this research increasingly are addressing the operational requirements of decision
support systems used by policy
makers, emergency managers and responders from international and federal to regional, state and
local jurisdictions. This forms
the basis for comprehensive risk assessments and better-informed mitigation planning, disaster
assessment and response
prioritization. Space-based geodetic measurements of the solid Earth with the Global Positioning
System, for example,
combined with ground-based seismological measurements, are yielding the principal data for modeling
lithospheric processes
and for accurately estimating the distribution of potentially damaging strong ground motions which is
critical for earthquake
engineering applications. Moreover, integrated with interferometric synthetic aperture radar, these
measurements provide
spatially continuous observations of deformation with sub-centimeter accuracy. Seismic and in situ
monitoring, geodetic
measurements, high-resolution digital elevation models (e.g. from InSAR, Lidar and digital
photogrammetry) and imaging
spectroscopy (e.g. using ASTER, MODIS and Hyperion) are contributing significantly to volcanic hazard
risk assessment, with
the potential to aid land use planning in developing countries where the impact of volcanic hazards to
populations and lifelines
is continually increasing. Remotely sensed data play an integral role in reconstructing the recent
history of the land surface and
in predicting hazards due to flood and landslide events. Satellite data are addressing diverse
observational requirements that are
imposed by the need for surface, subsurface and hydrologic characterization, including the delineation
of flood and landslide
zones for risk assessments. Short- and long-term sea-level change and the impact of oceanatmosphere processes on the coastal
land environment, through flooding, erosion and storm surge for example, define further requirements
for hazard monitoring
and mitigation planning. The continued development and application of a broad spectrum of satellite
remote sensing systems
and attendant data management infrastructure will contribute needed baseline and time series data,
as part of an integrated
global observation strategy that includes airborne and in situ measurements of the solid Earth. Multihazard modeling
0924-2716/$ - see front matter D 2005 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS).
Published by Elsevier B.V.
All rights reserved.
doi:10.1016/j.isprsjprs.2005.02.002
T Corresponding author. Tel.: +1 818 354 1835.
E-mail address: david.m.tralli@jpl.nasa.gov (D.M. Tralli).
ISPRS Journal of Photogrammetry & Remote Sensing 59 (2005) 185 198
www.elsevier.com/locate/isprsjprs
capabilities, in turn, will result in more accurate forecasting and visualizations for improving the
decision support tools and
systems used by the international disaster management community.
D 2005 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by
Elsevier B.V. All rights
reserved.
Keywords: Satellite remote sensing; Imaging spectroscopy; Interferometric synthetic aperture radar (InSAR); Global
positioning system (GPS);
risk mitigation; Disaster response; Decision support system (DSS); Natural hazards
1. Introduction
The last two decades have witnessed the increasing
use of remote sensing for understanding the geophysical
phenomena underlying natural hazards. The
scientific knowledge gained along with the ability to
disseminate timely geospatial information that can be
integrated with demographic and socioeconomic data
are contributing to comprehensive risk mitigation
planning and improved disaster response. As the
discipline of Earth science began to recognize the
interactions between the hydrosphere, atmosphere,
biosphere and solid Earth as a complex system, satellite
remote sensing uniquely has provided the synoptic
perspective for in situ measurements on local spatial
scales and variable temporal resolution. Observations
from Earth orbiting satellites are complementary to
local and regional airborne observations, and to traditional
in situ field measurements and ground-based
sensor networks in seismology, volcanology, geomorphology
and hydrology. The contributions of
satellite remote sensing to solid Earth science, ranging
from high-resolution topography (using e.g. Interferometric
SAR, Lidar and digital photogrammetry) and
geodesy to passive hyperspectral (such as ASTER,
MODIS and Hyperion) and active microwave imaging,
have transformed the discipline. This transformation

has helped to define a rapidly growing field of applied
research that increasingly will provide geospatial
information products addressing the operational
requirements of multi-hazard decision support tools
and systems. Policy makers, emergency managers and
responders from international and federal to state,
regional and local jurisdictions use these tools and
systems to generate scenarios, devise mitigation plans
and implement effective response measures.
Satellite remote sensing, in particular, is providing
a systematic framework for scientific knowledge of
the solid Earth that is the basis for better-informed
decision-making. The key to understanding the
Earths dynamics and system complexity is to
integrate observations at local, regional and global
scales, over a broad portion of the electromagnetic
spectrum with increasingly refined spectral resolution,
spatial resolution and over time scales that encompass
phenomenological lifecycles with requisite sampling
frequency. Advances in computational science and
numerical simulations are allowing the study of
correlated systems, recognition of subtle patterns in
large data volumes, and are speeding up the time
necessary to study long-term processes using observational
data for constraints and validation (Donnellan
et al., 2004). Integrating remotely sensed data into
predictive models requires measurements at resolutions
substantially superior to those made in the past
when the observational systems and the discipline of
natural hazards research were less mature than they
are today. Furthermore, assimilation of data and
model outputs into decision support systems must
meet operational requirements for accuracy, spatial
coverage and timeliness in order to have positive
impact on disaster risk management.
In the following sections, satellite remote sensing
systems and integrated observational and modeling
approaches that are being used to study and assess risks
due to earthquakes, volcanoes, floods, landslides and
coastal inundation are reviewed briefly. Only hazards
manifested through solid Earth processes are the focus
herein, although significant advances in the application
of satellite remote sensing to severe weather events and
wildfires, for example, indeed are being made but left
for review elsewhere. This paper is not intended as a
comprehensive treatment of the application of remote
sensing to the selected natural hazards but rather as a
recognition of the contributions of satellite remote
sensing to understanding underlying phenomena and
providing critical information for decision support by
emergency managers and the disaster response com186 D.M. Tralli et al. / ISPRS Journal of Photogrammetry & Remote Sensing 59 (2005) 185198
munity. As such, it is a summary of work reported by

the solid Earth science community in natural hazards
research and applications (NRC, 2003; SESWG

