Spatiotemporal Evolution and Driving Factors of Eco-Environmental Quality in the Shendong Mining Area Based on GEE and Long-Term Landsat Imagery

Abstract

The Shendong Mining Area, located in the transition zone between the northern Loess Plateau and the Mu Us Sandy Land, is a representative ecologically fragile region and desert coal base in China. Using Google Earth Engine (GEE) and Landsat imagery from 1999 to 2024, this study constructed a long-term remote sensing ecological index (RSEI) dataset and integrated the Theil–Sen median slope estimator, Mann–Kendall test, and Hurst exponent to examine the spatiotemporal evolution, future trajectories, and multi-stage driving mechanisms of eco-environmental quality (EEQ) at the mining-area and individual-mine scales. At the mining-area scale, RSEI showed pronounced interannual fluctuations and a weak downward trend, characterized by a two-stage, wave-like trajectory with a narrowing amplitude. The mean RSEI and standard deviation decreased from 0.5669 and 0.0855 during 1999–2010 to 0.5182 and 0.0501 during 2011–2024. Moderate and good grades predominated, with lower EEQ in the mining core and higher EEQ in peripheral buffer zones. Across 13 representative mines, ecological quality generally remained moderate but exhibited spatial and stage-dependent heterogeneity, with Liuta Mine showing the greatest variability. Slight degradation was the dominant trend, although local recovery occurred, and Hurst analysis revealed marked spatial differences in future trajectories. RSEI variations in the Shendong Mining Area exhibited pronounced spatial and stage-dependent associations with high-intensity mining, climatic water–heat conditions, topographic background, and ecological governance and restoration processes. No continuous and irreversible overall decline was observed; however, without an independent non-mining control, the findings represent integrated ecological responses within the mining area rather than causal estimates of mining’s net ecological effect.

1. Introduction

The Yellow River Basin serves both as a national ecological-security barrier and as a strategic energy-resource base; consequently, changes in its eco-environmental quality directly affect the coordinated advancement of ecological protection and high-quality development [1]. As implementation of the national strategy for ecological protection and high-quality development in the Yellow River Basin has deepened, the relationships among resource development, ecological conservation, and regional coordination have become a major topic in regional sustainability research [2]. The Shendong Mining Area lies in the transition zone between the northern Loess Plateau and the Mu Us Sandy Land. It is both a major coal-production base in China and a typical ecologically fragile region [3,4]. While long-term, high-intensity coal extraction has supported national energy security and regional economic development, it has also generated compound ecological problems, including surface disturbance, vegetation degradation, soil erosion, intensified surface thermal conditions, ground fissures, and mining subsidence [5,6]. Against the background of national strategy and coordinated development objectives, concepts such as eco-environmental self-restoration, natural recovery, and artificial restoration have emerged to explain ecological responses to high-intensity mining in aeolian-sandy regions, providing a new theoretical perspective on the disturbance–recovery process in mining areas [6,7]. Nevertheless, conclusions regarding the characteristics of eco-environmental change remain contested because studies differ in method, monitoring scale, observation period, temporal resolution, and analytical emphasis. Long-term, fine-scale monitoring of eco-environmental quality (EEQ) in the Shendong Mining Area is therefore essential for identifying ecological risks, evaluating restoration outcomes, and supporting green mine development.

Remote sensing provides broad spatial coverage, long time series, repeat observations, and strong spatial representation, and has therefore become an important tool for detecting mining disturbance, evaluating vegetation recovery, and monitoring ecological restoration [8]. Mining-area ecosystems are jointly constrained by vegetation, moisture, bare land, thermal conditions, and human activities; a single remote sensing indicator cannot fully characterize integrated changes in EEQ [9]. Within the remote sensing assessment framework used in this study, EEQ refers to the integrated state of the regional surface environment in terms of vegetation cover, surface moisture, surface dryness, and thermal conditions, and is quantified using RSEI. Eco-environmental degradation denotes an adverse change in this integrated state relative to a baseline over a given period. It is typically manifested by decreases in greenness and wetness, increases in dryness and heat, and, quantitatively, by a decline in RSEI or a shift toward a lower ecological grade. To avoid misclassifying short-term climatic fluctuations as degradation, we did not use a change in a single-year RSEI value as the sole criterion; instead, degradation was identified from long-term trends and their statistical significance. To overcome the limitations of single-indicator assessments, Xu proposed the remote sensing ecological index (RSEI), which integrates greenness, wetness, dryness, and heat through principal component analysis. This approach reduces the influence of subjective weighting while providing a comparatively comprehensive representation of regional EEQ [10]. RSEI and its modified forms have since been applied to urban agglomerations, watersheds, arid regions, grasslands, and mining areas, with research shifting from single-period assessment toward long-term dynamic monitoring, multisource remote sensing integration, evaluation of ecological restoration outcomes, and analysis of driving mechanisms [11].

Long-term RSEI analysis depends strongly on the continuity and consistency of remote sensing imagery. Google Earth Engine (GEE) integrates extensive remote sensing archives with cloud-based parallel computing, substantially improving the efficiency of image screening, preprocessing, and batch index calculation across large areas and multiple years. It has consequently become an important platform for regional eco-environmental monitoring and long-term assessment [12]. In studies of arid oases and ecologically fragile regions, combining GEE with RSEI is extending ecological quality assessment from descriptive mapping toward process-based interpretation [13].

In recent years, GEE-based index frameworks have been used to assess EEQ in different underground coal mines, providing an automated pathway for large-area, multi-mine monitoring. Such assessments must simultaneously account for land-cover differences, climatic and geomorphic conditions, and monitoring efficiency [14]. A fused remote sensing ecological index developed for the Jinjie Coal Mine Area in northwestern China showed that ecological quality in disturbed mining zones and their surroundings was lower than the area-wide level [15]. Research in northwestern Chinese mining areas has likewise shown that long-term mining alters the spatiotemporal pattern of ecological quality and produces pronounced regional heterogeneity in its drivers [16]. Moreover, eco-environmental change in mining areas is jointly influenced by climate, topography, land use, nighttime lights, mining subsidence, and ecological restoration projects. Single-factor correlations and empirical attribution alone cannot quantify the dominant factors or their interactions. The geographical detector identifies spatial stratification and its drivers and is therefore well suited to analyzing spatial heterogeneity in EEQ [17]. Parameter optimization and multivariate stratification further improve its capacity to explain complex spatial heterogeneity [18]. Previous studies have characterized ecological responses in Shendong and other coal bases from the perspectives of vegetation disturbance and recovery, integrated ecological quality assessment, mining subsidence, and eco-environmental self-healing. However, three limitations remain. First, most analyses focus on the mining area as a whole, a single mine, or a single ecological component, with little linkage between the mining-area and individual-mine scales; consequently, long-term differences among mines under disturbance and restoration remain poorly resolved. Second, long-term studies generally emphasize overall ecological quality change or vegetation disturbance and recovery, without fully examining the continuous transition in Shendong from mining modernization and capacity expansion, through high-intensity extraction, to governance and restoration. Stage-specific turning points and fluctuations in EEQ therefore remain insufficiently characterized. Third, attribution studies commonly use factors from a single period or the entire study period and pay limited attention to shifts in the explanatory power and interactions of climate, topography, and human activity across mining and restoration stages.

