Analysis of Spatiotemporal Variability and Drivers of Soil Moisture in the Ziwuling Region

Abstract

Understanding soil moisture’s spatiotemporal variations and the factors influencing it is crucial for the restoration and growth of vegetation across the Loess Plateau, particularly in the Ziwuling region. This study employs soil moisture remote sensing data, complemented by information on soil properties, environmental conditions, and topography, to examine soil moisture variability within the Ziwuling region between 2001 and 2020. Using trend analysis, geographic detectors, and multi-scale geographic weighting techniques, this research aims to elucidate the effects of driving factors on soil moisture’s spatiotemporal patterns. The findings indicate the following: (1) Over the study period, the mean soil moisture in the Ziwuling region exhibited a relatively stable declining trend, with an annual decrease of −0.00047 m³/(m³·a). Spatially, higher soil moisture levels were observed in the south-central area, while lower levels occurred in the northern, western, and eastern peripheries. (2) Geoprobe analysis illustrated that the normalized difference vegetation index (NDVI) had the most notable effect on the spatial distribution of soil moisture in the region. As a direct indicator of vegetation cover, NDVI strongly affects soil moisture distribution through ecological and hydrological processes. Following NDVI, average annual potential evapotranspiration and annual precipitation were identified as the next most influential factors. The combined effect of these factors on soil moisture surpassed that of individual factors, with the interaction between NDVI and annual precipitation being particularly pronounced, predominantly controlling the spatial variability of soil moisture in the Ziwuling region. (3) Different factors exhibited varying effects on soil moisture levels. Notably, slope and elevation consistently had negative impacts, whereas variables such as soil texture (loam and sand), land use, temperature, precipitation, NDVI, and slope aspect showed bidirectional influences. This study offers a comprehensive analysis of the spatiotemporal variability of soil moisture and its controlling factors in the Ziwuling region, ultimately offering a scientific basis to support ecological restoration and sustainable development initiatives on the Loess Plateau.

1. Introduction

Soil moisture is a vital component of the hydrological cycle in forest ecosystems, playing a key role in mediating the material and energy exchange among soil, vegetation, and the atmosphere [1,2]. Precipitation, groundwater, and irrigation water undergo various transformations within the soil, such as infiltration, retention, and redistribution, to become ecologically available water. This water is essential for plant root absorption and supports critical physiological processes, including photosynthesis and transpiration [3,4]. As the main source for vegetation uptake, soil moisture serves as an important indicator of plant growth and their capability to adapt to environmental settings [5,6,7,8,9]. In water-limited arid and semi-arid regions, the spatiotemporal distribution of soil moisture functions as a crucial regulator, influencing the ecological resilience and stability of these areas [10,11,12,13].

Currently, the primary methods for obtaining soil moisture information include manual observation, automatic sensor measurement, and multi-source remote sensing technology. Manual observation, primarily conducted through field measurements, offers high accuracy but is time-consuming and labor-intensive, which makes it difficult to support large-scale, long time-series data acquisition. Although automatic soil moisture monitors can quickly obtain point-level data, their high cost and limited deployment density make them unsuitable for large-scale regional coverage. In contrast, multi-source remote sensing technology, through remote sensing inversion and land surface process model simulation, can provide real-time, macroscopic, and rapid soil moisture products [14,15,16]. With the continuous advancement of remote sensing technology, it has been widely applied in soil moisture research, achieving a significant leap from point to area. This technology possesses remarkable advantages, capable of providing comprehensive, consistent, and quantitative observation results across a wide range of areas. It is of great importance for revealing the spatiotemporal dynamic changes in soil moisture on a large scale [17,18,19]. Research conducted in the Weihe River Basin reveals a significant declining trend in soil moisture from 2001 to 2020, indicating a clear drying pattern [20,21]. In contrast, soil moisture in the Mongolian Plateau shows an initial rise in moisture content, followed by a reduction with depth [22]. Based on remote sensing soil moisture data, these studies highlight the spatial complexity and heterogeneity of soil moisture changes and deeply analyze their driving mechanisms. It is found that spatiotemporal variation in soil moisture is the outcome of the combined effects of multiple factors, including climatic conditions (precipitation, evaporation, temperature), soil features (bulk density, soil texture, organic matter content), vegetation characteristics (land-use type, vegetation coverage, root distribution), topographic features (slope, aspect, elevation), and human activities. However, although existing studies have conducted in-depth discussions on soil moisture changes and their influencing factors from multiple perspectives [23,24], they have seldom quantified the impact intensity and spatial heterogeneity (such as positive promotion or negative inhibition) of different driving factors. In this study, we employ a geodetector and a multi-scale geographically weighted regression (MGWR) model. The geodetector identifies the primary factors driving spatiotemporal variations in soil water content and evaluates their interactive effects, whereas the MGWR quantifies the magnitude and direction of these influences [25,26,27]. The integration of these two approaches overcomes the limitations of conventional methods by discerning both positive and negative impacts, addressing spatial heterogeneity, and enabling quantitative analysis and spatial visualization of causal mechanisms [28].

The Ziwuling region on the Loess Plateau faces severe soil erosion and extensive damage to its native vegetation. Following a comprehensive initiative in 1999 that involved converting farmland back to forests and grasslands, alongside implementing vegetation restoration and reconstruction efforts, there has been a notable improvement in the region’s vegetation status [29,30,31,32]. Nevertheless, the emphasis on establishing artificial forests and grasslands has largely focused on achieving rapid growth and maximizing economic returns over the years. This approach has led to a gradual decline in soil moisture during the transition from initial to advanced stages of vegetation development in certain areas [33]. Moreover, as vegetation matures, soil water deficits intensify, resulting in an adverse water balance [34,35,36]. Compared with native flora, introduced Robinia pseudoacacia plantations particularly exacerbate soil moisture depletion [37].