Report, 2002), with reference to the latest results in
the literature and to seminal planning documents
developed by the Subcommittee on Disaster Reduction
(SDR), National Science and Technology Council of
the U.S. Committee on the Environment and Natural
Resources (CENR) (CENR, 2003), the Geohazards
Theme Report of the Integrated Global Observing
Strategy (IGOS) international partnership (IGOS Geohazards
Theme Report, 2004), and the Disaster
Management Support Group (DMSG) Report of the
Committee on Earth Observation Satellites (CEOS)
within IGOS (CEOS, 2003). The intent is to underscore
the view of the Earth as a complex system of forcings
and responses, the increasingly recognized potential of
satellite remote sensing to assess the consequences of
resultant changes, and the need for advanced observation
platforms and monitoring systems for sustained
measurements of the solid Earth relevant to disaster
management.
2. Earthquake hazards
Deformation at the Earths surface, predominantly
adjacent to tectonic plate boundaries, is the manifestation
of forces acting deep within its interior. Geodetic
and seismological measurements provide the
principal data for understanding mantle dynamics,
lithospheric processes and crustal response, and for
improving numerical modeling for forecasting catastrophic
events such as earthquakes and volcanic
eruptions. Major advances have been made in earthquake
research and risk mitigation. However, the
nature of significant seismic eventswith greater and
more widespread occurrence than volcanic hazards
and resultant loss impact that is much higher than that
of even more widespread landslide and subsidence
hazards (IGOS Geohazards Theme Report, 2004)
presents extant research needs. Research requirements
in earthquake science that will contribute to better
seismic risk management and forecasting on a global
basis include documenting the location, slip rates and
earthquake history of dangerous faults; understanding
the kinematics and dynamics of active fault systems
on interseismic time scales and applying this understanding
in constructing probabilities of occurrence;
characterizing the three-dimensional material properties
of fault systems, their response to deformation and
the physics of earthquake nucleation, propagation and
arrest; and predicting strong ground motions and nonlinear
surface layer response, including fault rupture,
landsliding and liquefaction (NRC, 2003).
Integrated ground measurements and satellite
remote sensing can help meet these various requirements
for baseline and time-series data. The Global
Positioning System (GPS), for example, used for
navigation and positioning in civilian and military
applications, provides the millimeter-level differential

accuracy that is used by regional ground deformation
networks to monitor interseismic ground deformation
and co-seismic displacements. GPS monitoring networks
are leading to better definition of off-fault
surface deformation rates; timely detection of diagnostic
changes in the fault environment; and constraints
on the extent of surficial fault creep and its significance
to potentially significant earthquakes. Accurate estimates
of the distribution of potentially damaging
ground motions from such earthquakes enable ground
motion modeling, structural design planning and risk
assessment for loss estimation. Earthquake forecasting
efforts are related intimately to generating this
increased understanding of the fundamental dynamics
of major faults, with fault segment definition leading to
a better description of the expected details of earthquake
faulting and rupturing. Moreover, GPS and
modern digital seismic data can be combined with
satellite remote sensing, such as interferometric synthetic
aperture radar (InSAR) to provide spatially
continuous deformation with sub-centimeter accuracy.
Fig. 1 shows a comparison of an InSAR-derived
deformation time series with GPS network time series
data, providing insight into the spatial and temporal
relationship between continuous GPS data and InSAR.
The availability of both temporal and spatial deformation
data allows for a greater level of understanding of
the dynamics of the area. InSAR techniques are an
integral component of an Earth observation capability
for seismic risk mitigation and response (CEOS, 2003).
Delineation of seismic source zones requires
understanding the geology, tectonics, and paleoseismic
and neotectonic features of the subject region.
Investigations of surface deformation, plate-boundary
interactions, frictional properties of faults, and
D.M. Tralli et al. / ISPRS Journal of Photogrammetry & Remote Sensing 59 (2005) 185198 187
mechanical properties of the Earths crust and lithosphere

help to determine what controls the spatial and
temporal characteristics of earthquakes. For example,
surface displacements due to the 2003 Bam, Iran,
earthquake were mapped using ENVISAT radar data
to reveal that over 2 m of slip occurred at depth on a
blind strike-slip fault, where no morphological features
were present (Talebian et al., 2004). Space-based
observations of the entire earthquake cycle, including
the aseismic accumulation of strain between events
(Fielding et al., 2004) are critical for learning about
the phenomenology and for forecasting potentially
hazardous earthquakes. Remote and in situ data that
support attendant scientific and engineering models
are necessary in order to understand the sourcerupture
process, fault plane geometry and thus infer
the resultant patterns of damage (see Lohman et al.,
2002). This is critical for effective risk management.
While there are in operation notably successful

dense GPS geodetic networks in regions prone to
potentially catastrophic seismic and/or volcanic
events, such as southern California (Hudnut et al.,
2002) and particularly Japan (Shimada and Bock,
1992), economic constraints limit the widespread
global deployment of these networks. Furthermore, a
lack of standard formats and established archives, plus
limited accessibility for the different kinds of deformation
data are major challenges for the integration of
local GPS data globally, and the integration of GPS
data with older, heritage deformation data sets (IGOS
Geohazards Theme Report, 2004). The ability of
geodetic data to resolve variations in slip patterns also
Fig. 1. (Left) InSAR time series inversion map of the radar line-of-sight deformation average velocity, overlying the
multi-look SAR amplitude
image (gray scale). Small black squares mark Southern California Integrated GPS Network (SCIGN) GPS site
locations. (Right) InSAR time
series (black triangles) at selected points. Plots compare the InSAR time series to SCIGN GPS (red*) time series for
indicated sites (produced
from European Space Agency remote sensing data, ERS-1 and ERS-2. Figure courtesy P. Lundgren, NASA/JPL; see
also Lanari et al., 2004).
188 D.M. Tralli et al. / ISPRS Journal of Photogrammetry & Remote Sensing 59 (2005) 185198
diminishes greatly with slip depth. Prediction of

strong motion velocities from geodetic data alone
offers limited spectral response, thus being a poor
replacement for actual strong motion recordings that
are critical to earthquake engineering. In either case,
the spectrum of ground displacement signatures
measurable by GPS and seismic networks is sampled
discretely.
Satellite remote sensing systems not only offer
spatially continuous information of the tectonic landscape
but also contribute to the understanding of
specific fault systems. Combined with ground network
data, remote sensing enables a better understanding
of displacements, and validation of slip
models that are cast in a regional setting of tectonic
strain (Cakir et al., 2003) and help constrain source
characterization (e.g. Lundgren and Stramondo,
2002). Satellite remote sensing observations are
providing insights into how stress is transferred
between fault systems from depth and to the surface,
how much energy is released by earthquakes and
other modes of deformation (Argus et al., in press)
and how faults fail mechanically.
In certain cases, earthquakes can produce global
gravity perturbations that are detectable through
analysis of satellite data from missions dedicated to
gravity field determination, such as the Challenging
Mini-satellite Payload (CHAMP), Gravity Recovery
and Climate Experiment (GRACE) and the Gravity
Field and Steady-State Ocean Circulation Explorer
(GOCE). Coseismic gravity and geoid changes differ
from other, larger and more coherent high-frequency
variations such as Earth tides. The coseismic effects
of great earthquakes such as the 1960 Chile, 1964
and 2002 Alaska, and 2003 Hokkaido events cause

