J. Geogr. Sci.

2011, 21(4): 689-704

DOI: 10.1007/s11442-011-0873-2

© 2011 Science Press Springer-Verlag

The scaling method of specific catchment area

from DEMs
YANG Xin1, TANG Guoan1, XIAO Chenchao2, GAO Yiping1, ZHU Shijie1
1. Key Laboratory of Virtual Geographic Environment, Nanjing Normal University, Ministry of Education, Nan-
jing 210046, China;
2. Institute of Remote Sensing and Geographic Information System, Peking University, Beijing 100871, China

Abstract: Specific Catchment Area (SCA) is defined as the upstream catchment area of a unit
contour. As one of the key terrain parameters, it is widely used in the modeling of hydrology,
soil erosion and ecological environment. However, SCA value changes significantly at dif-
ferent DEM resolutions, which inevitably affect terrain analysis results. SCA can be described
as the ratio of Catchment Area (CA) and DEM grid length. In this paper, the scale effect of CA
is firstly investigated. With Jiuyuangou Gully, a watershed about 70 km2 in northern Shaanxi
Province of China, as the test area, it is found that the impacts of DEM scale on CA are dif-
ferent in spatial distribution. CA value in upslope location becomes bigger with the decrease
of the DEM resolution. When the location is close to downstream areas the impact of DEM
scale on CA is gradually weakening. The scale effect of CA can be concluded as a math-
ematic trend of exponential decline. Then, a downscaling model of SCA is put forward by
introducing the scale factor and the location factor. The scaling model can realize the con-
version of SCA value from a coarse DEM resolution to a finer one at pixel level. Experiment
results show that the downscaled SCA was well revised, and consistent with SCA at the target
resolution with respect to the statistical indexes, histogram and spatial distribution. With the
advantages of no empirical parameters, the scaling model could be considered as a simple
and objective model for SCA scaling in a rugged drainage area.
Keywords: DEM; specific catchment area; scale effect; scaling model; Jiuyuan Gully

1 Introduction
Specific Catchment Area (SCA) is the upstream catchment area of a unit contour, and it can
be defined as the ratio of the upstream catchment area of contour and its length. The SCA is
not merely a specific numerical index, but also a basic index widely applied in landform
structure delineation, hydrological models and soil models. Many SCA integrated topog-
raphic indexes, like stream power index (SPI) and topographic wetness index (TWI), have
been commonly applied to the watershed network analysis (O’Callaghan and Mark, 1984;

Received: 2010-09-14 Accepted: 2011-02-20

Foundation: Key Project of National Natural Science Foundation of China, No.40930531; National Youth Science Foun-
dation of China, No.40901185; Specialized Research Fund for the Doctoral Program of Higher Education,
No.20093207120008
Author: Yang Xin (1976–), Ph.D, specialized in digital terrain analysis. E-mail: [email protected]

www.geogsci.com springerlink.com/content/1009-637X
690 Journal of Geographical Sciences

Jenson, 1994), soil moisture spatial analysis (O’Loughlin, 1986; Matı’as et al., 2006),
non-point source pollution analysis (Vieux, 1991), landslide monitoring and analysis (Duan
and Grant, 2000), geomorphologic structure researches (Lv et al., 1998a, 1998b), etc.
DEM resolution is a key factor compromising the accuracy and the efficiency of a terrain
representation. Like other terrain parameters derived from DEMs, the SCA is inevitably in-
fluenced by DEM resolution as well. Many researches focus on the influence of the DEM
resolution on the terrain analysis, such as DEM derivation, terrain analysis and watershed
parameters (Kienzle, 2004; Erskine et al., 2007; Tang et al., 2003; Wu et al., 2003; Yi et al.,
2007; Palph et al., 2007), the runoff and sediment simulation (Chaubey et al., 2005; Ren et
al., 2004; Chaplot, 2005; Li et al., 2009), the soil landscape models (Thompson et al., 2001;
Smith et al., 2006), the hydrological models (Sun et al., 2008; Le Coz et al., 2009), and the
water and soil erosion models (Wu et al., 2005; Svoray et al., 2009). All these studies turn
out to be of great help by clarifying the uncertainty issue of digital terrain analysis caused by
the DEM resolution through different aspects.
Although the effects of DEM resolution on terrain analysis are researched widely, most of
the achievements are made over the field of the rule or changes of terrain parameters ex-
tracted from multi-scale DEMs. Nevertheless, scaling is one of the key means to deal with
the effect of multi-scale DEMs on terrain analysis, and few efforts are made on studies of
scaling methods of terrain parameters.
Different from scale, scaling contains changes, transformation and conversion corre-
sponding to spatial distribution pattern, process over time or both its sensitivity to changes.
Li and Cai (2005) indicated that scaling can be regarded as two aspects: upscaling and
downscaling. Upscaling means extrapolating the observation, experiment and simulation
from microscopic scale to a larger scale process, and can be regarded as a procedure of gen-
eralization. On the contrary, the task of downscaling is to push down the scale from coarser
spatial and temporal resolution into more detailed information with finer spatial or temporal
resolution. Liu et al. (2007) pointed out that scale of DEM issues include geographic scale,
sampling scale, structure scale (horizontal resolution and vertical resolution) and analysis
scale. Among them, the vertical resolution, also called DEM resolution for short, as one of
the key factors representing the accuracy of terrain surface, attracts widespread attention in
scale issues of digital terrain analysis. Most of recent researches are focused on the rela-
tionship between DEM resolution and DEM derivatives.
In order to mitigate the effect of DEM resolution on the DEM derivatives, some scaling
models have been put forward. However, most of them focus on upscaling like automatic
generalization of relief (Wu and Zhu, 2001; Fei et al., 2006; Yang et al., 2008). Due to dif-
ficulties and complexities, there have been relatively few studies on downscaling of terrain
parameters. Nevertheless, for the purpose of weakening the effect of DEM resolution on
terrain parameters and overcoming the limitation of high resolution data, some downscaling
methods have been brought forward. Zhang et al. (1999) compared different slope algo-
rithms and built a slope-scaling model based on the fractal and statistics. By means of histo-
gram statistics, Chen et al. (2004) introduced a statistic method to convert the slope histo-
gram from 25 m to 5 m resolution on the Loess Plateau. Yang et al. (2006) performed a
downscaling procedure of slope gradient from 1:250,000 map scale of DEM to 1:100,000
map scale by means of histogram matching. Yang et al. (2007) proposed a slope down-
YANG Xin et al.: The scaling method of specific catchment area from DEMs 691