This paper utilizes a global high-resolution, continuous soil moisture dataset, generated by integrating multiple remote sensing data sources and machine learning techniques, to scrutinize the spatiotemporal variations in soil moisture in the Ziwuling region. This study also evaluates the evolving trends of soil moisture in the area. By combining geodetector analysis with multi-scale weighted regression, the research aims to identify the key factors impacting spatiotemporal soil moisture variations and quantify their significance. The ultimate goal is to propose effective strategies for managing soil moisture fluctuations in the Ziwuling region. Through this investigation, we have determined the drivers of spatiotemporal soil moisture variability and assessed their impacts using geographic detectors and multi-scale weighted regression analyses. These findings provide a solid foundation for monitoring soil moisture changes and implementing measures for ecological conservation and restoration in the Ziwuling region.

2. Materials and Methods

2.1. Study Area

The Ziwuling region, situated between Qingyang in Gansu and Yan’an and Tongchuan in Shaanxi, is a mountainous area within the interior of the Loess Plateau (Figure 1). The study area is located at 34°50′–36°35′ N and 108°05′–109°25′ E, covering an area of approximately 1.1 km² and situated at an altitude ranging from 800 to 1900 m. The overall terrain is high and low in the west and the east, respectively. Geologically, the study area belongs to the Ordos platform of the North China platform and is also one of the mountainous areas higher than the bedrock of the Loess Plateau. The landform is typical of the hilly and gully region of the Loess Plateau, with the surface extensively covered by loess, making it one of the most typical areas for the development of loess landforms. The geology mainly consists of systems: the underlying layer is composed of thick Triassic, Jurassic, and Cretaceous strata, consisting of sandstone, mudstone, shale, and conglomerate; the middle layer is composed of Paleogene red clay, subclay, and semi-cemented conglomerate; and the upper layer is composed of 120–150 m thick Quaternary Pleistocene eolian and pluvial loess or loess. The area belongs to a semi-arid monsoon climate, with an annual average temperature of about 11.5 °C and an annual average rainfall of about 567 mm. The soil type is mainly gray-brown soil, which is developed under the conditions of warm temperate semi-humid to semi-arid climate and forest. Due to the influence of natural landforms, vegetation, climate, and various human activities, the soil organic matter accumulation is relatively abundant, and soil calcification and clayization are not obvious. Overall, the soil is moist and fertile, suitable for forest growth. Therefore, the forest coverage rate in this area is relatively high, and the vegetation is dominated by natural secondary forests on the Loess Plateau, mainly including evergreen coniferous and deciduous broadleaf forests, mixed forests, and temperate coniferous forests.

2.2. Data Sources and Pre-Processing

The datasets utilized in this research span the period from 2001 to 2020. To ensure consistency, all datasets were uniformly resampled to a 1 km spatial resolution utilizing the nearest-neighbor interpolation method (

Table 1. Summary of data sources and characteristics.

Data Name	Source	URL	Spatial Resolution	Temporal Resolution	Processing
Soil Moisture	National Tibetan Plateau Data Center	http://data.tpdc.ac.cn	1 km	Monthly	Monthly values were averaged to attain annual mean values.
Soil Texture	Resource and Environment Science Data Center	https://www.resdc.cn/	1 km	-	Reclassification is used to analyze the influence of soil characteristics on soil moisture.
Land Use and Land Cover Change	Wuhan University CLCD	https://zenodo.org/records/12779975	30 m	Annual	Resample to a resolution of 1 km and reclassify.
Potential Evapotranspiration	National Earth System Science Data Center	http://www.geodata.cn/main/	1 km	Monthly	Average the monthly values to obtain the annual average evapotranspiration.
Temperature	National Tibetan Plateau Data Center	http://data.tpdc.ac.cn	1 km	Monthly	Monthly values were averaged to obtain annual mean temperature.
Precipitation	National Tibetan Plateau Data Center	http://data.tpdc.ac.cn	1 km	Monthly	Monthly values were averaged to obtain annual mean precipitation.
Normalized Difference Vegetation Index	NASA MOD13A3 Dataset	https://www.earthdata.nasa.gov	1 km	Monthly	Average the monthly values to achieve the annual average normalized difference vegetation index.
Digital Elevation Model	Geospatial Data Cloud	http://www.gscloud.cn	30 m	-	Mosaicked and clipped to extract topographic factors (slope, aspect) for the study area.

).

Through bibliometric analysis, 11 key evaluation indicators were comprehensively selected (

Table 2. Geographic detector factor selections.

Symbol	Factor	Factor
X1	Silt	Silt
X2	Sand	Sand
X3	Clay	Clay
X4	Land Use and Land Cover Change	LUCC
X5	Potential Evapotranspiration	PET
X6	Temperature	TEMP
X7	Precipitation	PREP
X8	Normalized Difference Vegetation Index	NDVI
X9	Aspect	Aspect
X10	Slope	Slope
X11	Digital Elevation Model	DEM

), with impact factors categorized by different levels. Silt (X1) and clay (X3) were divided into 10 categories based on their respective content levels; sand (X2) was divided into 13 categories based on its content level; land-use type (X4) was grouped into 6 main categories: forest, shrubland, grassland, cropland, water, and impervious surface; annual mean potential evapotranspiration (X5), annual mean temperature (X6), annual precipitation (X7), normalized difference vegetation index (NDVI) (X8), aspect (X9), slope (X10), and elevation (X11) were separated into 8 categories via the natural breaks classification method (Jenks).

2.3. Research Methodology

2.3.1. Trend Analysis

The Theil–Sen method was utilized to scrutinize soil moisture trends from 2001 to 2020 at the pixel scale. This method is widely recognized for its computational efficiency, robustness against measurement errors and non-normally distributed (leptokurtic) data, and its suitability for detecting trends in long-term time-series datasets. The mathematical expression is given as follows:

$$\mathsf{\beta } = \mathrm{median}\left({\displaystyle \frac{{\mathrm{x}}_{\mathrm{j}} - {\mathrm{x}}_{\mathrm{i}}}{\mathrm{j}-\mathrm{i}}}\right) \forall \mathrm{j} > \mathrm{i}$$

(1)

where β indicates the slope of the soil moisture trend; Median denotes the median function; j and i refer to the time-series indices; and x_i and x_j represent the soil moisture values in the ith and jth years, respectively.