global gravitational field changes that are sufficiently
large to be detected by GRACE, for example, based
on degree variance analysis using spherical harmonic
representation of dislocation theory (Sun and Okubo,
2004) and a normal mode technique comparing the
degree amplitude spectra of select earthquakes with
GRACE sensitivity (Gross and Chao, 2001). Such
gravity data also can provide important constraints
on the interpretation of seismological data (Tondi et
al., 2003), as for testing the shorter wavelength
features of three-dimensional tomographic models
based on the inversion of short period seismic waves.
Many parts of the lithosphere, where heterogeneity is
high, particularly near plate boundaries (Bowin,
1991), are poorly sampled by short period waves
and global gravity data from abovementioned satellite
systems are spatially continuous and uniquely
complementary.
Finally, observations of land cover, land use and of
the built environment, structures and lifelines specifically
are a critical component of risk management
and loss estimation methodologies. Integration of
high-resolution satellite remote sensing with InSAR
and airborne Light Detection and Ranging (LiDAR) is
being researched as a means to image, classify and
inventory the built environment through the extraction
of land cover and digital terrain models (Gamba and
Houshmand, 2002). This contributes to vulnerability
assessments and to rapid post-disaster damage assessment
(Rejaie and Shinozuka, 2004), through integration
with demographic data, infrastructure and
building stock databases in a geographic information
system (GIS). Rapid damage assessment is critical for
effective allocation of disaster response and relief
resources, including federal insurance assistance.
Available satellite remote sensing systems, from civil
space agencies and commercial imaging sources such
as IKONOS, OrbView and QuickBird, are witnessing
increased utilization in disaster management research
and operational domains.
3. Volcanic hazards
Subaerial volcanic eruptions occur often after long
intervals of dormancy and thus opportunities for direct
geophysical measurements are intermittent and sporadic.
While there are numerous indicators of subaerial
volcanic activity, in addition to surface deformation
and seismicitysuch as thermal emissions, and
changes in gravity, emission of gasses plus ash and
clastic eruptionslittle is known about the global
levels of these activities and how these phenomena
are related. Furthermore, the physical mechanisms
that cause surface deformation and those that control
the rates and styles of eruptions are poorly understood.
The ability to predict or otherwise forecast the
timing, magnitude, and style of volcanic eruptions on

the Earths land surface is an important yet generally
unmet objective in volcanic hazards assessment and
mitigation planning (SESWG Report, 2002).
An observation strategy that incorporates seismic

and in situ monitoring, geodetic measurements, highresolution
topography [i.e. Digital Terrain Elevation
Data (DTED) Level 1) and hyperspectral imaging can
contribute significantly to volcanic hazard risk assessment,
mitigation and response. Observational requirements
for volcanic hazards include the threedimensional
spatial distribution of seismicity; the
characteristic deformation of the volcanic edifice
using geodetic and gravimetric techniques that include
identification of faults, fractures, landslides and flank
instabilities, rift systems and calderas; characterization
of gas and ash emissions by species (SO2, CO2) and
flux; and characterizing and monitoring of thermal
features, their nature, location, temperature and
possible heat flux (CEOS, 2003).
Fundamental to the understanding of eruptive
systems are the identification and characterization of
active volcanoesnamely a comprehensive global
inventory. This would comprise not only geodetic
observations, with InSAR for example (Pritchard and
Simons, 2004), but spectroscopic observations of
debris flows and land surface, as with Hyperionthe
first spaceborne imaging spectrometeronboard the
NASA Earth Observing-1 (EO-1) satellite (Crowley et
al., 2003; Wright and Flynn, 2003), and ash and
emissions (see Fig. 2), as with the NOAA Geostationary
Operational Environmental Satellite (GOES)
(Ellrod et al., 2003) and the Advanced Spaceborne
Thermal Emission and Reflection Radiometer
(ASTER) (Watanabe and Matsuo, 2003; Pieri and
Abrams, 2004). Such measurements and new observational
tools are enabling a rapid growth in the
understanding of volcanic hazards worldwide.
Geodetic observations of volcanoes with GPS and
InSAR are yielding high-resolution digital elevation
models (DEMs) (Lu et al., 2003) and full vector
deformation rate maps that complement traditional
ground-based geodetic techniques. These high-resolution
measurements are required in order to reduce
ambiguities in inferences of magma chamber geometry
from outward structural changes (e.g. Lanari et
al., 2003). Computer models of a variety of flows are
increasingly being use in volcanic hazard assessment
to predict potential areas of devastation (Stevens et al.,
2003). Furthermore, DEMs are being used to predict
lava flow and lahar paths on remote volcanoes, with a
promising level of accuracy over distances of tens of
kilometers (Glaze and Baloga, 2003). The accuracy of
these models depends on the accuracy of the available
topographic data; measurements from the Shuttle

Radar Topography Mission (SRTM) provide invaluable
data, particularly in remote regions where highresolution
topographic data are unavailable (e.g.
Stevens et al., 2003). A monitoring strategy that
includes the use of such predictive flow models and
risk zonation could support land use planning,
particularly in developing countries where the impact
of volcanic hazards to populations and lifelines is
continually increasing.
Remote sensing indeed is defining a new paradigm
for volcanological observations (Pieri and Abrams,
2004). Imaging spectroscopy (or hyperspectral imaging)
in both the solar-reflected (0.42.5 Am) and
thermal portions (35 Am and 812 Am) of the
spectrum, permits the identification, separation, and
measurement of subtle variations reflecting the overlapping
molecular absorption and constituent scattering
signatures of materials present on the Earths
surface. Measurements of surface deposits and composition,
surface temperature, topography and surface
deformation, SO2 and ash detection and tracking (Fig.
2), and modeling are needed to better characterize,
understand and predict the volcanic hazards environment.
Measurements made by the Airborne Visible/
Infrared Imaging Spectrometer (AVIRIS) and by
Hyperion have been used to map subtle changes in
near-surface rock chemistry and, thereby, identify
zones of volcanic-debris-flow susceptibility on the
basis of rock strength inferred from specific mineralogical
indicators of hydrothermal alteration (Crowley
et al., 2003).
Global monitoring, including of remote areas, at
weekly time intervals with spaceborne systems would
enable the requisite sensitivity to low-level but more
nearly continuous processes for assessment of risks in
short-term early warning systems. In the event of an
eruption, shorter time intervals are desired, with
updating several times per day. However, in these
cases, only a spotlight view of a targeted area of the
globe is needed, for example to provide volumetric
estimation of eruptive lava outflow and source
modeling (Lundgren and Rosen, 2003), distinguishing
between surface deformation caused by magma
movement or fluid pressure build-up. A similar
rationale holds for the timing of spectroscopic
measurements that provide sensitivity to heat flux and

