Variation Characteristics of Two Erosion Forces and Their Potential Risk Assessment in the Pisha Sandstone Area

Abstract

Precipitation and wind, as the main external erosion forces in wind–water erosion crisscross regions, have profound impacts on water and soil loss. Meanwhile, with the intensification of climate change and human activities, the variation characteristics and risks caused by erosion forces need to be reassessed. In this study, we explored the time-varying characteristics, differences in action period and spatial distribution, and temporal evolution of risk for the compound events of two erosion forces, including precipitation and wind, in the Pisha sandstone area, one of the most seriously eroding and difficult-to-control areas in the Loess Plateau. The results indicated that: (1) the stationarity of regional precipitation was not destroyed, but the mean change existed in the five subseries divided by the detected change points in wind; (2) wind acted earlier than precipitation and increased from southeast to northwest, while precipitation did the opposite; and (3) precipitation-led erosion has become the main erosion type in this area. The above results reveal the evolution and dominant types of regional external erosion forces in a changing environment and thus have implications for regional erosion studies and policy adjustments.

1. Introduction

Water and wind erosion are the two main types of erosion that cause the loss of land and water resources and a decline in land production capacity [1,2,3]. In different locations and at different times, the degree of soil and water loss caused by water and wind is also different. Taking the spatial distribution of erosion types in China as an example, water erosion is widespread throughout the country [4,5], while wind erosion occurs mainly in the arid and semi-arid lands of northwestern, northern, and northeastern China [6,7]. The erosion types overlap in some local areas to form erosion crisscross regions in which two or more of these erosion forces interact temporally and spatially. Such areas are prone to erosion, and are also the focus of soil erosion control [8]. Among them, arid and semi-arid regions are prone to wind–water complex erosion, hindering the normal development of local production, human life, and ecology, thus forming wind–water erosion crisscross regions, which are widely distributed and are hotspots for erosion research [9,10,11].

Studies of soil erosion in such areas have tended to focus on two aspects: either adopting one or more models to evaluate individual types of erosion, or conducting specific experiments in a laboratory or the field. Approaches based on the former use of well-developed methods that focus on evaluating the rate of one particular type of erosion using the revised universal soil loss equation (RUSLE), revised wind erosion equation (RWEQ), etc. For instance, Du et al. [12] assessed the local erosion situation during the 2000s using the RWEQ and RUSLE. Meanwhile, approaches based on the latter are always small-scale studies investigating microscopic erosional processes for a given hypothesis. Yang et al. [13] estimated soil loss by wind erosion using beryllium-7 measurements in a wind tunnel experiment. Ren et al. [14] evaluated the effects of sand cover on runoff soil loss by conducting a series of rainfall simulation experiments. Guo et al. [15] measured the compound erosion from 2012 to 2014 in an agricultural field at the northern end of the Loess Plateau (LP). These previous studies have promoted our re-understanding of the distribution and mechanism of erosion in erosion crisscross regions. However, the limited number of experimental scenarios and the time length of model input data made these studies more targeted. In fact, the related experiments on soil erosion often needed to answer a specific question, that is, whether the setting of external forces is reasonable. Usually, researchers use the mean or maximum value of external forces to carry out relevant experiments. That is to say, it is very important to zonal erosion studies whether the time series of external forces are stationarity. Meanwhile, analyses on the temporal and spatial features of external forces urgently need to transition from static to dynamic analyses under the intensification of climate change and human activities, especially in intersecting erosion regions with severe erosion conditions where the vegetation habitat is no longer sufficiently supported [16,17].

After a long period of systematic management, the overall situation of soil erosion in the Loess Plateau, belonging to the main sand source of the Yellow River, has improved [12,18,19]. In the future, the focus of soil erosion control will shift to the ecologically weak areas in the region. Among them, the Pisha sandstone area, belonging to a typical wind–water erosion crisscross region, is the most seriously eroding and difficult-to-control area in the LP and is affected by the low strength of the rock structure, the interlacing of wind and water erosion in time and space, and the fragile ecological environment [20]. In addition, precipitation and wind are the primary drivers of water and wind erosion, respectively. The variation characteristics between them can be used to answer whether the erosion risks caused by external erosion forces have changed.

