1 Introduction

Great attention has been paid to sediment deposition and water quality degradation in the Three Gorges Reservoir. This is, because these factors are critical to the sustainable operation of the largest reservoir in the world. As the most significant sediment contributor, soil loss includes the transport of soil particles, as well as nutrients and pollutants (Schauble 1999). Soil loss is a serious problem arising from changes in land use, agricultural intensification, and possibly global climatic change (Yang et al. 2003; Amore et al. 2004). The sediments deposited into a reservoir reduce the water storage capacity necessary for flood prevention; in addition, non-point pollutants along with soil loss, degrade water quality, adversely affecting aquatic biota (Novotny and Olem 1994; Onyando et al. 2005). Thus, soil loss control is vital to the sustainable operations of the Three Gorges Reservoir. This problem has prompted the infusion of large investments into water and soil conservation programs in the region. The Three Gorges Reservoir Region measures around 84,000 km2, and the soil loss area is around 51,000 km2, accounting for 60.7% of the total area in the region. The total annual soil loss quantity is around 140 million tons, which accounts for 26% of the sediment delivery in the upper Yangtze River (Wang et al. 2007).

Watershed planning, conservation, and management are imperative to control soil loss and prevent further damage. Soil loss is a natural geomorphic process, which occurs in every part of the world (Butt et al. 2010). Its extent and magnitude are determined by natural factors, such as climate, soil, topography and vegetation; as well as human activities, including land use and management practices (Mutua et al. 2006). Natural factors are the potential conditions for soil loss, whereas human activities are the dominant conditions (Wang et al. 2005). For the conservation of soil resources in watersheds, priority should be given to promoting optimum land use and management practices. This should be done in accordance with natural factors in order to attain the maximum improvement in soil conservation with minimum capital investment (Mishra et al. 2007). The soil loss component determined by natural factors is called potential soil loss, and it can characterize the sensitivity of soil loss to human activities. Risk assessment on potential soil loss can assist in identifying critical soil loss-prone areas in watersheds for prioritization of land use and soil conservation practices (Wang et al. 2001). Watershed prioritization is the process of ranking the different areas of a watershed according to the risk of potential soil loss. Therefore, critical soil loss-prone areas must be identified in the Three Gorges Reservoir Region by assessing the risk of potential soil loss, to improve the water resource development, achieve sustainable land use, reduce sedimentation, and control non-point pollution. The objectives of the current study are to: (1) present a methodology, which integrates the revised universal soil loss equation (RUSLE) with the sediment delivery ratio (SDR) in a GIS environment to assess the spatially distributed risks of potential soil loss at a watershed scale; and (2) apply this methodology to identify critical areas for potential soil loss in the Xiangxi watershed, a typical watershed of the Three Gorges Reservoir Region.

2 Study Area

The Xiangxi watershed, the nearest one to the Three Gorges Dam (Fig. 1), lies between latitudes 30°57′ N and 31°34′ N as well as, longitudes 110°25′ E and 111°6′ E; it covers an area of 3,099 km2 (Ji et al. 2010). The terrain of the watershed is very steep. The elevation varies from 192 to 3,057 m, with a maximum height difference of 2,865 m. Parts with elevations above 1,200 m account for approximately 60% of the watershed, and areas with elevations below 800 m make up only 20%; meanwhile, parts with slopes above 15° account for 71.43%. The Xiangxi watershed has a humid subtropical climate with four distinct seasons. The mean annual rainfall is about 1,000 mm, 68% of which occurs between May and October. The steep terrain and abundant rainfall give rise to potential conditions for severe soil loss. The soil loss area accounts for 45.05% of the watershed. The annual soil loss is 11.63 million tons, along with about 116,300 tons of organic matter, 8,700 tons of nitrogen, and 4,700 tons of phosphorus (Hui et al. 2000; Sun 2008).