gas emissions (e.g. SO2 and CO2) (Prata et al.,
submitted for publication). Proper temporal resolution,
temperature change sensitivity of the order of 0.5
K and accurate measurements of gas emissions, along
with surface deformation maps, may allow the
forecasting of eruptions. For example, the Moderate
Resolution Imaging Spectroradiometer (MODIS)
onboard the NASA Terra and Aura spacecraft are

providing imagery of subaerial volcanoes on Earth
every 2 days (Watson et al., 2004). There is now an
online archive of eruptions going back nearly 5 years,
using MODIS data for global monitoring through
detection of thermal signatures (Wright et al., 2002,
Fig. 2. Mt. Etna is one of the worlds most active volcanoes and has been studied for centuries from the ground. On
November 3, 2002 Mt.
Etnas ash-laden plume was imaged by ASTER. The plume is seen blowing towards the southsoutheast, over the
city and airport of Catania,
Sicily. The previous day, the plume was blowing towards the northwest, and posed no hazard to Catania. The
eruption of Mt. Etna, Europes
most active volcano, began on October 27. The image covers an area of 50.8_76.5 km [image courtesy M. Abrams,
NASA/JPL].
2004). Retrieval algorithms are able to quantify

volcanic ash, ice, sulfates and SO2 using thermal
infrared (812 Am). Identification of outgassed
species near vents and craters provides information
on subsurface activity and processes and may
ultimately assist in forecasting eruptions. Thermal
measurements of land surface temperature, together
with simultaneous measurements of the changing
emissivity, provide additional constraints on magmatic
processes and volcanic activity. Even the
contribution of subaerial active volcanoes to the
Earths energy budget can be estimated (Wright and
Flynn, 2004).
4. Flood and landslide hazards
Floods are among the most devastating natural
hazards in the world, claiming the largest amount of
lives and property damage (CEOS, 2003). Remotely
sensed data play an integral role in reconstructing the
recent history of the land surface and in predicting
hazard events such as floods and landslides, subsidence
events and other ground instabilities. Reconstruction of
past erosion, deformation, and deposition and quantification
of tectonic, climatic, and biologic inputs
including human-induced changesto the evolving
landscape underpin the ability to develop a processbased
understanding of the Earths dynamic surface.
Since land-surface properties change through time,
remote sensing of such changes yields critical
temporal control on landscape evolution. The need
for higher spatial and temporal resolution data is
pinpointed by recognizing that destructive floods or
landslides can be launched by intense, short-lived
storm cells a few kilometers in extent. The height and
width of rivers, as well as rainfall intensity and
amounts, need to be measured hourly during storms.
Hossain and Anagnostou (2004) give an assessment
of the current state of passive microwave and infraredbased
satellite systems for flood prediction. Few data
exist on soil moisture, thickness, and strength, or on
vegetation cover, fire history, or detailed topography.
Synthetic aperture radar (SAR) and Landsat Thematic
Mapper (TM) data have been integrated to provide
information on land cover and the geomorphology of

slopes, to inventory and characterize landslide potential
in high relief areas (Singhroy et al., 1998), and to
monitor post-slide motion and characterize debris size
and distribution (Singhroy, 1995).
Contributions of spaceborne remote sensing to
flood warning, disaster assessment and hazard reduction
will rely on a broad-based program of remotely
sensed and in situ measurements of rainfall, river
heights, soil moisture, with vegetation change providing
critical indices for flood and landslide hazards.
Integration of remote sensing and in situ measurements
is needed, along with hydrologic models benefiting
from improvements in multi-scale observations of
climate and weather, from global to synoptic and
mesoscale to storm scales. For example, the SeaWinds
radar aboard QuikSCAT and MODIS optical data are
processed and combined with a GIS for monitoring
flood propensity and developing weekly surface water
anomaly maps (Fig. 3) that emphasize the sustained
excess moisture receipts most likely to cause river
flooding (see www.dartmouth.edu/~floods). Occasional
(510 years) quantification of soil composition
and thickness would suffice in areas governed by
gradual processes, but more frequent measurements
will be needed in areas affected by such dynamic
events as floods or landslides.
Diverse observational requirements are imposed by
the need for surface, subsurface and hydrologic
characterization, including the delineation of flood
and landslide zones for risk assessments (see Carrasco
et al., 2003) and mitigation planning, and zones prone
to subsidence due to groundwater interactions (see
Buckley et al., 2003). The types of measurements that
are needed to quantify, model, and predict flood
hazards include 1-m DEMs with 5 cm accuracy for
catchment geometry and hill-slope angles used for
water routing and landslide threshold assessments;
hourly measurements of rainfall intensity and duration
with 12 mm accuracy; and 12-hourly measurements
of soil moisture to assess infiltration and runoff
potential. Seasonal measurements of vegetation cover
and canopy structure provide for water interception
and soil strength assessment, while 5-m resolution
geologic mapping provides an overview of rock
strength, permeability and erosion potential. In conjunction,
multi-channel and multi-sensor data from
meteorological satellites are assimilated into numerical
weather prediction models to estimate precipitation
intensity, amount and coverage, winds and other
factors that impact the severity of flood hazards.
The principal contribution of Earth observation

data is to provide the morphological, land use and
geological detail to help determine how a landslide
failed and the cause of failure (CEOS, 2003). GIS is

being used increasingly for regional risk assessment,
including the integration of inventory mapping,
location of surface structures and roughness providing
information on flow emplacement parameters (i.e.
rate, velocity and rheology), and factors such as
lithology, location of faults, slope, vegetation and land
use. The remote sensors that increasingly will provide
flood and landslide hazard monitoring data include
InSAR, GPS (Malet et al., 2002; Gili et al., 2000),
visible and near infrared/ thermal infrared (VNIR/
Fig. 3. Wet soil conditions of the Tisza River Basin in Eastern Europe from analysis of QuickScat and MODIS data
[image courtesy S.V.
Nghiem, NASA/JPL and R. Brakenridge, Dartmouth University].
TIR) imaging, multi-parameter SAR, laser altimetry,

