Sie sind auf Seite 1von 29

Faculty of science

Department of Earth Science

(Remote sensing and Geographical Information System Division)


Applied Remote Sensing (Ersc. 771)

Project on:

Evaluation of Bush Encroachment Mapping Using Traditional


Supervised Classification and Spectral Mixture Analysis, a Case for
Borena Rangelands

Submitted to
Dagnachew Legesse (PhD)

By:
Yoseph Alemayehu (GSR/2035/00)

Zemenu Mintesnot (GSR/ 2037/00)

July 2008
Abstract

Bush encroachment is rated second to cause food insecurity in pastoral areas. Careful
management plan and practice requires knowledge of both the spatial and temporal
extent of the problem. This study was aimed at mapping the bush encroachment and
detecting the changes occur from 1986 through 2002. Landsat TM and ETM+ images,
p168r57, taken in January 1986 and February 2002 were used for this study. The
images were analyzed using supervised classification and spectral mixture analysis
(SMA). The mixture analysis was done based on endmembers spectral reflectance
derived from the scene. Change detection on out put images from SMA was conducted
by image differencing. The study showed that there exists substantial bush
encroachment expansion, about 14persent. This needs for immediate measure. Future
study needs to be conducted to evaluate the converted landcover type. Accuracy can be
achieved with utilization of reference data collected through field survey or available
reference data.

Key words: bush encroachment, change detection, supervised classification, spectral


mixture analysis.

II
Table of contents
Abstract ...............................................................................................................................II
Table of contents........................................................................................................... III
List of figures................................................................................................................ IV
List of tables................................................................................................................... V
1. Introduction..................................................................................................................... 1
1.1. Background .............................................................................................................. 1
1.2. Objectives ................................................................................................................ 2
1.2.2. General Objective ............................................................................................. 2
1.2.2. Specific objectives ............................................................................................ 2
1.3. Problem statement.................................................................................................... 3
1.4. Hypothesis................................................................................................................ 3
2 Literature Review............................................................................................................. 4
2.1. The concept of spectral mixture analysis................................................................. 4
2.1.1. Endmember Selection ....................................................................................... 6
2.2. Change detection...................................................................................................... 6
2.3. Image classification ................................................................................................. 6
3. Methodology ................................................................................................................... 8
3.1. Location of the study area........................................................................................ 8
3.2. General methodology............................................................................................... 8
3.2.1. Data acquisition and pre processing ................................................................. 9
3.2.2. Image processing .............................................................................................. 9
3.2.3. Spectral mixture analysis (SMA)...................................................................... 9
3.2.4. Change detection............................................................................................. 10
3.2.5. Supervised Classification................................................................................ 11
3.2.5.1. The processing stage ................................................................................ 11
3.2.5.2. The training and classification stage........................................................ 11
3.2.5.4. Interpretation............................................................................................ 12
4. Result and Discussion ................................................................................................... 13
4.1. Supervised classification........................................................................................ 13
4.2. Spectral Mixture Analysis...................................................................................... 16
4.2.1. Mixture Analysis............................................................................................. 16
4.2.2. Image differencing .......................................................................................... 17
5 Conclusions.................................................................................................................... 19
6. Recommendations......................................................................................................... 20
6. Recommendations......................................................................................................... 20
References......................................................................................................................... 21
Appendix............................................................................................................... 22

III
List of figures

Fig 1: General Methodology Flow chart


Fig 2: map showing major land covers in Borena range lands (1986)
Fig. 3: map showing major land cover in Borena range lands (2002)
Figure 4: gray scale images from the spectral mixture analysis, for bush endmember.
The images are 2002(a) and 1986 (b).
Figure 5: map showing percentage categories of changes in bush encroachment

IV
List of tables

Table 1: major land covers and their respective area coverage


Table 2: areas changed in their percentage composition

V
1. Introduction

1.1. Background

About 61-65 percent of the total area of Ethiopia is estimated to be occupied by pastoral
areas. These areas are home to 12-13 percent of the total population of the country. In
addition, out of the total estimated livestock population the pastoral areas constitute
approximately 30 percent of the cattle 52 percent of the sheep, 45 percent of the goats
and 100 percent of the camel (MOA, 2000). However the pastoral production system and
in particular the food security and lively hood situation is highly threatened because of
different man made and natural risks. Unwanted plant species encroachment is one of the
major challenges to dry land development among others.