To address these gaps, we constructed a long-term RSEI dataset for the Shendong Mining Area from 1999 to 2024 using GEE and analyzed it at both the mining-area scale and the scale of 13 individual mines. Theil–Sen median trend analysis, the Mann–Kendall test, the coefficient of variation, and the Hurst exponent were combined to identify the direction, fluctuation intensity, and future persistence of EEQ change. The study period was then divided according to the actual progression of mining and ecological governance, and stage-averaged climatic indicators and the optimal-parameter geographical detector were used to analyze stage-dependent shifts in the explanatory power and interactions of climatic, topographic, and human activity factors. This study provides new evidence of a two-phase, wave-like RSEI trajectory with narrowing amplitude, heterogeneous responses among individual mines, and a compound mechanism characterized by climatic constraint, topographic modulation, and amplification by human activity.

2. Materials and Methods

2.1. Study Area

The study area is located at the boundary between northern Yulin, Shaanxi Province, and southeastern Ordos, Inner Mongolia, in the transition zone between the southeastern margin of the Mu Us Sandy Land and the northern Loess Plateau (38.35–39.57° N, 109.70–111.48° E). Elevation is generally higher in the northwest and lower in the southeast. The area has a temperate arid to semi-arid continental monsoon climate, with long, cold winters, windy springs, and frequent rainstorms in summer. Stage-averaged precipitation is approximately 300–400 mm, whereas evaporation exceeds precipitation by more than fourfold. The dominant landforms are loess hills and gullies and aeolian-sandy terrain, where water erosion and wind erosion are intense. The dominant soil types are loessial soils and aeolian sandy soils, and the sparse native vegetation is dominated by psammophytic and xerophytic shrubs and grasses [19] (Figure 1).

Figure 1. Location and overview of the Shendong Mining Area: (a) location in China; (b) regional extent; (c) topography and coal mine distribution.

Established in late 1998, the Shendong Mining Area comprises 13 major mines: Buertai, Cuncaota No. 2, Liuta, Wulanmulun, Cuncaota No. 1, Huhewusu, Erlintu, Shigetai, Bulianta, Shangwan, Halagou, Huojitu, and Daliuta. The coal seams are stable, the geological structure is relatively simple, and the mines share the characteristics of shallow burial, thin bedrock, and thick unconsolidated overburden. Underground mining predominates, making Shendong an important coal production base in China. Since 2010, industrial development has promoted high-intensity, fully mechanized extraction and ultra-large longwalls in the area, with advance lengths of up to 6000 m, face lengths of 300–350 m, and advance rates exceeding 12 m d⁻¹. Annual production exceeds 10 million tonnes at the Daliuta, Bulianta, Buertai, Shangwan, Halagou, Shigetai, Liuta, and Huojitu mines. Coal reserves and available water resources are highly mismatched, forming a typical inverse distribution and imposing low eco-environmental carrying capacity. The coexistence of early disordered extraction and modern clustered, large-scale mining has produced a broad disturbance footprint and pronounced spatiotemporal variation in ecological impacts. Combined with arid stress, these pressures may drive adverse ecological succession. Long-term spatial monitoring of EEQ in Shendong is therefore practically important for ecological risk identification and the selection of restoration strategies [20].

2.2. Research Framework

Figure 2 summarizes the study design and analytical framework. First, Landsat 5 TM, Landsat 8 OLI, and ancillary datasets were preprocessed. Second, four ecological components—NDVI, WET, NDBSI, and LST—were extracted, normalized, and integrated through principal component analysis to construct RSEI. The spatiotemporal variation, trend, and persistence of RSEI were then analyzed. Finally, climatic, topographic, land-use, and human activity factors were incorporated into the geographical detector to identify the principal drivers of EEQ change in the study area.

Figure 2. Overall workflow of the study.

2.3. Data Sources and Preprocessing

The remote sensing data were obtained primarily from the Google Earth Engine (GEE) platform. Landsat 5 TM and Landsat 8 OLI imagery were selected to construct an annual long-term time-series dataset for 1999–2024. Landsat 5 TM imagery was used for the early period from 1999 to 2012, whereas Landsat 8 OLI imagery was used from 2013 onward. The Shendong Mining Area was formally established in late 1998 through the integration of the Shenfu and Dongsheng District mining areas. Accordingly, 1999 was selected as the first complete year after unified mining-area management for which sufficient growing-season Landsat observations were available. Therefore, 1999 represents the starting point of unified long-term monitoring rather than a pre-mining ecological baseline. To reduce seasonal effects, mean imagery from the vegetation-growing season (June–September) was used, with priority given to scenes with less than 10% cloud cover. Radiometric calibration, atmospheric correction, cloud and shadow masking, clipping, band calculations, and annual compositing were completed on GEE.

The driving factor dataset comprised climatic, topographic, land-surface, and human activity variables. The climatic variables included air temperature and precipitation; the topographic variables included elevation, slope, and aspect; and the human activity and land-surface variables included population density, land-use type, and nighttime light intensity. The data sources and original spatial resolutions are summarized in Table 1. Previous studies have shown that variations in eco-environmental quality are jointly associated with climatic conditions, land-cover change, and human activities, with pronounced interactions among these factors [21].

Table 1. Data sources and spatial resolution.

Data	Abbreviation	Spatial Resolution	Source
Landsat 5 TM	TM	30 m	USGS Earth Explorer/GEE (https://earthexplorer.usgs.gov)
Landsat 8 OLI	OLI	30 m	USGS Earth Explorer/GEE (https://earthexplorer.usgs.gov)
Land-Use Type	CL	30 m	China Land Cover Dataset (CLCD) (https://engine.piesat.cn/)
Air temperature	TMP	1 km	National Tibetan Plateau Data Center (https://data.tpdc.ac.cn/home)
Precipitation	PRE	1 km	National Tibetan Plateau Data Center (https://data.tpdc.ac.cn/home)
Elevation	DEM	30 m	United States Geological Survey (https://earthexplorer.usgs.gov)
Slope	Slope	30 m	United States Geological Survey (https://earthexplorer.usgs.gov)
Aspect	Aspect	30 m	United States Geological Survey (https://earthexplorer.usgs.gov)
Nighttime light intensity	NTL	500 m	Geospatial Data Cloud (http://www.gis5g.com)
Population density	POD	1 km	WorldPop global population spatial dataset (https://worldwidepop.com/)

Note: All URLs were accessed on 20 April 2026.

