Risk Assessment and Prediction of Soil Water Erosion on the Middle Northern Slope of Tianshan Mountain

Abstract

Soil erosion is a significant form of land degradation worldwide, leading to ecological degradation and a decline in agricultural productivity. The middle section of the northern slopes of Tianshan Mountain (MNSTM) in northwestern China is a high-priority area for soil water erosion prevention, and soil water erosion is a serious problem in the region. Despite this, there is a lack of research on soil water erosion in the MNSTM, and the trends and priority risk areas of soil water erosion remain unclear. Therefore, this study used the Revised Universal Soil Loss Equation (RUSLE) to quantitatively assess soil water erosion from 2001–2020 and predict it from 2030–2050. The study also used the Geodetector method to analyse the influencing factors of soil water erosion in the region. The results show that soil water erosion in the MNSTM has a fluctuating upward trend, increasing at a rate of 0.26 t hm⁻² y⁻¹ over the period 2001–2020 and reaching a maximum value of 39.08 t hm⁻² in 2020. However, soil water erosion in the region is mitigated under both RCP2.6 and RCP4.5 climate scenarios. Vegetation was found to have the highest degree of influence on soil erosion, indicating that its protection and management should be prioritised for future soil and water conservation efforts. The eastern part of the MNSTM was identified as the most vulnerable area to soil and water erosion, and in the context of global climate change, it is crucial to enhance the ecological restoration of the MNSTM to reduce the risk of soil water erosion. These findings can serve as valuable information for decision makers to develop effective strategies to prevent soil erosion and improve the ecological environment in the MNSTM.

1. Introduction

Soil erosion is a widespread phenomenon that results in the loss of soil by water or wind [1,2,3]. Currently, more than 1643 million ha of land worldwide are affected by soil erosion [4]. It is one of the most significant forms of land degradation in the world. Soil erosion leads to increased land desertification, reduced land productivity and loss of soil and water resources [5,6,7]. These problems significantly impede the achievement of environmental protection and sustainable development goals. With the intensification of global climate change in recent years, terrestrial ecosystems face a greater potential threat. Studies have shown that the climate in the arid zone of Central Asia has become warmer and wetter in recent years [8,9,10]. This change is expected to increase soil water erosion in the arid zone of Central Asia. The Middle North Slope of Tianshan Mountain (MNSTM) is located in the arid zone of Central Asia and is a key area for soil water erosion prevention in China. Unfortunately, soil water erosion in this area is very severe, making it a critical environmental problem.

Soil water erosion is usually estimated using both field observations and soil erosion model simulations [11,12,13], and although the amount of soil erosion obtained using field observation methods is more accurate, this approach is not applicable to soil erosion assessments on a regional scale [14]. In recent years, with the rapid development of remote sensing (RS) technology, it has become possible to achieve soil water erosion estimations on a large regional scale [15]. Several soil erosion models have been developed and applied, but their performance varies widely between regions. A meta-analysis by Ma et al. [16] studied soil erosion globally and found that the Revised Universal Soil Loss Equation (RUSLE) is found to have higher performance in the assessment of soil water erosion in arid areas, and the RUSLE model is also the most widely used soil water erosion model [17]. Angima et al. [18] used the RUSLE model to assess soil erosion in the central Kenyan highlands. Similarly, Ganasri et al. [19] used the RUSLE model to assess soil erosion in the Nethravathi basin. In addition, Lu et al. [20] combined remote sensing (RS) with the RUSLE model to study soil erosion in the Amazon Basin.

Global climate change will destabilise ecosystems, particularly in arid and semi-arid regions that are sensitive to climate change [21,22]. It has been shown that the temperature increase in the arid zones of Central Asia is higher than the global average, and this gap will continue to widen as the rate of global climate change accelerates [22,23]. In order to assess the impact of global climate change on soil erosion in arid zones, DouXin et al. [24] predicted soil erosion trends in Central Asia, but the results obtained were too coarse for the smaller regional of the MNSTM areas to help MNSTM soil and water conservation work; thus, it is necessary to use high-resolution remote sensing data for a fine-grained assessment of soil and water erosion.