“Bush encroachment” means the invasion of shrubs and bushes or trees in to former
grassy range lands. Currently this problem is highly pronounced in the Borena range
lands, which is a home for very large livestock population. Literatures showed that the
area under bush encroachment in the Borena range lands is about 40 percent
(Coppock1994, and Ahimed and Florian, 2000), and it is still progressing (Bruck Y.,
2003). The Oromiya Water Woks Design and Supervision Enterprise (OWWDSE) has
conducted a socio-economic analysis on the causes of food insecurity in the Borena zone
in 2007. And in its detailed analysis bush encroachment was ranked second next to
drought (OWWDSE, unpublished).

Coppock (1994) reported that about 15 woody plant species are considered to be
encroachers. Among these, according to Ahimed and Florian, 2000, Acacia brerispica,
Acacia bussei, Acacia drepanolobium, Acacia melifera, Acacia reficiens, Acacia seyal
and commiphora africana are the major encroaching species in the Borena range land.
The magnitude of encroaching varies from locality to locality and from species to
species. A.drepanolobium, A. reicience, and A. bussei are major and common
encroaching species found across all the range lands.

1
These encroachers have reduced the total area devoted to grazing for cattle. Their thorny
characteristics prohibit the access of livestock to areas where the stand is dense, suppress
grass growth and provide harbor for predators. The causes for this sort of deterioration
seem to be combination of over grazing, ban of range land burning during the former
regime (Ahimed and Florian, 2000). Indigenous knowledge and traditional systems were
neglected and considered backward by the decision makers. The Borena pastoralists had
traditionally used controlled fire as a range management tool. In the youth stage, the
Acacia species are susceptible for fire and can be killed with repetitive burning. Since
1970’s the practice of burning had been banned. Hence unwanted (unpalatable)
herbs/shrubs/bush species got the chance to grow. As they were not eaten by most
livestock, they over took other palatable grass species and begin to dominate. Some of
these woody species are indicators of desertification. It could be attributed to the
repeatedly occurring drought in the area.

Remote sensing and satellites imageries with temporal and synoptic view play a major
role in developing a global and local operational capability for monitoring land
degradation and desertification in dry lands (Khiry, 2007). Therefore, this study is
intending to improve the monitoring capability afforded by Remote Sensing to analyze
and map the bush encroachment change in Borena rangelands.

1.2. Objectives

1.2.2. General Objective

The general objective of this project is to assess the expansion of bush land in the Borena
range lands between 1986 and 2002.

1.2.2. Specific objectives

• To determine the total area converted to bush land between 1986 and 2002.

2
• To undertake classification of main land cover types in the study area using
supervised classification.
• To apply spectral mixture analysis
• To spatially locate bush encroached areas.
• To apply basic concepts of image classification and processing techniques those
were grasped during the theoretical sessions.
• To be well acquainted with image processing software (ERDAS IMAGINE)

1.3. Problem statement

Bush encroachment is one of the most serious problems in the Borena range lands.
Magnitude and location of areas under bush encroachment are not properly determined.
Proper management of range lands require to locate and quantify areas under bush
encroachment. Mitigation and control measures should be based on informed decision
making.

1.4. Hypothesis

Multi temporal satellite images taken from the same scene can be used to detect bush
encroachment changes occurred with in significant period of time.

3
2. Literature Review

2.1. The concept of spectral mixture analysis

The development in satellite technology as well as remotely sensed image acquisition and
analysis offer effective opportunities for monitoring land cover change in arid and semi
arid areas. Spatial resolution of an image is important in scaling observations. Each pixel
within an image provides only a single measurement of spectral response of an area. It is
however consisting of a multiple surface component. (Khiry, 2007) Usually mapping
land use and land cover has been accomplished using traditional (supervised and
unsupervised) techniques. However, it is difficult to find consistent classes with this
approach between images taken at different times (David etal, 2001). Sub pixel
classification interms of SMA is based on reflection proportion of the observed material.