Before the optimal-parameter geographical detector (OPGD) analysis, all raster datasets were transformed to a common coordinate system, clipped to the study area, and aggregated to an approximately 1 km regular grid to ensure one-to-one spatial correspondence between the dependent variable and the explanatory factors. The 500 m fishnet was used exclusively to define spatially systematic sampling locations for the internal consistency assessment of the RSEI, whereas the approximately 1 km grid served as the common spatial analysis unit for the OPGD analysis. The 500 m sampling procedure did not involve resampling the original 30 m RSEI and component-indicator rasters. Detailed aggregation procedures for continuous and categorical variables are described in Section 2.4.5. Similar long-term RSEI studies generally conduct trend analysis and driver detection on a common spatial grid [22].

2.4. Methods

2.4.1. RSEI Construction

The remote sensing ecological index (RSEI) integrates four components—greenness, wetness, dryness, and heat—to characterize EEQ. Greenness was represented by the normalized difference vegetation index (NDVI), which reflects vegetation cover and growth. Wetness was represented by the wetness component of the tasseled-cap transformation (WET), which reflects moisture in soil, vegetation, and the land surface. Dryness was represented by the normalized difference built-up and soil index (NDBSI), which captures surface drying associated with bare land and built-up areas. Heat was represented by land surface temperature (LST), which characterizes the surface thermal environment [10,23].

The coefficients used to calculate WET differ among sensors. WET for Landsat 5 TM (WET₁) and Landsat 8 OLI (WET₂) was calculated as follows:

$$\begin{array}{l}WE{T}_1=0.0315{\beta }_B+0.2021{\beta }_G+0.3102{\beta }_R+\\ 0.1594{\beta }_{NIR}-0.6806{\beta }_{SWIR1}-0.6109{\beta }_{SWIR2}\end{array}$$

(1)

$$\begin{array}{l}WE{T}_2=0.1511{\beta }_B+0.1973{\beta }_G+0.3283{\beta }_R+\\ 0.3407{\beta }_{NIR}-0.7117{\beta }_{SWIR1}-0.4559{\beta }_{SWIR2}\end{array}$$

(2)

where ${\beta }_B$, ${\beta }_G$, ${\beta }_R$, ${\beta }_{NIR}$, ${\beta }_{SWIR1}$, and ${\beta }_{SWIR2}$ represent the reflectance values of the blue, green, red, near-infrared, shortwave infrared 1, and shortwave infrared 2 bands, respectively.

The NDVI was calculated as follows:

$$NDVI=\left({\beta }_{NIR}-{\beta }_R\right)/({\beta }_{NIR}+{\beta }_R)$$

(3)

where ${\beta }_{NIR}$ represents the reflectance value of the near-infrared band in Landsat imagery, and ${\beta }_R$ represents the reflectance value of the red band.

The LST was calculated as follows:

$$LST={\displaystyle \frac{T}{1+(\frac{\lambda T}{\rho })\ln \epsilon }}$$

(4)

$$T={\displaystyle \frac{k_2}{\ln (\frac{k_1}{L}+1)}}$$

(5)

$$L=gain·DN+offset$$

(6)

where $\lambda$ is the central wavelength of the thermal infrared band; $\rho$ = 1.438 × 10⁻² m·K; and $\epsilon$ is the land surface emissivity. For TM and OLI data, the central wavelengths were set to 11.435 µm and 10.9 µm, respectively. $L$ represents the at-sensor radiance of the pixel in the TM/OLI thermal infrared band; $DN$ denotes the digital number of the pixel; $gain$ and $offset$ are the gain and bias values of the thermal infrared band, respectively; and $k_1$ and $k_2$ are calibration constants.

The NDBSI was calculated as follows:

$$NDBSI={\displaystyle \frac{SI+IBI}{2}}$$

(7)

$$SI={\displaystyle \frac{\left[({\beta }_{SWIR1}+{\beta }_R)-({\beta }_B+{\beta }_{NIR})\right]}{\left[({\beta }_{SWIR1}+{\beta }_R)+({\beta }_B+{\beta }_{NIR})\right]}}$$

(8)

$$IBI={\displaystyle \frac{\frac{2{\beta }_{SWIR1}}{\left({\beta }_{SWIR1}+{\beta }_{NIR}\right)}-(\frac{{\beta }_{NIR}}{{\beta }_{NIR}+{\beta }_R}+\frac{{\beta }_G}{{\beta }_G+{\beta }_{SWIR1}})}{\frac{2{\beta }_{SWIR1}}{{\beta }_{SWIR1}+{\beta }_{NIR}}+(\frac{{\beta }_{NIR}}{{\beta }_{NIR}+{\beta }_R}+\frac{{\beta }_G}{{\beta }_G+{\beta }_{SWIR1}})}}$$

(9)

where ${\beta }_B$, ${\beta }_G$, ${\beta }_R$, ${\beta }_{NIR}$, and ${\beta }_{SWIR1}$ represent the reflectance values of the blue, green, red, near-infrared, and shortwave infrared 1 bands in the TM and OLI data, respectively.

Before principal component analysis, NDVI, WET, NDBSI, and LST were normalized to eliminate dimensional differences among the indicators. To ensure interannual comparability, the four ecological indicators were normalized using temporally unified parameters derived from the complete 1999–2024 dataset. Principal component analysis (PCA) was then performed using the normalized indicators to obtain a single set of fixed PCA loadings, which was consistently applied to all annual datasets. The first principal component (PC1) was used to construct the initial RSEI, and a common PC1 normalization range derived from the complete study period was subsequently applied to scale the annual RSEI values to [0,1]. Values closer to 1 indicate higher eco-environmental quality, whereas values closer to 0 indicate lower eco-environmental quality [10,24].

$$N{P}_i={\displaystyle \frac{\left(P_i-P_{\min }\right)}{\left(P_{\max }-P_{\min }\right)}}$$

(10)

$$RSE{I}_0=\left\{\begin{array}{l}P{C}_1\left[f(WET,NDVI,LST,NDBSI)\right] V_{NDVI},V_{WET}>0\\ 1-P{C}_1\left[f(WET,NDVI,LST,NDBSI)\right] V_{NDVI},V_{WET}<0\end{array}\right.$$

(11)

$$RSEI={\displaystyle \frac{\left(RSE{I}_0-RSE{I}_{0\min }\right)}{\left(RSE{I}_{0\max }-RSE{I}_{0\min }\right)}}$$

(12)

where ${NP}_i$ is the normalized value of a given ecological indicator; $P_i$ is the indicator value at pixel $i$; and $P_{\max }$ and $P_{\min }$ are the temporally unified maximum and minimum values derived from the complete 1999–2024 dataset, respectively. $RSE{I}_0$ denotes the initial RSEI calculated using the fixed PCA loadings, whereas $RSE{I}_{0\min }$ and $RSE{I}_{0\max }$ are the common maximum and minimum PC1-based values derived from the complete study period and applied consistently to all years.