Fig. 1
figure 1

Location of the Xiangxi watershed

Full size image

Since the operation of the Three Gorges Dam in 2003, about 30 km of Xiangxi has become a stagnant river bay (Ji et al. 2010) and its water body has become eutrophic. Algae blooms almost every year (Cai and Hu 2006). This problem has aroused public concern over the water quality of the Three Gorges Reservoir. Although stagnant water flow induced by the impoundment of the Three Gorges Reservoir is an important factor in the eutrophication of Xiangxi Bay, excess nutrients that originate mainly from non-point pollution along with severe soil loss, are the intrinsic causes behind such phenomenon. Thus, the Xiangxi watershed was selected for a case study of potential soil loss risk assessment to provide a systematic guide for land use and soil conservation. The current study is an innovative demonstration for soil loss research in the Three Gorges Reservoir Region. It is also expected to contribute to eutrophication control in the Xiangxi Bay.

3 Data and Methods

3.1 Risk Assessment Model for Potential Soil Loss

Soil loss is a complex dynamic process by which productive surface soils are detached, transported, and accumulated into a water body. It involves two processes: soil erosion and sediment delivery. Soil erosion is the process, by which soils are detached from the ground surface; whereas sediment delivery is the process, by which the detached soil particles are transported into a water body (Jain et al. 2003). Models for soil erosion estimation that are available in literature can be grouped into two categories: the physical and the empirical models (Bhattarai and Dutta 2007). Physical models, such as WEPP (Nearing et al. 1989) and EUROSEM (Morgan et al. 1998), represent the essential mechanisms controlling soil erosion by solving the corresponding equations. Their use, however, requires many parameters related to each process. For example, WEPP requires as many as 50 input parameters. The lack of data sets for a model simulation constrains the application of these physical models. Simple empirical methods, such as the universal soil loss equation (USLE) (Wischmeier and Smith 1965) and the revised universal soil loss equation (RUSLE) (Renard et al. 1991), may not reflect actual erosion processes because these are based on coefficients that are computed or calibrated on the basis of observations. Nevertheless, they are still the most frequently used equations in erosion studies mainly because of their simple and robust structure, easily available data, and success at predicting average long-term erosion at watershed to catchment scales (Bartsch et al. 2002; Dabral et al. 2008; Ismail and Ravichandran 2008). The original RUSLE is a lumped model, which estimates the total amount of soil erosion in a watershed. However, advancements in the use of GIS and RS can make the RUSLE run in a spatially distributed manner to estimate the soil erosion amount in any spatial cell at the watershed scale. As such, the identification of critical soil erosion areas becomes feasible using the spatially distributed RUSLE. Thus, spatially distributed RUSLE model was chosen in this work to describe soil erosion, because its data requirements are not too complex or unattainable, and are relatively easy to parameterize with the use of GIS and RS. Detached soil particles are transported by overland flow; however, only a fraction reaches the river because of the reduced flow momentum at relatively flat regions and vegetation barriers, leading to deposition (Onyando et al. 2005). SDR, which measures the sediment transport efficiency, is commonly used for sediment delivery estimation; it reflects the relative amount of eroded soil that reaches a water body from the eroding sources compared with the total amount of soil detached from the same area (Mutua et al. 2006). Thus, soil loss into a river can be estimated by routing the soil erosion computed by the USLE/RUSLE with using SDR (Bhattarai and Dutta 2007). Upon the establishment of the SDR, many lumped empirical equations that estimate SDR have been developed, and are integrated with the lumped USLE/RUSLE to estimate the total amount of soil loss in a watershed. Recently, some spatially distributed SDR models have been developed and combined with the distributed RUSLE to estimate the soil loss amount in any spatial cell at the watershed scale. Thus, a spatially distributed SDR model was chosen to describe sediment delivery in this work.