SMA is a promising technique developed from the efforts of earth and land cover types
have shown characteristic patterns of reflectance within wave length across the EMS
(electro magnetic spectrum). In reality surfaces of land cover types are often composed of
a variety of mixtures of materials. For example a pixel composed of both soil and
vegetation will have a spectral response which depends on combination of the general
soil and vegetation spectra.(Borrison and Nicolav, no date) SMA provides a means for
determining the relative abundance of land cover materials present in any pixel based on
the spectral characteristics of the material. A combined spectrum thus can be decomposed
in to a linear mixture of its spectral and members that is the spectra of distinct material in
that IFOV (instantaneous field of view). (Leica Geosystems Geospatial Imaging, 2005)

SMA involves two steps. First defining the spectra for pure selected land cover type and
second each pixel is modeled as spatial mixture endmember spectra to determine the
physical abundance.

4
The following equation is solved for each pixel

DIV = Fi DNi
Where,
DN = brightness value of a given pixel for a specific wave length.
Fi = fractional abundance of particular endmember
DNi = intensity of the image endmember at particular band.
David et al, 2001)

SMA’s advantage can be summarized as follows (David et a,l 2001)


SMA endmember can be identified from image data, field data lab inventories or
from endmember fraction libraries. Therefore, time series of multiple geographic
location SMA fraction coverage is more readily comparable than the products from
classification based on DN.
™ It uses all the information in the multispectral band.
™ It allows a more detailed analysis of pixel contents.

The traditional method for inferring characteristics about vegetation cover from satellite
data is to classify each pixel in to specific land cover classes based on predefined
classification scheme. The goal of linear mixture models is to estimate the fractional
cover of each major landscape unit of interest (endmember) within image pixels. The
inputs to mixture models are endmember reflectance and an image of observation vectors
(pixel reflectance), and the output is a fraction image for each endmember along with an
image containing an error of fit (Khiry, 2007). These fraction images can then be used to
constrain additional spectral analyses, as input to biophysical and biogeochemical
models, or simply as a measure of land cover used to analyze spatial and temporal
changes (David et. al, 2001).

5
2.1.1. Endmember Selection

The selection of endmember is a critical component to successful application of mixture


modeling. Two general approaches exist for deriving endmembers, each with their own
advantages and shortcomings. The first uses reflectance spectra measured in the field or
laboratory. This method allows great control over the selection of endmember spectra,
but requires that raw image data be correctly converted to reflectance, an often difficult
task in remote sensing. It is also often difficult to obtain reference endmember spectra for all
cover endmembers. The second approach derives endmember spectra directly from the
image by extracting reflectance from relatively “pure” pixels. However, isolating a pure
pixel is often impossible given the great surface heterogeneity at the scale of most remote sensors.

2.2. Change detection

Though it is an elusive task, since different techniques (different maps of change, post-
classification or pre-classification) often produce different results (Bonaue, no date),
Change detection is a very common and powerful application of satellite based remote
sensing (RS).

Image differencing is one of the techniques used in change detection. It is a pre classification
technique. In this method the “after” image is subtracted from the “before” image, in a pixel
by pixel format. Even though it was found to be effective so far , it needs accurate geometric
co registration .

2.3. Image classification

Multispectral classification is the process of sorting pixels into a finite number of


individual classes, or categories of data, based on their data file values (Leica
Geosystems Geospatial Imaging, 2005b). Commonly known are the supervised and
unsupervised classification techniques.

6
In supervised classification partitioning of the feature space is realized by an ooperator
who defines the spectral characteristics of the classes by identifying sample areas
(training areas). There fore the operator needs to know where to find the classes of
interest in the area covered by the image.
In unsupervised classification clustering algorisms are used to partition the feature space
into a number of clusters. The choice of algorism depends on classification and the
characteristics of the images.

7
3. Methodology

3.1. Location of the study area

The study area is located in Oromiya region Borena administrative zone a boarding zone
of Ethiopia with Kenya. It is found at a distance of about 540 Km south east from Addis
Ababa. Geographically it is bounded 416595.55-351971.74 (UTM) N and 457487.82-
454583.38 (UTM) E. It comprises whole of Dire woreda and part of Yabello and Teltele
woredas.

The climate of the Borena lowlands is semi arid. Annual mean temperatures vary from
18-25 degrees Celsius, with little seasonal variation. Annual rainfall varies from 440-
1100mm, the average is 600mm. rain fall is bimodal; 59 percent of the annual
precipitation occurs during March to May and 27 percent in September to November. The
landscape is in general undulating, with a few scattered volcanic cones and rock outcrop.
The Borena lowlands have some surface water. However, Dawa and Genale rivers rising
from the Borena highlands and from the mountains of Bale Zone turn both soon to the
east into Somalia.