When β > 0, it indicates a rising trend in soil moisture over the specified period; when β = 0, it reflects a stable soil moisture level; and when β < 0, it represents a reducing trend in soil moisture across the given time scale.

2.3.2. Mann–Kendall (MK) Significance Test

The MK test was applied to evaluate the significance of soil moisture trends. This method is advantageous because it does not require the data to follow a specific distribution, is unaffected by missing values and outliers, and is well adjusted for analyzing trends in long-term time-series datasets. The correlation formula is expressed as follows:

$$\mathrm{Z} = \left\{\begin{array}{l}{\displaystyle \frac{\mathrm{S}-1}{\sqrt{\mathrm{S}\left(\mathrm{s}\right)}}} \mathrm{S} > 0\\ 0 \mathrm{S} = 0\\ {\displaystyle \frac{\mathrm{S} + 1}{\sqrt{\mathrm{S}\left(\mathrm{s}\right)}}} \mathrm{S} < 0\end{array}\right.$$

(2)

$$\mathrm{S}=\sum _{\mathrm{j}=1}^{\mathrm{n}-1}\sum _{\mathrm{i}=\mathrm{j}+1}^{\mathrm{n}}\mathrm{sgn}\left({\mathrm{x}}_{\mathrm{j}}-{ \mathrm{x}}_{\mathrm{i}}\right)$$

(3)

$$\mathrm{S}\left(\mathrm{s}\right)={\displaystyle \frac{\mathrm{n}\left(\mathrm{n}-1\right)\left(2\mathrm{n}-5\right)}{18}}$$

(4)

$$\mathrm{sgn}\left({\mathrm{x}}_{\mathrm{j}}-{ \mathrm{x}}_{\mathrm{i}}\right)=\left\{\begin{array}{l}1 {\mathrm{x}}_{\mathrm{j}}-{ \mathrm{x}}_{\mathrm{i}} > 0\\ 0 { \mathrm{x}}_{\mathrm{j}}-{\mathrm{x}}_{\mathrm{i}} =0\\ -1{ \mathrm{x}}_{\mathrm{j}}-{ \mathrm{x}}_{\mathrm{i} }< 0\end{array}\right.$$

(5)

where Z represents the test statistic for the trend test; x_j and x_i denote the soil moisture values in years j and i; n is the length of the time series; and sgn indicates the sign function.

A two-tailed trend test was conducted to determine the critical value, Z_1−α/2, from the standard normal distribution table at the chosen significance level. The null hypothesis was accepted when |Z| ≤ Z_1−α/2, indicating a non-significant trend, and rejected when |Z| > Z_1−α/2, signifying a significant trend. In this study, with a significance level of α = 0.05, the critical values Z_1−α/2 were ±1.96. Absolute Z values greater than 1.65, 1.96, and 2.58 correspond to trend significance at the 90%, 95%, and 99% confidence levels, respectively. The criteria for measuring trend significance are presented in

Table 3. MK significance test categories.

β	Z	Trend Properties
β > 0	2.58 < Z	Extremely notable growth
	1.96 < Z ≤ 2.58	Significant growth
	1.65 < Z ≤ 1.96	Slightly significant growth
	Z ≤ 1.65	Unremarkable growth
β = 0	0	No variation
β < 0	Z ≤ 1.65	Insignificant drop
	1.65 < Z ≤ 1.96	Slightly significant drop
	1.96 < Z ≤ 2.58	Significant drop
	2.58 < Z	Extremely significant drop

.

2.3.3. Stability Analysis

The coefficient of variation (Cv) is utilized to evaluate the stability and variability of soil moisture over time, reflecting both the degree of dispersion within the time series and the fluctuations around its mean level. It is calculated as follows:

$$\mathrm{Cv}={\displaystyle \frac{\mathsf{\sigma }}{\overline{\mathrm{x}}}}$$

(6)

where σ represents the standard deviation and x denotes the mean. A smaller Cv value indicates lower variability and risk, implying a smoother overall data trend with reduced fluctuations.

2.3.4. Geographic Detector

The geographic detector, meanwhile, is a statistical method employed to identify spatial variability in geographical features and to determine the key factors impacting them.

(1) Factor Detection

The factor detector was applied to quantitatively assess the influence of specific environmental factors on the spatial distribution of soil moisture in the Ziwuling region. This assessment was carried out using the q-value, which was determined according to the following formula:

$$\mathrm{q} = 1-{\displaystyle \frac{\sum _{\mathrm{g} = 1}^{\mathrm{k}}{\mathrm{N}}_{\mathrm{g}}{\mathsf{\sigma }}_{\mathrm{g}}^2}{\mathrm{N}{\mathsf{\sigma }}^2}}$$

(7)

where q represents the explanatory power of the independent variables, ranging from 0 to 1, with larger q-values suggesting robust explanatory power. The classifications of the variable Y (soil moisture) are denoted by k and g; N_g is the number of cells in class g; σ_g² is the variance of soil moisture within class g; N is the total number of cells in the study area; and σ² is the variance of soil moisture across the entire region. Based on a comprehensive review of the literature, 11 variables related to soil moisture were selected, including mean annual temperature (TEMP), annual precipitation (PREP), NDVI, mean annual potential evapotranspiration (PET), land use and land cover change (LUCC), sand, silt, clay, digital elevation model (DEM), slope, and aspect. These 11 environmental factors were subsequently analyzed to determine their influence on soil moisture.

(2) Interaction probing

Interaction probing is used to examine the combined effects of ecological factors on the spatial distribution of soil moisture. Initially, the q-values for two independent variables, X1 and X2, are calculated individually using factor detection to obtain q(X1) and q(X2). Then, the q-value representing the interaction between the two factors, represented as q(X1∩X2), is determined to evaluate whether the spatial variability of soil moisture is enhanced or diminished. On the basis of these results, the nature of the interaction is identified, allowing the association between the two factors to be categorized as described in

Table 4. Types of interaction between two variables.

