Sie sind auf Seite 1von 17

Water Resources Management

https://doi.org/10.1007/s11269-018-2074-6

Analysis of Runoff Aggregation Structures with Different Flow


Direction Methods under the Framework of Power Law
Distribution

Kahhoong Kok, et al. [full author details at the end of the article]

Received: 19 January 2018 / Accepted: 31 July 2018/


# Springer Nature B.V. 2018

Abstract
The main purpose of this study is to evaluate the performances of different flow direction
methods (FDM) in affecting the runoff aggregation structures within a watershed under the
framework of power law distribution. To this end, SFD (single flow direction method), MFD
(multiple flow direction method) and IFD (Infinite flow direction method) were applied for
determination of flow direction for water particles in Susu river basin, and consequently
assessed with respect to the variation of flow accumulation. The results indicate that different
patterns of flow accumulation were observed from each flow direction method. Effect of flow
dispersion on DEM is strongest with ascending order of SFD, IFD, MFD. However, contri-
bution of individual pixels into outlet follows descending order of SFD, IFD, MFD. Notably
MFD and IFD tend to make additional hydrologic abstraction from rainfall excess due to the
flow dispersion within the flow paths generated on DEM. This study also investigated the size
distribution of flow accumulation which is equivalent to the drainage area generated from the
selected FDM. They were fitted to the power law distribution and flow accumulation was
recognized as a scale invariance factor based on the parameter estimation for power law
distribution by maximum likelihood. It was also noticed that FDM affects the parameter
estimation of power distribution where highest exponent was estimated for MFD following
by IFD and SFD.

Keywords Flow accumulation . Runoff aggregation structures . SFD (single flow direction
method) . MFD (multiple flow direction method) . IFD (infinite flow direction method) . Power
law distribution

1 Introduction

The development of geographic information processing technology based on DEM


enables extraction of various geomorphological factors in the watershed. O'Callaghan
and Mark (1984) presented the concept of critical threshold area for channel formation.
This concept becomes the impetus for further developments of flow accumulation within
a watershed and since then various techniques for simulating the flow of water particles
on DEM have been presented relentlessly (Band 1986; Quinn et al. 1991; Lea 1992;
Costa-Cabral and Burges 1994; Tarboton 1997).
Kok K. et al.

In most hydrological practice, 8-flow direction method is mainly applied for determi-
nation of flow direction of water particles using DEM. Concept of 8-flow direction has
been utilized for developing the single flow direction method (SFD) (O'Callaghan and
Mark 1984) and the multiple flow direction method (MFD) (Quinn et al. 1991). In the
former case, it is known that it is hindered from simulation of the dispersion of the flow
in a gentle slope region, and the latter has a problem that the dispersion of the flow is
excessively calculated in the steep slope region. Tarboton (1997) related the difficulties
in simulating the dispersion of water flow to the assumption of 8 fixed directions of flow
especially when the flow direction of the water particles is discretized into eight fixed
directions separated by angle of 45 °. 8-flow direction is relatively suitable for simulation
of convergent topography such as a valley, but it has been reported that there is a
limitation for simulation of convex divergent topography such as a hillslope.
The deterministic infinite (∞) flow direction method (IFD) developed by Tarboton
(1997) aimed to improve the drawbacks noticed in 8-flow direction method as discussed
previously by using angle instead of 8 fixed directions to compute the flow direction.
IFD has been regarded as a mean which can efficiently simulate the saturation process
within a watershed by permitting the minimum flow dispersion at the same time escaping
from the limitations induced by 8-flow direction method. In this regard, Gregory et al.
(2001) analysed the statistical properties of drainage density using a stream network
extracted from DEM which demonstrated the utility of IFD as a topographic analysis
tool. In addition, Pack et al. (2001) developed a software namely, SINMAP for assess-
ment of slope safety and IFD has been adopted in the software for flow accumulation
instead of other algorithms such as MFD and SFD.
Many studies have been conducted for hydrological modelling using different flow accu-
mulation algorithms (Wolock and McCabe 1995; Vicente et al. 2014; Rampi et al. 2014).
However, little research has been done to assess the characteristics of the drainage structure
and runoff aggregation structure computed from different flow direction algorithms.
McNamara et al. (2006) proposed a methodology for classifying the watershed into multiple
regions based on the principle of channel initiation by relating the local slope of a pixel to its
corresponding drainage area and subsequently extracting the pixels which are susceptible to
soil losses. They suggested that runoff aggregation structure of a watershed determines the
boundary of erosion mechanism in which the regions which are less than the threshold
drainage area in the complementary cumulative curves are dominated by hillslope process.
On the contrary, the regions which are greater than the threshold drainage are dominated by
fluvial process. The threshold drainage area which separates the hillslope and fluvial process is
determined from fitting the distribution of drainage area to power law. However, there is a
notable point in their study which requires further considerations. Their study was conducted
based on the runoff aggregation structure computed from SFD only. It has been observed from
previous studies that different runoff aggregation structures for a watershed could be computed
using different flow direction methods (Wolock and McCabe 1995; Vicente et al. 2014; Rampi
et al. 2014). Thus, it is necessary to improve the estimation of threshold drainage area for
distinguishing between hillslope and fluvial process under the scenarios of different flow
direction methods.
The main purpose of this study is to extract the drainage structure of an actual watershed
from DEM using different flow direction methods (FDM), namely SFD, MFD, and IFD, and
quantitatively examine the difference of each method in order to recognized the most reliable
flow direction simulation technique. In our study, we focused on the theoretical review of each
Analysis of Runoff Aggregation Structures with Different Flow Direction...