and microwave imaging. SAR data can provide an allweather
flood mapping capability and can be useful
for the estimation of hydrological parameters such as
soil moisture (soil surface wetness), wet snow
mapping, and the monitoring of wetlands and flood
extent delineation (CEOS, 2003). InSAR offers the
capability for measuring displacements and providing
very high accuracy topographic mapping. However,
even with the integration of in situ measurements, the
ability for a predictive capability for the occurrence
and extent of landslide impact falls behind that for
mitigation planning (see CEOS, 2003). A thorough
evaluation of erosion hazards in the United States
related to coastal processes and flooding is provided
by the Heinz Center (2000), ranging from risk
assessment, economic impact and insurance programs
to management and policy. Satellite-based observations
will need to be augmented with extensive landbased
measurements and data from existing and
future, integrated hydrologic and geodetic arrays.
5. Coastal inundation
Atmospheric and oceanic processes have a significant
effect on coastal geomorphology. Sea level rise as
a consequence of global climate change represents an
enormous risk to coastal populations. Eleven of the
worlds 15 largest cities lie along the coast or on
estuaries. About 53% of the US population, for
example, lives near the coast (Small et al., 2000).
Any short-term or long-term sea-level change relative
to vertical ground motion is of great socioeconomic
concern, yet no accurate estimate of the vulnerability
of coastal communities exists.
The effects of sea level rise are spatially nonuniform
due to local coastal variables, such as
interactions between lithology, geomorphology, and
wave climate, currents and storm frequencies (Gornitz,
1991). Paleo-environmental and historical data
clearly have indicated the occurrence of such changes
in the past, and the potential impact under enhanced
greenhouse conditions (Mcinnes et al., 2003). Sea
level rise itself is not globally uniform, as Fig. 4

shows. The TOPEX/POSEIDON altimetric satellite
now in its 12th year of operation has established this
remarkable record with the most accurate measurements
of sea surface topography to date. Its successor
Jason-1, and the planned NASA Ocean Surface
Topography Mission (each a collaborative mission
between the US and French space agencies), as well
as ERS-1 and -2, and recently ENVISAT, of the
European Space Agency, all contribute to monitoring
this crucial quantity.
Sea level rise does not just passively inundate lowlying
coastal regions. Sea level rise, as a symptom of
climate change, and changes in storm frequency or
Fig. 4. Trends in sea level derived from TOPEX/POSEIDON data for the period 19932003. Inverse barometer (IB)
correction applied [see
http://www.sealevel.colorado.edu and Leuliette et al. (2004)].
intensity lead to greater erosion of the coasts and

accelerate the process whereby the ocean inundates
what was previously land surface. The Intergovernmental
Panel on Climate Change (IPCC) has outlined
major impacts of rising sea level on coastal communities
(IPCC, 2001), which include beach erosion,
inundation of land and increased flood and storm
damage. The IPCC reports that 1 cm rise in sea level
erodes beaches about 1 m horizontally; a 50-cm rise in
sea level will inundate 850019000 km2 of dry land;
and a higher sea level will provide a higher base for
storm surges. A 1-m rise in sea level would enable a
15-year storm to flood areas that today are only
flooded by 100-year storms (IPCC, 2001). Fig. 5
shows the effect of a 5-m sea level rise on the Florida
coast of the US. Flood damages would increase
accordingly and contribute to higher sediment deposits
at inlets, further exacerbating the inundation hazard
risk.
An example of an integrated observation strategy
consists of space measurements of ocean vector winds
(SeaWinds on QuickSCAT and ADEOS-2; ERS-1 and
-2 and now Meteosat) to assess the strength of storms
at sea from their surface wind vectors; in addition,
Tropical Rainfall Mapping Mission (TRMM) measurements
of the precipitation associated with storms,
with NASA Atmospheric Infrared Sounder (AIRS),
MODIS and other instruments observing the cloud
Fig. 5. The potential rise in sea level caused by melting of the Greenland ice sheets seriously jeopardizes low-lying
areas such as the Florida
coast. Red shows where land would be submerged for an estimated 5-m sea-level rise [image courtesy M. Kobrick,
NASA/JPL].
patterns above the storm, and passive radiometers

measuring the water vapor around the storm. By
studying changes in storm frequency and intensity, for
example, better estimates can be made of what future
climate change will bring to coastal regions. Integrated
SAR and TM imagery have been used to
monitor changes in coastal geomorphology and land

cover, flood and erosional damage, and to facilitate
planning and maintenance of mitigations (Singhroy,
1995). Advances to-date in the science and technology
of shoreline change mapping and projection of
future shoreline positions are elements of erosion
hazard zone identification under the FEMA National
Flood Insurance Program (Leatherman, 2003). The
U.S. Environmental Protection Agency has studied
the environmental impact and economic costs associated
with coastal inundation due to the greenhouse
effect and sea level rise (Titus et al., 1991). These
assessments have direct bearing on coastal land-use
planning for risk mitigation.
In addition to the continual effects of atmospheric
and oceanic processes on coastal geomorphology,
earthquakes underneath or near the ocean (particularly
at deep ocean trenches and island arcs) and less
commonly submarine landslides and volcanic eruptions
can generate tsunamis with the potential to
inundate the coast. The magnitude 9.0 Indonesian
earthquake of December 26, 2004 off the west coast
of northern Sumatra is a recent and indeed historic
example of extreme coastal inundation resulting from
generation of tsunamis. The devastating losses to life
and property from this great event are a reminder of
the potential of natural hazards to change the landscape
and calls attention to the need for integrated
monitoring systems, including in situ and orbiting
sensors, real-time communications, extremely fast
assessment, and prearranged communication lines to
those in the best position to warn populations
throughout the globe, where the socioeconomic
consequences of catastrophic disaster events are
widespread.
6. Conclusions
Integrated satellite-based observations and numerical
modeling are leading to new levels of understanding
of the complex solid Earth processes that
often lead to disasters. Satellite remote sensing data
and derived geospatial products increasingly are
complementing ground-based network data, and in
situ and field observations for disaster assessment
and response. Many advances in satellite remote
sensing have been and will continue to be made as
various resources are secured for technology and
infrastructure development, in efforts towards bridging
the transition from natural hazards research to
enhanced operational capability in disaster management.
The Advanced Synthetic Aperture Radar
(ASAR) operating at C-band, onboard ENVISAT,
launched by the European Space Agency in March
2002, is an enhanced capabilities continuation of the
ERS-1/2 that has as one of its mission objectives the
monitoring of earthquake and volcanic hazards. The
Japanese Advanced Land Observing Satellite