Above all, this study investigated the change characteristics of the precipitation and average wind speed in the Pisha sandstone area from 1960 to 2019 and revealed the probability combination change in the two external forces in the time domain. The aims of this paper were to (1) explore whether the stationarity of the two main external forces in the region is broken; (2) reveal the action period and spatial distribution of the above forces based on long-term data; and (3) evaluate the variability and risk associated with the coupled system of multiple erosion forces.

2. Materials and Methods

2.1. Study Area

The Pisha sandstone area (38.6°–40.3° N, 108.8°–111.7° E) is located in the northeastern part of the LP (Figure 1a). It is a typical wind–water erosion crisscross region. The zone covers an area of 1.67 × 10⁴ km², of which 1.5% is in Shanxi, 28.0% is in Shaanxi, and 70.5% is in Inner Mongolia [20]. The bedrock in this area is Pisha sandstone, a continental clastic rock characterized by a low degree of diagenesis, poor cementation between the sand grains, and low structural strength, with the result being that it is readily stripped by external forces. The terrain in the area has high relief with a large height difference (up to 845 m) and a fragmented topography (Figure 1b). Based on the characteristics of its topsoil and parent material, the area is divided into zone I (sand-covered Pisha sandstone), zone II (soil-covered Pisha sandstone), and zone III (bare Pisha sandstone) [21]. The climate is temperate continental, with the most intense rainfall from June to September. The mean annual precipitation is 394.9 mm and average wind speed is 2.49 m/s (1960–2019 data). Due to the limited habitat in this area, drought-tolerant shrubs (Hippophae rhamnoides Linn. and Caragana korshinskii Kom.) and grasses (Astragalus adsurgens Pall., Stipa capillata Linn., and Leymus chinensis (Trin.) Tzvelev) are the main pioneer plants. Pinus tabuliformis Carr. and Armeniaca sibirica (L.) Lam. are the main trees growing on the gentle slopes in the area.

2.2. Data Requirement and Preprocessing

The precipitation and average wind speed were considered to be the two main external forces causing erosion by water and wind factors. For the analysis, data for daily precipitation and average wind speed monitored by the 12 national surface meteorological stations (Figure 1b) were downloaded from the Resource and Environment Science and Data Center [22]. Then, daily precipitation and average wind speed for the period of 1960–2019 were converted into annual and monthly series for further analysis. In these processes, a Voronoi diagram was used to transform point meteorological data into areal data, from which the kriging interpolation results reflected the spatial distribution of the meteorological data in the study area. A digital elevation model (DEM) with 30 m resolution was derived from the ASTER GDEM V3 dataset provided by the Geospatial Data Cloud site, Computer Network Information Center, Chinese Academy of Sciences [23]. A slope with 30 m resolution was calculated from the DEM. Based on a Landsat 30 m remote sensing image, the land use data (30 m resolution) of 1980 and 2018 in the region were interpreted, including six first-class categories of arable land, woodland, grassland, water body, construction land, and unutilized land.

2.3. Data Analysis

2.3.1. Identifying Change Point by the Pettitt Test

The Pettitt test [24,25] is a method of detecting the change point in an observation sequence based on nonparametric statistics. The statistic $U_{t,N}$ is equivalent to a Mann–Whitney U test for the two samples X₁, …, X_t and X_t+1, …, X_N, which are from the same sample; the function is:

$$U_{t,N}=U_{t-1,N}+{\displaystyle \sum _{j=1}^N\mathrm{sgn}\left(x_t-x_j\right),t\in \left[2,N\right]}$$

(1)

where $U_{t,N}$ is the sample statistics for t; $U_{t-1,N}$ is the sample statistics for t–1; and $\mathrm{sgn}\left(x_t-x_j\right)$ is the sign function, defined as:

$$\mathrm{sgn}\left(x_t-x_j\right)=\left\{\begin{array}{ll}1& if{x}_t-x_j>0\\ 0& if{x}_t-x_j=0\\ -1& if{x}_t-x_j<0\end{array}\right.$$

(2)

where for the series, x_t is the sample value for t and x_j is the sample value for j.