scaling model in northern Shaanxi Province of the Loess Plateau region by means of statis-
tical and fractal methods. All these researches focus on slope gradient which is one of the
terrain parameters.
In spite of the difference from slope gradient, the existing slope scaling methods give us
some insights about the scaling of SCA, and other DEM derivatives as well. Pradhan (2006)
brought forward a downscaling model of topographic wetness index (TWI), which included
the downscaling model of CA. The method was performed based on several assumptions and
generated reasonable results. However, since the effect of DEM resolution on SCA has not
been demonstrated distinctly, his model is rather complicated to be easily understood. In
order to reduce uncertainty of SCA analysis, in this paper through studying the rules of
multi-scale SCA carefully, a relatively simple model for SCA downscaling was put forward,
which can be used in a complete watershed and get relatively reliable result of SCA on the
condition of coarse resolution DEM.

2 Definition and calculation of SCA

The Catchment Area (CA) describes an area draining to the outlet of that catchment. SCA is
defined as the upstream catchment area of a unit width of a contour (Moore et al., 1991), and
can be calculated through the limit of the ratio between the upstream area of a contour seg-
ment (CA) and the length of the segment (CL) as the latter approaches zero (Hutchinson and
Dowling, 1991):
CA
SCA = lim (1)
CL −> 0 CL

In fact, CA and SCA are distributed by the flow direction and flow line, which are con-
sidered as important characteristics in terrain analysis. A flow direction is the direction of
the maximum flow slope and is usually equal or approximate to the value of aspect. A flow
line is the flow path along with particular flow directions. In Figure 1, the segment of the
contour corresponding to CA is determined by two flow lines crossing the counter, and CL is
the length between the two intersection points.

Figure 1 Illustration of specific catchment area (the area of grey color denotes a specific catchment area, SCA)

There are many recognized techniques to compute flow direction, including D8 algorithm
(O’Callaghan and Mark, 1984), DEMON algorithm (Costa-cabral and Burges, 1994), Dinf
algorithm (Tarboton et al., 1997), multiple flow direction algorithm (Freeman, 1991; Quinn,
1991), random D8 algorithm (Fairfield and Leymarie, 1991), multi-level skeletonization
algorithm (Meisels, 1995), seed filling algorithm based on stack (Zhu et al. 2005) and adap-
692 Journal of Geographical Sciences

tive approach for a multiple-flow-direction algorithm (Qin et al., 2007). However, different
algorithms for flow direction often cause disagreement in some specific locations, especially
in mountainous area, which will greatly influence the value of SCA. In this study, as the
most widely used algorithm with the advantages of simplicity and efficiency, D8 was se-
lected to calculate the flow direction. Although Hutchinson and Dowling (1991) have stud-
ied the changes of SCA with two different DEM resolutions, they still have not given
mathematic equation between SCA and DEM resolution, and any scaling models either.
Therefore a downscaling model is needed when calculating SCA in large drainage area with-
out high DEMs resolution.

3 Study area and data processing

3.1 Study area

Jiuyuangou Gully is located in Suide County, northern Shaanxi of China (Figure 2). It covers
an area of 70 km2, with the altitude ranging from 832 to 1200 m. The loess crisscross gully
and hillock gully are fully developed in this area, known as the loess hilly and gully region
with the gully density of 7.2 km/km2 and average slope of 29.3°, belonging to a typical
semi-arid continental monsoon climate area.