The MNSTM, the most economically important region in Xinjiang, China [25], is facing significant challenges from soil water erosion, which has severely affected the development of local agriculture and livestock. Therefore, it is essential to identify the critical conservation areas of soil water erosion in the MNSTM and predict the trend of soil water erosion changes in the context of global climate change to support regional erosion control and ecological restoration [26]. However, despite the urgent need, few studies have focused on soil water erosion in the MNSTM. This study aims to address this research gap by (1) quantitatively assessing the spatial and temporal characteristics of soil water erosion in the MNSTM from 2001 to 2020, (2) predicting the trends of soil water erosion in the MNSTM under two climate scenarios, and (3) investigating the environmental factors affecting the spatial variation of soil erosion in the region and quantitatively assessing the risk of soil erosion. The results of this study will enable us to identify key areas for soil and water conservation management and provide scientific evidence for the rational allocation of local soil and water resources.

2. Materials and Methods

2.1. Study Area

The MNSTM is located in the Xinjiang Province, north-western China, in the central part of the Asian continent. Its geographical location is 84°50′–88°58′ E; 42°55′–44°22′ N; 334 km wide from east to west; and 161 km long from north to south, with a total area of about 28,010 square kilometres (Figure 1). The altitude is 472–5251 m. The climate is characterised by a temperate continental climate [27], with an average annual temperature of about 3.3 °C, an average summer temperature of 23.9 °C, and an average winter temperature of −12.5 °C. The average annual precipitation is 452.03 mm, with May–August accounting for about 67% of the total annual rainfall; annual evaporation is 1000–2000 mm [28,29]. Soil types mainly include grey-brown soil, black felt soil, grass felt soil, brown-grey soil, chestnut soil, calcareous soil, etc. The vegetation type is mainly herbaceous plants and trees, and the coverage of grasses is more than 80%. Mountainous areas have more vegetation, while plains have less. Human activities are mainly grazing [30].

2.2. Data Sources

2.2.1. Vegetation Data

Surface vegetation has a mitigating effect on soil water erosion [31,32], so it is necessary to obtain surface vegetation data as model input factors, and the rapid development of remote sensing technology makes it possible to obtain large-scale surface vegetation dynamics [33]. The normalized difference vegetation index (NDVI) is widely used to monitor the dynamics of surface vegetation. To obtain high spatial resolution NDVI data, we first collected Level-2 surface reflectance data from Landsat TM/ETM+ for 2001–2020 for calculating the NDVI at a 30 m spatial resolution. It is difficult to generate high-quality NDVI data with spatial and temporal continuity because Landsat remote sensing data are susceptible to interference from clouds and other factors [34]. Therefore, we also collected the MOD13Q1 NDVI product from the Moderate-resolution Imaging Spectroradiometer (MODIS) version 6 for 2001–2020, which has a spatial resolution of 250 m and a temporal resolution of 16 d. We performed a monthly maximum synthesis process for the MODIS NDVI. Both of the above remote sensing data were downloaded from the Google Earth Engine (https://earthengine.google.com/ (accessed on 4 October 2021)), and we used the ESTARFM method to fuse the two NDVI data for the same time period and finally obtained the 30 m spatial resolution of the 2001–2020 monthly NDVI data [35,36].

2.2.2. Other Data

The meteorological data for 2001–2020 used in this study were obtained from the ERA-5 land meteorological reanalysis data, a product with a spatial resolution of 0.1°, which proved to be very accurate in the Tien Shan Mountains [37]. We collected monthly total precipitation data. We collected future climate data for both RCP2.6 and RCP4.5 scenarios for soil erosion prediction. The future climate scenario data were obtained from Pan et al. [38].

Soil property data were obtained from the SoilGrids product, which provides information on soil organic carbon content, sand content, silt content, clay content, etc., at a 250 m spatial resolution. Terrain data were obtained from ASTER GDEM V2 with 30 m DEM data, which were downloaded from the Google Earth Engine (https://earthengine.google.com/ (accessed on 10 November 2021)).

The validation data for the Biome-BGC model came from the measured vegetation data. We designed 17 sampling plots of a 30 m × 30 m area size according to different altitudes and soil types, and set up three replicate 1 m × 1 m sampling squares at the diagonal ends and centre of each sampling plot to collect above-ground biomass data in each sampling square. Observations from the integrated soil erosion monitoring site in the study area were collected as validation data for the RUSLE model.