Interactions	Basis of Judgment
Nonlinear damping	q(X1 ∩ X2) < Min [q(X1),q(X2)]
Single-factor nonlinear damping	Min [q(X1),q(X1)] < q(X1 ∩ X2) < Max [q(X1),q(X2)]
Dual-factor intensification	q(X1 ∩ X2) > Max [q(X1),q(X2)]
Independent	q(X1 ∩ X2) = q(X1) + q(X2)
Nonlinear amplification	q(X1 ∩ X2) > q(X1) + q(X2)

.

(3) Risk detection

Risk detection was applied to estimate whether significant differences existed in the mean soil moisture values between the two subregions, using the t-statistic:

$${\mathrm{t}}_{{\overline{\mathrm{y}}}_{\mathrm{h} = 1} - {\overline{\mathrm{y}}}_{\mathrm{h} = 2}} = {\displaystyle \frac{{\overline{\mathrm{Y}}}_{\mathrm{h} = 1} - {\overline{\mathrm{Y}}}_{\mathrm{h} = 2}}{{\left[\frac{\mathrm{Var}\left({\overline{\mathrm{Y}}}_{\mathrm{h} = 1}\right)}{{\mathrm{n}}_{\mathrm{h} = 1}}+\frac{\mathrm{Var}\left({\overline{\mathrm{Y}}}_{\mathrm{h} = 2}\right)}{{\mathrm{n}}_{\mathrm{h} = 2}}\right]}^{\frac{1}{2}}}}$$

(8)

where ${\overline{Y}}_h$ is the mean soil moisture in sub-area h, nh is the number of samples in sub-area h, and Var denotes the variance of soil moisture values.

2.3.5. MGWR

To investigate the underlying mechanisms governing the spatial distribution of soil moisture in the Ziwuling region and to accurately characterize the spatial heterogeneity of various environmental influences, this study employed the MGWR model. This model offers a key advantage in capturing spatial correlations and non-uniformity by allowing each factor to have an independent optimal spatial bandwidth for its effect. The model was calibrated using a backward-fitting algorithm, with bandwidths adaptively adjusted to account for potential variations in the impacts of different factors on soil moisture across geographic scales. Compared to the conventional GWR model, MGWR overcomes the limitations associated with selecting a single optimal bandwidth, enabling a more flexible and realistic identification of the spatial scales of influencing factors. It effectively captures the contributions of processes operating at different scales to soil moisture spatial heterogeneity, thereby significantly enhancing both explanatory power and predictive accuracy regarding the driving forces behind soil moisture patterns in the region. The calculation formula is provided below:

$${\mathrm{y}}_{\mathrm{i}} ={ \mathsf{\beta }}_0\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right) + \sum _{\mathrm{j} = 1}^{\mathrm{k}}{\mathsf{\beta }}_{\mathrm{bwj}}\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right){\mathrm{x}}_{\mathrm{ij}} + {\mathsf{\epsilon }}_{\mathrm{i}}$$

(9)

where y_i is the fitted soil moisture value for sample i; x_ij is the value of the jth driving factor at sample i; β₀(u_i,v_i) is the local intercept term; β_bwj(u_i,v_i) denotes the spatially varying coefficient of the jth factor, adjusted using a specific regression bandwidth; and ε_i is the error term. Calculations were performed using MGWR 2.2 software.

3. Results

3.1. Characterization of Spatiotemporal Variability of Soil Moisture

3.1.1. Characterization of Temporal Changes in Soil Moisture

Since 2001, soil moisture in the Ziwuling region has consistently exceeded 0.12 m³/m³. In 2001, soil moisture levels were predominantly within the 0.18–0.20 m³/m³ range, covering 55.98% of the total study area. By 2020, soil moisture primarily fell within the 0.17–0.19 m³/m³ range, with the areal proportion of this interval increasing from 45.65% to 49.13% compared to 2001 (Figure 2).

Over the period 2001–2020, the average soil moisture in the study area exhibited a slightly fluctuating but overall declining trend, with an annual decrease rate of −0.00047 m³/(m³·a). During these two decades, soil moisture initially declined to 0.179 m³/m³ and then rose rapidly to a maximum of 0.201 m³/m³ in 2003, followed by a sharp decline to a minimum of 0.169 m³/m³ in 2009. Beginning in 2010, soil moisture fluctuated with an overall downward trend, characterized by alternating increases and decreases, before transitioning to an increasing trend after 2016. Except for 2003, there were no significant changes in soil moisture in the study area over time. Its extreme values occurred at the same time as the highest and lowest values of precipitation, but its interannual variation trend did not always align with precipitation (Figure 3).

3.1.2. Patterns of Spatial Variation in Soil Moisture

The average soil moisture in the Ziwuling region over multiple years was 0.178 over the years from 2001 to 2020, with approximately 80.28% of the area exhibiting values between 0.15 and 0.20 m³/m³. Spatial analysis indicated higher soil moisture levels in the central and southern parts of the region, while lower levels were observed in the northern, western, and eastern peripheries. Areas with higher soil moisture correspond to regions with the highest precipitation, but the spatial distribution patterns of the two are not completely consistent (Figure 4).

Theil–Sen trend analysis was employed on a per-pixel basis to examine soil moisture trends in the Ziwuling region from 2001 to 2020, producing a spatial distribution map of the annual soil moisture change rate (β) (Figure 5). Over the past 20 years, the annual change rate ranged from −0.012 to 0.013 m³/(m³·a), with the predominant slope falling between −0.01 and 0, covering 75.88% of the study area. Regions exhibiting a decreasing interannual soil moisture trend were mainly focused in the southern and central parts, encompassing 75.92% of the total area. Conversely, areas showing an increase in soil moisture accounted for only 24.08%, mainly located in the northwestern, central-eastern, and eastern peripheral zones. These results indicate an overall declining trend in regional soil moisture, although the decrease occurs at a relatively gradual pace.