Figure 2 The DEM of Jiuyuan gully (5 m resolution)

3.2 Data and its pre-processing

A DEM with 5-m resolution, generated from 1:10,000 scale map by the State Bureau of
Surveying and Mapping (SBSM), was used as the original data. It was produced with the
projection of Gauss-Krüger and the coordinate system of Xi’an 1980. The SCA derived from
the original data is considered more reliable and can be used as criterion data for the ex-
periment.
The state-of-the-art technologies for generating multi-scale DEMs include re-sampling,
image filtering (Yang et al., 2008), wavelet (Wu and Zhu, 2001) and 3d-dauglas analysis
(Fei et al., 2006). For the method of image filtering, it is difficult to choose an appropriate
scaling factor in specific drainage area. Using wavelet can generate multi-scale DEMs, but
the principle of wavelet is not clearly for interpretation and different wavelet functions may
YANG Xin et al.: The scaling method of specific catchment area from DEMs 693

have different multi-scale DEMs. In addition, for the method of 3D-dauglas, multi-scale
DEMs are influenced by searching directions and definition of base surface. Finally
re-sampling method is chosen in this paper for it is simple, practical and efficient. Consid-
ering the continuity of the terrain surface, we use the bilinear interpolation method for
re-sampling. The multi-scale DEMs were generated from the original 5 m DEMs, with the
resolutions ranging from 10 m to 100 m with 10 m increment. By comparing the mean
square errors of the re-sampled DEMs and the ones by SBSM (Table 1), it is shown that the
re-sampled DEMs are eligible for multi-scale analysis, since its accuracy is higher than that
of the SBSM DEMs.
The multi-scale SCA can be derived from DEMs in three steps (Figure 3): 1) calculating
the flow direction and flow accumulation based on the filled DEMs; 2) calculating flow
length and catchment area based on the flow direction and flow accumulation, respectively;
3) deriving the SCA with equation 1.

Table 1 A comparison of DEM accuracy between the re-sampled DEM and the SBSM DEM
Map scale DEM resolution (m) ME of SBSM DEM (m) ME of re-sampled DEM (m)
1:50,000 25 11.5 4.8
1:250,000 100 35.3 16.0

Figure 3 The process of SCA extraction

4 Scale effect of SCA

4.1 Scale effect of CA

The effect of DEM scale on CA was studied at first (Table 2). The minimum value oc-

Table 2 Statistics of CA at different DEM resolutions (m2)

DEM resolution (m) CA Min CA Max CA Ave
5 25 68976776 39019.2
10 100 69037400 77966.6
20 400 69080800 155329.2
30 900 69051600 230217.9
40 1600 69088000 311608.2
50 2500 69027500 391682.5
60 3600 69051600 461510.8
70 4900 68678400 537423.9
80 6400 68998400 608890.6
90 8100 68509800 682379.7
100 10000 67980000 770469.9
694 Journal of Geographical Sciences

curs in the first flow catchment site like the hilltops or the ridge areas, and it increases as the
DEM resolution going coarser and equals to the DEM grid cell area as well. As shown in
Table 2, the maximum values of CA are close or even equal to the corresponding total wa-
tershed area, which turns out to be insensitive to the DEM resolution. The average values of
CA, on the other hand, increase with the DEM resolution going coarser. It is clear that the
minimum and average values of CA are greatly affected by the DEM resolution.
The relationship of CA and DEM resolution can be denoted as follows:
2
CAmin1 ⎛ r1 ⎞
=⎜ ⎟ (2)
CAmin 2 ⎝ r2 ⎠
CAmax1 ≈ CA max 2 (3)
CAmean1 r1
= (4)
CAmean 2 r2
where CAmin1 and CAmin 2 are the minimum values of upstream catchments area for the
resolutions of r1 and r2. The ratio of CAmin1 and CAmin 2 equals to the square of the ratio of
the respective grid lengths. CAmax1 and CAmax 2 are the maximum values with the resolutions
of r1 and r2. They are not affected by the DEM resolutions. CAmean1 and CAmean 2 are the
average values of upstream catchments area for the resolutions of r1 and r2. The ratio of the
two values equals to the ratio of their respective grid lengths.
Figure 4 shows the accumulative histogram of CA at different DEM resolutions. The CA
varies regularly with the DEM resolution and can be described as the ‘trumpet-shaped’ scale
effect. The CA values at the outlet of the valley are the same for all DEM scales, which
mean the CA at the outlet of a valley is insensitive to the DEM resolution. The most obvious
difference appears in the front of the “trump”, where the CA values are relatively small,
which means their locations are in the upslope areas. It reflects that the effect of DEM reso-
lution on CA mainly act on the upstream slope area. In these areas, with DEM grid size in-
creasing, CA becomes larger; the proportion of CA minimal values gets smaller. Therefore,
it inevitably causes the rise of the accumulative CA value as shown in Figure 4. In a position
of watershed CA value is the product of accumulated grid number along flow direction and
DEM grid size, hence, when the CA value is less than the grid size, it will be ignored. In this
case, the small values of CA could be represented only with high-resolution DEM data.