2.3. Methods

2.3.1. RUSLE Model

The Revised Universal Soil Loss Equation (RUSLE) model allows the simulation of soil water erosion by using data acquired through remote sensing and geographic information technology as model inputs (Figure 2). The RUSLE model has the following structure:

$$SE=R\times K\times LS\times C\times P$$

(1)

where SE denotes soil water erosion (t hm² y⁻¹); R denotes the rainfall-runoff erosivity factor; K denotes the soil erodibility factor; LS denotes the topography factor; C denotes the cover-management factor; P denotes the support practice factor. Since there are no relevant engineering measures in the region, the P factor is set to 1 in this study.

The rainfall-runoff erosivity factor (R) reflects the driving force of rainfall on soil water erosion [39,40], and we used the formula in the study of Prasannakumar et al. to calculate the R-factor [41], which has been demonstrated to be applicable in Central Asia [11]. The formula was:

$$R={\displaystyle \sum }_{i=1}^{12}1.735\times {10}^{\left[1.5\times {\log }_{10}\left(p_i^2/p\right)-0.08188\right]}$$

(2)

where i represents the month in a year; P_i is the precipitation in the month i; P is the total annual precipitation.

The soil erodibility factor (K) is an indicator of the susceptibility of the soil to erosion, with a larger K indicating a more erodible soil [42]. The calculation formula is:

$$\begin{array}{c}K=\left\{0.2+0.3exp\left[-0.0256S_a\left(1-\frac{S_i}{100}\right)\right]\right\}\times {\left(\frac{S_i}{C_l+S_i}\right)}^{0.3}\times \left[1-\frac{0.25C}{C+exp\left(3.72-2.95C\right)}\right]\\ \times \left[1-\frac{0.7\left(1-0.01S_a\right)}{S_n+exp\left(-5.51+22.9S_n\right)}\right]\end{array}$$

(3)

where Sa, Si, Cl and C are the mass percentages of sand (0.02–2 mm), silt (0.002–0.02 mm), clay (<0.002 mm) and organic carbon in the soil, respectively.

The topographic factor (LS) represents the effect of the slope length factor (L) and slope steepness factor (S) on the amount of soil water erosion, and the formula for the topographic factor is referred to by McCool et al. and Liu et al. [43,44]. The equations are as follows:

$$L={\left(\lambda /22.13\right)}^{\left[\beta /\left(\beta +1\right)\right]}$$

(4)

$$\beta =\left(\sin \theta /0.0896\right)/\left[3\times {\left(\sin \theta \right)}^{0.8}+0.56\right]$$

(5)

$$S=\left\{\begin{array}{c}10.8\times \sin \theta +0.03,\theta <9°\\ 16.8\times \sin \theta -0.5,9^{°}\le \theta <{18}^{\circ }\\ 21.91\times \sin \theta -0.96,\theta \ge {18}^{\circ }\end{array}\right.$$

(6)

where λ denotes the slope length and θ denotes the inclination angle.

The cover-management factor (C) indicates the mitigation effect of surface vegetation on soil water erosion and is usually calculated by vegetation cover. In this study, the vegetation cover factor for the historical period was calculated using the NDVI obtained from remote sensing [45], and the C factor for the future climate scenario was obtained using model simulations [46]. The C factor was calculated as follows:

$$C=\left\{\begin{array}{c}1,fvc=0\\ 0.6508-0.3436\times {\log }_{10}fvc,0\le fvc\le 78.3\%\\ 0,fvc>78.3\%\end{array}\right.$$

(7)

where fvc denotes the vegetation cover.

2.3.2. Biome-BGC Model

The Biome-BGC model is an ecological process model [47,48], developed by the Numerical Terradynamic Simulation Group (https://www.ntsg.umt.edu/ (accessed on 1 November 2020)), which allows the simulation of terrestrial vegetation growth dynamics by inputting meteorological data. We used climate scenario data for 2030, 2040 and 2050 as model input data, which in turn enabled the simulation of future surface vegetation dynamics, and then we used the data obtained from model simulations as vegetation input data for the RUSLE model, which enabled the prediction of future soil erosion (Figure 2). The vegetation parameters of the model were referred to by Han et al.’s studies [49,50].