2.4.2. Theil–Sen Median Trend Analysis and Mann–Kendall Test

Theil–Sen median trend analysis is suitable for identifying trends in long-term time series and is more robust against data errors and outliers, thereby improving the reliability of trend estimation. This method characterizes the change trend by calculating the median slope between any two observations in the time series, and has been widely applied in studies of the spatiotemporal changes in RSEI-based eco-environmental quality [25]. The calculation formula is as follows:

$$S_{RSEI}=median\left({\displaystyle \frac{RSE{I}_j-RSE{I}_i}{j-i}}\right),1999\le i<j\le 2024$$

(13)

where median denotes the median function; $RSE{I}_j$ and $RSE{I}_i$ are the index values corresponding to the $j$ and $i$ observations in the time series, respectively; and $S_{RSEI}$ represents the slope of RSEI change, indicating the change trend of the index. When $S_{RSEI}$ < 0, the index shows a decreasing trend, whereas when $S_{RSEI}$ > 0, the index shows an increasing trend. A larger absolute Sen value indicates a greater magnitude of change, while an absolute Sen value closer to 0 indicates a smaller magnitude of change.

The Mann–Kendall significance test is used to determine the significance of trends in time series and does not require the data to follow a normal distribution, making it suitable for long-term time-series trend analysis. Therefore, this study employed the Mann–Kendall significance test to analyze the significance of temporal changes in EEQ grades in the Shendong Mining Area from 1999 to 2024. The calculation formula is as follows:

$$S={\displaystyle \sum _{i=1}^{n-1}{\displaystyle \sum _{j=i+1}^{n}sign}}\left({\displaystyle \frac{RSE{I}_j-RSE{I}_i}{j-i}}\right)$$

(14)

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

(15)

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

(16)

where $RSE{I}_j$ and $RSE{I}_i$ are the index values corresponding to the $j$ and $i$ observations in the time series, respectively; sign denotes the sign function; and n is the total number of observations. $Z$ is the standardized test statistic. When using the statistic $Z$ for trend significance, testing at the significance level of α = 0.05, |Z| > 1.96 indicates that the trend passes the 95% significance test, whereas |Z| ≤ 1.96 indicates that the trend is not significant. This method has been widely applied in long-term RSEI-based studies of eco-environmental quality [26].

The combined results of the Mann–Kendall significance test and Theil–Sen slope estimation were used to determine the significance of RSEI change trends, as shown in Table 2.

Table 2. Classification criteria for RSEI trend changes.

|Z|	SRSEI × 10−5
	>5	−5~5	<−5
>1.96	Significant increase	Almost unchanged	Significant decrease
≤1.96	Slight increase		Slight decrease

2.4.3. Coefficient of Variation (CV)

The coefficient of variation (CV) was used to measure the relative fluctuation degree of the RSEI time series. The fixed threshold method was used to describe differences among different mines, and the CV was classified into four levels: low fluctuation I (CV < 0.15), moderate fluctuation II (0.15 ≤ CV ≤ 0.30), relatively high fluctuation III (0.30 < CV ≤ 0.50), and high fluctuation IV (CV > 0.50). The natural breaks method was further used to identify spatial differences within the study area. A larger CV value indicates stronger interannual fluctuations in EEQ and weaker ecosystem stability, whereas a smaller CV value indicates more gradual interannual RSEI changes and relatively stable ecosystem conditions [27].

$$CV=\delta /\mu$$

(17)

where $\delta$ is the standard deviation and $\mu$ is the mean value. A larger $CV$ value indicates more intense RSEI variation and greater eco-environmental vulnerability, whereas a smaller $CV$ value indicates weaker RSEI fluctuation and stronger ecosystem resistance to disturbance.

2.4.4. Hurst Exponent

The Hurst exponent derived from rescaled range analysis can quantitatively characterize the long-term dependence of time-series information and has been widely used to analyze the future persistence of EEQ, vegetation change, and regional ecological processes.

For a given time series {RSEI_(t)}, t = 1, 2, 3, …, n, the mean sequence overline RSEI_(T) and cumulative deviation X(t,T) are defined as follows:

$$RSE{I}_{(T)}={\displaystyle \frac{1}{T}}{\displaystyle \sum _{t=1}^{T}RSE{I}_{(t)}}$$

(18)

$$X(t,T)={\displaystyle \sum _{t=1}^T\left(RSE{I}_{(t)}-RSE{I}_{(T)}\right)}$$

(19)

If there exists an H value such that the range R of the cumulative deviation and the standard deviation S of the time series satisfy R/S∝T^H, the time series is considered to exhibit the Hurst phenomenon. When H > 0.5, the series exhibits persistence, indicating that its future direction of change is likely to be consistent with the historical trend. When H < 0.5, the series exhibits anti-persistence, suggesting that its future direction may be opposite to the historical trend. When H = 0.5, the series approximates a random process without a clear persistence pattern. To reduce classification uncertainty around the theoretical threshold of H = 0.5, values between 0.45 and 0.55 were treated as near-random, whereas H < 0.45 and H > 0.55 were classified as anti-persistent and persistent behavior, respectively. In this study, the Hurst exponent was combined with the Theil–Sen median trend to assess the potential persistence or reversal of RSEI changes [22,28].

The Hurst exponent was further combined with the Theil–Sen median trend to classify the future persistence of RSEI changes, as shown in Table 3.

Table 3. Classification criteria for the future persistence of RSEI changes.