Figure 4 Accumulative histogram of the CA at different DEM resolutions

YANG Xin et al.: The scaling method of specific catchment area from DEMs 695

Figure 5 clearly shows the representation of CA at different DEM scales. Since the area
for each individual unit of the terrain representation is the square of the DEM resolution, the
minimum CA must be the square of the DEM resolution. Therefore, when the DEM resolu-
tion is 100 m, the minimum CA is 10,000 m2. Figure 5 shows that the CA values in almost
97.5% of the area are under 10,000 m2 for the 5-m resolution (Figure 5a). However, for the
100-m resolution, as compared, the CA values which are less than 10,000 m2 have almost
disappeared (Figure 5b). It could be further concluded that: 1) The effect of DEM scale on
CA in the area of upstream slope is more serious than it is in the gully area; 2) the effect of
DEM scale on CA reduces from upstream to downstream. Therefore, the CA scaling model
must take the spatial distribution of the scale effect into consideration.

Figure 5 Comparison between the CA at different DEM scales (Light gray color means CA value less than
10,000 m2, while dark grey color denotes CA value equal to or greater than 10,000 m2)

A logarithmic relationship between the CA at the location i (CAi) and the percentage of
that flow accumulative value (Pi) is found by fitting the curves in Figure 4.
Pi = alnCAi + b (5)
where a and b are the key coefficients corresponding to different DEM scales. Pi is the ratio
of the upstream cumulative CA at the location i and the total area. It also equals to the ratio
of the flow accumulation number Ni at location i and the total upstream grid number N0 at
outlet, since grid size is fixed:
Pi = Ni /N0 (6)
Therefore, the upstream CA at location i can be expressed as:
Ni / N 0 −b
CAi = e a (7)
It shows that CAi is closely related to Pi, which is consistent with Figure 4. When a point is
in the upstream area, the CA value is smaller and the impact of DEM scale is serious; when
it is in the downstream area, the CA value is larger, and turns out to be less affected by DEM
scale. The effect of DEM scale on CA, which could be distinguished as the ‘trump type’, can
be described in a negative exponential relationship with Pi.

4.2 Scale effect of SCA

SCA is defined as the ratio of a contour’s upstream catchment area and its length. The unit
contour length equals to the length of DEM grid width or its 2 length in the diagonal di-
696 Journal of Geographical Sciences

rections. Table 3 shows the basic information of SCA at each DEM scale. Different from CA,
the minimum value of SCA is not equal to the grid cell size. It is normally smaller than the
grid size, since the real value is also related to the length of unit contour. If the flow direc-
tion is the diagonal direction, the unit contour length is 2 of the grid length, and SCA is
1/ 2 of the CA. The maximum of SCA, appearing at the stream outlet, varies significantly
with the DEM resolution. When DEM resolution becomes coarser, the SCA at stream outlet
gradually decreases as the consequences of the unit contour length (the unit contour length
increases with the grid length increases).

Table 3 Statistics of SCA at different DEM scales (m2)

Resolution (m) SCA Min SCA Max SCA Ave
5 3.54 13777710 6515.5
10 7.07 6900950 6483.5
20 14.14 3445980 6442.5
30 21.22 2295990 6353.1
40 28.29 1722920 6469.9
50 35.36 1367800 6543.9
60 42.43 1150260 6401.0
70 49.50 980210 6381.8
80 56.58 860000 6307.9
90 63.65 761220 6259.9
100 70.72 679800 6467.1

The rule of multi-scale SCA can be concluded as follows:

SCAmax1 1 r
= = 2 (8)
SCAmax 2 r1 / r2 r1
SCAmin1 r1
= (9)
SCAmin 2 r2
SCAmean1 = SCA mean 2 (10)
where SCAmax1 and SCAmax 2 are the maximum values appear at the stream outlets with the
resolution of r1 and r2. The ratio of them is the reciprocal of the ratio of their grid lengths.
SCAmin1 and SCAmin 2 are the minimum values appear at the first water flow position with the
resolution of r1 and r2. The ratio of them is the ratio of their grid lengths. SCAmean1 and
SCAmean 2 are the average values, which are not affected by DEM scale. The minimum value
of SCA increases and the maximum value decreases with the increasing of DEM grid size,
and they all have a trend to approach average value, which is different from the scale effect
of CA.
The proportion of the SCA low value gradually increases with the decrease of DEM reso-
lution (Figure 6), but its amplitude is less than that of CA. For the resolution of 5 m and 100
m, the differences of the accumulative percentages of the minimum values within 0.5%, ob-
viously less than that in the case of CA, 7.2%.
YANG Xin et al.: The scaling method of specific catchment area from DEMs 697

Figure 6 Accumulation histogram of SCA at different DEM resolutions

Compared with the value of CA which has a regular trend with DEM resolution, the scale
effect of SCA shows an unstable characteristic as shown in Figure 6. The difference is ap-
parently larger in certain resolutions like 20 m to 30 m, and 40 m to 50 m.
The unit contour length is a constant value for each scale, so CA is the main factor af-
fecting the value of SCA. Moreover, the scale effect of CA can be distinctively observed in a
regular pattern. In this way, once the CA downscaling model is properly established, it could
be utilized in the downscaling of the SCA.