The MK test statistic (Zc) was utilized to generate a significance classification map of soil moisture trends, as shown in Figure 6. Approximately 21.64% of the study area exhibited increasing soil moisture, including highly significant increases (3.19%, p < 0.01), notable increases (2.01%, p < 0.05), slight increases (1.37%), and non-significant increases (15.07%). These areas were primarily located in the northwestern, eastern, and northern peripheries. In contrast, about 70.17% of the region showed decreasing soil moisture, categorized as highly significant considerable decreases (17.12%, p < 0.01), marked decreases (11.53%, p < 0.05), slight decreases (6.90%), and unremarkable decreases (34.63%). These areas were predominantly clustered in the northern, southern, and central parts of the study region.

Both the annual mean change rate and the Cv are key parameters for characterizing the soil moisture’s interannual variability in the study area. The Cv indicates multi-year stability, classified into three categories: weak variation (Cv = 0–0.15), moderate variation (Cv = 0.15–0.30), and strong variation (Cv > 0.30). As shown in Figure 7, the Cv values for soil moisture across the Ziwuling region ranged up to 0.113. Notably, 99.94% of the area exhibited Cv values between 0 and 0.1, indicating a high degree of stability and consistent interannual variability in soil moisture.

3.2. Spatial Driving Force Analysis of Soil Moisture

3.2.1. Factor Detector Analysis

Factor detector analysis was conducted to evaluate the influence of 11 factors on soil moisture in the Ziwuling region, with the relative impact of each factor quantified by its q-value. Significant variances were noted in the explanatory power of individual factors. Among the contributing factors (

Table 5. Detection q-values of soil moisture factors in the Ziwuling region.

Factor	q-Value	Sort
X8	0.455	1
X5	0.325	2
X7	0.267	3
X2	0.250	4
X3	0.236	5
X1	0.225	6
X4	0.137	7
X11	0.119	8
X6	0.098	9
X9	0.006	10
X10	0.001	11

), X8 (NDVI) had the highest q-value (0.455), indicating the strongest influence on spatial variation in soil moisture and establishing NDVI as the primary controlling factor. X5 (annual mean potential evapotranspiration) ranked second (q = 0.325), reflecting its role in depleting soil water and modifying moisture content. X7 (annual precipitation) was third (q = 0.267). Among soil texture factors, sand content (X2) exhibited the greatest driving effect (q = 0.250), followed by clay content (X3) (q = 0.236), while silt content (X1) had the weakest influence (q = 0.225). X4, X11, and X6 showed relatively lower explanatory power (q = 0.137, 0.119, and 0.098, respectively). Among topographic factors, elevation (X11) had the strongest effect, whereas slope (X10; q = 0.006) and aspect (X9; q = 0.001) contributed negligibly to spatial variation and did not pass significance tests (p > 0.01).

3.2.2. Analysis of Interaction Detector

The interaction detector analysis for driving factors of soil moisture spatial differentiation in the Ziwuling region is presented in Figure 8. All factor interactions exhibited enhancement effects, with most combinations showing bilinear enhancement and others displaying nonlinear enhancement, indicating that combined factors improve the explanatory power for the dependent variable (soil moisture, Y). The q-values of interactions between dominant environmental factors and other factors exceeded those of individual factors, demonstrating bilinear enhancement and confirming that spatial differentiation arises from multi-factor interactions. The interaction pair with the highest average explanatory power was X8 ∩ X7 = 0.657, where X8 (NDVI) and X7 (annual precipitation) ranked first and third in the individual factor analysis, establishing their interaction as the primary driver of spatial differentiation. Moreover, interactions involving the top three individual factors (X5, X7, X8) with all other driving factors yielded substantially elevated q-values: interactions of X5 and X7 with all factors except slope and aspect exceeded 0.35, while all interactions involving X8 surpassed 0.45, highlighting NDVI, annual potential evapotranspiration, and annual precipitation as dominant drivers. Secondary high-explanatory interactions included X8∩X2 = 0.578, X8∩X1 = 0.569, and X8∩X3 = 0.569, indicating that specific soil textures notably amplify NDVI’s explanatory power. Additionally, interactions between soil texture factors and X4 (land use), X6 (temperature), and X11 (elevation) maintained high explanatory power (q > 0.32).

3.2.3. Risk Detector Analysis

After excluding under-exhibited fluctuations with X1 (silt), X2 (sand), X3 (clay), X4 (land use), and X9 (aspect), peaking at silt 32–34%, sand 59–62%, clay 21–23%, water bodies, and aspects 267.51–312.represented soil texture categories and associated sampling points, risk detector analysis was conducted to classify sand (X2) into eight categories using the natural breaks (Jenks) method. Similarly, clay (X3) and silt (X1) were classified into seven categories with the same method.

Analysis of the 20-year risk detector results identified optimal ranges for soil moisture spatial distribution (Figure 9). Soil moisture 14°. It decreased with increasing X5 (potential evapotranspiration, peak at 71.13–75.85 mm·a⁻¹) and X6 (temperature, peak at 7.68–8.46 °C) and increased with X7 (precipitation, peak at 625.04–656.64 mm), X8 (NDVI, peak at 61.2–69.4%), and X11 (elevation, peak at 1564–1828 m). Soil moisture showed a non-significant increase with X10 (slope, peak at 23.78–42.30°).

3.3. Multi-Scale Geographically Weighted Analysis

Collinearity analysis was performed to remove variables exhibiting multicollinearity, and all retained factors had variance inflation factors (VIFs) below 7.5 (

Table 6. VIF values for each factor.

Symbol	Factor	VIF
X1	0.455	2.578
X2	0.325	2.850
X4	0.267	1.128
X6	0.250	6.951
X7	0.236	1.703
X8	0.225	1.813
X9	0.137	1.006
X10	0.119	1.015
X11	0.098	6.027

). Using the MGWR model, location-specific relationships between driving factors and soil moisture were analyzed, producing spatial distributions of regression coefficients for each factor (Figure 10). In these spatial maps, the absolute values of regression coefficients represent the magnitude of each factor’s impact on soil moisture, while their signs (positive/negative) indicate the direction of the effect.

