Spatio-Temporal Variations in Soil Erosion and Its Driving Forces in the Loess Plateau from 2000 to 2050 Based on the RUSLE Model

Abstract

Assessing the spatio-temporal variability and driving forces of soil erosion on the Loess Plateau is critical for ecological and environmental management. In this paper, the Revised Universal Soil Loss Equation (RUSLE) model, the patch-generating land use simulation, and the Geographical Detector are used to investigate the spatio-temporal variations of the Loess Plateau’s soil erosion from 2000 to 2050. The results showed that: (1) The primary categories of soil erosion from 2000 to 2020 were moderate, mild, and slight, and the average level of soil erosion exhibited a decreasing and then an increasing tendency during the last 20 years. (2) Soil erosion was directly impacted by changes in land use, with cropland and forest being the primary land use and land cover changes in the study region. Cropland and construction land being turned into woodland between 2000 and 2020 resulted in a significant decrease in the severity of soil erosion. Projected soil erosion is expected to increase significantly between 2020 and 2050 due to arable land being converted into construction land. (3) The key variables impacting the spatial distribution of soil erosion were LUCC (Land-Use and Land-Cover Change), NDVI (Normalized Difference Vegetation Index), and slope, and the interplay of these variables may increase their ability to explain soil erosion. Grasslands with an NDVI ranging from 0.9 to 1, rain ranging from 0.805 to 0.854 m, a slope above 35°, and a terrain elevation ranging from 1595 to 2559 m were identified as having a high risk of soil erosion. Soil erosion prevention and management efforts should focus on the ecological restoration of upland areas in the future.

1. Introduction

As a natural occurrence of material movement on the surface of the ground [1,2,3,4], soil erosion can be viewed as a dynamic process of soil movement that is influenced by both natural and manmade influences. Soil erosion reduces soil fertility, raises riverbeds, silts up-river channels, intensifies the occurrence of natural disasters such as droughts, floods, landslides, and mudslides, leads to environmental degradation, and hinders economic and social development. China’s most serious place for soil erosion is the Loess Plateau [5], which is also crucial for water and soil conservation and ecological construction [6,7]. A comprehensive understanding and knowledge of the natural and topographical conditions, soil erosion laws, and land use suitability of the study area is the key to promoting rational land use, soil erosion prevention and control, and developing ecological construction plans according to local conditions [8,9,10].

On the Loess Plateau, there is currently a base of study on soil erosion, such as in the current situation and estimation of future conditions on soil erosion on the Loess Plateau. Chen et al. [11] used the RUSLE model in the Huangshui watershed to quantify the spatio-temporal variation in soil erosion rates. Based on an evaluation of the spatiotemporal characteristics of soil erosion modulus on the Loess Plateau from 1901 to 2016, Mu et al. [12] used the RUSLE model. The findings showed that human activities were the primary cause of changes in soil erosion modulus on the Loess Plateau. According to earlier research, historical spatio-temporal changes have been the main focus of soil erosion investigations, often neglecting future soil erosion changes. The patch-generating land use simulation (PLUS) model [13] is a recently created model that combines transformation analysis strategies exploring the driving mechanisms of land use change (e.g., the Logistic-CA model) and pattern analysis strategies based on the probability of occurrence and competition mechanisms (e.g., the CLUE-S model) with a dominant component. These strategies, combined with seed growth mechanisms and multi-objective algorithms, enable a more thorough examination of land use change [14]. Meanwhile, the PLUS model has a greater advantage in simulating and predicting soil erosion conditions under future LUCC development scenarios [15]. In identifying soil erosion drivers on the Loess Plateau [16], we investigated soil erosion in the Loess Hills and Gullies of northern Shanxi and pointed out that the vegetation cover type on steep slopes tended to be homogeneous, and the primary cause of soil erosion was changes in land use patterns. Keo et al. [17] investigated the 50-year variation in rainfall erosive force and its impact on Loess Plateau soil erosion and discovered that it is not the dominant factor influencing riverine sand transport. By identifying their spatially stratified heterogeneity, the geographical detector method can identify the driving forces behind geographic phenomena [18,19].

The warm temperate forest and farming subregion in the northeast of the Loess Plateau [20,21,22,23,24] mainly includes the southeast area, which comprises Henan Province’s northwest region, Hebei Province’s southwest region, and Shanxi Province. The three provinces are located in the eastern Loess Plateau, across the Yellow River Basin. The Ecological Restoration Plan of Territorial Space in the three provinces pointed out, while the 13th Five-Year Plan was in effect, that the three provinces have made significant achievements in solidly promoting ecological restoration of land spaces and strengthening resource protection, but there are still problems such as large areas of land desertification, serious local soil erosion, and a lack of arable land reserve resources. There are not many studies on the spatio-temporal distribution patterns of soil erosion in this study area, and there are not many studies on the impact of various factors and how they interact with one another in studies that combine soil erosion and future land use changes. Therefore, from 2000 to 2020, the RUSLE model will be used to examine the spatio-temporal changes in soil erosion in the study area, which also combines it with the PLUS model in order to more accurately calculate the state of soil erosion in the investigation’s area between 2025 and 2050. With the aid of the geographical detector, it also investigates the main contributing factors and interactions between the various types of land use and soil erosion in the research region. The goal is to create a scientific foundation for ecological conservation and restoration in the mild temperate forest farming subregion in the northeast of the Loess Plateau.