The RUSLE provides long-term estimates of the average annual soil erosion for a given site as a product of five multiplied factors: the rain and runoff erosivity (R), the soil erodibility (K), the slope steepness and length (SL), the vegetation cover (C), and the support practice (P) (Dabral et al. 2008). Among these factors, R, K, and SL are determined by precipitation, soil properties and topography, respectively; these are natural watershed properties minimally altered by human activities. C is determined by vegetation, which is influenced by natural conditions and human activities. However, the native vegetation and regeneration capacity of a watershed are dominated by natural conditions; thus, C can also be regarded as a natural factor. P is determined by human activities including land use and management practices (Wang et al. 2005). The SDR describing the sediment delivery process strongly depends on the topographic and vegetation cover characteristics of an area (Sivertun and Prange 2003) that are controlled by the natural properties of the watershed. In brief, among these six factors, only P is dominated by human activities; the other five (R, K, SL, C, and SDR) are dominated by the natural properties of a watershed. Potential soil loss characterizes the soil loss component induced by natural watershed factors; thus, its risk should be a product of the five natural factors, which is expressed as follows:

$$ PR = R \cdot K \cdot SL \cdot C \cdot SDR $$
(1)

where PR is the risk of potential soil loss determined by the five risk factors. These factors greatly vary within the various sub-areas of a watershed because of the spatial variations in rainfall, soil, topography, and vegetation. Such variability has promoted the use of data-intensive distributed approaches. The use of GIS and RS can make Eq. 1 run in a spatially distributed manner at the watershed scale. Using a distributed approach allows the discretization of the watershed into homogeneous grid cells that represent the spatial heterogeneity of watershed rainfall, soil properties, topography and vegetation cover, as well as the spatial variability and interaction of erosion and sediment delivery processes (Bhattarai and Dutta 2007). In the GIS-assisted approach, each factor of the model is described in the form of digital maps; and these five digital layers of the equation are overlaid to obtain a spatially distributed risk of potential soil loss in a watershed (Fistikoglu and Harmancioglu 2002). This allows the identification of critical areas of potential soil loss for prioritization in land use and implementation of soil conservation practices.

3.2 Calculation of Risk Factors

Generation of the R Factor

R was estimated using a monthly rainfall model (Wischmeier and Smith 1978), whose data requirements are easier to attain than those of the classical EI30 model (Wischmeier and Smith 1965). The monthly rainfall model is given by:

$$ R = \sum\limits_1^{12} {\left[ {1.735 \times {{10}^{1.5 \times \lg P_i^2/P - 0.8188}}} \right]} $$
(2)

where P i (mm) represents the long-term mean jth monthly precipitation, and P (mm) represents the long-term mean annual precipitation.

The R values of 25 rainfall stations were calculated using Eq. 2, based on the long-term mean annual and monthly precipitation of these rainfall stations. Their locations are shown in Fig. 2a. The point R values of the rainfall stations are shown in Table 1.

Fig. 2
figure 2

Input data sets of the Xiangxi watershed. a rainfall stations; b soil types; c DEM; d average NDVI in 2009

Full size image
Table 1 R factor values of rainfall stations
Full size table

Generation of the K Factor

K was estimated using the model developed in the erosion-productivity impact calculator (Williams and Renard 1983), which requires only the percentage composition of soil particles and organic carbon. This model is more applicable in China. The equation is expressed as:

$$ \begin{array}{*{20}{c}} {K = \left\{ {0.2 + 0.3\exp \left[ {0.0256SAN\left( {1 - SIL/100} \right)} \right]} \right\}{{\left( {\frac{{SIL}}{{CLA + SIL}}} \right)}^{0.3}}} \\ {\left( {1.0 - \frac{{0.25C}}{{C + \exp \left( {3.72 - 2.95C} \right)}}} \right)\left( {1.0 - \frac{{0.7SN1}}{{SN1 + \exp \left( { - 5.51 + 22.9SN1} \right)}}} \right)} \\ \end{array} $$
(3)

where SAN, SIL, and CLA are the volume percentages of sand, silt, and clay, respectively; and C is the weight percentage of organic carbon; SN1 = 1−SAN/100.

The K factor value of every soil type in the Xiangxi watershed was calculated using Eq. 3. The results are shown in Table 2. The soil type vector map with the soil attributes of the Xiangxi watershed was obtained from the 1:1,000,000 Soil Database of China (Shi et al. 2002),Footnote 1 the result is shown in Fig. 2b.