flow direction method and the variability of the flow accumulation accomplished by respective
flow direction method. The novelty of the study is an attempt has been conducted to
investigate the characteristics of runoff aggregation structure computed from different flow
direction methods by fitting them to power law distribution. In addition, we have attempted to
simulate the runoff which imitates closely to the actual water particle flow by considering the
dispersion effect and cohesion effect of the flow according to the landscape of the watershed.

2 Theoretical Background

2.1 Descriptions of Flow Direction Methods

The first group of FDM is called as single flow direction (SFD) methods in which the flow
from a given cell is only allowed to pass to one neighboring downslope cell. The typical
examples of SFD are deterministic D8 method proposed by O'Callaghan and Mark (1984) and
the random single direction method (Rho8) proposed by Fairfield and Leymarie (1991). Rho8
is done to prevent long and parallel flow paths as produced in D8 methods. Thus, it is able to
simulate more realistic-looking flow networks than D8 method. However, Rho8 possess the
same disadvantage as D8 has in which both of them are not able to model flow dispersion.
The second group is multiple flow direction (MFD) method in which the flow from a given
cell is allowed to be distributed to more than one neighboring downslope cells. The examples
of MFD which allows the flow to be distributed to maximum two neighboring cells are Digital
Elevation Model Network (DEMON) presented by Costa-Cabral and Burges (1994) and
deterministic infinite (D∞) flow direction method (IFD) developed by Tarboton (1997).
Gruber and Peckham (2008) proposed the Mass Flux (MF) method which permits the flow
to be distributed to maximum of four neighbouring cells while maximum of eight neighbor
cells are available in the models of Triangular Multiple Flow direction (MD∞) developed by
Seibert and McGlynn (2007), Divergent Flow (FD8) method proposed by Freeman (1991) and
the Digital Terrain Models (DTM) presented by Quinn et al. (1991). We adopted D8 as SFD
and two MFD with maximum two (D∞) and eight (DTM) neighboring cells for the analysis in
our study and their theoretical backgrounds will be briefly presented in the following sections.

2.1.1 8-Flow Direction Method

The 8-flow direction method is operated based on a 3 × 3 pixel window. In Eq. (1), Zj
represents the elevation of each pixel, and the direction of flow is determined by
calculating the slope, Sj with eight neighboring pixels around the elevation of the pixel
located at the center where L represents the distance between the center pixel and the
center of neighboring pixels.
Z o −Z j
Sj ¼ ; j ¼ 1; 2; …; 8 ð1Þ
L
Two types of flow direction methods can be derived from 8-flow direction method. SFD is
formed if the flow is only allowed to be distributed to one neighboring cell (D8) and MFD is
derived if more than one maximum neighboring cell is allowed. SFD is a relatively simple
method which calculates Sj between the target pixel and neighboring eight pixels as presented
in Eq. (1), and the neighboring pixel with the steepest descent slope is selected to receive all
Kok K. et al.

the flow from the target pixel (O’Callaghan and Mark 1984). Unlike SFD, Quinn et al. (1991)
proposes the MFD that estimates Sj as presented in Eq. (1) for eight directions around the target
pixel and the flow is distributed with a constant ratio to the directions with lower elevations
than the target pixel. The flow is divided to the low-elevation pixels according to the weights
depending on the degree of steepness Sj of the pixel where pixel with higher steepness will
receive more flow from the target pixel.
The flow directions computed from the above-mentioned two methods are fixed in the
axial and diagonal directions. In other words, the flow is restricted which can only flow
axially and diagonally. Therefore, in the case of SFD, the flow in the convergent
topography is comparatively well simulated but it has been known that there is a limit
to simulate flow in the divergent topography. On the contrary, the MFD can consider the
dispersion effect of the flow in the divergent terrain, but it has been reported that the
convergence terrain is overly dispersed.