2. Material and Methods

2.1. Study Area

The area of research is the warm temperate forest and farming subregion in the northeast of the Loess Plateau (hereinafter referred to as the forest and farming subregion of Loess Plateau), one of the secondary national natural resource zones studied [25,26] (Figure 1). The research zone is situated in the Loess Plateau’s eastern region, geographically between 30° N and 40° N and 110° E and 120° E. It has four distinct seasons and a continental monsoon climate. Winters are chilly and dry, while autumns are moderate and cool, and summers are hot and rainy. The average annual temperature from south to north is 10.5~16.7 °C. Everywhere, the annual precipitation is between 350 and 1230 mm, with an uneven seasonal distribution. Additionally, 60% of the yearly rainfall occurs from June to August, when precipitation is relatively concentrated. The study region’s interior is undulating, with longitudinal river valleys and a landscape consisting of mountains, hills, platforms, and plains. The world’s thickest loess cover is found in the Loess Plateau area. The loess layer has an average thickness of 50–100 m and a maximum thickness of over 300 m, and it declines in thickness from northwest to southeast. Loess, black clay, brown clay, and other soil types are available; 50% of the weight of loess is made up of powdery grains, which are loose and porous, easily loosened by water, and vulnerable to erosion from runoff and rainfall. The study area covers 81,735 km², which includes two three-order zones, a dry-land area in the northern Loess Plateau and forested dry-land area in the northern Loess Plateau, and nine four-order zones, including dry-land plots in Luoyang City (IV12-1), dry-land plots in Pingdingshan City (IV12-2), cultivated vegetation plots in the eastern part of Jinzhong City (IV11-1), forested dry land plots in the eastern part of Linfen (IV11-2), wooded plots in Jincheng City (IV11-3), wooded plots in Changzhi (IV11-4), temperate deciduous scrub plots in Anyang (IV11-5), Jiyuan bush land dry land plots (IV11-6), and Handan City, Cao Ceng District (IV11-7) [27,28,29].

2.2. Data Sources

The information utilized in this study includes meteorological, LUCC (land use and land cover changes), terrain elevation, NDVI, soil properties, and socioeconomic data. Terrain elevation data (30 m) were retrieved from the Geospatial Data Cloud http://www.gscloud.cn/ (accessed on 20 April 2024) and used to determine the slope length slope factor LS in the RUSLE model for 2000 to 2050. The soil type data were a 1:1 million scale map of Chinese soils, including sand, silt, and clay data of different soil types, which were used to obtain the K-factor for soil erodibility. NDVI (1 km) data in China were used to calculate surface vegetation cover and management factor C. Precipitation data at 1 km resolution were used to calculate rainfall erosivity factor R. Data for factors R, K, and C (2000–2020) were obtained from the Chinese Academy of Sciences Resource and Environment Science Data Center http://www.resdc.cn (accessed on 20 April 2024). The rainfall erosivity factor R for the period 2025–2050 was obtained from Shouzhang Peng’s team at the National Qinghai-Tibet Plateau Science Data Center. In this study, the C factor for 2025 to 2050 was used as the average value of the C factor for 2000 to 2020 based on previous experience [30]. Data on land cover types from 2000 to 2020 were obtained from the team of Huang Xin of the School of Remote Sensing at Wuhan University [31] and included cropland land, woodland, grassland, water areas, construction land, unused land, and wetland. The conservation practice factor P for the years 2000 to 2020 was calculated using data with a spatial resolution of 30 m. The distributions of future land use generated from the PLUS model were used to construct P for 2025–2050.

2.3. Methods

2.3.1. Technological Routes

In this research, the RUSLE model, terrain elevation, rain, NDVI, soil type, and LUCC data were used to investigate soil erosion in the research domain between 2000 and 2020. In addition to performing future soil erosion studies utilizing constant and variable factors to forecast soil erosion in the research region for the years 2025–2050, the PLUS model was also used to simulate land use change in the study area. The various land uses in the research area, the main drivers of soil erosion, and their interactions were investigated using the geographical detector. Figure 2 provides a summary of the study’s approach. The subsections provide a thorough discussion of the datasets and methods.

2.3.2. Revised Universal Soil Loss Equation (RUSLE) Model

In order to compute soil erosion in the Loess Plateau’s forest farming subregion, the RUSLE model was utilized, and the research area’s features and data were used to develop the methodology for each factor. The RUSLE model’s equations are as follows:

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

(1)

where $A$ is the annual quantity of soil erosion (t/ha per year), $R$ is the precipitation erosivity factor, $K$ is the soil erodibility factor, $L$ is the slope length factor, $S$ is the surface vegetation cover and management factor, and $P$ is the factor for soil and water conservation methods.

2.3.3. Slope Length Factor (LS)

The slope’s length ($L$) and steepness ($S$) are indicators of how the topography and geomorphology of the region affect soil erosion [32,33]. The general method of Ganasri and Ramesh [34] was used to calculate $L$ and $S$, which is calculated by the formula:

$$L=(\frac{\lambda }{22.13}{)}^m$$

(2)

$$m=\frac{\beta }{(1+\beta )}$$

(3)

$$\beta =\frac{\mathit{sin}\theta }{0.089}\times \frac{1}{\left[3.0\times (sin\theta {)}^{0.8}+0.56\right]}$$

(4)

$$S=\left\{\begin{array}{ll}10.8sin\theta +0.03& \theta ⩽5^{\circ }\\ 16.8sin\theta -0.5& 5^{\circ }<\theta <{10}^{\circ } \\ 21.97sin\theta -0.96& \theta ⩾{10}^{\circ }\end{array}\right.$$

(5)

where $\lambda$ is the slope’s horizontal projection length, which is simplified by substituting the grid length (30 m), $m$ is the slope’s length factor index, $\beta$ is the ratio of the fine gully to surface erosion, and $\theta$ is the slope angle (°).

2.3.4. Precipitation Erosivity Factor (R)

$R$ is the potential capacity that might lead to soil erosion, and annual precipitation data were used to determine $R$. In this experiment, the following method was used to calculate the precipitation erosion force factor using the yearly precipitation algorithm developed by Zhang and Fu [35].

$$R_a=\alpha P_a^{\beta }$$

(6)

where $R_a$ is the $a$ year precipitation erosion force ((MJ·mm)/(ha·h)), $P_a^{\beta }$ is the $\mathrm{a}$ year precipitation (mm), $\alpha$, $\beta ,\mathrm{a}\mathrm{n}\mathrm{d} \mathrm{a}$ are parameters in the model. $\alpha$ is 0.0534, and $\beta$ is 1.6548.