Table 2 K factor values of different soil types
Full size table

Generation of the SL Factor

SL was calculated using the upslope area contributing method (Mitasova et al. 1996) for each cell based on the watershed DEM. This method takes into account the steepness, the slope length and the upstream water contribution area, and is suitable for modeling increased erosion in areas with concentrated water flow. It also drives the RUSLE model to act in a semi-distributed manner. The equation is given as:

$$ S{L_i} = \left( {m + 1} \right){\left[ {{A_i}/{a_0}} \right]^m}{\left[ {\sin {\theta_i}/{b_0}} \right]^n} $$
(4)

where A i (m2m−1) is the upslope contributing area per unit contour width, θ i (deg) is the slope, m and n are parameters, a 0  = 22.1(m) is the length, and b 0  = 9% is the slope of the standard USLE plot (Mitasova et al. 1996). The typical values for m and n are 0.4–0.6 and 1.0–1.4, respectively, depending on the prevailing type of flow. Lower values for m and n should be used for areas with prevailing dispersed flow, such as areas well covered with vegetation. Higher values should be used for areas with more turbulent types of flow caused by existing rills or disturbed areas (Mitasova et al. 2002). In the current study, the exponent constants for the entire watershed were considered for simplicity; the values m = 0.45 and n = 1.2 were used because the Xiangxi watershed primarily has natural vegetation.

The basic input data for the SL map consisted of the DEM of the Xiangxi watershed, with a cell resolution of 3 arc second in geographic unit; this was obtained from the 1:250,000 Topographic Data Base of the National Fundamental Geographic Information System of China (NFGIS, PRC). The data was projected in UTM-WGS84 coordinates, and the cell-size was 85.91278 m. The result is shown in Fig. 2c.

Generation of the C Factor

C can be inferred using NDVI-derived remotely sensed data (De Asis and Omasa 2007), and some NDVI-derived methods have been developed and applied in soil erosion studies (Van der Knijff et al. 2000; Lin 2002; Lin et al. 2008; Chou 2010). In the current work, the C factor was estimated using the model based on the vegetation cover ratio developed for the soil erosion study on the Wangjiaqiao watershed of the Three Gorges Reservoir Region (Cai et al. 2000). The equation is expressed as:

$$ \left\{ {\begin{array}{*{20}{c}} {C = 1} \hfill & {Vc = 0} \hfill \\ {C = 0.6508 - 0.3436\lg Vc} \hfill & {0 < Vc \leqslant 78.3} \hfill \\ {C = 0} \hfill & {Vc > 78.3\% } \hfill \\ \end{array} } \right.\% $$
(5)

where Vc (%) is the vegetation cover ratio calculated using the NDVI-based method developed for the vegetation cover study on the Minjiang catchment of the upper Yangtze River basin (Ma et al. 2001), whose latitudinal zone is the same as that of the Xiangxi watershed. The equation is given as

$$ Vc = {108}.{49}NDVI + {0}.{717} $$
(6)

The MODIS NDVI products are extensively used worldwide for monitoring vegetation conditions and displaying land cover and land cover changes because of their high temporal and spatial resolution and freely available data sets. In all, 23 MODIS/Terra 16-Day L3 NDVI images with a cell resolution of 250 m in the Xiangxi watershed in 2009Footnote 2 were selected to derive the annual average NDVI image (Fig. 2d). Using these data, the vegetation cover ratio map was generated according to Eq. 6.