5 The scaling model

Since the minimum value of CA in a watershed equals to the area of the corresponding grid,
the CA value would change with the DEM resolution. Before establishing the downscaling
model, we define the number that a coarse-scale grid contains fine-scale grids as the factor
Ns (Figure 7). Pradhan (2006) has used Ns to build a downscaling model for CA. However,
that model is rather complicated to be easily understood. This paper proposes a simplified
downscaling model of SCA based on the study of the scale effect of SCA.

Figure 7 Concept map of scale effect factor

Ns is the scale factor, equal to the squared ratio (Rf) of the coarse grid (R) and the fine grid (R*) at two different
DEM scales.

Based on the effect of DEM scale on CA, there are four different relations:
(i) The ratio of the two minimum values of CA is the squared ratio of their respective
resolutions (equation 2), and equals to Ns. It indicates that the minimum value is totally af-
fected by the DEM scale.
(ii) The maximum values keep stable (equation 3). It indicates that the maximum value is
independent to the DEM scale.
(iii) An exponential relationship (equation 5) exists between the CA (at any position of a
698 Journal of Geographical Sciences

watershed) and Pi (equation 6). It shows that the effect of DEM scale on CA at the position i
is also affected by the position parameter Pi.
(iv) A larger variance between two DEM scales would cause greater difference between
the two CA values. It shows that the impact of DEM scale on CA, in a specific location, is
related with the scale factor (Ns).
Thus, a CA downscaling model can be established:
CAi , scaled = CAi fi (11)
where i distinguishes positions in the watershed, CAi is the CA at position i with current
DEM resolution, CAi , scaled is the downscaled value or revised value of CA at position i. fi is
a coefficient correction, which meets the following conditions:
• If i is located at the top of a hill (flow accumulation equals to one), by which Ni = 1,
then fi = 1/Ns, which meets equation 2.
• If i is located at the outlet of a watershed, that is Ni = N0, then fi = 1, which meets
equation 3.
• If i is located at somewhere else, that is located between the top of a hill and a water-
shed outlet, by which 1< Ni < N0, then 1/Ns≤fi≤1, and fi can be expressed as the function of
the scale factor Ns and the location factor Pi:
Ni
1− P 1−
⎛ 1 ⎞ i ⎛ 1 ⎞ N0
fi = ⎜ ⎟ =⎜ ⎟ (12)
⎝ Ns ⎠ ⎝ Ns ⎠
It shows that the smaller the fi is, the larger the revised amount to CAi will be. Ni increases
gradually as a point closer to downstream, and fi increases sequentially, which means the
revised amount of CAi lessening and turning to zero when a point is close to a drainage out-
let. On the contrary, fi becomes lower and approaches to 1/Ns as Ni reduces. It supports the
regulation shown in Figure 4. In this case, equation 11 can be expanded as:
Ni
1−
⎛ 1 ⎞ N0
CAi , scaled = CAi ⎜ ⎟ (13)
⎝ Ns ⎠
Therefore, the downscaling model of SCA is:
CAi , scaled
SCAi , scaled = (14)
Lscaled
where SCAi , scaled is downscaled SCA at a target DEM resolution. Lscaled is the downscaled
unit contour length at a target DEM resolution, and can be expressed as:
L
Lscaled = (15)
Rf
where L is unit contour length at original DEM resolution, and Rf is the ratio of the DEM
resolution before and after downscaling.

6 Scaling results
6.1 Scaling results of CA

Table 4 shows all scaling results from each individual scale to a 5-m resolution. While DEM
YANG Xin et al.: The scaling method of specific catchment area from DEMs 699

Table 4 Statistics of CA after downscaled (m2)

Scaling resolution(m) CA Min CA Max CA Ave
10→5 25 69037400 45290.9
20→5 25 69080800 65214.7
30→5 25 69051600 83840.0
40→5 25 69088000 107229.2
50→5 25 69027500 130186.9
60→5 25 69051600 146918.8
70→5 25 68678400 166978.5
80→5 25 68998400 177029.7
90→5 25 68509800 207931.5
100→5 25 67980000 225776.5

Figure 8 Accumulative histogram of CA after scaling

resolution is 5 m, the minimum, the maximum and the average values of CA are 25 m2,
68,976,780 m2, and 39,019.20 m2, respectively. After the downscaling, the minimum and
maximum values are almost the same as the value at target resolution, while the CA average
value has a tendency of getting higher. This is due to the fact that the downscaling model
could revise the CA value at each grid cell, yet it could not change the DEM resolution. The
number of DEM grid cells is unchanged so that the average value of CA gradually increases.
Compared with Figure 4 which shows that the proportion of low value area decreases
with the DEM resolution going coarser, Figure 8 displays that the histograms of CA at each
resolution are revised well and close to target resolution after downscaled. It can also be
found that a better result could be generated if the target scale is closer to the original scale.