2.1.2 Infinity (∞) Flow Direction Method

Infinity flow direction (IFD) method has been developed as the alternative of 8-flow direction
method. The different point in IFD is that the center of each pixel represents the elevation of
the corresponding pixel and when these centers are connected to each other, 8 triangular facets
can be formed and the slope of each facet is computed using Eq. (2) which can be character-
ized as the slope vector as:
eo −e1 e1 −e2
S1 ¼ ; S2 ¼ ð2Þ
d1 d2
where S1 and S2 are the slope vector for x and y components and e0, e1, and e2 are the elevations
of the pixels which located at the vertex of that particular facet. d1 and d2 are distances between
the vertex. The direction, r and magnitude, S of the slope vector are defined as:
  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
S2
r ¼ tan−1 ;S ¼ S 21 þ S 22 ð3Þ
S1

If the r value computed from Eq. (3) is 0, indicating that it is beyond the range of tan−1(d2/d1).
Equation (4) is applied to calculate the slope of the respective facet in order to prevent
excessive dispersion of the flow.
8
< r < 0; r ¼
>  0; S ¼ S 1  
−1 d 2 −1 d 2 ðe0 −e2 Þ
r > tan ; r ¼ tan ; s ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð4Þ
>
: d1 d1 d 21 þ d 22

The process as above-discussed is the core concept of IFD which selects the steepest
downward slope among the 8 facets and the corresponding r and s compute the flow direction
from the upstream central pixel. Equation (5) is applied to transform r in order to consistently
display the point-by-point flow direction throughout the watershed.
rg ¼ a f r þ ac ðπ=2Þ ð5Þ

where rg (0 and 2π) is the east origin and the flow direction is measured in the form of radian
in counterclockwise direction starting from the east origin. There are two types of simulations
for flow accumulation based on IFD: 1) drainage to one downstream pixel and 2) drainage to
Analysis of Runoff Aggregation Structures with Different Flow Direction...

two downstream pixels. In the former case, the estimation of the flow accumulation is
consistent with the SFD. However, in the latter case, the flow is proportioned between these
two neighbor cells according to how close the flow direction angle (α1, α2) is to the direct
angle to those cells.

2.2 Power Law Distribution

The fundamental relations of power law distribution of drainage area (A) can be converted into
a more general form of power function, f(A) (Newman 2005; Clauset et al. 2009).
f ðAÞ ¼ CA−α ð6Þ
where C is a normalization constant with α being a scaling exponent. Equation (6) goes to
infinity as A approaches to zero implying the existence of the lower limit for the definition
domain. Within the range of Amin < A < ∞, the complete form of f(A) can be derived as:
 
α−1 A −α
f ðAÞ ¼ ð7Þ
Amin Amin
Equation (7) is a full mathematical expression for probability density function of power law
distribution, which gives rise to a closed form of statistical moments for
8
< ∞ ;α < m þ 1

E½A  ¼ ∫Amin A f ðAÞdA ¼
m m α−1 ð8Þ
: Amin m
;α > m þ 1
α−1−m

where m represents the order of statistical moment. It can be seen that Eq. (8) depends on the
magnitude of α. Especially all of the statistical moments for A go to infinity in the case of α
being less than two. This is a typical feature of critical behavior for the complex system
without representative scale. An analytical expression for exceedance probability F(A) also can
be derived from Eq. (9)
 
A −ðα−1Þ
F ðAÞ ¼ ð9Þ
Amin

It is interesting that F(A) is also a power function as well as f(A) and both of them being in
a linear form on a double logarithmic paper. The analysis framework of power law
distribution of drainage area (A) has been thoroughly explained in Kok et al. (2018) and
readers are advised to refer to the literature for further details related to the concept of
power law distribution.

3 Materials and Methodology

3.1 Overview of Target Area

Susu river basin is located at Cameron Highland, Malaysia. It is located at west coast of
Peninsular of Malaysia as illustrated in Fig. 1. Susu reservoir is located at downstream of
the river basin for hydropower generation and flood control purposes. There are three
major rivers located within the river basin which are Bertam River, Telom River, and
Kok K. et al.

Susu River Basin

Fig. 1 Location of Susu river basin, the filled DEM with the outlet of Susu river basin

Lemoi River. A gravity dam with dimension of 88 m height and 500 m length has been
constructed to impound Bertam river, leading to creation of Susu reservoir. The contrib-
uting drainage area of Susu reservoir is about 220km2 as shown in Fig. 1.

3.2 Analysis of DEM

The DEM data used in this study is Shuttle Radar Topography Mission (SRTM) DEM
produced by the National Aeronautics and Space Administration (NASA) and the National
Geospatial-Intelligence Agency (NGA) with grid resolution of 30.7 m × 30.7 m. A GIS
software, SAGA which was developed in Hamburg University has been applied in this study
for the analysis of the DEM data (Conrad et al. 2015). SAGA is an open source free software
which offers topographic tool kit that can be operated sequentially by combining various
simulation techniques for the topography analysis of a watershed. Therefore, in this study, we
investigated the variability of the drainage structure of Susu river basin by applying the three
flow direction methods (SFD, MFD, IFD) as above-mentioned utilizing the SAGA software.
Figure 1 shows the DEM of the study area where the sink has been removed using SAGA. The
blue dot indicates the location of the outlet point of the Susu river basin.