2.3.5. Surface Vegetation Coverage and Management Factors (C)

$\mathrm{C}$ represents the erosion resistance of vegetation cover and field management mode, and its magnitude is based on 0–1. It measures the ratio of soil erosion under established plant cover or field management to soil erosion on bare land. The more significant the soil erosion associated with the value, the lower the erosion resistance under cover and management conditions. Based on previous research, Cai et al. [36] determined the correspondence between fractional vegetation coverage, $f,$ and its value. The $C$ values can be calculated using the following equation [31].

$$C=\left\{\begin{array}{cc}1& ,0\le f\le 0.1\%\\ 0.6508-0.3436lgf& ,0.1\%\le f\le 78.3\% \\ 0& ,f\ge 78.3\%\end{array}\right.$$

(7)

$$f=\frac{NDVI-{NDVI}_{\mathit{soil}}}{{NDVI}_{veg}-{NDVI}_{\mathit{soil}}}$$

(8)

2.3.6. The Soil and Water Conservation Practice Factor (P)

The primary data were collected through a questionnaire survey. At present, agricultural production activities in China are mainly based on smallholder households. Therefore, the smallholder household (0–3 ha) was the unit of analysis in this study. Household heads were interviewed because they are the key decision-makers of key agricultural practices. A multi-stage random sampling technique was used to select households. Towns, villages, and smallholder households were similarly selected in four cities in Henan Province and two districts in each city. A total of 850 farm households were selected to participate in the research, and 765 questionnaires were valid, with an effective response rate of 90%. The survey was conducted during the winter holidays (December 2014–February 2015) by fellow students who had previously been interviewed. The research was conducted through face-to-face interviews.

The P-factor, which is always between 0 and 1, is the comparison between soil loss with and without certain soil and water conservation techniques. Based on the results of the actual survey of the study area as well as previous research [37,38], the corresponding P-factors for arable land, woodland, grassland, waters, construction land, fallow land, and wetland were set to 0.4, 1, 1, 0, 1, 1, and 0, respectively.

2.3.7. Soil Erodibility Factor (K)

The $K$ is one of the required parameters of the RUSLE model to reflect the sensitivity of soil to erosion, and soil texture data are applicable for calculating the soil erodibility factor. In this paper, the formula used to calculate the soil erodibility factor in the EPIC model, a model designed particularly to measure soil erosion by Rao et al. [39].

$$\begin{gathered} K=0.1317\left\{0.2+0.3exp\left[-0.0256Sa\left(1-\frac{\mathit{Si}}{100}\right)\right]\right\}\times {\left(\frac{\mathit{Si}}{Cl+Si}\right)}^{0.3}\left[1-\frac{0.25C}{C+exp(3.72-2.95C)}\right] \\ \times \left[1-\frac{0.7Sn}{Sn+exp(-5.51+22.9Sn)}\right] \end{gathered}$$

(9)

$$Sn=1-\frac{Sa}{100}$$

(10)

where $Sa$ represents the amount of sand, $\mathit{Si}$ is the percentage of silt, $Cl$ is the percentage of clay, $C$ is the percentage of organic carbon, and $Sn$ is the percentage of soil that is not sand.

2.3.8. Patch-Generating Land Use Simulation (PLUS) Model

The PLUS model developed by Liang et al. [13] is based on traditional methods for simulating land usage, such as the CA-Markov model [40], the CLUE-S model [41], and the FLUS model [42], combined with a random forest model. This incorporates a CA model with a variety of stochastic seed methods and a rule-mining technique for land expansion analysis. The PLUS model outperforms other models in modeling and predicting soil erosion under future scenarios [43].

The PLUS model primarily comprises two modules: land expansion analysis strategy (LEAS) and CA based on multi-type random patch seeds (CARS).

In order to determine the probability of development for each land use type, the LEAS module first extracts the initial land use and then applies the random forest classification (RFC) algorithm to examine the relationship between the growth of various land use types and multiple drivers. This probability is determined by the following formula [13]:

$$P_{i,k}^d(x)=\frac{{\sum }_{n=1}^{M}I\left[h_n\left(x\right)=d\right]}{M}$$

(11)

where $x$ is a vector made up of several different drivers. The nth decision tree’s $h_n\left(x\right)$ prediction type is used for the vector $x$. $d$ can take a value of 0 or 1, where 1 represents the transformation of other land use types on grid $i$ to this k-type land use type and 0 represents other transformations; $M$ represents the overall number of decision trees; $I\left[h_n\left(x\right)=d\right]$ represents the decision tree’s indicator function; and $P_{i,k}^d(x)$ represents the likelihood that $k$ land use type will expand into a space unit $i$.

The CARS module simulates the spatial diversity of intricate natural geographic phenomena. It is a scenario-driven land use simulation program that simulates land use distribution patterns, primarily by figuring out the likelihood that various land use types will evolve. The overall probability of the conversion of land use type $k$ can be estimated using the following equation [13].

$$O{P}_{i,k}^{d=1,t}=P_{i,k}^{d=1}\times {\Omega }_{i,k}^t\times D_k^t$$

(12)

Land use type $P_{i,k}^{d=1}$ is represented by cell $k$ and cell growth probability $i$. ${\Omega }_{i,k}^t$ is the cell $i$ domain effect. According to the following formula [13], $D_k^t$ represents the demand’s effect on future land use:

$${\Omega }_{i,k}^t=\frac{con\left(c_i^{t-1}=k\right)}{n\times n-1}\times w_k$$

(13)

$$D_k^t=\left\{\begin{array}{l}{D}_k^{t-1}\left(\left|G_k^{t-1}\right|\le \left|G_k^{t-2}\right|\right)\\ \left.D_k^{t-1}\times \frac{G_k^{t-2}}{G_k^{t-1}}(0⟩G_k^{t-2}⟩G_k^{t-1}\right)\\ \left.D_k^{t-1}\times \frac{G_k^{t-1}}{G_k^{t-2}}\left(G_k^{t-1}\right.⟩G_k^{t-2}>0\right)\end{array}\right.$$

(14)

In Equations (13) and (14), $con$ is the first $n\times n$ window in which the first $k$ land use type occupied in the last iteration is the total number of grid cells. $w_k$ has a default value of 1 and determines how much importance each sort of land use is given. $G_k^{t-1}$ and $G_k^{t-2}$ are the land use type $k$ in the first $t-1$ and $t-2$ iterations of the current demand and the discrepancy between anticipated demand.