Sen × 10−5	Hurst Exponent
	0~0.45	0.45~0.55	0.55~1
<−5	Shift from degradation to improvement	Near-random change	Persistent degradation
>5	Shift from improvement to degradation	Near-random change	Persistent improvement
−5~5	Almost unchanged	Near-random change	Almost unchanged

2.4.5. Geographical Detector

The geographical detector was used to identify spatial stratification and its driving mechanisms. Climatic, surface, and human activity factors were treated as independent variables and RSEI as the dependent variable to analyze the explanatory power of individual factors and pairwise interactions. Because anomalous endpoint-year values may inflate or suppress the ecological relevance of climatic factors, the temporal representation of each explanatory variable was determined from its data properties and ecological meaning. Stage-averaged precipitation and temperature represented the persistent hydrothermal background of each stage. Changes in population density and nighttime lights between the first and last year of each stage represented changes in human activity intensity. The dominant land-use type within each stage represented the overall land-cover pattern. Elevation, slope, and aspect changed little during the study period and were treated as static topographic background variables. The q value measures how strongly the spatial stratification of a variable explains spatial differences in RSEI within the same stage; it does not provide a direct comparison of temporal effect magnitudes among variables.

The stages were defined by jointly considering mining development, ecological governance and policy milestones, and turning points in the RSEI time series. Candidate boundaries were first identified from transitions among mine construction and capacity expansion, high-intensity mechanized extraction, and green mine and ecological restoration programs. Relevant planning and governance milestones were then incorporated, and consistency was checked against peaks, troughs, and multiyear directions of RSEI change around each candidate year. The study period was ultimately divided into five continuous, non-overlapping stages: 1999–2004, 2005–2010, 2011–2015, 2016–2020, and 2021–2024. These corresponded, respectively, to early mining modernization, large-scale expansion and formation of a cluster of 10-million-tonne mines, high-intensity mechanized mining disturbance, green mine development and active restoration, and stable mining restoration coordination. This division is a rule-based periodization corroborated by historical processes and remote sensing changes rather than an automatic statistical breakpoint procedure. Because the original spatial resolution of the drivers ranged from 30 m to 1 km, an approximately 1 km regular grid was used as the common spatial analysis unit. This scale matches the original resolution of the temperature and precipitation data and avoids pseudo-replication and scale mismatch caused by mechanically downscaling 1 km climate data to 30 m. RSEI and other continuous variables were aggregated as within-grid means. Categorical variables, including land use, were assigned by the dominant category; aspect was first classified by direction and then assigned by the dominant within-grid class. This procedure ensured one-to-one correspondence between the dependent variable and the drivers in each spatial unit. Optimal-parameter discretization was subsequently used to divide each independent variable into 3–16 classes. A larger q value indicates greater explanatory power for the spatial differentiation of RSEI. The OPGD was finally used to quantify the explanatory power and interactions of the factors across stages [29,30].

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

(20)

where $q$∈ [0, 1], and a larger $q$ value indicates stronger explanatory power; $L$ is the number of strata for the factor; $N_h$ is the number of sample units in stratum h; $N$ is the total number of sample units in the entire study area; ${\delta }_h^2$ is the variance of RSEI in stratum h; and ${\delta }^2$ is the variance of RSEI across the entire study area.

3. Results

3.1. Internal Consistency Assessment of the RSEI

Correlations between RSEI and its ecological components can be used to determine whether the model’s explanatory direction is ecologically meaningful [31]. To assess whether the ecological interpretation of the RSEI was directionally consistent, we constructed a regular 500 m × 500 m fishnet for the RSEI applicability assessment and extracted corresponding values from the original 30 m RSEI, NDVI, WET, NDBSI, and LST rasters at spatially systematic sampling locations. The 500 m fishnet was used only to define systematic sampling locations; the indicator rasters were not resampled to 500 m. This approach reduced spatial redundancy and point overlap among adjacent 30 m pixels while maintaining an even sample distribution across the study area.

A total of 4889 valid samples were obtained for correlation analysis and density scatterplot visualization. Figure 3 presents the Pearson correlation coefficients and the corresponding density scatterplots, whereas Table 4 reports the Spearman rank correlation coefficients. As shown in Figure 3, the absolute correlations with RSEI decreased in the order NDBSI (−0.922) > NDVI (0.795) > WET (0.763) > LST (−0.740), and all were significant. Spearman’s rank correlation indicated that NDVI and WET were significantly and positively correlated with RSEI, suggesting that greater vegetation cover and surface wetness generally corresponded to higher EEQ. NDBSI and LST were negatively correlated with the RSEI, indicating that higher dryness and surface thermal conditions were associated with lower RSEI values [32].

Figure 3. Correlations between RSEI and its components: (a) NDVI; (b) WET; (c) NDBSI; (d) LST.

Table 4. Spearman correlation coefficients between RSEI and its component indicators.

Variable	RSEI	NDVI	WET	NDBSI
NDVI	0.7751 **
WET	0.7092 **	0.3944 **
NDBSI	−0.9160 **	−0.7008 **	−0.6473 **
LST	−0.7799 **	−0.3951 **	−0.5150 **	0.6407 **

** p < 0.01.

These correlations confirm the expected directional consistency between the RSEI and its component indicators. However, because the RSEI was constructed directly from these four components, the results should be interpreted as an internal consistency assessment rather than an independent validation of the model. Compared with a single remote sensing indicator, RSEI provides a more integrated representation of the multidimensional characteristics and dynamics of the regional eco-environment and offers objective technical support for monitoring, restoration, and scientific decision making in mining areas [33].

3.2. RSEI Dynamics at the Mining-Area Scale

3.2.1. Temporal Change Characteristics

As shown in Figure 4, RSEI in the Shendong Mining Area exhibited marked interannual fluctuation from 1999 to 2024, with an overall two-phase, wave-like pattern and narrowing amplitude. The division into two stages was not based arbitrarily on statistical variation in the RSEI curve alone; it primarily reflected changes in mining practices and ecological governance in Shendong and was corroborated by the time-series characteristics of RSEI. To ensure that the stages did not overlap, 1999–2010 was defined as the mining-modernization and capacity-expansion stage, whereas 2011–2024 was defined as the transition from high-intensity mining to governance and restoration. The boundary in Figure 4 was therefore placed between 2010 and 2011.

Figure 4. Interannual variation in RSEI in the study area from 1999 to 2024.

From 1999 to 2004, during the initial construction and commissioning of modern mines, regional EEQ was jointly influenced by climatic conditions, engineering construction, and localized mining disturbance. From 2005 to 2010, as a cluster of 10-million-tonne mines gradually formed, the extent and intensity of disturbance increased; however, concurrent revegetation and soil- and water-conservation measures attenuated the signal of persistent regional degradation. From 2011 onward, high-intensity mechanized mining and ecological restoration proceeded in parallel. RSEI fluctuated around a lower mean, with a weak trend and distinctly narrowing interannual amplitude, indicating that the mining area entered a dynamic adjustment stage governed by the interaction between mining disturbance and ecological recovery. The overall linear fit was y = −0.0034x + 7.3782 (R² = 0.1315), indicating a weak downward trend in EEQ. The low goodness of fit, however, shows that RSEI did not undergo unidirectional degradation but fluctuated in stages under the combined influence of climate, mining disturbance, land-use conversion, and ecological governance. For 1999–2010, the fitted slope was −0.0048 (R² = 0.0407), with strong within-stage fluctuation; RSEI peaked at 0.6756 in 2009 and reached a minimum of 0.4249 in 2005. For 2011–2024, the fitted slope was only −0.0009 (R² = 0.0053), indicating low-amplitude fluctuation around a lower mean rather than Persistent decrease during the transition from high-intensity mining to governance restoration.