Generation of the SDR Factor

The ratio of soil loss to surface soil erosion is termed as SDR. It implies that the gross erosion in a cell multiplied by its SDR value becomes the sediment yield contribution of that cell to the nearest river channel. Ferro and Minacapilli (1995) have hypothesized that the SDR in cells is a strong function of the travel time of eroded particles from given cells to the nearest downstream channel. Travel time strongly depends on the topographic and land cover characteristics of an area (Ferro and Minacapilli 1995). Based on their study, the empirical relationship below was adopted in the current study in order to generate a spatially distributed SDR factor at a watershed scale with modest input parameter requirements.

$$ SD{R_i} = \exp \left( { - \beta {t_i}} \right) = \exp \left( { - \beta \frac{{{l_i}}}{{\sqrt {{{s_i}}} }}} \right) = \exp \left[ { - \beta \left( {\sum\limits_{j = 1}^{{N_i}} {\frac{{{\lambda_{ij}}}}{{\sqrt {{{s_{ij}}}} }}} } \right)} \right] $$
(7)

where t i is the travel time (h) of eroded particles along the hydraulic path from the ith overland cell to the nearest river channel, and β is a coefficient considered as constant for a given watershed, which lumps together the effects of roughness and runoff along the hydraulic path. The travel time t i is assumed to increase as the hydraulic path length l i increases, and as the square root of the slope s i of the hydraulic path decreases. If the hydraulic path from cell i to the nearest channel cell traverses N i cells, the flow length of the jth cell located along hydraulic path i is λ ij , and the slope of the jth cell is s ij . Travel time t i can be estimated by obtaining the sum of the travel times through each of the N i cells located along hydraulic path i (Ferro and Porto 2000). The hydraulic path length and the slope were derived in ArcGIS using the DEM of the Xiangxi watershed (Fig. 2c) as basis. The sensitivity analysis of coefficient β showed that the sediment yield amount is not very sensitive to the value of β, which varied from 0.1 to 1.5, while the sediment yield amount varied by only 10%. Therefore, for simplicity, the β value can be taken as equal to 1 in the computation (Bhattarai and Dutta 2007).

4 Results and Discussion

4.1 Risk Maps

The R factor map was created by interpolating the point R values of the rainfall stations (Table 1) to every grid cell with a resolution of 85.91278 m using the Kriging interpolation method, the result is shown in Fig. 3a. The K factor map was generated based on the K factor value of every soil type (Table 2) and the soil type vector map (Fig. 2b), and was resampled into a raster grid image with a cell-size of 85.91278 m using the nearest neighbor algorithm, the result is shown in Fig. 3b. The SL factor map was generated using Eq. 4 after the slope and the unit contributing area were derived from the DEM (Fig. 2c) in ArcGIS, the result is shown in Fig. 3c. Once the vegetation cover ratio was calculated based on the 2009 average NDVI image (Fig. 2d) using Eq. 6, the C factor map was created according to Eq. 5 using the ArcGIS raster calculator, and a resampled cell size of 85.91278 m, the result is shown in Fig. 3d. The SDR factor map was generated according to Eq. 7 after the hydraulic path length and the slope were derived in ArcGIS using the DEM (Fig. 2c) as basis, the result is depicted in Fig. 3e. The aforementioned risk factor maps were multiplied together using the ArcGIS raster calculator in order to generate a spatial risk map of potential soil loss, with a cell-size of 85.91278 m and raster values ranging from 0 to 478.18. The result is depicted in Fig. 4.

Fig. 3
figure 3

Risk factor maps of potential soil loss in the Xiangxi watershed. a R factor; b K factor; c SL factor; d C factor; e SDR factor

Full size image
Fig. 4
figure 4

Risk map of potential soil loss in the Xiangxi watershed

Full size image

4.2 Classification of Potential Soil Loss Risk

Classifying the risk of potential soil loss into several categories by defining the ranges for different risk classes is necessary to identify the critical areas of potential soil loss clearly and easily. The ranges of the four classes are chosen after the mean and standard deviation of the computed soil loss potential risk values are determined. According to Sivertun and Prange (2003), only areas with risk values of more than two standard deviations above the mean risk value are classified as critical areas, whereas areas with risk values between one and two standard deviations above the mean are considered sub-critical areas. In addition, areas with risk values between zero and one standard deviation above the mean are called low-risk areas, and those with risk values below the mean value are classified as minimal-risk areas (Sivertun and Prange 2003). The values shown in Table 3 were chosen to define these risk classes.