2.3.9. Geographical Detector

The geographical detector is a technique for detecting the regionally stratified heterogeneity of processes, which reveals the driving factors behind them [44]. Four units make up the geographical detector: factor detector, risk detector, interaction detector, and ecological detector. The geographical detector measures the degree of spatial stratified heterogeneity, which is a phenomenon where the sum of the intra-layer variance is less than the sum of the inter-layer variance; it measures value $q$.

The $q$ value increases as the effect of various environmental factors on the spatial variation of soil erosion increases [45]. $q$ is used to measure the extent of such influence. Its expression is as follows:

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

(15)

$$SSW={\sum }_h^L{N}_h{{\sigma }_h}^2$$

(16)

$$SST=N{\sigma }^2$$

(17)

where $h=$1, 2 … is the classification or partition of the independent variable. $N_h$ and $N$ are the number of cells in layer $h$ and the whole zone, respectively. ${\sigma }_h$ and $\sigma$ are the variances of the Y-values of layer $h$ and the whole region, respectively. $SSW$ is the sum of the within-stratum variance. $SST$ is the total variance of the region. $q$ is the explanatory power of the independent variable on the dependent variable, with a value range of [0, 1].

The geodetector also has the function of interaction detection, which may examine how multiple factors interact, i.e., to determine if factors X1 and X2 combined strengthen or weaken the ability of variable Y to explain itself and also to calculate whether these factors are apart from one another. The calculation technique determines the explanatory power q1 and q2 of the two factors, respectively, and then calculates the g value of the new polygon created by the tangent of the two variable polygons superimposed.

Table 1. Types of interaction between two covariates.

Judgment	Interaction
$\mathrm{q}(\mathrm{X}1\cap \mathrm{X}2)<\mathrm{M}\mathrm{i}\mathrm{n}\left(\mathrm{q}\right(\mathrm{X}1),\mathrm{q}(\mathrm{X}2)$	Non-linear weakening
$\mathrm{M}\mathrm{i}\mathrm{n}\left(\mathrm{q}\right(\mathrm{X}1),\mathrm{q}(\mathrm{X}2\left)\right)<\mathrm{q}(\mathrm{X}1\cap \mathrm{X}2)<\mathrm{M}\mathrm{a}\mathrm{x}\left(\mathrm{q}\right(\mathrm{X}1),\mathrm{q}(\mathrm{X}2\left)\right)$	Single-factor nonlinear attenuation
$\mathrm{q}(\mathrm{X}1\cap \mathrm{X}2)>\mathrm{M}\mathrm{a}\mathrm{x}\left(\mathrm{q}\right(\mathrm{X}1),\mathrm{q}(\mathrm{X}2\left)\right)$	Two-factor enhancement
$\mathrm{q}(\mathrm{X}1\cap \mathrm{X}2)=\mathrm{q}\left(\mathrm{X}1\right)+\mathrm{q}\left(\mathrm{X}2\right)$	Independent
$\mathrm{q}(\mathrm{X}1\cap \mathrm{X}2)>\mathrm{q}\left(\mathrm{X}1\right)+\mathrm{q}\left(\mathrm{X}2\right)$	Non-linear enhancement

depicts the connection between the two factors in comparison to the single factor’s explanatory power [13]. The basis of interaction is shown in Table 1. To assess if there is a substantial variation in the average value of the characteristics between the factor regions, the risk detector is often utilized. By doing this, high-risk sites for soil erosion may be found.

NDVI, rainfall, slope, and terrain elevation were used as land use type drivers to identify the dominant land use type drivers in the Loess Plateau forest management sub-region. When soil erosion is used as the dependent variable, LUCC is added as the influencing factor to complete the exploration of the main driving factors of soil erosion in the research area. According to the requirements of the geographical detector, combining the data discretization approach suggested by Wang et al. [44] resulted in the discretization of continuous-type data. Among them, land use data were classified according to categories, and data on vegetation cover were separated into eight groups (<0.3, 0.3~0.4, 0.4~0.5, 0.5~0.6, 0.6~0.7, 0.7~0.8, 0.8~0.9, 0.9~1); according to the equal spacing approach, average annual precipitation data were split into nine categories, while slope data were sorted into eight categories (<5°, 5°~10°, 10°~15°, 15°~20°, 20°~25°, 25°~30°, 30°~35°, >35°). The above-mentioned attribute values are provided as the operational data of the geographical detector. Elevation data were sorted into eight groups using the equal natural break point approach.

2.4. Data Classification and Factor Analysis

2.4.1. Significance of Data per Issue

The data for all research included in this analysis were averaged across a five-year period using data from the current year, the two years prior to it, and the two years following it. For instance, the figures from 2005 represent the average of five years from 2003, 2004, 2005, 2006, and 2007, making the overall trend obvious. The five historical periods of data were 2000, 2005, 2010, 2015, and 2020, and the three future periods of data were 2025, 2030, and 2050.