6.2 Scaling results of SCA

From Table 3 we know that the SCA minimum value increases while the maximum value
decreases when the DEM resolution becomes coarser. Table 5 shows both the minimum and
maximum values can be revised to the target scale with the downscaling model. However,
the average value of SCA turns higher after downscaled. As mentioned earlier, it is because
the model converted SCA value into a relatively reasonable one, but did not change its reso-
lution. In this case, the coarser the resolution, the fewer the DEM grid number will be, and
the average value of SCA will increase gradually.
700 Journal of Geographical Sciences

Table 5 Statistics of SCA after downscaled (m2)

Resolution(m) SCA Min SCA Max SCA Ave
5 3.54 13777710.00 6515.49
10→5 3.54 13794170.00 7301.28
20→5 3.54 13695030.00 10195.23
30→5 3.54 13653590.00 12880.52
40→5 3.54 13642040.00 16624.98
50→5 3.54 13108440.00 20302.26
60→5 3.54 13767400.00 22760.39
70→5 3.54 13655920.00 25932.47
80→5 3.54 13542320.00 26939.31
90→5 3.54 13701960.00 32399.99
100→5 3.54 13596000.00 35480.58

Figure 9 shows the spatial pattern of SCA at original resolution and its downscaling
process. The spatial distribution of SCA is very simple when the resolution is 100 m, with
only three classes in the histogram. After downscaled to 5 m, the SCA values are revised to
be lower in most slope areas and the spatial and histogram distribution becomes more simi-
lar to that of the destination scale. The spatial pattern of downscaled SCA seems more rea-
sonable than it was before.
It can also be seen that in some areas within the watershed, the downscaled SCA is yet
lower than the SCA at a 5-m resolution. It is because that although the downscale model
realizes the revision of SCA value in each DEM grid, however, the heterogeneity within
each grid is not considered in this model. When the original resolution and target resolution
are different largely, downscaling may not generate a reasonable result for lacking of het-
erogeneity details.

7 Conclusions and discussion

The study on Jiuyuangou Gully shows that the SCA changes regularly with DEM resolution.
The effects of DEM resolution on SCA can be summarized as follows: (1) The CA minimum
value equals to the DEM grid size and increases with the raise of the grid size. The ratio of
the minimum values at two different resolutions is the same as the ratio of the corresponding
resolutions. The maximum value of CA equals to the area of the entire watershed, it keeps
stable. The effect of DEM resolution on CA reduces from upstream to downstream. There is
an exponential function of CA in each grid and its accumulative frequency of flow in the
watershed. (2) For SCA. The minimum value is the ratio of the unit grid area and the grid
length in diagonal direction. It increases with the DEM resolution becoming coarser. The
ratio of the minimum values at two different scales equals to the ratio of their grid lengths.
The maximum value of SCA is the ratio of total drainage area and a diagonal length of an
individual grid. It decreases with the rise of DEM grid size. The ratio of SCA maximum
values at two different scales is the reciprocal ratio of their grid length. Both the minimum
and the maximum values approach to the average value with the decreasing of DEM resolu-
tion.
YANG Xin et al.: The scaling method of specific catchment area from DEMs 701

Figure 9 The comparison of SCA downscaled (The classification of SCA legend is designed according to the
minimal value of SCA at different DEM resolutions)

According to the regulation of SCA with DEM resolution and by introducing scale factor
and position factor, a downscaling model of SCA is put forward. This model can revise SCA
value from a coarse resolution to a finer one. Experiments in Jiuyuangou Gully show that
SCA after downscaling gets more reasonable than it was before. In aspect of eigenvalue,
statistic histogram and spatial distribution of the SCA after downscaled are well matched to
target scale. However, since the heterogeneity within the original gird is ignored in this
model, the downscaled SCA in upstream area is generally lower than it is in target scale.
Besides, larger difference between the original and target resolutions leads to higher scaling
702 Journal of Geographical Sciences

error.
This downscaling model includes a position factor Pi, which is the ratio of Ni (upstream
catchment grid numbers at point i) and N0 (upstream catchment grid number at the outlets of
a drainages area). Because the flow accumulation at any position must contribute to the flow
accumulation at the catchment outlet, the scaling model is only applicable to a complete
drainage area. Besides, since this model is established based on the region of a hilly area, the
result might be unreliable when it is applied to a gently relief region. The accuracy of the
scaling model will be seriously affected when the resolution of original scale is too large to
the resolution of target scales.