Table 3 Risk classification of potential soil loss
Full size table

The classified potential soil loss risk map (Fig. 5) of the Xiangxi watershed was created through the ArcGIS “reclassify” tool, using the defined classification ranges as basis. After the classification and clipping off of the water areas, the proportion (in square kilometers) and percentage of the classes in the Xiangxi watershed were analyzed. The results (Table 4) suggest that the 138.84 km2 critical area (4.48% of the watershed area) and 187.49 km2 sub-critical region (6.05% of the watershed area) should be prioritized in terms of land use and soil conservation.

Fig. 5
figure 5

Classified risk map of potential soil loss in the Xiangxi watershed

Full size image
Table 4 Proportions of risk classes in the Xiangxi watershed
Full size table

The Xiangxi watershed can be divided into five sub-watersheds: Gufuhe, Xiangpinghe, Xiangping-Xiakou, Gaolanhe, and Xiakou down (Fig. 6). The distribution of critical and sub-critical areas in these sub-watersheds was analyzed. The results (Table 5) revealed that the overall percentage of critical and sub-critical areas in the Xiangpinghe and Gufuhe sub-watersheds was higher than the average of the Xiangxi watershed. This indicate that soil loss are more sensitive to human activities in these areas compared with the three other sub-watersheds, where the overall percentage of critical and sub-critical areas is lower than the average of the watershed.

Fig. 6
figure 6

Sub-watersheds of the Xiangxi watershed

Full size image
Table 5 Proportions of critical and sub-critical areas in the sub-watersheds
Full size table

4.3 Verification of Critical and Sub-critical Areas

The critical and sub-critical areas of potential soil loss identified by the risk assessment model are subject to errors. These are due to the inaccuracies inherent in each input data theme and the limitations of the methods used in deriving the values for each risk factor. Thus, a verification of the critical and sub-critical areas is necessary to determine their validity. Agricultural land is the most soil loss-prone area, because it is harvested regularly and is classified as open, unprotected land (Sivertun and Prange 2003). This means that the variations of agricultural land in critical and sub-critical areas can induce corresponding dissimilarities in soil loss.

The Xiangxi watershed has a hydrological station called Xingshan (see Fig. 6 for its location), which controls the sediment discharge of the Gufuhe and Xiangpinghe sub-watersheds. The sediment variations at the Xingshan hydrological station in 1985, 1995, and 2000 are shown in Table 6. The agricultural land in critical and sub-critical regions in the Gufuhe and Xiangpinghe sub-watersheds in 1985, 1995, and 2000 were extracted from the land use vector maps in the scale of 1:100,000 (Fig. 7)Footnote 3 (Liu et al. 2005). The results (Table 6) showed a good relationship between the variations of agricultural land area in critical/sub-critical regions and the variations of sediment discharge at the Xingshan hydrological station. The area of agricultural land in critical and sub-critical regions in 1995 was smaller than that in 1985, and the sediment discharge and concentration in 1995 was also less than those in 1985. In addition, the sediment discharge and concentration in 2000 increased along with the area increase of agriculture land in critical and sub-critical regions compared with those observed in 1995. This meant that the sediment discharge and concentration increased with the area increase of agricultural land in critical and sub-critical regions. Therefore, the results of the critical and sub-critical area identification of potential soil loss in the Xiangxi watershed are reasonable. However, because the results in this study are based on relatively rough DEM and NDVI images, better performance can be achieved if finer DEM and NDVI images are used.

Table 6 Variations of the agricultural land in critical/sub-critical regions and the sediment at the Xingshan hydrology station
Full size table
Fig. 7
figure 7

Agricultural land in the Gufuhe and Xiangpinghe sub-watersheds. a 1985; b 1995; c 2000

Full size image

4.4 Distribution of Critical and Sub-critical Areas

Soil loss includes the transport of soil particles, nutrients and pollutants. Thus, areas with severe soil loss are often the critical areas of non-point source pollution. Areas close to rivers are often critical, because most of the soil particles within these areas can reach water bodies, and the likelihood of accumulation in the ground surface is small (Sivertun and Prange 2003). These areas are commonly called non-point pollution sensitive zones, where land use plays a more important role in river water quality. The risk map of potential soil loss in the Xiangxi watershed verifies this viewpoint: areas with a high risk of potential soil loss are those located near the river channel. This implies that the parameter of cell distance to the nearest river channel is crucial in the risk of potential soil loss.