2.4. Data Analysis

2.4.1. Trend Analysis

In this study, the annual trends of soil water erosion on the pixel scale and the MNSTM were analysed using a linear regression trend analysis [51]. The equation of this method is:

$$k_{slope}=\frac{n\times {{\displaystyle \sum }}_{i=1}^n\left(i\times A_i\right)-{{\displaystyle \sum }}_{i=1}^{n}i{{\displaystyle \sum }}_{i=1}^n{A}_i}{n\times {{\displaystyle \sum }}_{i=1}^n{i}^2-{\left({{\displaystyle \sum }}_{i=1}^n{A}_i\right)}^2}$$

(8)

where k_slope indicates the slope of the trend, n represents the length of time to be analysed and Ai indicates the amount of soil water erosion at year i. k_slope > 0 indicates an increasing trend, and k_slope < 0 indicates a decreasing trend.

2.4.2. Analysis of Spatial Characteristics and Influencing Factors

In this study, we used the spatial autocorrelation analysis tool in ArcGIS 10.8 software to determine whether there is a significant spatial aggregation of soil water erosion, and then used the hotspot analysis tool to identify areas of high and low value aggregation of soil water erosion. The principles and formulas of these two methods are detailed in the study by Said et al. [52].

We used the Geodetector method to quantitatively identify the influencing factors of soil water erosion, which focuses on quantifying the degree of influence of each factor by identifying the differences in the spatial distribution of soil water erosion and the spatial distribution of each influencing factor [53,54], and we completed this analysis using the Geodetector software (http://www.geodetector.cn/ (accessed on 15 November 2021)). The equation of the method is:

$$q=1-\frac{{{\displaystyle \sum }}_{h=1}^L{N}_h{\sigma }_h^2}{N{\sigma }^2}$$

(9)

where q denotes the degree of influence of the influencing factor, q ranges from 0 to 1 and the larger q is the higher degree of influence. N denotes the number of samples, ${\sigma }^2$ denotes the variance and L denotes the number of influencing factors.

2.4.3. Soil Erosion Sensitivity

In order to further analyse the compound impact of each environmental factor on soil water erosion and the degree of risk of soil water erosion occurring in the region, we introduced the soil water erosion sensitivity evaluation index, and the higher the degree of sensitivity, the higher the risk of soil water erosion occurring in the region represented [55,56]. The equation of the method is:

$${\mathrm{SS}}_j=\sqrt[4]{{{\displaystyle \prod }}_i^4{S}_i}$$

(10)

where SS_j is denoted as the soil erosion sensitivity index on the jth pixel cell and S_i is the element i soil erosion sensitivity index value. The natural interruption method was used to divide the sensitivity index values into five intervals. The five intervals were defined as: not sensitive, mildly sensitive, moderately sensitive, highly sensitive and extremely sensitive.

3. Results

3.1. Accuracy of RUSLE Model and Biome-BGC Model

Figure 3 shows the comparison of the RUSLE and Biome-BGC models with the measured data. To verify the accuracy of our soil water erosion calculation, we collected soil water erosion data from field observations in the study area as well as measured data. The results show that the soil water erosion results obtained from the RUSLE model simulation are highly accurate, with an R² of 0.88 and an RMSE of 4.3481 t hm⁻², which can justify the use of the RUSLE model to simulate soil water erosion for the MNSTM. We verified the simulation accuracy of the Biome-BGC model by using the above-ground biomass of the vegetation at the sampling sites as the measured data, and obtained an R² of 0.85 and an RMSE of 6.169 gC m⁻², which indicates that the vegetation data obtained by using the Biome-BGC model are reasonable.