References
Chaplot V, 2005. Impact of DEM mesh size and soil map scale on SWAT runoff, sediment, and NO3-N loads pre-
dictions. Journal of Hydrology, 312(1–4): 207–222.
Chaubey I, Cotter A S, Costello T A et al., 2005. Effects of DEM data resolution on SWAT output uncertainty.
Hydrological Processes, 19: 621–628.
Chen Yan, Tang Guoan, Qi Qingwen, 2004. Analysis of slope change atlas drawing from digital elevation models
of different spatial scales. Journal of Huaqiao University (Natural Science), 25(1): 79–82.
Costa-cabral M C, Burges S J, 1994. Digital Elevation Model Networks (DEMON): A model of flow over hill-
slopes for computation of contributing and dispersal areas. Water Resources Research, 30(6): 1681–1692.
Duan J, Grant G E, 2000. Shallow landslide delineation for steep forest watersheds based on topographic attrib-
utes and probability analysis, In: Wilson J P, Gallant J C (eds.). Terrain Analysis: Principles and Application.
John Wiley & Sons Press, 311–330.
Erskine R H, Green T R, Ramirez J A et al., 2007. Digital elevation accuracy and grid cell size: Effects on esti-
mated terrain attributes. Soil Science Society of America, 71(4): 1371–1380.
Fairfield J, Leymarie P, 1991. Drainage networks from grid elevation models. Water Resources Research, 27(5):
709–717.
Fei Lifan, He Jin, Ma Chenyan et al., 2006. Three dimensional Douglas-Peucker algorithm and the study of its
application to automated generalization of DEM. Acta Geodaetica et Cartographica Sinica, 35(3): 278–284.
(in Chinese)
Freeman T G, 1991. Calculating catchment area with divergent flow based on a regular grid. Computer & Geo-
sciences, 17(3): 413–422.
Hutchinson M F, Dowling T I. A continental hydrological assessment of a new grid-based digital elevation model
of Australia. Hydrological Processes, 1991, 5(1): 45–58.
Jenson S K, 1994. Application of hydrologic information automatically extracted from digital elevation models. In:
Beven K J, Moore I D (eds.). Terrain analysis and Distributed Modelling in Hydrology. Chichester: UK John
Willy & Sons, 35–48.
Kienzle S, 2004. The effect of DEM raster resolution on first order, second order and compound terrain deriva-
tives. Transactions in GIS, 8(1): 83–111.
Le Coz M, Delclaux F, Genthon P et al., 2009. Assessment of Digital Elevation Model (DEM) aggregation meth-
ods for hydrological modeling: Lake Chad basin, Africa. Computers & Geosciences, 35(8): 1661–1670.
Li Shuangcheng, Cai Yunlong, 2005. Some scaling issues of geography. Geographical Research, 24(1): 11–18. (in
Chinese)
Li Tiejian, Wang Guangqian, Xue Hai et al., 2009. The spacial scale effects of runoff-sediment characteristics in
the hilly and gully area of the Loess Plateau. Science in China (Series E), 39(6): 1095–1103. (in Chinese)
Liu Xuejun, Lu Huaxing, Ren Zheng et al., 2007. Scale issues in digital terrain analysis and terrain modeling.
Geographical Research, 26(3): 433–442. (in Chinese)
YANG Xin et al.: The scaling method of specific catchment area from DEMs 703