3.3 Parameter Estimation for Power Law Distribution Using Maximum Likelihood

The usual power functions can be generally fitted to the observations by the least square.
However, in the case of power law distribution, the least square based fitting methods are
likely to result in biased results (Newman 2005; Clauset et al. 2009). As an alternative, Clauset
et al. (2009) proposed the fitting of maximum likelihood instead of least square. Therefore, we
adopted the approach of maximum likelihood in this study to estimate the parameter of Amin
Analysis of Runoff Aggregation Structures with Different Flow Direction...

and α as presented in Eq. (7). The procedures of parameter estimation using maximum
likelihood are elaborated as following. Firstly, let consider the set containing n observations
for A (Ai≥Amin, i = 1,2,…, n) where they are following a power law distribution. Based on Eq.
(7) the likelihood of A can be estimated as:
 −α
n α−1 Ai
pðAjαÞ ¼ ∏ ð10Þ
i¼1 Amin Amin

where ∏ is product operator. Taking logarithm of both sides of Eq. (10), setting ∂ ln p(A|α)/
∂α = 0 and solving for α the maximum likelihood estimate α^ can be derived as follows:
  −1
n Ai
^ ¼ 1 þ n ∑ ln
α ð11Þ
i¼1 A min

The application of Eq. (11) to real world observation requires a priori determination of Amin. To
this end, Clauset et al. (2009) suggested Kolmogorov-Smirnov (KS) statistics D for determi-
nation of Amin:
minD ¼ max jS ðAÞ−F ðAÞj ð12Þ
A ≥ Amin

Where S(A) and F(A) denote exceedance probabilities, the former from the observation
and the latter from Eq. (9). Assuming certain Amin can produce a pair of α^ and D by Eqs.
(11) and (12). Therefore, if this estimation process is repeatedly performed on all of the
assumptions for Amin or all of Ai, a final pair of α ^ and Amin can be obtained by
minimizing the D. Clauset et al. (2009) demonstrate the applicability of Eqs. (11) and
(12) to power law distribution fitting by numerical experiments and in this study we
adopted the methodology as discussed above to fit the drainage area, A of the study area
to power law distribution.

4 Results and Discussions

4.1 Applications of Different Flow Direction Methods

Three flow direction methods (SFD. MFD, IFD) have been applied to estimate the flow
accumulation for the study area as shown in Fig. 1. We assume that the flow accumulation is
similar to upstream drainage area (A) in the case of the SFD. However, the flow accumu-
lation computed from MFD and IFD cannot be assumed to be similar to the upstream
drainage area due to the characteristic of dispersed flow. In fact, the accumulated flow from
MFD and IFD is more related to concept of aggregated flow rather than drainage area.
Figure 2a, b, and c show the estimated flow accumulation (m2) of each pixel contributes to
the predefined outlet and different results are obtained from different flow direction
methods. It can be seen from Fig. 2(a) that sharp and clear stream lines (or stream networks)
are observed for the case of SFD. However, the stream lines computed from IFD and MFD
as shown in Fig. 2(b) and (c) are not as clear as SFD and the stream lines look like as they
were drawn on a paper using ink which showing that the streamlines are heavily dispersed
compared to SFD. Figure 2(b) and (c) also visually confirmed that the flow accumulated by
IFD and MFD is influenced by characteristic of flow dispersion.
Kok K. et al.

Fig. 2 Flow accumulation in (a) SFD; (b) IFD; (c) MFD

Figure 3b, c, and d illustrates the flow contribution of each pixel to the predefined outlet for
the watershed delineated using different flow direction methods which is expressed in per-
centage of flow of each pixel. The contribution of each pixel refers to the extent to which
rainfall excess at a given point (or pixel) in the watershed reaches the outlet point through the
flow path. If the contribution of any pixel is 50%, it represents that half (0.5) of the excess
rainfall at that particular point (with excess rainfall of 1) is drained in the form of a surface
runoff through the outlet of the watershed, while the other half is lost within the boundary of
the watershed through the dispersion of the flow. It is interesting to note that the contributing
area grows in the order of SFD, IFD, and MFD as the spatial extent contributing to the outlet

Fig. 3 a) Watershed boundary derived from different FDM; Flow contribution of each pixel in b) SFD, c) IFD,
d) MFD
Analysis of Runoff Aggregation Structures with Different Flow Direction...