2.4.2. Classification of Erosion Grades

The Soil Erosion Classification and Grading Standard (SL190-2007) [46], which assigns six categories of soil erosion intensity: slight, mild, moderate, strong, very strong, and extreme erosion, is used to classify erosion grade in the present research. Slight erosion refers to minimal soil loss, often barely noticeable. It can be caused by light rain or wind action. In this category, mild erosion means that soil loss is still relatively low, but it is more noticeable than light erosion. It can be caused by moderate rainfall or other factors. Moderate erosion means that soil loss becomes more significant. It is caused by moderate rainfall, runoff, or other erosive processes. Vegetation cover may begin to decline. Strong erosion here means that soil loss is significant. It can lead to gullies, rills, and significant damage to the landscape. Vegetation cover is severely affected. Very strong erosion means that this level is associated with severe soil loss, often resulting in deep gullies, extensive land degradation, and reduced soil fertility. Vegetation cover is severely affected. At this highest level, extreme erosion means that soil loss is catastrophic. It leads to severe land degradation, loss of topsoil, and significant environmental impacts. Vegetation cover is almost non-existent.

3. Results

3.1. Spatio-Temporal Variation of Soil Erosion from 2000 to 2020

The spatial distribution of annual rainfall erosivity was investigated, and the spatial distribution of annual rainfall erosivity factors in 2000, 2005, 2010, 2015, and 2020 was obtained by grid calculation (Figure 3). The results showed that the overall spatial distribution of rainfall erosivity in the study area increased from northwest to southeast. In 2000, the high rainfall erosivity was mainly distributed in the eastern and southern regions, but in 2020, the degree of erosivity decreased significantly, and most of the rainfall erosivity was distributed in the southeastern corner of the study area. The area of low rainfall erosivity expanded from the northwest corner of the study area to the central area. Globally, the mean rainfall erosivity was estimated to be 2190 MJ mm ha⁻¹ h⁻¹ yr⁻¹ [47], with the highest values in South America and the Caribbean countries, Central east Africa, and southeast Asia. China has a mean value of 1600 MJ mm ha⁻¹ h⁻¹ yr⁻¹ but exhibits high variability with zero erosivity in the arid northwest areas (Taklimakan desert) and extreme erosivity (>15,000) in the southeastern coastal zones. Regional studies conducted by Zhu and Yu [48] and Qin et al. [49] show very similar spatial patterns compared to our rainfall erosivity distribution in China.

According to the geographical distribution of soil erosion (Figure 4), there are significant differences between the research area’s north and south. The northern portion of IV11-1, IV11-3, and IV11-5 is where the majority of the regions with the higher soil erosion intensity level are located, where the landscape is more undulating, and the vegetation cover is sparse. In the southern regions IV12-1 and IV12-2, the terrain is flatter, the slope is even milder, the soil thickness is deeper, and the vegetation is mainly temperate broad-leaf forest, which is less prone to erosion. In general, in the past 20 years, the greening process of the country has been accelerated by key provincial and municipal afforestation projects, such as forest rehabilitation, natural forest protection, and the Three-North Shelter Forest Program (TNSFP), and the area and intensity of soil erosion have been doubly reduced. This shows that over this time, the ability to conserve soil and water has substantially increased, and forest rehabilitation has been effective, which is consistent with Li et al. [50]. The results of He et al. [51] are compatible with the regional distribution pattern of the severity of soil erosion in this area, and the temporal variation pattern is basically the same as that of Mu et al. [12].

The average soil erosion modulus (t/ha per year) in the field of study is 4866, 3435, 3423, 3474, and 3787 in 2000, 2005, 2010, 2015, and 2020, respectively. Soil erosion modulus classes vary significantly over time, and the percentage of area eroded by each class is shown in

Table 2. Statistics on the level of soil erosion in the forest farming subregion of the Loess Plateau between 2000 and 2020.

Year	Erosion Grade	Modulus/ (t/ha per Year)	Proportion/ (%)	Effective Area/(103 km2)
2000	Slight	<200	60.69	2397.9
	Mild	200–2500	9.65	381.3
	Moderate	2500–5000	10.94	432.4
	Strong	5000–8000	8.58	339.0
	Very strong	8000–15,000	7.74	305.9
	Extreme	>15,000	2.40	94.7
2005	Slight	<200	60.85	2411.6
	Mild	200–2500	10.65	422.0
	Moderate	2500–5000	12.07	478.2
	Strong	5000–8000	8.60	340.9
	Very strong	8000–15,000	6.79	269.0
	Extreme	>15,000	1.04	41.2
2010	Slight	<200	60.25	2388
	Mild	200–2500	10.47	415.0
	Moderate	2500–5000	11.90	471.6
	Strong	5000–8000	9.39	372.3
	Very strong	8000–15,000	7.04	279.1
	Extreme	>15,000	0.95	37.8
2015	Slight	<200	59.40	2356.4
	Mild	200–2500	10.81	428.9
	Moderate	2500–5000	11.38	451.4
	Strong	5000–8000	9.75	386.8
	Very strong	8000–15,000	7.62	302.3
	Extreme	>15,000	1.05	41.5
2020	Slight	<200	57.45	2279.8
	Mild	200–2500	10.01	397.4
	Moderate	2500–5000	11.54	457.8
	Strong	5000–8000	10.00	397.0
	Very strong	8000–15,000	9.03	358.5
	Extreme	>15,000	1.96	77.8

. The soil erosion intensity categorization in the research area during the five years had the largest percentage of the entire area (60.69%, 60.85%, 60.25%, 59.40%, and 57.45%, respectively). The percentages of soil erosion mild and above are in the order of 9.65%, 10.65%, 10.47%, 10.81%, and 10.01%. Erosion that is slight or mild affects more than 67.46% of the total area. The stability rate (a constant percentage of erosion severity) of slight erosion in the research region from 2000 to 2020 is 75.32%, 81.18%, 71.12%, and 81.45%, respectively (Figure 5). There tends to be an increase and, subsequently, a reduction in the percentage of land subject to mild to moderate erosion. The proportion of the area that has very strong and extreme erosion shows first a decreasing and then an increasing trend. The percentage of strongly eroded land that has been significantly degraded is steadily increasing.