The regression coefficients of different driving factors ranged from −1.517 to 1.244 (

Table 7. Standardized regression coefficient of different influencing factors in the MGWR model.

Factor	Mean	Standard Deviation	Minimum	Median	Maximum
X1	0.000	0.002	−0.006	0.001	0.004
X2	−0.030	0.206	−0.727	−0.029	0.545
X4	−0.003	0.002	−0.006	−0.004	0.002
X6	−0.017	0.328	−1.517	0.028	0.892
X7	−0.177	0.418	−1.166	−0.127	0.515
X8	0.661	0.218	−0.157	0.664	1.244
X9	0.000	0.010	−0.018	−0.001	0.021
X10	−0.007	0.004	−0.015	−0.007	−0.000
X11	−0.015	0.003	−0.022	−0.015	−0.011

). Based on mean coefficient magnitudes, NDVI exerted the strongest impact on soil moisture, consistent with the geodetector results, followed by annual precipitation and annual mean temperature. Silt content and aspect showed the weakest effects on spatial distribution. The standard deviations of the regression coefficients reflected variability in the relationships between soil moisture and explanatory variables: annual precipitation and annual mean temperature exhibited the highest variability, followed by NDVI and sand content, while silt content and land-use type demonstrated the lowest variability.

Multi-scale geographically weighted regression revealed location-specific directions of factor effects on soil moisture (Figure 10). Silt content had positive effects in eastern and central Ziwuling but negative effects elsewhere. Sand content exerted positive impacts in the western, northwestern, northeastern, and southern peripheral areas, with negative effects in other regions. Land-use type predominantly exhibited negative effects. Soil moisture response to annual mean temperature was dual-natured, showing positive effects in western, central, northern, and southern peripheral regions. Annual precipitation had positive effects in northern-central and southern zones but negative impacts elsewhere. NDVI displayed positive effects across most areas, except in the central, northern, northwestern, and southern peripheral regions. Aspect showed negative effects in eastern and central Ziwuling but positive effects elsewhere. Slope and elevation consistently demonstrated negative effects, with slope intensity increasing westward and elevation intensifying toward the east and south.

4. Discussion

4.1. Spatiotemporal Analysis of Soil Moisture

In terms of time, the soil moisture in the Ziwuling region experienced significant fluctuations from 2001 to 2020, showing a slow downward trend. The dominant range of soil moisture gradually shifted from 0.18–0.20 m³/m³ to 0.17–0.19 m³/m³. Studies have found that in the semi-arid loess region, the suitable soil moisture range for maintaining net photosynthesis in Pinus tabulaeformis and Platycladus orientalis is 10–18% and 8–18%, respectively. The effective water range for Leymus chinensis communities and Robinia pseudoacacia forests is between 22.65% and 28.13% [38,39]. It is noteworthy that the “ideal” or “optimal” soil moisture content is not a fixed value but rather is influenced by various factors such as soil texture, climatic conditions, and vegetation type. In ecologically fragile areas like the Loess Plateau, due to intense evaporation and uneven precipitation distribution, soil moisture is often in a deficit state. Furthermore, groundwater in this region is buried deep, which makes it very difficult to provide effective replenishment for the consumption of upper soil moisture. Prolonged water deficit can easily lead to the formation of dry layers in the soil, resulting in soil moisture content often being lower than theoretical “ideal” values such as field capacity [40,41,42]. Therefore, for this region, the soil moisture range that maintains ecosystem stability is more practically meaningful for reference than a certain absolute “ideal value” [43]. This process can be understood as a dynamic equilibrium, which can be described through the “compensation” state of soil moisture: when the soil water storage increases, it is in a positive compensation state, reflecting that the water is replenished and restored; when the water storage remains unchanged, it indicates that the water balance has been achieved; and when the water storage decreases, it is in a negative compensation state, meaning that the water consumption exceeds the replenishment [44]. This study indicates that the overall migration of soil moisture in the region towards lower values is a specific manifestation of the negative compensation state of soil moisture. This water migration within different zones is primarily attributed to the extensive implementation of the Three-North Shelterbelt Program and a series of projects aimed at preventing soil erosion and restoring ecological vegetation, such as converting farmland back to forests, in the Ziwuling region. These large-scale artificial forestry ecological projects have effectively increased forest coverage in the area. While these initiatives substantially increased forest coverage [45,46], they also led to heightened transpiration rates, resulting in the depletion of soil moisture reserves [47]. Land-use changes emerged as a primary driver of this decline in moisture content [48]. Notably, a peak in moisture levels was observed in 2003 (0.201 m³/m³), directly correlated with above-average precipitation, which serves as the principal natural source for recharging soil water on the Loess Plateau [49,50]. Fluctuations after 2004 can be attributed to interactions between climate and vegetation, including increased evapotranspiration [51] and higher water consumption driven by enhanced NDVI [52]. The observed recovery trend after 2016 may be linked to stabilized vegetation growth, where mature canopies aid in rainfall interception and developed litter layers enhance water retention [53]. These findings underscore the importance of continuous long-term monitoring efforts.

Spatially, 76.92% of the region experienced a decrease in soil moisture levels from 2001 to 2020, consistent with degradation patterns observed across the Loess Plateau. Jiao et al. [54] reported that 72.64% of the plateau showed reduced moisture levels (1998–2000 vs. 2008–2010) based on ECV SM-derived SWI. Interestingly, this decline coincided with a significant increase in NDVI across 80.99% of the plateau, reflecting the success of vegetation restoration efforts under the Grain for Green Program (GGP). However, the accelerated evapotranspiration rate (+2.8694 mm/yr during 2001–2020) and water deficits caused by climate warming exceeded the capacity for precipitation recharge, highlighting the hydrological risks associated with heavy reliance on vegetation expansion in arid and semi-arid ecosystems. Within areas experiencing declining soil moisture (70.17% of the region), although 34.63% showed statistically insignificant decreases, these persistent negative trends may contribute to the formation of dry soil layers over time [55]. These dynamics illustrate an ecological “water resource dilemma”: while vegetation restoration mitigates soil erosion, excessive vegetation cover or inappropriate species selection (e.g., high-water-consumption trees) can lead to severe moisture depletion [56,57,58,59,60]. Therefore, optimizing vegetation allocation is essential to balance ecological protection with water sustainability [61,62,63].