When t is 1, the sample statistic is calculated from:

$$U_{1,N}={\displaystyle \sum _{j=1}^N\mathrm{sgn}\left(x_1-x_j\right)}$$

(3)

For the null hypothesis of the Pettitt test, no change points exist in the series. When |U_t,N| takes a maximum value, the corresponding X_t is considered to be a possible change point. U_t,N > 0 indicates a shift down in the level from the beginning of the series. Similarly, U_t,N < 0 indicates a shift up in the level. The significance probabilities associated with |U_t,N| are given by:

$$p=2\exp \left[-6U_{t,N}{}^2/\left(N^2+N^3\right)\right]$$

(4)

where p is the significance level; U_t,N is the sample statistics for t; and N is the sample number. Effective mutation points exist in the data for p ≤ 0.5 in this method.

2.3.2. Testing Trend Using Modified Mann–Kendall Test

The trends presented by variables over time are generally divided into three categories: downward trend, upward trend, and no trend. Such trends often need to be explained in conjunction with the significance level, and are further classed as significant and non-significant trends. In this study, the modified Mann–Kendall trend test was selected to test the series trend.

The Mann–Kendall test [26] is a widely used non-parametric test for detecting trends in time series, and is recommended by the World Meteorological Organization. When this method is used to calculate the trends in a time series, it appears that autocorrelated data cause the results to fluctuate. Here, the modified Mann–Kendall trend test, which is suitable for autocorrelated data [27], was adopted.

The test statistic S is calculated from:

$$S={\displaystyle \sum _{k=1}^{n-1}{\displaystyle \sum _{j=k+1}^n\mathrm{sgn}\left(x_j-x_k\right)}}$$

(5)

where x_j and x_i are the observed j and k, n > j > k, and:

$$\mathrm{sgn}\left(x_j-x_k\right)=\left\{\begin{array}{ll}1& if{x}_j-x_k>0\\ 0& if{x}_j-x_k=0\\ -1& if{x}_j-x_k<0\end{array}\right.$$

(6)

In the modified Mann−Kendall trend test, the variance is modified as follows:

$$V^{\ast }(S)=VAR(S)\cdot \left(n/n_S^{\ast }\right)=\left\{\left[n(n-1)(2n+5)\right]/18\right\}\cdot \left(n/n_S^{\ast }\right)$$

(7)

where n is the actual number of observations and n/n_S^* corrects for autocorrelated data. The function is given as:

$$n/n_S^{\ast }=1+2/\left[n(n-1)(n-2)\right]\cdot {\displaystyle \sum _{i=1}^{n-1}\left[(n-i)(n-i-1)(n-i-2){\rho }_S(i)\right]}$$

(8)

where n is the actual number of observations and ρ_S(i) is the autocorrelation function of the ranks of the observations, which is calculated after subtracting a suitable nonparametric trend estimator. Only significant values of ρ_S(i) are used in Equation (8).

The statistic Z is calculated from:

$$Z=\left\{\begin{array}{cc}\left(S-1\right)/\sqrt{V^{\ast }(S)}& if\begin{array}{c}S>0\end{array}\\ 0& if\begin{array}{c}S=0\end{array}\\ \left(S+1\right)/\sqrt{V^{\ast }(S)}& if\begin{array}{c}S<0\end{array}\end{array}\right.$$

(9)

A positive (negative) value of Z indicates that the sequence to be tested has an upward (downward) trend. The statistic Z obeys the standard normal distribution. In the two-tailed test, the null hypothesis is rejected if the statistic |Z| > Z_1–α/2. The value of Z_1–α/2 is obtained from the standard normal distribution table. The significance level of the test is often 0.001, 0.01, 0.05, or 0.1 (

Table 1. Test value at the given significance level.

Significance Level	Test Value	Degree of Confidence
0.001	3.291	99.90%
0.01	2.576	99.00%
0.05	1.96	95.00%
0.1	1.645	90.00%

).

2.3.3. Distribution Fitting of Samples Considering Non-Stationarity

The distribution fitting of samples considering non-stationarity follows these steps: (a) whether the non-stationarity exists in the study sequence is judged by combining the analysis methods of Section 2.3.1 and Section 2.3.2; (b) if the stationarity of the sequence is broken, the sequence is restored; otherwise, the third step is carried out; (c) parameters of alternative distributions (

Table 2. Probability functions adopted in this study.