The statistical relationship between the cell distance to the river channel and the mean risk value of potential soil loss was analyzed (Fig. 8a). A negative exponentially descending statistical relationship was observed between the cell distance to the river channel and the mean risk value of potential soil loss within the region away from the river channel, whereas a positive ascending statistical relationship exists within the region near the river channel. This inferred that another parameter may have induced this statistical phenomenon. The cell distance to the river channel mainly influenced the SDR value, and the SDR equation showed that the slope also affected the SDR value. The statistical relationship between the slope and the mean risk value (Fig. 8b) indicated that the mean risk value linearly increased as the slope increased. The statistical relationship between the cell distance to the river channel and the mean slope value (Fig. 8c) showed that within the region near the river channel, the slope sharply increased as the cell distance increased; whereas the slope within the region away from the river channel fluctuated around 20° with a relatively narrow range. Therefore, the slope played a dominant role within the region near the river channel, resulting in an increase in the mean risk value as the cell distance grew; in comparison, the cell distance played a dominant role within the region away from the river channel, resulting in a decrease in the mean risk value as the cell distance increased (Fig. 8a–c).

Fig. 8
figure 8

Statistical relationships between a cell distance and risk; b slope and risk; c cell distance and slope; d annual precipitation and risk; e soil erodibility and risk; f vegetation cover ratio and risk

Full size image

Apart from the cell distance and the slope, three other natural impact factors (rainfall, soil properties, and vegetation cover) also affected the risk of potential soil loss. The statistical relationships between the mean risk value and the three other parameters (long-term mean annual precipitation, soil erodibility, and vegetation cover ratio) were also analyzed (Fig. 8d–f). No obvious regularity in the three statistical relationships was observed. Therefore, in the Xiangxi watershed, the cell distance to the river and the slope played the most important roles in the spatial risk distribution of potential soil loss.

Furthermore, a statistical analysis between the critical/sub-critical area distribution and the two most important parameters (the cell distance and the slope) was conducted. The results are shown in Tables 7 and 8. Table 7 showed that 93.87% of the critical areas and 82.91% of the sub-critical areas are distributed in the zone with a cell distance below 2,000 m, whereas the zone with a cell distance below 2,000 m accounts for only 37.02% of the entire watershed. In the zone with a cell distance below 1,000 m, 77.47% of the critical areas and 55.39% of the sub-critical areas are found, whereas the zone with a cell distance below 1,000 m accounts for only 20.33% of the entire watershed. The statistical relationship between the cell distance and the mean risk (Fig. 8a) showed that the mean risk values in the zones with a cell distance below 2,000 m were greater than 7.96, and those in the zones with a cell distance below 1,000 m were greater than 12.55. These two risk values were also greater than the mean risk value of the entire Xiangxi watershed (7.53). Therefore, in the Xiangxi watershed, the region with a cell distance below 2,000 m can be classified as a non-point pollution sensitive zone, and the region with a cell distance below 1,000 m can be considered as a non-point pollution sensitive core zone, to which more attention should be paid in terms of land use and soil conservation. Table 8 showed that 74.93% of the critical areas and 70.01% of the sub-critical areas were distributed in the zone with slopes above 25°, whereas the zone with slopes above 25° accounts for 41.19% of the entire watershed. The statistical relationship between the slope and the mean risk (Fig. 8b) showed that the mean risk values in the zones with slopes above 25° were greater than 8.38, which was also greater than the mean risk value of the entire Xiangxi watershed (7.53). Therefore, the critical and sub-critical areas of potential soil loss were found mainly in the zone with slopes above 25°, where forests and grass with soil conservation functions should be given priority in land use.