When examining rates over 5-year periods, from 2000 to 2005, the proportion of very strong and extreme erosion areas decreased (0.95 and 1.36 percentage points, respectively); the percentage of erosion area at other intensity levels increased, and in general, the rate of soil erosion was trending downward. Specifically, very strong erosion replaced intense erosion at 0.36% of the area, and the region of soil erosion at all levels changed to the surrounding 75.32% of the slightly eroded area remains unchanged, 8.94% is converted to mild erosion, 7.36% is converted to moderate erosion, 4.28% is converted to strong erosion, 3.21% is converted to very strong erosion, and 0.90% is converted to extreme erosion. From 2005 to 2010, the transformation of soil erosion at every level was different from that of the previous period, except for the increase in the percentage of strong erosion and very strong eroded areas. The proportion of soil erosion area at all other intensity levels shows a decreasing trend, and the number of transformations from high-intensity levels to low-intensity levels increases significantly, especially the transformation of mild erosion to slight erosion (41.29% transformed to slight erosion), demonstrating a general downward trend in the severity levels of soil erosion over this time. From 2010 to 2015, the proportion of slight and moderate erosion declined considerably, while the percentage of extreme, strong, very strong, and mild erosion showed an increasing trend. Except for the decline in the area of slight and mild erosion and the growth in the area of other intensity categories, the transformation of the soil erosion area from 2015 to 2020 is comparable to that of the preceding time period. The percentage indicates a rising trend; specifically, the unchanged erosion intensity of slight erosion area rises to 81.45%, and the transfer of high-intensity to low-intensity area is mainly from mild and moderate erosion to slight erosion transformation, converting 40.65% and 34.33%, respectively. This suggests that the soil erosion intensity levels as a whole show a decreasing trend in this time period. It is complicated how soil erosion zones change over time and at various echelon levels.

3.2. Soil Erosion Characteristics of Different Land Use Types

The majority of land use categories in the study area—more than 72%—are comprised of cropland and woodland. Other land use types make up a smaller proportion (Figure 6). Analyzing the shifts in land use patterns in the forest farming subregion of the Loess Plateau between 2000 and 2020, large changes in cropland land, woodland, waters, and wetland areas have occurred, and the area ratios of woodland and wetland have increased slightly, from 28.91% and 4.22% to 34.34% and 6.88%, respectively. Arable land to water area ratios drastically declined, from 43.33% and 21.95% to 21.95% and 16.07%, respectively, indicating the continued occurrence of the phenomenon of land formation and deforestation around bodies of water (

Table 3. Statistics of soil erosion intensity in China from 2000 to 2020 land use type area.

	2000	2005	2010	2015	2020
Cropland land/(%)	43.33	42.53	40.80	40.73	41.60
Woodland/(%)	28.91	30.23	32.15	32.53	34.34
Grassland/(%)	1.25	1.37	0.80	0.81	0.53
Waters/(%)	21.95	20.62	20.19	19.01	16.07
Construction Land/(%)	0.34	0.48	0.52	0.56	0.58
Wetland/(%)	4.22	4.76	5.55	6.36	6.88

).

The state of soil erosion in relation to different forms of land usage (

Table 4. Average soil erosion by land use type.

	2000	2005	2010	2015	2020
Cropland land/ (t/ha per year)	4242	2225	2005	2654	2160
Woodland/ (t/ha per year)	4866	3435	3423	3474	3787
Grassland/ (t/ha per year)	2296	2026	2122	2086	2123
Waters/ (t/ha per year)	3408	2581	2154	2909	2398
Construction Land/ (t/ha per year)	1142	4471	1748	1358	1019
Wetland/ (t/ha per year)	1703	1043	1743	1672	1805

) shows that more than 87% of the soil erosion in the study area is concentrated in woodland and cropland. This is mostly because the majority of the woods in the research region are situated in terrain with steep hills and mountains, and once the vegetation is destroyed, the vegetation cover decreases rapidly, and large amounts of soil loss can easily occur under heavy rainfall. Although the intensity of soil erosion is generally lower on cropland, the final cumulative result is greater because of the substantial amount of cropland. Grassland and construction land are quite tiny, with an average area share of 1.19% and 0.50% from 2000 to 2020, corresponding to a 20-year average soil erosion of 2131 t/ha per year and 1143 t/ha per year. The major drop in regional erosion in 2010 was mostly caused by an increase in woodland and wetland areas and a marked decrease in other land uses. As a result, the study area’s degraded hills should be the focus of soil erosion control efforts, and if appropriate, a mix of engineering, biological, and agricultural approaches should be implemented to restore the degraded hills’ ecological health.

Combining Table 3 and Table 4, it can be noticed that between 2000 and 2020, the area ratio of cropland marginally declined, from 43.33% to 41.60%, corresponding to a decrease in erosion from 4242 t/ha per year to 2160 t/ha per year. On the other hand, the ratio of woodland to construction land increased from 28.91% and 0.34% to 34.34% and 0.58%, and the amount of soil erosion caused by forest area decreased from 4866 t/ha per year to 3787 t/ha per year. As a result, soil erosion improved significantly over this period. Over a period of 20 years, there has been a modification to the local land use pattern, which has effectively reduced soil erosion.

3.3. Projected Variations of Land Use Types and Soil Erosion

Land usage projections for the years 2025, 2030, and 2050 are based on the PLUS model, using 2020 as the starting point. From the outcomes, a chord diagram of the land use change transfer matrix between 2020 and 2050 was created (Figure 7). Land use continues the trend from 2000 until 2020, with cropland, woodland, and grassland decreasing by 2554.0, 670.6, and 82.2 km², respectively, and construction land, waters, and wetland increasing by 20.7, 467.0, and 2818.6 km², respectively. The main change is the conversion of cropland to construction land, followed by the transformation of woodland, water areas, etc., to construction land and unused land. Cropland has decreased in size while other land uses have increased, which verified that the urbanization procedure expanded with the constant enhancement of social and economic levels. This trend in land use change also shows that decisions to return cropland to grassland, woodland, and water were implemented successfully in order to suit the demands of urban development and the surrounding ecosystem.