Distribution	Probability Density Function	Parameters
Gamma (GAM)	$f(x\left|\alpha ,\beta \right.)=\frac{1}{{\beta }^{\alpha }\mathsf{\Gamma }\left(\alpha \right)}{x}^{\alpha -1}{e}^{-\frac{x}{\beta }}$	α, β
exponential (EXP)	$f\left(x\left|\mu \right.\right)=\frac{1}{\mu }{e}^{-\frac{x}{\mu }}$	μ
Weibull (WBL)	$f\left(x\left|\alpha ,\beta \right.\right)=\frac{\beta }{\alpha }{\left(\frac{x}{\alpha }\right)}^{\beta -1}{e}^{-{\left(\frac{x}{\alpha }\right)}^{\beta }}$	α, β
generalized extreme value (GEV)	$\left\{\begin{array}{cc}f\left(x\left|k,\mu ,\sigma \right.\right)=\left(\frac{1}{\sigma }\right)\exp \left(-{\left(1+k\frac{\left(x-\mu \right)}{\sigma }\right)}^{-\frac{1}{k}}\right){\left(1+k\frac{\left(x-\mu \right)}{\sigma }\right)}^{-1-\frac{1}{k}}& k\ne 0\\ f\left(x\left|0,\mu ,\sigma \right.\right)=\left(\frac{1}{\sigma }\right)\exp \left(-\exp \left(-\frac{\left(x-\mu \right)}{\sigma }\right)-\frac{\left(x-\mu \right)}{\sigma }\right)& k=0\end{array}\right.$	k, μ, σ
generalized Pareto (GP)	$\left\{\begin{array}{cc}f\left(x\left|k,\sigma ,\right.\theta \right)=\left(\frac{1}{\sigma }\right){\left(1+k\frac{\left(x-\theta \right)}{\sigma }\right)}^{-1-\frac{1}{k}}& k\ne 0\\ f\left(x\left|0,\sigma ,\right.\theta \right)=\left(\frac{1}{\sigma }\right)e^{-\frac{\left(x-\theta \right)}{\sigma }}& k=0\end{array}\right.$	k, σ, μ
Pearson type III (P–III)	$\begin{array}{cc}f\left(x\left|\alpha ,\beta ,\xi \right.\right)=\frac{1}{{\beta }^{\alpha }\mathsf{\Gamma }\left(\alpha \right)}{\left|x-\xi \right|}^{\alpha -1}{e}^{-\frac{\left|x-\xi \right|}{\beta }}& \begin{array}{c}\alpha =\frac{4}{{\gamma }^2}\\ \beta =\frac{1}{2}\sigma \left|\gamma \right|\\ \xi =\mu -2\frac{\sigma }{\gamma }\end{array}\end{array}$	μ, σ, γ

) are estimated by maximum likelihood; (d) the K–S test is used to test the hypothesis of sample distribution; (e) the optimal distribution of variables is screened based on Akaike information criterion (AIC) and minimum root mean square error (RSME).

The AIC is calculated from:

$$AIC=2k+n\ln (\mathrm{SSR}/n)$$

(10)

where AIC is the Akaike information criterion; k is the number of parameters; n is the number of samples; SSR is the sum of squared residuals.

The RSME is given by:

$$RMSE=\sqrt{\frac{{{\displaystyle \sum _{i=1}^n\left(F_e\left(x_i\right)-F_t\left(x_i\right)\right)}}^2}{n}}$$

(11)

where RMSE is the root mean square error; $F_e\left(x_i\right)$ is the empirical CDF at x_i; $F_t\left(x_i\right)$ is the theoretical CDF at x_i; n is the number of samples.

2.3.4. Coupled Scenarios of Multiple Erosion Forces

According to the analysis methods of Section 2.3.3, the optimal distribution of variables (annual precipitation and average wind) was screened. Then, the theoretical values corresponding to different probabilities (0.25 and 0.75) under the specified distribution could be calculated (Figure 2a) and used as the critical values of the nine coupled scenarios of multiple erosion forces (Figure 2b).