(ALOS) includes a panchromatic stereo imager for
digital elevation mapping, a visible and near I/R
radiometer and phased array L-band SAR (PALSAR).
The PALSAR system is designed to provide
continuous images of land deformation. Moreover,
tools to produce information products through
integration, assimilation, modeling and realistic computational
simulations must continue to be developed,
addressing issues of data access continuity,
completeness, interoperability and validation.
The solid Earth science research community must
continue to demonstrate the potential of these remote
sensing systems and derived products for operational
decision-making that impacts the ability to reduce
losses to life and property. Assimilation of science,
model outputs and satellite data into decision support
tools and systems through applications, validation and
performance benchmarking is a critical step. Policyand
decision-makers, emergency managers and responders,
in turn, will use the enhanced decision
support systems, geospatial information products,
model-based forecasts and visualizations in long-term
planning of emergency services and lifelines, comprehensive
disaster assessment and response prioritization.
Cost-effective approaches will be necessary,
with the participation of the commercial sector in
distinct elements of an overall observational architecture,
as the resources available are limited. International
partnerships and cross-agency relationships can
be expected to increasingly enable civil space
agencies to develop a broad range of observations
and shared data sets that respond to the needs of the

operational and policy communities in disaster management
while addressing key questions in the solid
Earth research community.
Acknowledgements
This review paper was written at the Jet Propulsion
Laboratory, California Institute of Technology, under
contract to the National Aeronautics and Space
Administration, and was guided in part by the NASA
Solid Earth Science Working Group (SESWG)
Report, bLiving on a Restless Planet,Q 2002; the
CEOS hazard reports and the CENR Interim Report of
the Subcommittee on Disaster Reduction. We thank
the reviewers of the initial manuscript for their
comments and recommendations; and Mike Abrams,
Paul Lundgren and Son Nghiem for sharing some data
analyses and discussion.
References
Argus, D., Heflin, M., Peltzer, G., Crampe, F., Webb, F., in press.
Interseismic strain accumulation and anthropogenic motion in
metropolitan Los Angeles, J. Geophys. Res.
Bowin, C., 1991. The Earths gravity-field and plate-tectonics.
Tectonophysics 187 (13), 6989.
Buckley, S.M., Rosen, P.A., Hensley, S., Tapley, B.D., 2003. Land
subsidence in Houston, Texas, measured by radar interferometry

and constrained by extensiometers. J. Geophys. Res. 108 (B11)
(Art. No. 2542).
Cakir, Z., Chabalier, J.B., Armijo, R., Meyer, B., Barka, A.,
Peltzer, G., 2003. Coseismic and early post-seismic slip
associated with the 1999 Izmit earthquake (Turkey), from
SAR interferometry and tectonic field observations. Geophys. J.
Int. 155 (1), 93 110.
Carrasco, R.M., Pedraza, J., Martin-Duque, J.F., Sanz, M.A.,
Bodoque, J.M., 2003. Hazard zoning for landslides connected
to torrential floods in the Jerte Valley (Spain) by using GIS
techniques. Nat. Hazards 30 (3), 361 381.
CENR, 2003. Reducing Disaster Vulnerability Through Science and
Technology. Interim Report of the National Science and
Technology Council, Subcommittee on Disaster Reduction,
July.
CEOS, 2003. The Use of Earth Observing Satellites for Hazard
Support: Assessments and Scenarios. Final Report of the CEOS
Disaster Management Support Group (DMSG), November.
Crowley, J.K., Hubbard, B.E., Mars, J.C., 2003. Analysis of
potential debris flow source areas on Mount Shasta, California,
by using airborne and satellite remote sensing data. Remote
Sens. Environ. 87 (23), 345 358.
Donnellan, A., Rundle, J., Ries, J., Fox, G., Pierce, M., Parker, J.,
Crippen, R., DeJong, E., Chao, B., Kuang, W., McLeod, D.,
Matuura, M., Bloxham, J., 2004. Illuminating Earths interior
through advanced computing. Comput. Sci. Eng. 6 (1), 36 44.
Ellrod, G.P., Connell, B.H., Hillger, D.W., 2003. Improved
detection of airborne volcanic ash using multispectral infrared
satellite data. J. Geophys. Res., [Atmos.] 108 (D12) (Art. No.
4356).
Fielding, E.J., Wright, T.J., Muller, J., Parsons, B.E., Walker, R.,
2004. A seismic deformation of a fold-and-thrust belt imaged by
SAR interferometry near Shahdad, SE Iran. Geology 32 (7),
577 580.
Gamba, P., Houshmand, B., 2002. Joint analysis of SAR, LIDAR
and aerial imagery for simultaneous extraction of land cover,
DTM and 3D shape of buildings. Int. J. Remote Sens. 23 (20),
4439 4450.
Gili, J.A., Corominas, J., Rius, J., 2000. Using global positioning
system techniques in landslide monitoring. Eng. Geol. 55 (3),
167 192.
Glaze, L.S., Baloga, S.M., 2003. DEM flow path prediction
algorithm for geologic mass movements. Environ. Eng. Geosci.
9 (3), 225240.
Gornitz, V., 1991. Global coastal hazards from future sea level rise.
Glob. Planet. Change 89 (4), 379398.
Gross, R.S., Chao, B.F., 2001. The gravitational signature of
earthquakes. In: Sideris, M.G. (Ed.), Proc. IAG International
Symposium Gravity, Geoid and Geodynamics (GGG2000),
Banff, Alberta, Canada, July 31August 4, 2000, vol. 123.
Springer-Verlag, New York, pp. 205 210.
Heinz Center, 2000. Evaluation of Erosion Hazards. Report to the
Federal Emergency Management Agency, The H. John Heinz III
Center for Science, Economics and the Environment, Washington
D.C. http://www.heinzcenter.org (accessed 22 Dec.
2004).
Hossain, F., Anagnostou, E.N., 2004. Assessment of current
passive-microwave and infrared-based satellite rainfall remote
sensing for flood prediction. J. Geophys. Res. 109 (D7) (Art.
No. D07102).
Hudnut, K.W., Bock, Y., Galetzka, J.E., Webb, F.H., Young,
W.H., 2002. The Southern California Integrated GPS Network
(SCIGN). In: Fujinawa, Y., Yoshida, A. (Eds.), Seismotectonics
in Convergent Plate Boundary. TERRAPUB, Tokyo,
Japan, pp. 167 189.
IGOS (Integrated Global Observing Strategy) Geohazards Theme
Report, 2004. http://www.dup.esrin.esa.it/igos-geohazards/pdf/
igos_report.zip (accessed 22 Dec. 2004).
IPCC, 2001. Climate change 2001-impacts, adaptation and vulnerability.