point changes according to the flow direction methods. The changes noticed in the contributing
area are strongly related to the degree of dispersion expressed in the algorithm of respective
flow direction method.
Table 1(a) summarizes the main findings computed from each flow direction method as
illustrated in Fig. 3b, c, and d. Here, the second column represents the sum of the areas of all
the pixels having the contribution exceeding 0% (i.e., the total contribution area) and the
statistics shown in the table allows the growth of the contribution area according to the
dispersion of the flow described above to be quantitatively determined. It is noted from
Table 1(a) that different contribution of each pixel is obtained for different types of flow
direction methods. Mean contribution of 100% is acquired for SFD which indicating that all
the pixels within the delineated watershed will contribute to the predefined outlet point.
However, not all the pixels within the watershed delineated from MFD and IFD will contribute
to the outlet point as mean contribution of 96.31 and 99.55% were calculated respectively. The
minimal contribution of a single pixel to the outlet is 1.401 × 10−45% (MFD) and 1.541 ×
10−3% (IFD).
In hydrology, excess rainfall is defined as the rainfall amount which has been subtracted the
hydrological losses such as evapotranspiration or infiltration from the total rainfall. Thus, the
statistics in Table 1(a) can be interpreted as the excess rainfall which falls on the watershed
delineated from SFD will be fully (100%) drained to the predefined outlet point while only 96
and 99% of the excess rainfall will be drained for the case of MFD and IFD. It is estimated that
around 4 and 1% of the excess rainfall will be lost in MFD and IFD (through dispersion)
during the process of draining the flow from its origin to the outlet. These losses explain that
even though greater contributing area was computed from MFD, lower flow accumulation
(Fig. 2c) has been calculated for MFD as compared to SFD. The flow losses due to dispersion
inform us that considering hydrological abstractions in only vertical directions may lead to
miscalculations in hydrological simulations. Horizontal hydrological abstractions as portrayed
by flow dispersion in MFD and IFD provide us the efficient tools for improving and refining
the techniques for hydrological simulations.
As elaborated in the previous paragraph that flow losses have been observed in MFD and
IFD as flow is allowed to be dispersed according to the topography of the watershed. However,
only small differences were observed between SFD and IFD in term of contributing area and
contribution of each pixel. In fact, the flow accumulation for MFD is even lesser (about
0.10 km2 lesser) than SFD at the outlet point even though largest contributing area has been

Table 1 Contribution of each pixel to outlet by different FDM

(a) Contribution of each cell to outlet for Susu River Basin


Flow Direction Method Contributing Area (km2) Contribution of Each Pixel
Minimum (%) Mean (%) Maximum (%)
SFD 226.75 100 100 100
MFD 235.32 1.401 × 10−45 96.31 100
IFD 227.71 1.641 × 10−3 99.55 100
(b) Contribution of each cell to outlet for small scaled watershed in Korea
Flow Direction Method Contributing Area (km2) Contribution of Each Pixel
Minimum (%) Mean (%) Maximum (%)
SFD 8.35 100 100 100
MFD 9.06 9.99 × 10−5 57.02 100
IFD 8.48 2.30 × 10−3 37.42 100
Kok K. et al.

estimated for MFD. The Susu river basin of high steepness contribute to the high contribution
of each pixel (>90%) estimated in MFD and IFD even though flow dispersion has been taken
into consideration. The range of local slope is from 0.00014 to 3.072 in which about 78% of
the pixels consist of local slope greater than 0.2. The high steepness of the watershed
diminishes the effects of flow dispersion in MFD and IFD, leading to no significant differences
were observed for the accumulated flow in the outlet point between the SFD, MFD, and IFD.
The effects of flow dispersion are not prominent for overall scale of the watershed.
However, it is prominent at local scale within the watershed. For example, the catchment
boundary at the outlet derived using MFD extended further than the one derived using SFD
and IFD as illustrated in Fig. 3(a). This may not yield significant variations for hydrological
simulation of lump model. But significant variations are expected for hydrological simulation
of distributed model which performing the simulations on the basis of pixel by pixel. Thus,
applications of flow direction methods which considering flow dispersion are encouraged as
distributed modeling are prevailing nowadays. Apart from that, the stream networks derived
from flow direction methods which consider flow dispersion look more natural than the one
derived from flow direction methods without considering flow dispersion. We understood that
excess rainfall will be directed to nearest stream networks in the form of surface runoff which
is definitely in the form of liquid. Liquid does not move like an object which follows certain
fixed directions or angles. In fact, the surface runoff neither flow according to fixed angles nor
fixed angles. It will definitely flow according to the gravity and at the same time it will be
dispersed to its surroundings during the motion depending on the topography of the surface.
Therefore, the flow accumulated from SFD seems to be over idealized as it neglects the natural
dispersion characteristic of water flow. MFD and IFD offer alternatives for flow accumulation
which imitate more closely to the actual flow characteristic in reality even though some
scholars found that the flow was over dispersed notably in convergent regions of a watershed
(Quinn et al. 1991; Wolock and McCabe 1995; Vicente et al. 2014; Rampi et al. 2014)
Table 1(b) and Fig. 4 show the results of terrain analysis for a small-scaled watershed
located at Korea which occupies catchment area of about 9 km2 and DEM resolution of 30 m
has also been adopted for carrying out the related terrain analysis. The results obtained are
similar to the results obtained for Susu river basin whereby the contributing area estimated
from MFD is the largest and followed by IFD and SFD. Apart from that, similar additional
hydrological extraction has also been observed in MFD and IFD as the estimated mean
contribution of each pixel is less than 100%. However, it seems that the topography charac-
teristic of the watershed induces greater influences on terrain analysis using different FDM.
The estimated mean contribution of each pixel in MFD and IFD is 57 and 37% respectively.
Low contribution of each pixel obtained for the small-scaled watershed may be attributed to its
gentle basin slope which allows more flow dispersion. In contrast, Susu river basin is
characterized with steeper basin slope, resulting in restricted flow dispersion and thus high
mean contribution of each pixel is obtained even though MFD and IFD are adopted.