Table 5. Structure of land use and soil erosion in the period 2025 to 2050.

Land Use Type		2025	2030	2050
Arable land	Area/(103 km2)	35.16	34.87	32.60
	Soil erosion/ (t/ha per year)	2442	2412	2483
Woodland	Area/(103 km2)	27.52	27.07	26.85
	Soil erosion/ (t/ha per year)	4885	4984	4761
Grassland	Area/(103 km2)	32.37	34.92	24.15
	Soil erosion/ (t/ha per year)	1439	1321	1254

shows the soil erosion status for various land use categories, and Figure 8 shows the geographical distribution of soil erosion levels in the research region from 2025 to 2050. It can be seen that soil erosion (t/ha per year) is 4885, 4984, and 4761 in 2025, 2030, and 2050, respectively. Soil erosion increased in all three years compared to soil erosion in 2020. Between 2020 and 2050, soil erosion is likely to peak and then start to decline due to an increase in wetland and water bodies and a decrease in construction land. Soil erosion will increase significantly as the area of cropland decreases between 2020 and 2050. Therefore, the implementation of decisions such as returning unused land to grassland, woodland, and waters should be adopted by 2050, based on the protection of arable land. These steps can successfully stop the growth of construction land and reduce soil erosion.

3.4. The Geographical Detector-Based Quantitative Attribution Analysis of Soil Erosion and Land Use Types

The explanatory power of each factor contribution to land use type in descending order is LUCC (0.081) > terrain elevation (0.013) > slope (0.006) > rainfall (0.005) from the forest farming subregion of the Loess Plateau. The results show that among the selected factors, LUCC is the dominant factor in determining land use type. An analysis of the extent to which land use types are influenced by the interaction of the two drivers was conducted using the interaction detector in the geographical detector (Figure 9). The most powerful explanation is in vegetation cover and elevation, and the interaction of the two factors enhances the explanatory power of the spatial distribution of land use types in the forest farming subregion of the Loess Plateau. There is a stronger association between NDVI and LUCC, as seen by the interaction values between NDVI and the variables being much higher than those between other factors. The interaction between terrain elevation and NDVI was the most important factor, with an explanatory power of 0.112. The results suggest that the forest farming subregion of the Loess Plateau should pay more attention to vegetation cover in areas with high elevation and in places with significant geographic relief to boost the planting of plants.

In the research region, there are a variety of factors that have varying degrees of explanatory power for soil erosion. The q-values for the five variables are rainfall (0.021), NDVI (0.179), terrain elevation (0.103), slope (0.129), and LUCC (0.199). Among them, the LUCC factor with the highest q-value and the strongest explanatory power was the dominant factor affecting the research area’s soil erosion’s geographic spread. The interaction detector’s findings demonstrate that the degree of synergy between the factors affecting soil erosion is significantly greater than the single factor’s capacity for explanation. Among these, the most significant element determining the explanatory power of the geographical distribution of soil erosion is the synergistic influence of LUCC and other parameters. Particularly, the non-linearly increased interaction between the two factors strengthens the inhibitory influence on soil erosion. The interaction with rain is around 10 times larger than the single factor. This means that soil erosion is very different between land use types with different amounts of rainfall or between land use types with the same amount of rainfall. Local soil erosion can be effectively controlled through rational and orderly land use. This result is in line with other researchers’ findings [43,44,45] about how different areas’ land use impacts soil erosion.

Soil erosion is a high-risk area if the LUCC is grassland, the NDVI is 0.9 to 1, the slope is greater than 35°, the terrain elevation is 1595–2559 m, and the rain is 0.805–0.854 m when the average value of soil erosion intensity reaches the maximum (Figure 10). In summary, soil erosion is a complex process influenced by a combination of factors. It is therefore recommended that the process of prevention and restoration not only focus on the dominant factors but also require regional coordination and integrated management. For example, in sensitive areas such as slopes and grasslands, measures such as reforestation are taken to increase vegetation cover and strengthen local capacity for water and soil conservation.

4. Discussion

4.1. Major Finding and Result Comparison

In this study, the PLUS model was used to forecast changes in land use in the forest farming subregion of the Loess Plateau from 2025 to 2050. The land usage prediction indicates that the soil erosion in 2025, 2030, and 2050 will be 48.85, 49.84, and 47.61 t/ha per year, respectively, and minor erosion will continue to be the main erosion degree. The results showed that the study area had abundant LUCC, and the NDVI regression slope values were >0, generally consistent with previous studies [52]. The findings also indicate that soil erosion would worsen as a result of future precipitation’s large rise. As a consequence, shrubs may be managed as a priority in the Loess Plateau’s ecological system going forward, and at the same time, afforestation and grass planting, afforestation and natural closure, and grass planting and natural closure are effective methods for conserving soil and water [53].

The primary land types in the Loess Plateau were cropland and woodland. All land types showed a decrease in erosion in 2020; the decrease in cultivated land erosion was the most obvious. Reclamation in the study area resulted in a decrease in the cropland area [12]. Many scholars have studied the spatial distribution pattern and temporal variation characteristics of vegetation coverage in the Loess Plateau, and the results all show that the average annual vegetation coverage in the Loess Plateau shows an upward trend from 2000 to 2020. Overall, the vegetation cover of the Loess Plateau is higher in the north and lower in the south [54,55]. Wei et al. showed that runoff and erosion show clear differences under different rainfall and land use patterns, and this study further supports that conclusion [56].