Table 7 Statistical analysis of the relationship between risky areas and cell distance
Full size table
Table 8 Statistical analysis of the relationship between risky areas and slope
Full size table

4.5 Data Discussion

Four primary input data themes were used in assessing the risk of potential soil loss: the precipitation data, soil data, DEM, and MODIS NDVI images. The R factor was derived from the precipitation data; K from the soil data; C from the MODIS NDVI images; and DEM was used to derive SL and SDR. The precipitation data was obtained from the long-term record of rainfall gauges. The soil data were obtained from the 1:1,000,000 Soil Database of China. This database has been created based on the data of the second soil survey of China, and is also the most authoritative soil data in China at present. The DEM was obtained from the 1:250,000 Topographic Data Base of the NFGIS, PRC, which was generated based upon the contour, the elevation point, the bathymetric line, the water-depth point and the hydrographic net map provided by the Chinese State Bureau of Surveying and Mapping (CSBSM). The DEM data originate from the terrestrial survey map and has been validated by CSBSM; thus, the data quality is higher than some freely available DEM datasets originating from satellite radar remote sensing, such as the SRTM DEM and the ASTER GDEM. In these RS DEM data, local anomalies such as sinks and peaks are inevitable, and should be filled to ensure a proper delineation of basins and streams. The analysis of local anomalies on the SRTM DEM and the ASTER GDEM of the Xiangxi watershed, showed that there were large number of sinks and peaks. If these are not filled, a derived drainage network may be discontinuous. Moreover, a comparison between the drainage networks derived from these three DEM datasets and the one in the satellite photo image with a 10 m spatial resolution of the Xiangxi watershed showed that the spatial accuracy of the DEM obtained from the NFGIS, PRC at the scale of 1:250,000 was the highest. Global MODIS vegetation indices are designed to provide consistent spatial and temporal comparisons of vegetation conditions. Their NDVI products contain atmospherically corrected bi-directional surface reflectances masked for water, clouds, and cloud shadows. In the current study, MODIS/Terra 16-Day L3 NDVI images were obtained from Version-5 MODIS/Terra Vegetation Indices products, whose uncertainties are well defined over a range of representative conditions. In all, three products with different spatial resolutions of 1,000, 500 and 250 m, respectively, are available. Of these, the 250 m spatial resolution product was selected in the current study.

However, the resolutions of DEM and NDVI are relatively rough, thereby limiting the spatial accuracy of the current study. In future, better data provision in the form of remote sensing and LIDAR data for an improved digital terrain model and vegetation data should be considered to improve risk map resolutions.

5 Conclusions

The current study aimed to identify the critical areas of potential soil loss in the Xiangxi watershed by developing a risk assessment model for potential soil loss based on the RUSLE and SDR. The developed model consists of five multiplied risk factors, including the rain and runoff erosivity, soil erodibility, slope steepness and length, vegetation cover, and sediment delivery. The model, integrated in a GIS environment, can be used to facilitate fast and efficient risk assessment on potential soil loss with modest data set requirements. It can also be used to identify critical areas with high potential for soil loss. Thus, the model can serve as a useful tool in land use planning and soil conservation practices for large-scale watersheds.

Precipitation data, soil data, DEM and MODIS NDVI images were used to generate a risk map of potential soil loss in the Xiangxi watershed through the ArcGIS platform. The risks of potential soil loss were classified into four classes based on the mean and standard deviation of the risk values. The results showed that the critical and sub-critical areas accounted for 4.48% and 6.05%, respectively, of the entire watershed area. These areas should be given more importance in terms of land use and management practices. The identification of critical and sub-critical areas was verified by analyzing the relationships between the variations of the agricultural land area and those of sediment discharge. Furthermore, the critical and sub-critical area distribution was analyzed. The critical and sub-critical areas of the Xiangxi watershed assemble in the zone with a cell distance below 2,000 m, or in the zone with slopes above 25°.

The rough DEM data and NDVI images limited the spatial accuracy of this study. However, the resolution can be improved easily using the developed GIS-implemented model if more accurate input data are available. In future, better data provision in the form of remote sensing and LIDAR data for an improved DEM and NDVI data should be used.