4.2 Variability of Runoff Aggregation Structures

The complementary cumulative distribution of flow accumulation which is equivalent to


drainage area (A) computed from different FDM for the study watershed is shown in
Fig. 5a. The P[.] in the vertical axis defines the probability of exceedance which satisfies
the designated condition stated within the parentheses. The complementary cumulative distri-
butions of the drainage area were also fitted to power law distribution using the approach of
Analysis of Runoff Aggregation Structures with Different Flow Direction...

Fig. 4 a) Boundary derived for small-scaled watershed; Flow contribution of each pixel in b) SFD, c) IFD,
d) MFD

maximum likelihood as depicted in Eqs. (11) and (12). The power-law fitted exceedance
probabilities F(A) are shown in Fig. 5a depicts the parameters estimation for fitting the flow
Kok K. et al.

accumulation computed from different FDM to power law distribution. It can be seen clearly
from Fig. 5a that the flows accumulated from three different FDM are following power law
distribution as the linear relationships observed in the distribution curves exceeded about 2 to 3
logarithmic scale intervals of the horizontal axis which are similar to the behavior of typical
power law distribution as stated in Gutenberg-Richter’s law or Zip’s law (Bak 1996).
It is noted that all the estimated α ^ values are less than 2 and according to Eq. (8) the
statistical moment for all orders will not be defined if the value of α ^ is less than 2. This
represents that the flow accumulation is a scale invariance topographical factor in which its
representative scale cannot be quantitatively determined irrespective of the FDM. Kim et al.
(2016) found that this behavior has been observed in several watersheds in Korea whereby the
flow accumulations were computed based on SFD. Therefore, it can be concluded that the
three FDM considered in this study are appropriate to describe the runoff aggregation
structures within the watershed.
It can be visually confirmed from Fig. 5a that all the distribution curves show a straight line
pattern at the center, but different behaviors in the vicinity of the origin and at the tail of the
curves. The behaviors observed at the tail of the curves could be interpreted as being due to the
finite size effect of the sample (Newman 2005; Clauset et al. 2009). In other words, as it moves
towards downstream, the data of large value becomes insufficient in the sample space, and the
distribution curve of the corresponding section (downstream, tail of the curve) deviates from
the linear characteristic as shown in the central part of the curves. However, if the current outlet
point was extended to further downstream which occupying larger watershed area, the linear
distribution behavior of the slope would also be extended towards the downstream direction
(Kim et al. 2016).
Moglen and Bras (1995) categorized the complementary cumulative distribution curve of A
into three regions: 1) inverse S-region in the vicinity of the origin, 2) straight line in the middle
part, and 3) sharp fall at the tail of the curve. They explained that the distribution of A with an
inverted S-shaped behavior near the origin which different from the middle part was due to the
difference in the runoff aggregation structure between the surface and the stream networks. In
other words, transition from region 1 to region 2 represents changes of runoff aggregation
structure from hillslope to valley. The landscape evolution model, SIBERIA developed by
Willgoose et al. (1991) defines the transition point as the place where the dominant erosion
process change from hillslope process (region close to origin) to fluvial process (region far
from origin). Therefore, the estimated value of Amin can be recognized as the threshold which
distinguishes the runoff aggregation structure between the hillslope and the valley and Amin
changes according to the FDM applied for flow accumulation. MFD yielded smallest value of
Amin (3312.10 m2 with α ^ of 1.652) while largest Amin was estimated in SFD (23,184.60 m2 with
^ of 1.502). The Amin estimated in IFD is 11,828.90 m2 with α
α ^ of 1.554. This variation
indicates that the transition of runoff aggregation structure occurs at smaller drainage area in
MFD when comparing to SFD and IFD and this transition threshold plays an important role in
assessing soil losses susceptibility within a watershed. Thus, applications of different FDM are
encouraged for determining the optimal drainage area threshold for distinguishing the runoff
aggregation structure.

4.3 Comparison of Approaches for Parameter Estimation of Power Law Distribution

^ were compared to the results concluded from previous studies where the
The values of α
parameter was computed using different approaches. Most of the previous studies computed
Analysis of Runoff Aggregation Structures with Different Flow Direction...

Fig. 5 a Power law distribution of drainage area computed from different FDM; b Power law distribution of
drainage area fitted by regression analysis
Kok K. et al.