Table 5 further shows that during the mining-modernization and capacity-expansion stage (1999–2010), mean RSEI was 0.5669, the standard deviation was 0.0855, and the coefficient of variation was 0.1510. EEQ was therefore relatively high overall but fluctuated strongly, indicating sensitivity to natural climatic variability and localized mining disturbance.

Table 5. Stage-specific statistical characteristics of interannual RSEI variation.

Period	Mining Stage	Min	Max	Mean	SD	CV
1999–2010	Mining modernization and expansion stage	0.4249	0.6756	0.5669	0.0855	0.1510
2011–2024	Intensive mining–restoration transition stage	0.4389	0.5881	0.5182	0.0501	0.0967

Compared with the preceding stage, mean RSEI decreased to 0.5182 during the transition from high-intensity mining to governance and restoration (2011–2024), with a maximum of 0.5881, indicating an overall reduction in EEQ. However, the standard deviation decreased to 0.0501 and the coefficient of variation to 0.0967, demonstrating a pronounced narrowing of interannual fluctuation. EEQ did not deteriorate continuously during this stage but fluctuated around a lower mean.

Overall, RSEI in Shendong was characterized by high fluctuation and relatively high values during 1999–2010, followed by low fluctuation around a lower mean during 2011–2024. This pattern suggests that changes in mining intensity and restoration practices around 2011 may have exerted stage-specific effects on EEQ. Although the overall downward trend was weak, the persistence of a lower mean indicates that EEQ remained jointly constrained by the cumulative effects of mining disturbance and the lagged response of ecological restoration. Subsequent governance should therefore focus on ecological risks in high-intensity mining zones and in areas corresponding to low-RSEI years.

3.2.2. Spatial Distribution

After uniform normalization, RSEI ranged from 0 to 1. Following previous RSEI studies, EEQ was divided at fixed intervals of 0.2 into Grade I (very poor, 0–0.2), Grade II (poor, 0.2–0.4), Grade III (moderate, 0.4–0.6), Grade IV (good, 0.6–0.8), and Grade V (excellent, 0.8–1.0). Fixed thresholds ensure a consistent assessment criterion among years and support direct comparison of long-term eco-environmental change. EEQ in Shendong exhibited pronounced patchiness and spatial heterogeneity.

As shown in Figure 5, RSEI in Shendong generally followed a spatial pattern of lower values in mining core areas and higher values in peripheral buffers, accompanied by a weak declining trend amid fluctuation. During 1999–2003, EEQ was generally moderate to good, with green areas concentrated in the northwest and along parts of the periphery. Low-value areas expanded markedly in 2004–2005. Red (Grade I, very poor) and orange (Grade II, poor) areas occurred mainly in the central and northeastern zones with concentrated mining activity, including the coal-mining cores of Dongsheng District, Jungar Banner, and Shenmu City, where vegetation degradation and exposed soil were prominent. Improvement was evident during 2006–2009, as green areas (Grades IV and V) expanded and RSEI reached its study-period maximum in 2009. This pattern was closely associated with the initiation of ecological restoration projects, including the conversion of cropland to forest and grassland and mine-land reclamation. Ecological quality declined again in 2010, possibly because of the combined effects of drought and expanded mining. During 2011–2024, the landscape fluctuated in a predominantly yellow-orange matrix interspersed with green, reflecting a dynamic interaction among mining disturbance, ecological governance, and natural recovery.

Figure 5. Spatial distribution of RSEI grades in the Shendong Mining Area from 1999 to 2024.

As shown in Figure 6, the mean area proportions of the very poor, poor, moderate, good, and excellent grades during 1999–2024 were 5.73%, 17.89%, 36.94%, 29.22%, and 10.22%, respectively. Moderate and good areas together accounted for 66.16% and formed the dominant ecological quality classes, whereas very poor and poor areas together accounted for 23.62% and were concentrated in zones of mining activity, construction land, and exposed ground. The area of low-quality grades increased markedly in 2005 and 2010, corresponding to low RSEI values; the area of high-quality grades expanded substantially in 2009, corresponding to the maximum RSEI over the full period. Grade proportions continued to fluctuate during 2011–2024. Ecological quality improved in 2021, whereas low-grade areas expanded again in 2022. Mean RSEI across the full 26-year period was 0.5407, indicating a weak overall decline accompanied by pronounced interannual fluctuations. At the mining-area scale, the annual mean RSEI did not exhibit a monotonic trajectory, although the pixel-wise trend analysis presented in Section 3.4 revealed widespread slight degradation. Taken together, the temporal variability in the annual mean RSEI and the spatial coexistence of degradation and improvement indicate that the eco-environment remained in a state of dynamic adjustment rather than dynamic equilibrium. This adjustment was characterized by spatially heterogeneous degradation and recovery. Sustained disturbance control and ecological restoration are therefore required to consolidate existing restoration gains and promote the expansion of high-quality ecological areas.

Figure 6. Interannual variation in mean RSEI and area proportions of different RSEI grades in the Shendong Mining Area.

3.3. RSEI Dynamics at the Individual-Mine Scale

As shown in Figure 7 and Table 6, RSEI across the 13 representative mines remained near 0.5 during 1999–2024, indicating generally moderate EEQ at the individual-mine scale, although fluctuation intensity and stage-dependent change varied substantially among mines. Liuta and Wulanmulun exhibited markedly stronger RSEI fluctuations than the other mines during 2009–2017; the reasons are examined further in Section 4. Across the full period, Liuta had the highest mean RSEI (0.6571) but also the largest standard deviation (0.2728), indicating relatively high ecological quality coupled with the strongest interannual fluctuation and high sensitivity to mining disturbance and external environmental change. By contrast, the RSEI curves for Buertai and Cuncaota No. 1 were comparatively smooth, with full-period standard deviations of 0.0628 and 0.0688, respectively. Their stronger temporal stability indicates that interannual disturbance had a smaller effect on EEQ.

Figure 7. Interannual variation in RSEI for the 13 representative mines.

Table 6. Stage-specific RSEI statistics for the 13 representative mines.