3. Results

3.1. Variation Analysis and Stationarity Correction of Annual External Erosion Forces

The annual variation in the precipitation and average wind in the Pisha sandstone area from 1960 to 2019 (Figure 3) shows that the annual precipitation increased at a rate of 0.43 mm/a, while the annual average wind speed decreased at a rate of 0.13 (m/s)/10a. The highest annual precipitation (656.8 mm) occurred in 1967 and the lowest (167.2 mm) in 1965. The highest annual average wind speed (3.18 m/s) occurred in 1960 and the lowest (2.05 m/s) in 1986. In addition, a nonparametric test (MMK) was used to test the trend of two annual series, which showed a non-significant upward trend (Zmmk = 0.72) in precipitation and a non-significant downward trend (Zmmk = −1.35) in the average wind speed. The change point test results of the two annual series showed that although there were no significant change points in annual precipitation, three change point levels were detected in the annual average wind speed: a first-order change point in 1981 (p < 0.001), second-order points in 1976 (p < 0.05) and 2004 (p < 0.01), and a third-order change point in 2014 (p < 0.05).

The annual average wind speed from 1960 to 2019 was divided into five subsequences according to the detected change points, and then the mean, Cv, and slope of the linear regression were used to compare the differences in the subsequences (

Table 3. Internal variability of annual average wind speed.

Statistical Parameters	Period
	P1	P2	P3	P4	P5
Mean /(m/s)	2.96	2.58	2.19	2.30	2.58
Cv	0.05	0.05	0.04	0.04	0.02
Slope of linear regression (m/s)/a	−0.0024	−0.0721	−0.0041	−0.0041	−0.0115

Note: the significance of the regression equations was greater than 0.05.

). The results showed weakly variable, non-significant decreases (p > 0.05) in the subseries, so the mean change was considered to be the main internal difference between the subseries. In order to keep the sequence consistent, the annual average wind speed was restored according to the mean difference in the subsequences (Figure 4). Subsequently, trend and change point tests were carried out on the modified series, and the test results failed to pass the significance test of 0.05 reliability, indicating that the series modified by the reduction method met the requirements of stationarity.

3.2. Differences in the Action Period and Spatial Distribution of External Erosion Forces

The intra-annual distribution of precipitation in the Pisha sandstone area from 1960 to 2019 (Figure 5) shows that the regional precipitation is composed of a single-peaked distribution, with June to September as the main peak, accounting for 75.8% of the annual precipitation. According to the detected change points, the intra-annual distribution of the average wind speed in the Pisha sandstone area from 1960 to 2019 was divided into five subsequences. In comparison, we found that the order of the monthly average wind speed in each period was consistent with the annual average wind speed, with March to June as the main peak. The variation coefficients of the monthly precipitation and average wind speed were calculated, and the monthly precipitation showed strong variation with a coefficient of variation of 1.08, while the average wind speed showed moderate variation with a coefficient of variation range of 0.11–0.17.

The multi-year average annual precipitation and average wind speed values of 12 meteorological stations in and around the study area were calculated, and the spatial distribution maps of the monitoring values (Figure 6) were obtained by kriging interpolation. At the spatial scale, the multi-year average annual precipitation in the basin ranged from 356.3 to 436.2 mm, with an obvious distribution feature of decreasing from southeast to northwest. Meanwhile, the multi-year average annual average wind speed in the basin ranged from 1.59 to 3.06 m/s, with an obvious distribution feature of increasing from southeast to northwest. The spatial distribution of the multi-year average annual precipitation and average wind speed were divided into seven classes by the natural breaks, and the distribution of the data in each class was counted (Figure 6). The results show that the spatial distribution of the multi-year average annual precipitation was more uneven than that of the multi-year average annual average wind speed, and the variation coefficients of the multi-year average annual precipitation and average wind speed were 0.55 and 0.19, respectively. The main distribution interval of the multi-year average annual precipitation was 395.5–403.6 mm, and that of the multi-year average annual average wind speed was 2.10–2.27 m/s.

3.3. Temporal Evolution of the Risk of Compound Events Consisting of Multiple External Erosion Forces

In this paper, six sampling distribution functions—GAM, EXP, WBL, GEV, GP, and P–III—were used to fit the annual precipitation and average wind speed, and then the optimal distribution of the variables was optimized (

Table 4. Optimization of the distribution of annual elements.