3.2. Spatial and Temporal Variation of Soil Water Erosion from 2001 to 2020

Figure 4 shows the annual spatial and temporal characteristics of soil water erosion in the MNSTM from 2001 to 2020, and it can be seen that the number of pixels showing a decreasing trend in the region is larger than the number of pixels showing an increasing trend, with 62.8% and 37.2% of the total pixels showing decreasing and increasing trends, respectively, and the highest percentage of the trend type has a decreasing rate between 0 and 0.1 t hm⁻² y⁻¹, accounting for about 51% of the total number of pixels. Although the percentage of pixels showing a decreasing trend is higher, the total annual soil water erosion in the MNSTM shows an increasing trend due to the higher values of pixels showing an increasing trend. The areas with an upward trend are mainly located in the central region with relatively high elevation, while the pixels with a downward trend are mainly located in the northern region with lower elevation. The values provide an indication of both erosion and soil accumulation (positive values). This is largely dependent on the condition of the underlying surface and differences in landscape type. From the time variation, the MNSTM soil water erosion showed a fluctuating upward trend in the last 20 years, with an increase rate of about 0.26 t hm⁻² y⁻¹, with the lowest soil water erosion in 2008, with a value of 23.56 t hm⁻², reaching the lowest value in 2007, and the highest soil water erosion in 2020, with a value of 39.08 t hm⁻².

3.3. Soil Water Erosion under RCP2.6 and RCP4.5 Climate Scenarios

Figure 5 shows the spatial distribution of soil water erosion under the RCP2.6 and RCP4.5 scenarios for 2030–2050. Under the RCP2.6 scenario, the average soil water erosion for 2030–2050 is 18.29 t hm⁻² and reaches its highest value of 27.53 t hm⁻² in 2040, with a significant concentration in the central-eastern region in all three years of high values. The average soil water erosion for 2030–2050 under the RCP4.5 scenario is 10.66 t hm⁻² and will be highest in 2050 with a value of 11.99 t hm⁻². Soil erosion in the RCP4.5 scenario is lower than in the RCP2.6 scenario. As with the RCP2.6 scenario, most of the high erosion areas in the RCP4.5 scenario are located in the eastern region, while the low erosion areas are concentrated in the northern region. As the topography of the MNSTM is characterised by a high south and low north, this results in a lower risk of erosion in the northern region.

Table 1. Area of change in soil water erosion for different classes compared to the reference time period under RCP2.6 and RCP4.5 scenarios.

Time Period	2030		2040		2050
Climate Scenarios	RCP2.6	RCP4.5	RCP2.6	RCP4.5	RCP2.6	RCP4.5
0–2 t hm−2	11.2%	17.46%	11.74%	14.29%	13.04%	12.25%
2–25 t hm−2	13.34%	19.41%	3.95%	14.96%	13.55%	18.55%
25–50 t hm−2	3.37%	−7.5%	4.36%	−0.77%	0.79%	−1.8%
50–80 t hm−2	−11.84%	−0.77%	−9.03%	−11.5%	−11.04%	−11.66%
80–150 t hm−2	−12.49%	−11.5%	−11.6%	−13.13%	−12.62%	−13.42%
>150 t hm−2	−0.36%	−13.13%	4.12%	−0.33%	−0.18%	−0.38%

shows the changes in the area of soil water erosion for each class compared to the base period (2020). We found an increasing trend in the area of soil water erosion occurring at lower levels (<50 t hm⁻²) for both climate scenarios, while the area of soil water erosion occurring at higher levels (>50 t hm⁻²) has decreased.

3.4. Regional Assessment of Soil Erosion Risk

The rainfall, soil, topography and vegetation of the last 20 years from 2001 to 2020 were selected as the environmental factors affecting soil water erosion, and the five environmental factors were graded using natural interruptions, and the grading criteria were determined as shown in

Table 2. Indicators to assess soil water erosion sensitivity.

	Not Sensitive	Marginally Sensitive	Moderately Sensitive	Highly Sensitive	Extremely Sensitive
Rainfall	<250	250–300	300–350	350–400	>400
Soil	0–0.015	0.015–0.023	0.023–0.027	0.027–0.030	>0.030
Topography	0–5	5–12	12–18	18–25	25–32
vegetation	<0	0–0.770	0.770–0.871	0.871–0.953	0.953–1
Degree	1	2	3	4	5

.