Mine Units	Mean	SD	CV	Mean	SD	CV	Mean	SD	CV	Grade
	1999–2010			2011–2024			1999–2024
Liuta Mine	0.6948	0.3011	0.4333	0.6248	0.2414	0.3863	0.6571	0.2728	0.4152	III
Wulanmulun Mine	0.6256	0.1806	0.2886	0.5745	0.1101	0.1916	0.5981	0.1491	0.2492	II
Shigetai Mine	0.5853	0.1442	0.2463	0.5450	0.1068	0.1960	0.5636	0.1270	0.2254
Shangwan Mine	0.5695	0.0754	0.1324	0.4812	0.1144	0.2377	0.5220	0.1077	0.2064
Huojitu Mine	0.5726	0.0962	0.1680	0.5012	0.1050	0.2095	0.5342	0.1071	0.2005
Bulianta Mine	0.5702	0.1007	0.1765	0.4843	0.0904	0.1866	0.5239	0.1044	0.1993
Cuncaota No. 2 Mine	0.6010	0.1462	0.2433	0.5354	0.0927	0.1732	0.5657	0.1248	0.2205
Halagou Mine	0.5733	0.1010	0.1761	0.4994	0.0845	0.1691	0.5335	0.0995	0.1865
Huhewusu Mine	0.5550	0.0754	0.1359	0.4792	0.0778	0.1624	0.5142	0.0855	0.1663
Erlintu Mine	0.5333	0.0538	0.1009	0.4488	0.0819	0.1824	0.4878	0.0820	0.1680
Daliuta Mine	0.5625	0.0387	0.0689	0.4700	0.0805	0.1713	0.5127	0.0794	0.1549
Cuncaota No. 1 Mine	0.5273	0.0863	0.1636	0.5307	0.0490	0.0924	0.5291	0.0688	0.1300	I
Buertai Mine	0.5383	0.0775	0.1439	0.5444	0.0465	0.0855	0.5416	0.0628	0.1160

Stage-wise comparison showed that mean RSEI at most mines was higher during mining modernization and capacity expansion (1999–2010) than during the transition from high-intensity mining to governance and restoration (2011–2024). Liuta, Wul-anmulun, Bulianta, Shangwan, Huojitu, Halagou, and Huhewusu all declined to varying degrees in the later period, indicating sustained pressure from increased mining intensity on mine-scale EEQ. Nevertheless, the standard deviation decreased at most mines in the later period, and interannual RSEI variation narrowed and became more stable.

Overall, mine-scale EEQ in Shendong exhibited spatial heterogeneity and stage-dependent differentiation. Liuta was a high-fluctuation, disturbance-sensitive unit, reflecting an unstable response between mining disturbance and ecological recovery. Buertai and Cuncaota No. 1 were low-fluctuation, stable units with comparatively gradual changes in ecological quality. Mean RSEI decreased slightly at most mines during the high-intensity mining stage, but fluctuation narrowed later and no pronounced, persistent adverse succession occurred at the regional scale. Thus, although mining disturbance remained negatively associated with mine-scale EEQ, the ecological response was not controlled by mining alone; it was jointly modulated by disturbance, restoration, topographic setting, and climatic conditions [34].

3.4. Trends in Eco-Environmental Quality

Theil–Sen median trend analysis and the Mann–Kendall test were implemented in MATLAB R2021b to evaluate EEQ trends in the Shendong Mining Area from 1999 to 2024. Based on the slope and significance results, change was classified as significant degradation, slight degradation, stable, slight improvement, or severe improvement. Figure 8 shows an overall pattern dominated by slight degradation but accompanied by marked localized improvement, indicating pressure on ecosystem stability [35]. Areas of significant and slight improvement accounted for 1.17% and 23.17%, respectively, or 24.34% in total. Stable areas accounted for only 0.79%, indicating that the mining-area ecosystem was generally dynamic. Slightly and significantly degraded areas accounted for 59.57% and 15.30%, respectively, or 74.87% in total. Slight degradation was the dominant change type, indicating that most degraded areas had not undergone severe deterioration.

Figure 8. Spatial distribution of change trends in eco-environmental quality in the Shendong Mining Area from 1999 to 2024.

Spatially, EEQ change in Shendong was characterized by improvement along the periphery, degradation in the interior, and a predominance of slight degradation overall. Improvement was concentrated along the western and northern margins and in localized southern areas, forming green bands or patches that closely corresponded to peripheral forest and grassland and artificially enclosed conservation areas. Stable areas were extremely limited and scattered within the study area, with no contiguous pattern, indicating that the ecosystem was strongly affected by human activity. Degradation was widespread across the central and eastern zones, particularly in concentrated industrial and mining areas, eastern towns, and transport corridors, and appeared mainly as extensive light-orange patches of slight degradation and scattered red patches of severe degradation. Its close spatial correspondence with coal faces, industrial sites, and settlements indicates a negative spatial association between concentrated mining activity and EEQ in the study area.

3.5. Potential Persistence of Future Eco-Environmental Trends

Previous studies have shown that RSEI combined with trend analysis can identify long-term and stage-dependent changes in EEQ in mining areas [36]. On this basis, we combined the Theil–Sen median trend with the Hurst exponent to assess near-future EEQ trajectories. As shown in Figure 9a, persistence predominated: areas with H > 0.55 accounted for 52.36%, indicating that more than half of the study area may continue in its existing direction of change. Areas with H < 0.45 accounted for 21.67% and exhibited strong anti-persistence, whereas near-random areas with 0.45 ≤ H ≤ 0.55 accounted for 25.97%.

Figure 9. Spatial distribution of the Hurst exponent and future trends of RSEI. (a) Spatial distribution of the Hurst exponent. (b) Future change trends.

After overlaying the RSEI trend and the Hurst exponent (Figure 9b), future ecological trajectories were classified as Persistent decrease, Persistent improvement, random change, a Decrease to increase, a Increase to decrease, or approximately unchanged. Overall, degradation risk exceeded improvement potential. Persistent decrease was the dominant type, accounting for 39.01%, and formed large contiguous areas in the central, eastern, and northern mining zones, suggesting that existing degradation trends may continue. Random change accounted for 24.70% and was widely interspersed across the central zone and surrounding transition areas, indicating substantial uncertainty. Areas shifting from degradation to improvement accounted for 15.52% and occurred mainly as patches in the west, southwest, and parts of the center, indicating positive effects of ecological governance and natural recovery. Persistent improvement accounted for 10.49% and was concentrated in the west, southwest, and several belt-shaped areas, suggesting continuity of the improvement trend. Approximately unchanged areas and areas shifting from improvement to degradation accounted for 5.59% and 4.69%, respectively, and were sparsely distributed. Although the latter proportion was small, localized reversal risk remains. Degradation-related types together accounted for 43.70%, substantially exceeding the 26.01% represented by improvement-related types, and random-change areas were also extensive. Future management should prioritize persistently degrading zones in the center, east, and north; consolidate restoration gains in the west and southwest; and strengthen dynamic monitoring and risk control in random-change zones and areas shifting from improvement to degradation.