3.2. General methodology

The Processing in this study composes a supervised classification and spectral mixture
analysis (SMA) technique. The general methodology flow chart is given in Figure 1. Two
year data obtained from satellite imagery of the study area were analyzed by classifying
the image in to major land cover types. In addition SMA technique was applied to
evaluate the percentage of bush (unpalatable vegetation) encroachment in the study area.

8
3.2.1. Data acquisition and pre processing

Two cloud free Land Sat scene (p168r57) covering the study area were selected for
analysis. The TM and ETM+ respectively were taken in January 1986 and February
2002. This time is a dry season for the study area and it helps to discriminate the woody
ever green vegetation from dry grass or photosynthetically less active vegetation.

3.2.2. Image processing

The whole task of digital image processing revolves around increasingly spectral
separablity of the object features in the image. Accordingly the two images were
geometrically registered taking the ETM+ 2002 image as a reference. Subset of the study
area was selected using the ETHIO GIS vector layer as an AOI (Area of Interest) layer.
To apply the spectral mixture analysis atmospheric calibration was required to balance
the reflectance. This calibration was done by using the internal average relative
reflectance method.

3.2.3. Spectral mixture analysis (SMA)

In order to assess the vegetation composition of pixels multi temporal spectral mixture
analysis was done. This method involves image preprocessing, image endmembers
selection, image fraction production and classification of the fraction image. To
determine the endmembers in the image tasseled cap transformation and separability
merging was done to enhance the visual capability to select endmember from the images.
The endmembers used in this study are image endmembers, because they are easily
accessed and they are representatives of the spectra measured at the same scale as image
data. Three endmembers; vegetation (bush), bare ground and dry grass were defined.
The general mathematically model for linear mixing can expressed as

9
DN = ∑Em * f,

∑f = 1
Where,

DN = Relative radiance for each pixel


EM = Relative Radiance of endmember
f = fraction of the endmember in the specified pixel.

(David et. al, 2001)

Generally Sf = 1, this means that the fraction of the endmember of each pixel in the
image sums to 100%. An endmember greater than 100% occurs as the pixel has spectra
similar to that of the particular endmember but with higher reflectance. A negative value
of fraction indicates that an endmember is unnecessary to model in that particular pixel.
Visual interpretation of the vegetation fraction was done to asses the land cover change.
Changes in vegetation fractions were visually interpreted by displaying fractions.

3.2.3.1. Signature derivation

Signature derivation and evaluation is an important part in the subpixel classification.


There are two ways to derive a signature, manual and automatic. Manual derivation is
used when whole pixel Material of interest can be used as training set. In this case, an
interactive method was applied. 30 signatures were taken from each scene. These pixels
were believed to represent pure bush reflectance. To minimize error these signatures were
merged to produce a single representative scene derived signature of pure pixel.

3.2.4. Change detection

The images from the spectral mixture analysis were in gray scale. The pixels of the image
represent the percentage composition of each end member. A visual interpretation was
done to evaluate the magnitude of encroachment in both images. Image differencing was

10
conducted to locate the changes in bush composition. The out put image from Image
Differencing was classified using RGB slice to separate between the degrees of bush
encroachment.

3.2.5. Supervised Classification

Both images were classified separately in to major land cover types using supervised
classification. Classes of major land cover types were identified. These are bush, wood
land, farm land, bare land, grass (pasture land), water body and unknown cover.

Three steps are included in this process. These are the preprocessing and processing
stage, training and classification stage, and the interpretation stage.

3.2.5.1. The processing stage

Normalized difference vegetation index (NDVI), tasseled cap transformation and


principal component analysis were computed to aid the training stage. The greenness
band from the tasseled cap and the second principal component were used to carefully
locate and select representative AOI (training areas).

3.2.5.2. The training and classification stage

Intensive training was made to separate between the wood land and bush land relative to
the other land cover classes. Maximum likely hood classifier was used and AOIs are
given based on seed pixels. The training areas were assigned based on previous
knowledge of the area. No reference data/ image were used. This can be considered as a
limitation during the training stage.

11
3.2.5.4. Interpretation

The change in land cover was visually interpreted. The total land converted to bush
(encroached land covers) were computed from the data found during classification.