Lv Guonian, Qian Yadong, Chen Zhongming, 1998a. Automated extraction of the characteristics of topography
from grid digital elevation data. Acta Geographica Sinica, 53(6): 562–568. (in Chinese)
Lv Guonian, Qian Yadong, Chen Zhongming, 1998b. Study of automated extraction of shoulder line of valley
from grid digital elevation data. Scientia Geographica Sinica, 18(6): 567–573. (in Chinese)
Matı’as L Ruffo, Germa’n A Bollero, David S Bullock et al., 2006. Site-specific production functions for variable
rate corn nitrogen fertilization. Precision Agriculture, 7: 327–342.
Meisels A, Raizman S, Karniel I A, 1995. Skeletonizing a DEM into a drainage network. Computer & Geo-
sciences, 21(1): 187–196.
Moore I D, Grayson R B, Ladson A R, 1991. Digital terrain modeling: A review of hydrological, geomorphologi-
cal, and biological applications. Hydrological Processes, 5(1): 3–30
O’Callaghan J F, Mark D M, 1984. The extraction of drainage networks from digital elevation data. Computer
Vision, Graphics, and Image Processing, (28): 323–344.
O’Loughlin E M, 1986. Prediction of surface saturation zone in natural catchments by topographic analysis. Water
Resources Research, 22(5): 794–804.
Palph K Straumann, Ross S Purves, 2007. Resolution sensitivity of a compound terrain derivative as computed
from Lidar-based elevation data. In: Lecture Notes in Geoinfomation and Cartography. Berlin & Heidelberg:
Springer, 87–109.
Pradhan N R, Tachikawa Y, Takara K, 2006. A downscaling method of topographic index distribution for match-
ing the scales of model application and parameter identification. Hydrological Processes, 20: 1385–1405.
Qin Chengzhi, Zhu Axing, Pei Tao et al., 2007. An adaptive approach to selecting a flow-partition exponent for a
multiple-flow-direction algorithm. International Journal of Geographical Information Science, 21(4):
443–458.
Quinn P F, Beven K, Chevallier P et al., 1991. The prediction of hillslope flow paths for distributed hydrological
modelling using digital terrain models. In: Terrain Analysis and Distributed Modelling in Hydrology. Chiches-
ter: John Willy & Sons, 63–83.
Ren Xiyan, Zhang Xuesong, Hao Fanghua et al., 2004. Effects of DEM resolutions on simulated runoff and
sediment yield. Research of Soil and Water Conservation, 11(1): 1–4. (in Chinese)
Saulnier G M, Obled C, Beven K, 1997. Analytical compensation between DTM grid resolution and effective
values of saturated hydraulic conductivity within the TOPMODEL framework. Hydrological Process, 11(9):
1331–1346.
Smith M P, Zhu A X, Burt J E et al., 2006. The effects of DEM resolution and neighborhood size on digital soil
survey. Geoderma, 137(1/2): 58–69.
Sun Liqun, Hu Cheng, Chen Gang, 2008. Effects of DEM resolution on the TOPMODEL. Advances in Water
Science, 19(5): 699–706. (in Chinese)
Svoray T, Markovitch H, 2009. Catchment scale analysis of the effect of topography, tillage direction and un-
paved roads on ephemeral gully incision. Earth Surface Processes and Landforms, 34(14): 1970–1984.
Tang Guoan, Zhao Mudan, Li Tianwen et al., 2003. Modeling slope uncertainty derived from DEMs in Loess
Plateau. Acta Geographica Sinica, 58(6): 824–830. (in Chinese)
Tarboton D G, Bras R L, Rodriguez-Iturbe I, 1997. A new method for the determination of flow directions and
unslope area in grid digital elevation models. Water Resource Research, 33(2): 309–319.
Thompson J A, Bell J C, Bulter C A, 2001. Digital elevation model resolution: Effects on terrain attribute calcula-
tion and quantitative soil-landscape modeling. Geoderma, 100: 67–89.
Vieux B E, 1991. Geographic information systems and non-point source water quality and quantity modeling. In:
Beven K J, Moore I D (eds.), Terrain Analysis and Distributed Modelling in Hydeology. Chichester, UK: John
Willy & Sons, 49–62.
Wolock D M, Price C V, 1994. Effect of digital elevation model map scale and data resolution on a topogra-
phy-based watershed model. Water Resource Research, 30(11): 3041–3052.
704 Journal of Geographical Sciences

Wu Fan, Zhu Guorui, 2001. Multi-scale representation and automatic generalization of relief based on wavelet
analysis. Geomatics and Information Science of Wuhan University, 26(2): 170–174. (in Chinese)
Wu S, Li J, Huang G, 2005. An evaluation of grid size uncertainty in empirical soil loss modeling with digital
elevation models. Environmental Modeling and Assessment, 10: 33–42.
Wu Xianfeng, Liu Changming, Wang Zhonggen, 2003. Effect of horizontal resolution of raster DEM on drainage
basin characteristics. Journal of Natural Resources, 18(2): 148–154. (in Chinese)
Yang Qinke, David Jupp, Guo Weiling et al., 2008. Genaralizing the fine resolution DEMs with filtering method.
Bulletin of Soil and Water Conservation, 28(6): 58–62. (in Chinese)
Yang Qinke, Li Rui, Liang Wei et al., 2006. Re-scaling lower resolution slope by histrogram matching. In: Pro-
ceedings of International Symposium on Terrain Analysis and Digital Terrain Modeling.
Yang Xin, Tang Guoan, Zhang Yong, 2007. Scale effect and transformation model of slope based on DEMs. Pro-
ceedings of SPIE, 6753(22): 1–15.
Yi Weihua, Zhang Jianming, Kuang Yongsheng et al., 2007. Effect of horizontal resolution on the watershed fea-
tures derived from DEM. Geography and Geo-Information Science, 23(2): 34–38. (in Chinese)
Zhang Xiaoyang, Drake Nick A, Wainwright John et al., 1999. Comparison of slope estimates from low resolution
DEMs: Scaling issues and a fractal method for their solution. Earth Surface Processes and Landforms, 24(9):
763–779.
Zhou Qiming, Liu Xuejun, 2006. Digital Terrain Analysis. Beijing: Science Press. (in Chinese)
Zhu Qing, Tian Yixiang, Zhang Yeting, 2005. The extraction of catchment and subcatchment from regular grid
DEMs. Acta Geodaetica et Cartographica Sinica, 34 (2): 129–133. (in Chinese)