4.2. Regional Future Soil Erosion and Recommendations

The Loess Plateau is one of the most eroded locations in the world as a result of low soil erosion resistance and long-term illogical human land use and development [57,58]. In this study, the forest farming subregion of the Loess Plateau’s soil erosion situation between 2000 and 2020 was evaluated. The actual erosion modules were 48.66, 34.35, 34.23, 34.74, and 37.87 t/ha per year (Table 2), and they had a declining trend at first, followed by an increase. The initiative the Chinese government initiated in 1999 to return farmland to grassland resulted in an 11.92% decrease in soil erosion between 2000 and 2020. By altering how land is used and raising the percentage of woodland and grassland, soil erosion was effectively reduced [59,60]. Due to the execution of these initiatives, the surface environment has changed, and the expansion of the woodland and grassland areas will significantly lessen soil erosion. Between 2010 and 2020, there was an upward tendency in the total erosion, which accelerated in 2020. This might be a result of global warming, which has significantly increased precipitation in northern China [61,62].

Using 19 downscaled General Circulation Models (GCMs), the projections estimate global mean precipitation erosivity values for 2050 and 2070. The mean estimate for 2050 is in the range of 2765–2822 MJ mm ha⁻¹ h⁻¹ yr⁻¹. For 2070, the range is 2782–2942 MJ mm ha⁻¹ h⁻¹ yr⁻¹. These values represent an increase of 27–34.3% compared to the 2010 baseline [63]. Between 2020 and 2060, overall wind erosion in Central Asia will decrease, but the area of increasing wind erosion will increase compared to the current period. Water erosion in Central Asia is increasing, and the area of increasing water erosion will increase [64]. Water erosion in the Loess Plateau is higher than in Central Asia because the Loess Plateau has more gullies, higher slopes, lower vegetation cover, and higher precipitation than in Central Asia [65]. For example, the Loess Plateau has a higher population and urban density than Central Asia. The vegetation cover in the Loess Plateau has increased significantly due to the conversion of agricultural land into forests and grasslands and the ecological construction that has taken place under human intervention [66,67]. Therefore, the contribution of vegetation and management practices is higher than that of precipitation in most areas of the Loess Plateau [68]. Due to artificial ecological restoration, water erosion on the Loess Plateau is decreasing under climate change, although precipitation is currently increasing slightly [69].

The diversity in land use type, which is the component most impacted by human activities, has a significant impact on the study region [70]. The two main ways that land use affects soil erosion are as follows: first, disturbing the surface soil affects soil compaction, which in turn influences how erosive rain and surface runoff are [71]; second, altering land cover modifies the direct impact of precipitation on the surface and regulates the process of surface runoff, which in turn modifies the severity of soil erosion [72]. The patch areas of various land use groups have varying effects on the amount of soil erosion; the amount of soil erosion is positively connected with the patch areas of cropland, whereas patches of woodland and grassland are adversely connected with the quantity of soil erosion [73]. In contrast to numerous studies, soil erosion in the forest farming subregion of the Loess Plateau has been estimated as: grassland > woodland > cultivated land > construction land > waters > unused land [59]. Numerous studies have shown that due to greater anthropogenic disturbance, cropland and construction land often have greater soil erosion than grassland and woodland sites, whereas cropland and construction sites have less surface soil protection than grassland and woodland sites.

In this study, the primary flaw is that it only forecasts the P-factor; it does not account for the R-factor or the C-factor. The forecast for future precipitation lacks numerous scenarios and is not thorough enough to take into account changing climatic conditions. Future predictions of soil erosion are, therefore, too inaccurate. The soil erosion model is impacted by the amount of resolution since the spatial resolution of the data we utilized is only 1 km, which is a bit low. The more precise the soil assessment is, the greater the resolution [74]. Future estimations of soil erosion should consider data with improved resolution. Although the study has several limitations, there have not been many such investigations conducted in the vicinity of or inside the study area. By conducting a thorough examination and predicting historical soil erosion scientifically, this research offers a scientific foundation for regional environmental conservation.

5. Conclusions

The following findings are made from this study’s calculations of the spatio-temporal variations in soil erosion in the forest farming subregion of the Loess Plateau from 2000 to 2050 and analysis of the elements that affect soil erosion.

The forest farming subregion of the Loess Plateau experiences mostly slight soil erosion, with the proportion of slight erosion to the total area being no less than 67.46%. In the research region, soil erosion is distributed in a highly diverse manner, with very strong and extreme erosion mainly in the wooded plots of Jincheng City, the temperate deciduous scrub plots of Anyang City, and the cultivated vegetation plots in the eastern part of Jinzhong City. There is a small distribution in the eastern part of the wooded plots in Changzhi City and a small distribution in the southwestern part of the shrubland dry-land plots in Jiyuan, and in the northwestern part of the wooded dry-land plots in eastern Linfen City.

From 2000 to 2020, there was more change in land use, with cropland, woodland, and grassland decreasing by 2554.0, 670.6, and 82.2 km², respectively. Soil erosion has significantly increased as a result of the conversion of arable and construction land to woodland. Therefore, by 2050, certain environmental protection measures should be adopted based on the protection of arable land, which successfully restricts the growth of constructing land to mitigate the rate at which soil erosion is escalating. Soil erosion (t/ha per year) in 2025, 2030, and 2050 is 4885, 4984, and 4761, respectively, with an increasing trend in soil erosion in all three years compared to soil erosion in 2020.

In the forest farming subregion of the Loess Plateau, the analysis of the numerous factors influencing land use types using the geographical detector revealed that vegetation cover is a dominant factor in land use type. The ability of vegetation covering and altitude to explain the kind of land usage is 0.081 and 0.013, respectively, when acting as a single factor and rises to 0.112 when acting as a two-factor superposition. The two elements’ interplay has the most explanatory potential, leading to the recommendation that vegetation cover be enhanced in locations with higher relief. The kind of land use is the primary determinant of the geographical distribution pattern of soil erosion. The interaction detector demonstrates that the interaction between the plant cover and land use type parameters has a bigger impact on soil erosion than each component acting alone. Areas with LUCC of grassland, NDVI of 0.9 to 1, a slope greater than 35 degrees, a terrain elevation between 1595 and 2559 m, and rain between 0.805 and 0.854 m are areas at high risk of soil erosion. In this context, the prevention and management of areas at high risk of soil erosion should focus on the combined effects of many factors.