4.2. Driving Factor Analysis

4.2.1. Soil Texture and Topography

Among soil texture factors, sand exhibited the greatest explanatory capacity for soil moisture, in line with the findings of Dong Jinyi et al. [22], followed by silt and clay. The negative impact of sand on soil moisture in most regions can be attributed to the high porosity and low water retention of sandy soils. These soils have rapid infiltration rates and are prone to water loss, resulting in lower water content compared to other soil types. In contrast, clay-rich soils typically have low porosity but high water retention capacity due to their fine particles and large specific surface area, which enables them to adsorb and hold more water. Consequently, under similar conditions, clayey soils generally maintain higher moisture content. Soils with a moderate proportion of fine particles exhibit intermediate porosity and water-holding capacity, allowing them to retain adequate water while facilitating proper aeration, thereby promoting plant growth. In regions along the Meridian Ridge experiencing reduced soil moisture, sand content exceeds 40%, with most areas surpassing 50%, likely reflecting the diminished water-holding capacity associated with high sand content.

Variations in slope orientation, gradient, elevation, and land use also significantly influence soil moisture levels by affecting local temperature, precipitation, and potential evapotranspiration. These factors collectively shape hydrothermal conditions, which in turn govern the soil moisture content’s spatial distribution and variability.

4.2.2. Potential Evapotranspiration

The study of various adaptations of PET to changes in the spatial distribution of soil moisture revealed a negative correlation between PET and soil moisture levels. Analysis of annual average variations showed that regions experiencing minimal declines in PET maintained higher soil moisture content compared to other areas. This pattern is attributable to the increased demand for soil moisture by plants to sustain normal physiological functions as potential evapotranspiration rises, which consequently reduces soil moisture content. In particular, in arid regions or areas with limited water availability, intensified plant competition further exacerbates the decline in soil moisture levels.

4.2.3. Temperatures

Temperature exerts a dual impact on soil moisture, mediated by environmental factors. In arid regions or during dry seasons, elevated temperatures can increase soil water evaporation, leading to soil desiccation and negatively affecting soil moisture. Conversely, in areas with heavy precipitation, rising soil temperatures enhance the movement and redistribution of soil moisture, thereby increasing water storage in the soil and producing a positive effect on soil moisture [64,65].

4.2.4. Precipitation

Precipitation has a significant influence on soil moisture content. In regions such as the Loess Plateau, characterized by arid and semi-arid conditions, increased precipitation directly elevates soil moisture levels. For example, in the Meridian Ridge area, total precipitation reached a peak of 800 mm in 2003, a markedly higher value compared to the preceding two decades. As a result, soil moisture content also peaked during this period. The intensity and duration of precipitation play a critical role in soil water infiltration. Heavy precipitation events can trigger surface runoff, limiting soil water infiltration, whereas prolonged light rainfall promotes gradual infiltration and enhances water storage within the soil. The Loess Plateau is prone to intense precipitation, which contributes to substantial soil erosion. This process depletes soil nutrients and reduces the soil’s water retention capacity, indirectly diminishing overall soil water content. Therefore, precipitation exhibits a dual nature of positive and negative effects on soil moisture in the study area, which is consistent with the relationship between soil moisture spatiotemporal variation and rainfall, reflecting that soil moisture dynamics are also comprehensively influenced by factors other than precipitation.

4.2.5. NDVI

The geodetector and MGWR model analyses revealed that NDVI significantly influences spatial soil water variability in the Ziwuling region. This finding aligns with the research of Zhang Yannan et al. on the Yellow River Basin, which identified NDVI as a primary driver of spatiotemporal soil water variability in that region [66]. Our study observed a positive impact of NDVI in certain areas, supporting the soil and water conservation benefits of vegetation restoration. However, in the central and northwestern margins, a negative effect was noted, possibly due to excessive vegetation restoration increasing water consumption.

Evapotranspiration, identified as the second most influential factor, shows a year-on-year increase, resulting in a compounded water depletion effect in regions with high NDVI [67,68,69]. This pattern was further confirmed through interaction analysis conducted in this study. The combined impact of environmental variables was found to exceed the individual effects of each factor, highlighting that the collective influence of environmental factors plays a critical role in regulating fluctuations in soil moisture levels and ecosystem vitality. Therefore, a holistic consideration of these interactions is essential for developing scientifically robust strategies and interventions to safeguard ecological balance.