Cumulative Area Distribution for SFD


10
Power Law Fitting
by Least Square Approximation
(Regression Analysis)
1
F(A) = 101.54A-0.47
R² = 0.9934
0.1
P[A>A*]

0.01

F(A) = 347.84A-0.556
R2=0.903
0.001

0.0001

Amin = 23184.6m2 Amax = 100608888m2


0.00001
100 1,000 10,000 100,000 1,000,000 10,000,000 100,000,000 1,000,000,000
Drainage Area, A (m2)

Cumulative Area Distribution for IFD


10
Power Law Fitting
by Least Square Approximation
(Regression Analysis)
1

F(A) = 85.506A-0.495
0.1 R² = 0.9914
P[A>A*]

0.01

F(A) = 278.96A-0.577
0.001 R² = 0.9058

0.0001

Amin = 11828.9m2 Amax = 100620216m2


0.00001
100 1,000 10,000 100,000 1,000,000 10,000,000 100,000,000 1,000,000,000
Drainage Area, A (m2)

Cumulative Area Distribution for MFD


10
Power Law Fitting
by Least Square Approximation
1 (Regression Analysis)

F(A) = 57.403A-0.538
0.1 R² = 0.987

0.01
P[A>A*]

0.001
A= 170.47A-0.613
R² = 0.9127
0.0001

0.00001

Amin = 3312.1m2 Amax = 100875096m2


0.000001
100 1,000 10,000 100,000 1,000,000 10,000,000 100,000,000 1,000,000,000
Drainage Area, A (m2)

(b)
Fig. 5 continued.
Analysis of Runoff Aggregation Structures with Different Flow Direction...

the α ^ for the flow accumulated using SFD. The α ^ for SFD obtained in this study is 1.502
which is greater than value in the study conducted by Rodriguez-Iturbe et al. (1992) where α ^
of 1.45 was obtained in their study. (Kim et al. 2016) related the variation of the estimated
parameter was due to the different approach has been adopted for parameter estimation of
power law distribution. Fitting by least square approximation has been extensively adopted for
estimation of the parameter prior to the emergence of power law distribution fitting by
maximum likelihood method which was developed by Clauset et al. (2009). In other words,
the studies conducted before that were using crude and subjective method in assuming the
value of Amin by extracting the first data showing a linear relationship as Amin through judging it
from the scatter plots of the distribution curve. Subsequently, the α ^ value was estimated by
performing fitting by least square approximation starting from that point of Amin.
In order to investigate the effect of the fitting approaches on estimation of the parameter,
fitting by least square approximation has been conducted as well to estimate the α ^ value by
using the Amin estimated from maximum likelihood as the starting point which portrays the
linear relationship and the results are illustrated in Fig. 5b. The dotted black line in Fig. 5b
indicates that least square approximation was conducted excluding the data at the tail of the
curve while orange solid line indicates the least square approximation including the data at the
tail of the curve. And it can be clearly seen from Fig. 5b that inclusion of the data at the tail of
the curve leads to the increment of α ^ and α^ of 1.47 (for SFD) was obtained when the data at
the tail were omitted which was quite similar to the value obtained by Rodriguez-Iturbe et al.
(1992). The α ^ value for SFD obtained in this study using maximum likelihood method is about
1.502 which is just slightly higher than the value estimated from least square approximation.
Therefore, parameter estimation of power law estimation using maximum likelihood is
validated.
However, it should be taken into consideration of the quality of the DEM data as well as the
FDM applied for flow accumulation for estimating the parameter of power law estimation. The
resolution of the DEM data affects the size distribution of the drainage area and automatically
the estimation of the parameter is affected as well. The effects of FDM on the parameter
estimation are also illustrated in Fig. 5b. Different α ^ and Amin have been estimated for
respective FDM and the results obtained from least square approximation show a similar trend
with maximum likelihood where MFD yielded highest value following by IFD and SFD.

5 Conclusion

In this study, three flow direction methods namely SFD, MFD, and IFD have been applied to
simulate the flow direction as well as the flow accumulation in the actual watershed of Susu
river basin. It is found that different patterns of flow accumulation were observed resulted from
applications of different FDM. The effect of flow dispersion on DEM is strongest with
ascending order of SFD, IFD, MFD and contribution of individual pixels into outlet follows
descending order of SFD, IFD, MFD. In addition, we have examined the characteristics of the
runoff aggregation structures resulted from each method by fitting the flow accumulation to
power law distribution using maximum likelihood. The drainage area which is equivalent to
flow accumulation has been recognized as a scale invariant factor and its representative scale
cannot be quantitatively determined irrespective of the FDM. The threshold value of Amin
which distinguishes hillslope from fluvial process varied with FDM. Maximum likelihood
method provides an improved alternative for parameter estimation of power law as it avoids
Kok K. et al.

subjective assumption in determining the starting points of linear relationship (Amin). It is


noticed also that the exponent changes according to the FDM where MFD shows the largest
exponent following by IFD and SFD.
Lastly, we realize that watershed is a complex system which consists of different landforms
and the characteristics of runoff are greatly affected by the landforms conditions. Therefore,
different flow direction methods should be adopted according to the terrain condition in order
to describe the actual flow condition in respective terrain condition and it is suggested to adopt
SFD in convergent terrains and MFD/IFD in divergent terrain.

Acknowledgements This research was supported by Basic Science Research Program through the National
Research Foundation of Korea (NRF) funded by the Ministry of Education of Korea Government
(NRF2017R1D1A1B03033970).

Compliance with Ethical Standards

Conflict of Interest None.