Contribution of working group II to the third assessment
report of the intergovernmental panel on climate change (IPCC).
In: McCarthy, James J., Canziani, Osvaldo F., Leary, Neil A.,
Dokken, David J., White, Kasey S. (Eds.), Cambridge University
Press, UK. (1000 p. Available at http://www.grida.no/
climate/ipcc_tar/wg2/ (accessed 22 Dec. 2004)).
Lanari, R., Berardino, P., Borgstrom, S., del Gaudio, C., de
Martino, P., Fornaro, G., Guarino, S., Ricciardi, P., Sansosti, E.,
Lundgren, P., 2003. The use of IFSAR and classical geodetic
techniques for caldera unrest studies: application to the Campi
Flegrei uplift event of 2000. J. Volcanol. Geotherm. Res. 133
(14), 247 260.
Lanari, R., Lundgren, P., Manzo, M., Casu, F., 2004. Satellite radar
interferometry time series analysis of surface deformation for
Los Angeles, California. Geophys. Res. Lett. 31 (23) (Art. No.
L23613).
Leatherman, S.P., 2003. Shoreline change mapping and management
along the US East Coast. J. Coast. Res., 5 13. (Sp. Iss. 38).
Leuliette, E.W., Nerem, R.S., Mitchum, G.T., 2004. Calibration of
TOPEX/Poseidon and Jason altimeter data to construct a
continuous record of mean sea level change. Mar. Geod. 27
(12), 79 94.
Lohman, R.B., Simons, M., Savage, B., 2002. Location and
mechanism of the Little Skull Mountain earthquake as constrained
by satellite radar interferometry and seismic waveform
modeling. J. Geophys. Res., [Solid Earth] 107 (B6) (Art. 2118).
Lu, Z., Patrick, M., Fielding, E.J., Trautwein, C., 2003. Lava
volume from the 1997 eruption of Okmok volcano, Alaska,
estimated from spaceborne and airborne interferometric synthetic
aperture radar. IEEE Trans. Geosci. Remote Sens. 41 (6),
14281436.
Lundgren, P., Rosen, P.A., 2003. Source model for the 2001 flank
eruption of Mt. Etna volcano. Geophys. Res. Lett. 30 (7), 1388.
Lundgren, P., Stramondo, S., 2002. Slip distribution of the 1997
Umbria-Marche earthquake sequence: joint inversion of GPS
and synthetic aperture radar interferometry data. J. Geophys.
Res., [Solid Earth] 107 (B11) (Art. No. 2316).
Malet, J.P., Maquaire, O., Calais, E., 2002. The use of Global
Positioning System techniques for the continuous monitoring of
landslides: application to the Super-Sauze Earthflow (Alpes-deHaute-Provence, France). Geomorphology 43 (12), 33 54.
Mcinnes, K.L.,Walsh, K.J.E., Hubbert, G.D., Beer, T., 2003. Impact
of sea-level rise and storm surge on a coastal community. Nat.
Hazards 30 (2), 187 207.
NRC (National Research Council), 2003. Living on an active earth:
perspectives on earthquake science. Committee on the Science
of Earthquakes. National Academies Press, Washington, D.C.
432 pp.
Pieri, D., Abrams, M., 2004. ASTER watches the worlds
volcanoes: a new paradigm for volcanological observations
from orbit. J. Volcanol. Geotherm. Res. 135 (12), 13 28.
Prata, A.J., Watson, I.M., Rose, W.I., Realmuto, V., Bluth, G.J.S.,
Sevranckx, R., submitted for publication. Volcanic sulfur
dioxide measurements derived from infrared satellite measurements.
Geophys. Res. Lett.
Pritchard, M.E., Simons, M., 2004. An InSar-based survey of
volcanic deformation in the central Andes. Geochem. Geophys.
Geosyst. 5 (Art. No. Q02002).
Rejaie, A., Shinozuka, M., 2004. Reconnaissance of Golcuk 1999
earthquake damage using satellite images. J. Aerosp. Eng. 17
(1), 20 25.
SESWG (Solid Earth Science Working Group) Report, 2002. Living
on a Restless Planet. S. C. Solomon, Chair, NASA Jet
Propulsion Laboratory, Report JPL 400-1040, November.
Shimada, S., Bock, Y., 1992. Crustal deformation measurements in
central Japan determined by a Global Positioning System fixedpoint
network. J. Geophys. Res. 97 (B9), 1243712455.
Singhroy, V., 1995. SAR integrated techniques for geohazard

assessment, Natural Hazards: monitoring and assessment
using remote sensing techniques. Adv. Space Res. 33 (3),
205 290.
Singhroy, V., Mattar, K.E., Gray, A.L., 1998. Landslide characterization
in Canada using interferometric SAR and combined
SAR and TM images. Adv. Space Res. 21 (3), 465476.
Small, C., Gornitz, V., Cohen, J.E., 2000. Coastal hazards and the
global distribution of population. Environ. Geosci. 7 (1), 3 12.
Stevens, N.F., Manville, V., Heron, D.W., 2003. The sensitivity of a
volcanic flow model to digital elevation model accuracy: experiments
with digitized map contours and interferomegtric SAR at
Ruapehu and Taranaki volcanoes, New Zealand. J. Volcanol.
Geotherm. Res. 119 (14), 89105.
Sun, W., Okubo, S., 2004. Coseismic deformations detectable by
satellite gravity missions: a case study of Alaska (1964, 2002) and
Hokkaido (2003) earthquakes in the spectral domain. J. Geophys.
Res., [Solid Earth] 109 (B4) (Art. No. B04405).
Talebian, M., Fiedling, E.J., Funning, G.J., Ghorashi, M., Jackson,
J., Nazari, H., Parsons, B., Priestley, K., Rosen, P.A.,Walker, R.,
Wright, T.J., 2004. The 2003 Bam (Iran) earthquake: rupture of
a blind strike-slip fault. Geophys. Res. Lett. 31 (11) (Art. No.
L11611).
Titus, J.G., Park, R.A., Leatherman, S.P., Weggel, J.R., Greene,
M.S., Mausel, P.W., Brown, S., Gaunt, G., Trehan, M., Yohe,
G., 1991. Greenhouse effect and sea level rise: the cost of
holding back the sea. Coast. Manage. 19 (2), 171 204.
Tondi, R., De Franco, R., Biella, G., 2003. Deep and shallow solidearth
structures reconstructed with sequential integrated inversion
(SII) of seismic and gravity data. Space Sci. Rev. 108 (12),
115130.
Watanabe, H., Matsuo, K., 2003. Rock type classification by multiband
TIR of ASTER. Geosci. J. 7 (4), 347358.
Watson, I.M., Realmuto, V.J., Rose, W.I., Prata, A.J., Bluth, G.J.S.,
Gu, Y., Bader, C.E., Yu, T., 2004. Thermal infrared remote
sensing of volcanic emissions using the moderate resolution
imaging spectroradiometer. J. Volcanol. Geotherm. Res. 135 (1
2), 75 89.
Wright, R., Flynn, L.P., 2003. On the retrieval of lava-flow surface
temperatures from infrared satellite data. Geology 31 (10),
396 893.
Wright, R., Flynn, L.P., 2004. Space-based estimate of the volcanic
heat flux into the atmosphere duing 2001 and 2002. Geology 32
(3), 189 192.
Wright, R., Flynn, L.P., Garbeil, H., Harris, A., Pilger, E., 2002.
Automated volcanic eruption detection using MODIS. Remote
Sens. Environ. 82 (1), 135 155.
Wright, R., Flynn, L.P., Garbeil, H., Harris, A., Pilger, E., 2004.
MODVOLC: near-real-time thermal monitoring of global
volcanism. J. Volcanol. Geotherm. Res. 135 (12), 29 49.