Soil erosion sensitivity in the middle part of the northern slope of Tianshan Mountain shows a high spatial distribution in the east and a low spatial distribution in the west, and the largest distribution area is the light sensitive area and the non-sensitive area, which account for 28.99% and 26.96% of the area, mainly in Shawan City, Hutubi County, Manas County and Changji City; the medium sensitive area and the highly sensitive area account for 21.53% and 16.94%, mainly distributed in Fukang City, Darban City and Midong District; 5.56% of the area of extremely sensitive areas, mainly distributed in the eastern part of Darban City and the southern part of Urumqi County (Figure 6). The reason why the extremely sensitive areas of soil water erosion are mainly distributed in the east is because the rainfall in the eastern region is larger and the terrain is more undulating, thus the eastern region is the key area for future soil and water conservation work.

4. Discussion

4.1. Environmental Factors Affecting the Spatial Differentiation of Soil Water Erosion

The spatial distribution of soil water erosion is influenced by rainfall, vegetation, soil and topography. In order to quantitatively evaluate the influencing factors of soil water erosion, we first calculated the average values of soil water erosion and each influencing factor from 2001 to 2020, and then graded the values of each factor using the natural intermittent grading method and inputted them into the Geodetector model.

Table 3. Degree of influence of environmental factors on soil water erosion.

	Rainfall	Soil	Slope Length	Slope	Vegetation
q	0.047	0.018	0.063	0.149	0.239
p	0	0	0	0	0

The p-value is < 0.01, which means that the q-value is highly significant.

shows the results of the Geodetector model. The driving ability of different influencing factors on the difference in the spatial distribution of soil water erosion exists differently, and the magnitude of their influencing ability is ranked as follows: vegetation > slope > slope length > rainfall > soil. The degree of influence of vegetation on soil water erosion is the largest, with a q-value of 0.239, indicating that vegetation has the strongest influence on the difference in the spatial distribution of soil water erosion; the degree of influence of soil on soil erosion is the smallest, with a q-value of 0.018, indicating that soil has the weakest influence on the difference in the spatial distribution of soil water erosion. This result is consistent with the findings of Yuan et al. [57] and Zhao et al. [58] in the Loess Plateau region of China, probably because the soils in both the Loess Plateau region and the MNSTM are susceptible to erosion and therefore, the spatial differentiation of soil water erosion in the region is less influenced by soil. While Guo et al. [59] used the Geodetector model to analyse the factors influencing soil erosion in the karst landscape region of southwest China, they found that topography and land use had the highest degree of influence on regional soil erosion. The results of combining the MNSTM soil water erosion spatial distribution with the Geodetector model show that, because the Tianshan Mountains are located in the arid zone of Central Asia, the spatial heterogeneity of vegetation cover is high, and at the same time, the altitude difference and topography in the region are more complex; thus, the vegetation cover and topography (slope and slope length) become the main control factors affecting the MNSTM soil water erosion spatial distribution.

The Geodetector model interaction detector mainly analyses the extent of the compound influence of two different influencing factors and determines whether the interaction of the two influencing factors enhances or weakens the effect on soil water erosion [60]. The results of the interaction detector showed that the interactions of the influencing factors were mostly non-linearly enhanced, and only the interactions between the two factors of slope and slope length were mutually enhanced. The two-factor interaction between vegetation and slope and vegetation and slope length had the highest driving effect on the spatial variation of soil erosion, with q-values of 0.579 and 0.380, respectively; the two-factor interaction between rainfall and soil had the lowest driving effect on the spatial variation of soil erosion, with q-values of 0.105 only (

Table 4. Interaction of soil water erosion influences.

q (Factor 1)	q (Factor 2)	q (Interaction)	Types
X1 = 0.047	X2 = 0.018	X1 ∩ X2 = 0.105	Enhance, Non-linear
X1 = 0.047	X3 = 0.063	X1 ∩ X3 = 0.117	Enhance, Non-linear
X1 = 0.047	X4 = 0.149	X1 ∩ X4 = 0.214	Enhance, Non-linear
X1 = 0.047	X5 = 0.239	X1 ∩ X4 = 0.309	Enhance, Non-linear
X2 = 0.018	X3 = 0.063	X2 ∩ X3 = 0.085	Enhance, Non-linear
X2 = 0.018	X4 = 0.149	X2 ∩ X4 = 0.194	Enhance, Non-linear
X2 = 0.018	X5 = 0.239	X2 ∩ X5 = 0.290	Enhance, Non-linear
X3 = 0.063	X4 = 0.149	X3 ∩ X4 = 0.182	Enhance, bi
X3 = 0.063	X5 = 0.239	X3 ∩ X5 = 0.380	Enhance, Non-linear
X4 = 0.149	X5 = 0.239	X4 ∩ X5 = 0.579	Enhance, Non-linear

X1 is rainfall; X2 is soil; X3 is slope length; X4 is slope; X5 is vegetation.