References

Bak P (1996) How nature works. Copernicus/Springer–Verlag, New York


Band LE (1986) Topographic partition of watersheds with digital elevation models. Water Resour Res 22(1):15–
24
Clauset A, Shalizi CR, Newman MEJ (2009) Power-law distributions in empirical data. SIAM Rev 51(4):661–
703
Conrad O, Bechtel B, Bock M, Dietrich H, Fischer E, Gerlitz L, Wehberg J, Wichmann V, Böhner J (2015)
System for automated geoscientific analyses (SAGA) v. 2.1.4. Geosci Model Dev 8:1991–2007. https://doi.
org/10.5194/gmd-8-1991-2015 Download
Costa-Cabral M, Burges SJ (1994) Digital elevation model networks (DEMON): a model of flow over hillslopes
for computation of contributing and dispersal areas. Water Resour Res 30(6):1681–1692
Fairfield J, Leymarie P (1991) Drainage networks from grid digital elevation models. Water Resour Res 27:709–
717
Freeman GT (1991) Calculating catchment area with divergent flow based on a regular grid. Comput Geosci
17(3):413–422
Gregory ET, Catani F, Rinaldo A, Bras RL (2001) Statistical analysis of drainage density from digital terrain data.
Geomorphology 36:187–202
Gruber S, Peckham S (2008) Land-surface parameters and objects in hydrology. In: Hengl T, Reuter HI (eds)
Geomorphometry: concepts, software, applications. Elsevier, Amsterdam, pp 171–194
Kim JC, Kang H, Jung KS (2016) Analysis of drainage structure for river basin on the basis of power law
distribution. J Korea Water Resour Assoc 49(6):495–507
Kok KH, Lariyah MS, Jung KS, Kim JC (2018) Application of geomorphologic factors for identifying soil loss
in vulnerable regions of the Cameron Highlands. Water 10(4):No. 396
Lea NL (1992) An aspect driven kinematic routing algorithm, overland flow: hydraulics and erosion mechanics.
Parsons AJ, Abrahams AD (eds), Chapman & Hall, New York
McNamara JP, Ziegler AD, Wood SH, Vogler JB (2006) Channel head locations with respect to geomorphologic
thresholds derived from a digital elevation model: a case study in northern Thailand. For Ecol Manag 224:
147–156
Moglen GE, Bras RL (1995) The importance of spatially heterogeneous erosivity and the cumulative area
distribution within a basin evolution model. Geomorphology 12:173–185
Newman MEJ (2005) Power laws, Pareto distributions and Zipf's law. Contemp Phys 46(5):323–351
O'Callaghan JF, Mark DM (1984) The extraction of drainage networks from digital elevation data. Comput Vis
Graphics Image Process 28:324–344
Analysis of Runoff Aggregation Structures with Different Flow Direction...

Pack RT, Tarboton DG, Goodwin CN (2001) Assessing terrain stability in a GIS using SINMAP, in 15th annual
GIS conference. GIS 2001, Vancouver, British Columbia, February 19–22
Quinn P, Beven K, Chevallier P, Planchon O (1991) The prediction of hillslope flow paths for distributed
hydrological modeling using digital terrain models. Hydrol Process 5:59–79
Rampi LP, Knight JF, Lenhart CF (2014) Comparison of flow direction algorithms in the application of the cti for
mapping wetlands in Minnesota. Wetlands 34(3):513–525
Rodriguez-Iturbe I, Ijjasz-Vasquez EJ, Bras RL, Tarboton DG (1992) Power law distributions of discharge, mass,
and energy in river basins. Water Resour Res 28(4):1089–1093
Seibert J, McGlynn BL (2007) A new triangular multiple flow direction algorithm for computing upslope areas
from gridded digital elevation models. Water Resour Res 43(4):W04501
Tarboton DG (1997) A new method for the determination of flow directions and upslope areas in grid digital
elevation models. Water Resour Res 33(2):09–319
Vicente ML, Bielsa CP, Montero TL, Lamban LJ, Navas A (2014) Runoff simulation with eight different flow
accumulation algorithms: recommendations using a spatially distributed and open-source model. Environ
Model Softw 62:11–21
Willgoose G, Bras RL, Rodriguez-Iturbe I (1991) A coupled channel network growth and hillslope evolution
model, 1. Theory. Water Resour Res 27(7):1671–1684
Wolock DM, Mccabe GJ Jr (1995) Comparison of single and multiple flow direction algorithms for computing
topographic parameters in TOPMODEL. Water Resour Res 31(5):1315–1324

Affiliations

Kahhoong Kok 1 & Lariyah Mohd Sidek 2 & Kwansue Jung 1 & JooCheol Kim 3

* JooCheol Kim
kjoocheol@hanmail.net
1
Chungnam National University, Building E2-132, 99 Daehak-ro, Yuesong-gu, Daejeon 34134, South Korea
2
Sustainable Technology and Environment Group (STEG), College of Engineering, Jalan IKRAM-UNITEN,
Universiti Tenaga Nasional, 43000 Kajang, Malaysia
3
International Water Resources Research Institute, Chungnam National University, Building E2-132, 99
Daehak-ro, Yuesong-gu, Daejeon 34134, South Korea

Das könnte Ihnen auch gefallen