Delineation
Abstract

Introduction
around 40 percent.


selection.
additional band.
procedure.
0099-1112I0OI6612-1461$3.00/0
and Remote Sensing


techniques using a quantitative spatial accuracy assess- - -
ment is needed.
Landsat TMdata,
Y1.E
PI'%
(C)

boundary mapping.
The Study Area

Methods
conditions.
Materials

Small Pool
Dam
(105 features)
features)
Max depth 8 to 10 m
Max depth 1to 4m
Area 6 to 18 ha
Variable shape
Variable shape
Max depth 1 to 8 m
Area 2.8 to 14.5 ha
Marginal trees
< 5% crown cover
Marginal trees
< 5% crown cover
Marginal trees
< 10% crown cover
Nil

resulting images
digital
eight
GCPS were
water bodies
likelihood of a
Aedal photograph lnterpretatlon warped using a linear geometric transformation, nearest-neighThe aerial photographs were scanned at 300 DPI using a Micro- borhood intensity interpolation, and a pixel
resampling size of

20WX)
1-
~ ~ z ~ q ~ ~ g ~ ~ ~ ~ s~~s$>~g2>~- ~- -$
s~:a:~~
t-4

J


variation.

Density Slicing

Mean Pixel Value
Lagoon 1
Lagoon 2
Lagoon 3
Lagoon 4
Lagoon 5
Lagoon 6
River 1
River 2
River 3
River 4
Dam 1
Dam 2
Standard Deviation
AREAS.

Band Value Value
1464 December 2000

irrigation.

(Table 4).
Band 1
Band 2 Band 7
Band 3
Water Not water
Ground truth Image
Band 4
I
I
R~ver Dam Not Water
Lagoon Small

Band 2
9 00 5 ; K i
Q5"m
.I 0 h
0 255 0 255
Band 4
g5;m
95q-J
.I P
rn
0 255 0255
Band 6
5 0 255
9yrl
8 0 255
-0
8K

pixel ranges.

17.799.65.0
26.496.74.8
38.096.54.9
493.677.142.6
596.981.764.5
794.474.445.4

I!

pixel mode.
Classification Data
Percentage
Dam 1642919385.0%
River 2626419304586.2%
Lagoon 77717395081.8%
Not Water 2079886639074297.7%
Classification Data
Dam
River
Lagoon
Not Water
Total
Percentage Correct
lrr~gatedP asture
II
mm

r
Small unclassi
polygons

Conclusion

of the water area.
Or SPOT.
Acknowledgments
References
pp. 114-122.
316 p.
44235-252.

Read PE&RS on-line!

ASPRS web site.
59(9):1389-1395.
283-290.
Adelaide, 12 p.
(on a CD).
11:1427-1439.

11:1409-1413.
Grids & Datums

Abstracts
Software ~evieew
Calendar
Classifieds
Book Review
and many more
prs. org

Literature

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Literature

Hochgeladen von

Copyright:

Verfügbare Formate

Water Body Detection and

bodies with that derived from digitized aerial photography and

School of Science and Technology, Charles Sturt University,

Photogrammetric Engineering & Remote Sensing

O 2000 American Society for Photogrammetry

and Remote Sensing

have used a random sampling procedure to check the classification

Figure 1. The location of (A) the City of Wagga Wagga in Australia,

The Study Area

water for variable but usually short periods after floods or

Polygon label Description

Figure 2. Aerial photograph of the Wagga Wagga region

Single-band density slicing and the multispectral maximumlikelihood

Figure 3. Discharge hydrograph for the Murrumbidgee River at Wagga Wagga

able variation in water brightness.

Results and Discussion

Figure 4 shows a section of the density slicing results for each

Minimum Water Maximum Water

a reasonable representation of the water boundaries but

Figure 4. Results of density slicing each Landsat TM band.

Figure 5. Individual band histograms and water training

Overall Water Class Water Class

Lagoon, and Dam classes) with results above 80percent. The

is presented in Table 5. This table shows that the

Band 5 dens~tys lice Max~mumL ~kehhcodc lassficstion

Figure 6. Water bodies identified by the band 5 density slice

Figure 7. The affect of misregistration

and small polygons on the accuracy of

reduced the likelihood of an accurate classification.

of wetland systems: the Queensland approach, Workshop Proceedings,

Read PE&RS on-line!

Scientist, Environment Australia, Jabiru, Australia, 52, 45 p.

Grids & Datums

and many more

PHOTOGRAMMETRIC ENGINEERING 81 REMOTE SENSING December 2000 1467

Water Body Detection and

TMdata can be used to map water bodies accurately. Density

slicing of the single mid-infrared band 5 proved as successful

School of Science and Technology, Charles Sturt University,

density slicing of Landsat TM or MSS data to delineate water

O 2000 American Society for Photogrammetry

and Remote Sensing

have used a random sampling procedure to check the classification

aerial photography to determine how well the Landsat

Figure 1. The location of (A) the City of Wagga Wagga in Australia,

The Study Area

Polygon label Description

Figure 2. Aerial photograph of the Wagga Wagga region

Aedal photograph lnterpretatlon warped using a linear geometric transformation, nearest-neigh-

Single-band density slicing and the multispectral maximumlikelihood

Figure 3. Discharge hydrograph for the Murrumbidgee River at Wagga Wagga

able variation in water brightness.

using all 12 "water" training areas (Table 2). Classified pixels

Results and Discussion

Figure 4 shows a section of the density slicing results for each

Minimum Water Maximum Water

a reasonable representation of the water boundaries but

poor user's accuracies, of less than 5 percent.

Figure 4. Results of density slicing each Landsat TM band.

Figure 5. Individual band histograms and water training

Overall Water Class Water Class

Lagoon, and Dam classes) with results above 80percent. The