Annual Elements	Distribution	K–S Test		RSME	AIC
		h	p_Value
Precipitation	GAM	0	0.82	0.027	−428.4
	EXP	1	0.00	0.227	−175.8
	WBL	0	0.74	0.032	−409.3
	GEV	0	0.93	0.026	−434.5
	GP	1	0.00	0.163	−213.4
	P–III	0	0.92	0.025	−436.0
Average wind speed	GAM	0	0.84	0.032	−408.3
	EXP	1	0.00	0.302	−141.8
	WBL	0	0.75	0.033	−405.8
	GEV	0	0.93	0.028	−424.8
	GP	1	0.00	0.345	−123.6
	P–III	0	0.92	0.028	−421.5

). The results of the distribution optimization show that the annual precipitation followed the P–III distribution and the annual average wind speed followed the GEV distribution. Meanwhile, the PP plots of the above elements (Figure 7) show that the empirical and theoretical CDF of each element were well distributed around the 1:1 line. Then, theoretical values under 0.25, 0.5, and 0.75 probabilities were calculated based on the optimal distribution of the selected forces (

Table 5. Theoretical values of annual elements.

Cumulative Probability	Precipitation/mm	Average Wind Speed/(m/s)
0.25	323.3	2.89
0.5	389.0	2.96
0.75	460.1	3.03

).

Based on the theoretical values under the probabilities of 0.25 and 0.75, nine combination scenarios were constructed. The frequency of different combination scenarios (

Table 6. Frequency of the different combination scenarios in each decade.

Sort	1960s	1970s	1980s	1990s	2000s	2010s	Sum
P1W1	−	1	2	3	4	1	11
P1W2	2	−	−	−	−	−	2
P1W3	1	1	−	−	−	−	2
P2W1	1	2	7	5	5	5	25
P2W2	−	3	−	−	−	−	3
P2W3	3	−	−	−	−	−	3
P3W1	3	2	1	2	1	4	13
P3W2	−	−	−	−	−	−	−
P3W3	−	1	−	−	−	−	1

) was calculated, and the results show eight actual occurrence types of regional scenarios in the whole period, with five actual occurrence categories of regional scenarios in the 1960s, six in the 1970s, and a sharp reduction to three after the 1980s due to the stepwise variation of the actual monitored wind speed, which was mainly concentrated in the interval of [0, 0.25]. At the same time, when the annual precipitation or average wind speed fell into the interval of (0.75, 1], it was considered that the risk of soil and water loss caused by the above regional erosion force was greater than that of the general scenario, and this broad category included four types (P1W3, P2W3, P3W1, and P3W3) in the whole period. The frequency ranking of the above scenarios shows that P3W1 was the main scenario, with 13 occurrences, and that this type of event had the highest probability of occurring in the 2010s, at 40%. Therefore, regional precipitation-led erosion is still the main target for regional erosion control.

4. Discussion

4.1. Causes of Temporal and Spatial Differences between Two External Erosion Forces

It was found that the external forces causing erosion by wind in the western Pisha sandstone area was more intense than in the east, and that water erosion force was most intense in the east. The northwesterly monsoon is affected by the Siberian high-pressure system (the ‘Siberian High’) in the winter, while the southeasterly monsoon is affected by the Western Pacific Subtropical High in the summer, influencing the wind speed and precipitation distribution on a large scale [28,29,30]. The elevation of the land gradually increases from east to west in this area, which indirectly affects the local meteorological factors [31]. The specific heat capacities of bare and sand-covered land are smaller than for the soil cover, and vegetation on the former type of land are always less prolific than on the latter, which mainly affect the local wind speed [32].

In our research, non-stationarity analysis of the external forces found that only the stationarity of the wind was broken. The average wind speed monitored in 1982–2014 was distinctly smaller than in 2015–2019, which is consistent with a study conducted in northwestern China by Ge et al. [33]. While the precipitation in the Pisha sandstone area showed tiny fluctuations, similar related precipitation results were reported by Zhang et al. [19]. In summary, the wind was more sensitive than the precipitation in this area. The results of the present study are related to the implementation of soil and water conservation measures such as the Three-North Shelter Forest Program implemented in 1978 [34] and the changes in the temperature gradients between middle and high latitudes [35]. “Non-stationarity” is a typical statistical concept that cannot be ignored when carrying out studies involving long-term sequential data, which can be used to prove that under the impact of the above reasons, zonal wind has changed. Hence, when carrying out wind erosion-related experiments, it is necessary to adjust the experimental settings according to the current regional conditions.