TM (1986) and ETM+ (2002)


Imagery

Stacking Separate Bands

Geo Rectification

Image subsetting

Image Enhancement
Supervised Classification

Selection of Endmembers

Change analysis For Bush


Linear Mixture Model

Fraction Image of
Components

Analyzing of Fraction Image

Map of Change in Fraction Image


(Image Differencing)

Fig 1: General Methodology Flow chart

12
4. Result and Discussion

In this study two separate out puts were found. These are from the supervised
classification of the images and the spectral mixture analysis.

4.1. Supervised classification

The image processing phase has helped the training phase for it has enhanced the visual
separablity of the images. The tasseled cap transformation has helped to distinguish
between the bare land and green features in the scene.

The two images were classified in two seven classes with over all accuracy of 95 and 94
percent for the 1986 TM image and 2002 ETM+ image respectively.(the out put images
are given in Fig. 2 and Fig. 3) The error matrix is given in the appendix. At the training
stage intensive training was given two discriminate between wood land and bush land
(lands encroached with bush). The intensive training has increased the accuracy of the
classification process.

Major land cover types Area coverage (in hectare)


1986 2002
Unknown feature 2678.88 2546
Bare Land 218516 124136
Grazing land 596178 903590
Water body 50.9281 29.4847
Farm land 239505 138379
Wood land 216926 167711
Bush land 603229 689083

Table 1 major land covers and their respective area coverage

13
The result found shows there exists an increase in the magnitude of land encroached
with bush. In sixteen year (1986-2002) 14.23 percent increase was observed. This rate is
equivalent to about one percent annual increment.

Fig. 2 map showing major land covers in Borena range lands (1986)

14
Fig. 3: map showing major land cover in Borena range lands (2002)

15
4.2. Spectral Mixture Analysis

4.2.1. Mixture Analysis

Spectral mixture analysis unmixed pixel values in to their component end member.
Generally three end members were selected. Bush, grass (dry), and bare land were used
as major components of the scene. The analysis gave three end member composition gray
scale imagery for both images.

Only the image for the bush endmember was further analyzed for change detection. The
image values ranges -0.56 to 1.74 and -0.6 to 1.43 for the 1986 and 2002 images
respectively. Under ideal accuracy in spectral mixture analysis values should range from
0 to 1(David etal, 2001). Values below zero indicate that the pixel was unnecessary for
that specific end member. On the other hand values above one indicates that the pixel
contains material that has similar reflectance but with relatively higher values. Visual
examination showed that pixel values that fall below zero are derived from area that
were considered bare and unknown features during the supervised classification stage.
Similarly areas that were thought to be wood land showed a pixel value above one. The
output images from the SMA are shown in figure 4 (a) and (b). the gray scale images
helps to visualize the direction of change in bush density.

16
(a) (b)
Figure 4: gray scale images from the spectral mixture analysis, for bush endmember.
The images are 2002(a) and 1986 (b).

4.2.2. Image differencing

Image differencing was the technique applied to detect the change in bush encroachment.
The output of the process is gray scale image. The DN (pixel values) can be negative,
indicating decreasing pixel value, positive, indicating an increase in pixel value, or zero
indicating no change in the “after” image.

The out put form the image differencing was clustered in to groups using a level slicing
method. The slicing was done based on pixel values. The image developed after image
level slicing is given in figure 5. The pixel values indicate the percentage composition
change of bush in that specific pixel. The following table shows areas of land with
percentage increase categories.

0-5% 5-10% 10-15% 15-20% 20-25% 25-35% 35-40% 40- >50%


50%

284946 291614 130328 68424 24731 32020 6545 13140 2918

Table 2 areas changed in their percentage composition

17
From the table it can be seen that a large acreage of land is being encroached with bush.

Spectral Mixture analysis provides the data in more detail than the traditional supervised
classification. The method gives additional information other than merely indicating
areas of change. It showed the composition change in each pixel.

Figure 5 map showing percentage categories of changes in bush encroachment

18
5 Conclusions
From this study the following conclusions have been made.

¾ Bush encroachment (both conversion of land covers in to bushland and


percentage increase in an already encroached areas) are significant and calls for
measures.
¾ The techniques used to enhance the image (the tasseled cap transformation and
the PCA) enhances the visual interpretability of the images.
¾ It was possible to detect the bush encroachment with both supervised
classification and spectral mixture analysis.
¾ The results from the spectral mixture analysis provide more detail information
than the ordinary supervised classification.