). By comparison with the study of Chao et al. [61], it was found that land use type has the highest degree of influence on soil erosion in the karst landscape, while the MNSTM has a high spatial heterogeneity of topography and vegetation, and therefore, vegetation and topographic conditions have the highest degree of influence on soil water erosion.

4.2. Impact of Climate Change on Soil Erosion

Climate change has a direct or indirect impact on soil water erosion, which is primarily driven by precipitation. Consequently, changes in precipitation patterns play an important role in determining changes in soil water erosion [62,63]. Favis-Mortlock and Boardman found that for every 7% increase in rainfall, soil water erosion would increase by 26% and for a 20% increase in rainfall, soil water erosion would increase by 37% [64]. Hu et al. [65] predicted soil water erosion on the Loess Plateau of China under RCP2.6 and RCP4.5 scenarios and found that soil erosion increased under RCP2.6 and decreased under RCP4.5. Additionally, rainfall and temperature also have an indirect impact on soil water erosion by affecting the surface vegetation cover, which is a key factor in determining soil and water conservation capacity [66]. For arid regions with fragile terrestrial ecosystems, the suitability of future climates for the growth of surface vegetation is a critical factor that impacts regional security [67,68]. According to Sheffield et al. [69], with global warming, there will be a significant decrease in rainfall at mid-latitudes in the Northern Hemisphere. Although this will reduce the amount of soil water erosion, this may also lead to the reduction of surface vegetation and the occurrence of land desertification [70]. A study on predicting soil erosion in Turkey found that climate change in some regions increased vegetation growth and that increased above-ground biomass led to a reduction in soil erosion [71]. This could also be the reason for the reduction in soil erosion in the MNSTM found in this study under both climate scenarios. On the other hand, global climate change is a highly complex and controversial process, and modelling and projections of the future climate are subject to a high degree of uncertainty, which may also lead to some biases in our projections. In future studies, we will use climate scenario data from multi-model ensembles to avoid the uncertainty in the experimental results caused by the bias in the data itself. There is a growing body of research suggesting that the implementation of ecological restoration projects can be effective in mitigating the negative impacts of climate change on ecological environments. China began a series of ecological restoration programs in 1980, which led to significant improvements in soil water erosion, soil wind erosion and land desertification [72,73,74], proving the effectiveness of these ecological restoration projects, and should continue to strengthen ecological restoration to protect the ecological environment in the future.

5. Conclusions

In this study, we used the RUSLE model to evaluate the spatio-temporal characteristics of soil water erosion in the MNSTM from 2001 to 2020 and to project future soil water erosion under two climate scenarios. Our results showed an increasing trend of soil erosion in the MNSTM over the last two decades, with an annual increase rate of 0.26 t hm⁻² y⁻¹ between 2001 and 2020, and then we projected a decrease in future soil erosion under both the RCP2.6 and RCP4.5 climate scenarios. We also used the Geodetector method to investigate the factors influencing soil water erosion and our results showed that vegetation has the greatest influence on soil water erosion. As the MNSTM is located in the arid zone of northwest China, where the ecosystem is fragile and vulnerable to climate change, the protection and management of vegetation can effectively prevent the escalation of soil water erosion and the onset of land desertification, thus protecting the ecological environment. Finally, we assessed the level of soil erosion risk in the region and identified the eastern region as the most vulnerable. Therefore, priority areas for soil conservation should focus on the east. Our study presents a methodology for predicting soil erosion at a small regional scale using the Biome-BGC model, which can take into account the impact of climate change on vegetation and can facilitate a more accurate prediction of soil water erosion, serving as a reference for predicting soil water erosion in other regions. We also analysed the main risk areas for soil erosion in the MNSTM, providing a scientific reference for regional ecological protection and soil and water conservation work.