4.2.6. The Interaction Among Various Factors

The outcomes of the geographical detector interaction suggest that the spatial differentiation of soil moisture in the Ziwuling region is not driven by a single factor independently but rather by the nonlinear synergistic effects of multiple factors. This not only confirms the effectiveness of geographical detectors in revealing the multi-factorial driving of ecological processes but also reveals the complexity of the water cycle process in this region. The interaction between NDVI and annual precipitation has the highest explanatory power among all factor combinations (q-value: 0.657). This suggests that the impact of vegetation on soil moisture is strongly dependent on precipitation conditions. In areas with sufficient precipitation, vegetation grows more vigorously and can significantly regulate precipitation redistribution and soil moisture through processes such as canopy interception, transpiration, and altering the relationship between surface runoff and infiltration [70,71]. On the contrary, under drought conditions, vegetation may also become a strong consumer of water. This synergistic effect indicates that there is a complex equilibrium relationship among increased precipitation, vegetation restoration, and soil moisture increase. The interactions between NDVI and silt, sand, and clay particles all exhibit high explanatory power (q-values are all greater than 0.56). This clearly indicates that the impact of vegetation on soil moisture is deeply regulated by soil physical properties. For example, soils with higher contents of silt and clay typically have better water retention capacity, which may buffer the water depletion caused by vegetation transpiration to some extent; meanwhile, sandy soils, despite their good infiltration performance, have poor water retention capacity and may be more prone to deep soil desiccation when there is high vegetation coverage [72]. This reveals the feedback mechanism of the “vegetation-soil” system: soil texture affects water availability, which in turn affects vegetation growth, while vegetation, through root distribution and transpiration, in turn affects the spatial pattern and dynamics of soil moisture. The study by Li, T., et al. also points out that in arid environments, besides soil moisture content itself, other soil physicochemical features (e.g., soil particle composition, i.e., texture) also play a non-negligible role in determining the biochemical and physiological features of plant communities [73]. The interaction between mean annual potential evapotranspiration, annual precipitation, and topographic factors such as slope and elevation also exhibited a two-factor enhancement effect, highlighting the essential role of topography in redistributing climatic elements. Specifically, elevation and aspect alter the energy supply and actual rate of land surface evapotranspiration by influencing temperature and humidity. For instance, Ma, Y.J., et al. demonstrated in a study conducted in the Qinghai Lake watershed that rising elevation leads to decreased temperature and variations in shortwave radiation, resulting in increased soil moisture with higher altitude [74]. Slope gradient controls surface runoff generation and groundwater recharge, thereby creating distinct moisture surplus or deficit environments across different landforms. Research by Fu, Z., et al. in karst hillslopes further revealed that topographic features such as slope govern the interactions between surface runoff and subsurface flow at the soil–bedrock interface, as well as their responses to rainfall patterns. Even within regions sharing the same macroclimatic conditions, this topography-driven redistribution of water and heat can lead to significant micro-scale heterogeneity in soil moisture, underscoring the profound influence of underlying surface characteristics on hydrological processes [75]. In conclusion, the spatial variation in soil moisture in the Ziwuling region is shaped by complex interactions among vegetation dynamics, climatic processes, soil properties, and topographic patterns.

4.3. Summary

In conclusion, the notable increase in vegetation cover and the relatively stable, albeit slightly declining, trend in soil moisture levels in the Ziwuling region can be ascribed to multiple factors. These include changes in land use, the water demands of vegetation, the impact of vegetation cover on soil evapotranspiration, variations in soil texture, and the interplay of climatic and topographic factors. The findings of this study indicate a stable or marginally declining trend in soil moisture within the Ziwuling region, consistent with prior research by Li, Jian, and Chen in the Loess Plateau [76,77]. These studies highlight structural inadequacies in existing artificial forest stands in certain areas of Ziwuling. Future efforts should focus on optimizing stand structures and implementing appropriate management strategies for artificial forest vegetation in the region.

During ecological restoration and vegetation construction in the Loess Plateau, it is essential to consider local topography, climate conditions, soil characteristics, and the ecological adaptability of vegetation. The scientific and rational selection of vegetation types and configurations is critical to achieving coordinated development of vegetation protection and soil moisture management, thereby promoting sustainable ecological, economic, and social development. In the future, adjusting forest structure and fostering natural regeneration through thinning and replanting can enhance forest quality and ecological benefits in the Ziwuling region.

This study relied fundamentally on multi-source remote sensing technology, which provided long-term, large-scale data coverage. Remote sensing enabled the acquisition of consistent and continuous data on soil, environmental, and topographic factors across the entire region from 2001 to 2020, making it possible to conduct large-scale spatiotemporal trend analysis and driving factor investigation. Based on these remote sensing data, we successfully applied statistical methods such as geographical detectors and MGWR to reveal the influence of various factors on soil moisture. It can be stated that remote sensing served not only as the primary data source but also as a critical tool supporting large-scale spatial analysis in this study. However, this research was also subject to certain limitations inherent to remote sensing data and methodology. First, the spatial resolution of the soil moisture data utilized was 1 km, which, while suitable for analyzing regional-scale variations, may not fully capture fine-grained heterogeneity in topographically complex or vegetation-diverse local areas. Second, although NDVI was identified as the most influential factor in the spatial variation in soil moisture in the Ziwuling region, the correlation between NDVI and soil moisture—as well as differences in soil water content among various vegetation types (e.g., forest vs. grassland)—were not thoroughly examined. Moreover, the specific processes underlying vegetation growth and soil moisture consumption require further mechanistic investigation. Lastly, the impact of human activities, such as afforestation practices and land management policies, on soil moisture was not quantified in this research. Future studies could integrate higher-resolution remote sensing data and ground-based observations to enable a more detailed analysis of the link between vegetation and soil moisture, particularly the influence of different vegetation types on soil moisture dynamics.

5. Conclusions

(1) From 2001 to 2020, the average soil moisture in the research region showed a fluctuating but declining trend, although this decrease was statistically non-significant, with an annual reduction rate of −0.00047 m³/(m³·a). Spatially, the Ziwuling region had a multi-year average soil moisture value of 0.178 m³/m³, with soil moisture levels between 0.15 and 0.20 m³/m³ accounting for 80.28% of the total study area. The central and southern parts of the region exhibited higher soil moisture values compared to the northern, western, and eastern peripheries.

(2) The geodetector’s factor detection results indicated that NDVI significantly influenced the spatial variation in soil water in the Ziwuling region. NDVI, as a direct measure of vegetation cover, plays a critical role in regulating soil water distribution through various processes. Following NDVI, average annual potential evapotranspiration and annual precipitation were identified as the next most influential factors. The combined effect of these factors on soil moisture patterns was found to be more significant than the effect of any single factor. Notably, the interaction between NDVI and annual precipitation had the greatest impact on the spatial differentiation of soil moisture in the Ziwuling region.

(3) Various environmental factors exert distinct effects on soil moisture levels. Specifically, slope and elevation generally have negative impacts on soil moisture, whereas factors such as air temperature, NDVI, and slope direction demonstrate bidirectional effects, influencing soil moisture either positively or negatively depending on local conditions.