4.2. Impacts and Adjustment of the Measures According the Results

Through the analysis of regional land use change (Figure 8), this study found that with the acceleration of urbanization and the implementation of the “Grain for Green Program (GFGP)” in 1999, all kinds of land use in the region have undergone corresponding changes, manifested in the fluctuations in area and the transformation of use types. The proportion of construction land and woodland increased by 4.5% and 0.8%, respectively, while the proportion of other land types decreased by 0.2–1.8% and was less than half of the increase rate of construction land. At the same time, the main land types being turned into grassland were unutilized land and arable land. In the present study, P3W1, a scenario with an annual precipitation that fell into the interval of (0.75, 1] and an annual average wind speed that fell into the interval of [0, 0.25], was the main scenario of the combination scenarios containing soil and water loss risk caused by the related regional erosion forces. Combined with regional land use change, in years with an abundance of water, specific weak links such as regional urban soil erosion and the stability of grassland that belongs to the area changed from unutilized land or farmland still need to be further studied and corresponding measures should be formulated.

In the present study, the two external erosion forces showed obvious spatial differentiation, as shown in Figure 6, which implies that the spatial distribution of the related erosion will be different in this area. Soil erosion is related to external erosion forces, but the response of external forces differs due to underlying surface conditions such as topography, land use, and soil [36,37]. To facilitate further erosion studies in the area, we overlaid the results of the grading of the external forces with the slope of the area (Figure 9a). Then, this study found that the average slope difference in the seven-level classification was the largest, and the slope under this level of annual precipitation was 6.1° larger than that of the annual average wind speed. We argue that the changes in geomorphological features caused by water erosion may be greater than those caused by wind erosion. When further comparing the land use change in the seven-level classification of external forces (Figure 9b,c), we found that the ratio of easily eroded land types such as arable and unutilized land decreased and the ratio of high soil and water conservation functional land types such as forests increased in this class for both external forces. At the same time, the ratio of better soil and water conservation functional land types such as grassland showed inconsistent changes in the seven-level classification of external forces, and the reason for this phenomenon was that the ratio of grassland converted to urban area in the annual average wind speed class 7 was large, and the transfer rate was 45%. To sum up, the erosion in the region undoubtedly decreased, even in areas where the external erosion force was large. Thus, constantly reduced sediment delivery, which can reflect the improvement of regional erosion in local basins, has been reported, although our research indicated that regional precipitation-led erosion has been the main target for regional control [38,39]. By increasing the surface cover, the GFGP has increased soil infiltration and reduced soil erosion [40]. The construction of check dams, one of the main projects of the Ministry of Water Resources of the People’s Republic of China implemented in 2003, has the function of retaining runoff and sediment and controlling gully erosion [41]. Combinations of the above management measures and changes in regional external forces have contributed to enhancing the capabilities of water storage and soil conservation, which have improved the local ecological living environment. However, to prevent potential erosion of the region, the corresponding measure of years with an abundance of water still need to be evaluated in terms of their validity, since P3W1 had been at the highest probability of occurring in the 2010s, at 40%.

5. Conclusions

Clarifying the variability characteristics between the primary drivers of water and wind erosion in time and space can contribute to the improvement of the scientific setting of external erosion forces and the adjustment of relevant erosion control measures during the changing environment affected by intensified climate and human activities. In the present study, the variability characteristics in the time and space of long-term precipitation and wind speed series in a typical wind–water erosion crisscross region located in the northeastern part of the LP were investigated using statistical analyses. In addition, the changing erosion risk caused by the regional erosion forces were revealed using the frequency changes in each decade of different combination scenarios constructed by the theoretical values under 0.25 and 0.75 probabilities, both in precipitation and modified wind speed series.

The key conclusions are summarized as follows: Interannually, the stationarity of the regional average wind speed over the last 60 years was disrupted compared to that of the precipitation, as evidenced by the presence of four significant change points (p < 0.05), namely, 1976, 1987, 2004, and 2014 within the series. Meanwhile, the mean change in the subseries was the main variation type that destroys the stationarity of the annual average wind speed series. Intra-annually, the precipitation and wind speed were both mainly distributed from March to June. Spatially, the precipitation decreased from southeast to northwest, while the wind speed showed the opposite. P3W1 was the main scenario of the combination scenarios containing soil and water loss risk, and had the highest probability of occurring in the 2010s, at 40%. This is why, considering only the primary drivers, regional precipitation-led erosion has become the main risk erosion type.