19
6. Recommendations

Bush encroachment change detection requires well organized ground reference data in
addition to the utilized satellite images. Classification accuracy can be increased if
training areas are given based on reliable and on site reference data.

Further study needs to be undertaken to map the land cover conversion probabilities. In
addition the invasion can be separately mapped across location for different invading
species. This, however, requires satellite images with better spatial, temporal and spectral
resolution than utilized in this study.

The study assessed that the conversion of the land cover to a bush land in the study area
is significant. The highly noxious species has been invading a wide range of new areas.
This calls for a rapid and sound control and mitigation measures.

20
References

Ababu, A. and Menberu, L. 2007 Awareness raising In drylands coordination group

Ahmad, J. and F. Menzel, 2000 Bush encroachment in the Borena Range lands and how
to turn harmful bush in to use full bush. BLPDP and ORBAD, Negelle Borena
Borrison D. and Nikolov B.(2001) Vegetation and soil spectral mixture analysis, new
south Wales university, Australia

Bruck, Y. 2003 Food security Situation in the pastoral areas of Ethiopia, Oxfam Addis
Coppock, D.L., 1994 The Borena Platue Of southern Ethiopia: Synthesis of
pastoral
David et al. (2001). Per Pixel Analysis of Tree Structure, Vegetation Indices, Spectral
Mixture Analysis and Canopy Reflectance Modelling. Stanford University,
Stanford.
Khiry M.A., (2007). Spectral Mixture Analysis for Monitoring and Mapping of
Desertification Process in Semi-arid Areas in North Korodofan state, Sudan. PhD
Thesis. Dresden
Leica Geosystems Geospatial Imaging, 2005, ERDAS IMAGINE field guide, Leica
Geosystems
Leica Geosystems Geospatial Imaging, 2005, ERDAS IMAGINE Spectral Analysis tour
guide, Leica Geosystems
MAO 1998, Natural resources and regulatory Agro-ecological zones of Ethiopia, Addis
OWWDSE, (Unpublished). Socioeconomic study of Borena Development
corridor proceedings No. 23 Norway Research, development and change, 1980-
1991 system study no 5 ILKA Addis Ababa, Ethiopia
Yusuf.A. and Virginia Atmopawiro (no date) subpixel and maximum likelihood
classification of land sat ETM+ images for detecting illegal logging and mapping
rainforest cover types in East Kalimantan, Indonesia.

21
Appendix

ERROR MATRIX (1986)


-------------

Reference Data
--------------
Classified
Data Bare Land Pasture La Water Unknown Feature
---------- ---------- ---------- ---------- ----------
Bare Land 5828 123 0 0
Pasture La 313 9331 0 0
Water 0 0 28 0
Unknown Fe 0 0 0 3287
Farm Land 13 0 0 0
Bush Land 0 0 0 0
Wood Land 0 0 0 3

Column Total 6154 9454 28 3290

Classified
Data farm Land Bush Land Wood Land Row Total
---------- ---------- ---------- ---------- ----------
Bare Land 14 0 0 5965
Pasture La 1 0 0 9645
Water 0 0 0 28
Unknown Fe 0 0 0 3287
Farm Land 1229 0 0 1242
Bush Land 0 6912 492 7404
Wood Land 0 464 6923 7390

Column Total 1244 7376 7415 34961

ERROR MATRIX (2002)


-------------

Reference Data
--------------
Classified
Data Bush land Wood Land Bare Land Grazing La
---------- ---------- ---------- ---------- ----------
Bush land 7620 115 0 0
Wood Land 4 4416 0 0
Bare Land 0 0 2858 0
Grazing La 18 0 0 6670
Farm Land 0 0 0 0
Water Body 0 0 0 0
Unknown Fe 0 0 0 0

Column Total 7642 4531 2858 6670

22
Classified
Data Farm Land Water Body Unknown Feature Row Total
---------- ---------- ---------- ---------- ----------
Bush land 0 0 3 7738
Wood Land 0 0 0 4420
Bare Land 1 0 0 2859
Grazing La 0 0 0 6688
Farm Land 2075 0 0 2075
Water Body 0 218 0 218
Unknown Fe 0 0 4443 4443

Column Total 2076 218 4446 28441

23
24

Das könnte Ihnen auch gefallen