3.6. Detection of Drivers of Eco-Environmental Quality

Spatial differentiation in EEQ is generally shaped by the combined effects of climate, surface conditions, and human activity [37]. Eight independent variables were selected as potential drivers, with RSEI as the dependent variable. A 1 km × 1 km fishnet yielded 1200 sample points. Because anomalous endpoint-year values may affect the explanatory power of climatic factors, the study period was divided, on the basis of interannual RSEI variation, mining disturbance stages, and ecological governance processes, into five substages within the two major phases: 1999–2004, 2005–2010, 2011–2015, 2016–2020, and 2021–2024. A separate geographical detector analysis was conducted for each substage. A larger q value indicates stronger explanatory power for the spatial differentiation of RSEI. The OPGD was used to quantify the explanatory power and interactions of the factors across stages [38]. Compared with equal-interval periodization, this division more closely represents transitions between peaks and troughs in ecological quality, changes in mining practices, and the progression of ecological restoration projects, thereby helping to reveal shifts in the driving mechanisms.

The significance results in Table 7 show that the q values of all drivers were significant at p < 0.05 in every stage, indicating that the selected variables significantly explained spatial differences in RSEI and that the detector results were reliable. Several factors were also significant at p < 0.01, indicating more stable explanatory effects. Single-factor detection revealed pronounced differences in explanatory power among stages, reflecting the joint regulation of EEQ by hydrothermal conditions, topographic setting, and human activity. Using stage-averaged climatic indicators provided a more stable representation of stage-specific climatic effects and avoided inflating or suppressing the inferred mechanisms through anomalous single-year values. During 1999–2004, stage-averaged temperature (X2) had the highest explanatory power (q = 0.2672), followed by land use (X7), stage-averaged precipitation (X1), and elevation (X3), with q values of 0.1820, 0.1673, and 0.1289, respectively. Pronounced interannual RSEI fluctuation during this stage indicates that the thermal background, land-use pattern, and topography jointly shaped spatial differences in ecological quality. During 2005–2010, stage-averaged precipitation (X1) had the highest explanatory power (q = 0.1299); however, q values were generally low, suggesting that RSEI fluctuation reflected multiple factors and that no single factor had strong independent explanatory power.

Table 7. Single-factor detection results (q values).

Period	Climatic Factors			Land Surface Factors			Human Activities
	X1	X2	X3	X4	X5	X6	X7	X8
1999–2004	0.1673 **	0.2672 **	0.1289 **	0.0486 **	0.0432 **	0.0500 **	0.1820 **	0.0072 *
2005–2010	0.1299 **	0.0693 **	0.0315 *	0.0289 *	0.0197 *	0.0253 **	0.0185 *	0.0266 **
2011–2015	0.3663 **	0.3627 **	0.1622 **	0.1018 **	0.0373 **	0.0701 **	0.0166 **	0.0327 **
2016–2020	0.0531 **	0.0898 **	0.0352 *	0.0392 **	0.0415 **	0.0300 *	0.0085 *	0.0458 **
2021–2024	0.0381 **	0.0795 **	0.1005 **	0.0555 **	0.0388 *	0.0917 **	0.0103 **	0.0389 **

Note: X1, X2, X3, X4, X5, X6, X7, and X8 denote stage-average precipitation, stage-average air temperature, elevation, slope, aspect, population density change, land use, and nighttime light intensity change, respectively. * and ** indicate significance at p < 0.05 and p < 0.01, respectively.

During 2011–2015, the q values of stage-averaged precipitation (X1) and stage-averaged temperature (X2) were 0.3663 and 0.3627, respectively, substantially exceeding those of the other factors. Spatial differences in RSEI during the initial stage of high-intensity mining were therefore primarily constrained by the hydrothermal background. Elevation (X3) and slope (X4) had q values of 0.1622 and 0.1018, respectively, indicating simultaneous topographic modulation. During 2016–2020, stage-averaged temperature (X2) had the highest explanatory power (q = 0.0898). The q values of nighttime light intensity change (X8), stage-averaged precipitation (X1), and aspect (X5) were 0.0458, 0.0531, and 0.0415, respectively. Their convergent explanatory power indicates multifactor regulation during the green mine development and active restoration stage. During 2021–2024, the q values of elevation (X3) and population-density change (X6) were 0.1005 and 0.0917, respectively, whereas that of stage-averaged temperature (X2) was 0.0795. Thus, during stable mining-restoration coordination, climatic dominance weakened while the relative importance of spatial variation in topographic setting and human activity increased. Overall, the driving mechanism shifted from joint climatic and land-use control in the early period, to pronounced hydrothermal dominance during high-intensity mining, and then to coordinated multifactor regulation during restoration.

As shown in Figure 10, interaction detection generally indicated either bivariate enhancement or nonlinear enhancement, demonstrating compound control of stage-dependent RSEI variation in Shendong. During 1999–2004, the interaction between stage-averaged temperature and elevation produced the highest q value (0.501). Interactions of stage-averaged precipitation with elevation and with population-density change produced q values of 0.452 and 0.411, respectively, indicating that the effects of the early hydrothermal background were amplified jointly by topography and human activity. During 2005–2010, the interaction between stage-averaged precipitation and elevation was strongest (q = 0.355); interactions of stage-averaged precipitation with stage-averaged temperature and with aspect yielded q values of 0.340 and 0.343, respectively, indicating sensitivity of ecological fluctuation to moisture conditions and their coupling with topography. During 2011–2015, the interaction between stage-averaged precipitation and temperature had a q value of 0.592, while that between stage-averaged precipitation and slope had a q value of 0.584. Both exceeded the corresponding single-factor effects, indicating that the ecological response during the initial high-intensity mining stage was jointly controlled by the hydrothermal background and topography. During 2016–2020, stage-averaged precipitation and temperature produced the strongest interaction (q = 0.443). During 2021–2024, the interaction between stage-averaged temperature and elevation had a q value of 0.427, indicating that topographic–climatic coupling remained important during the stable post-restoration stage.

Figure 10. Interaction detection results of RSEI driving factors across different ecological evolution stages in the Shendong Mining Area.

In summary, the driving mechanism of stage-dependent RSEI variation in Shendong can be described as follows: hydrothermal conditions provided the baseline constraint on ecological quality fluctuation, topography modulated the spatial response, and human activity amplified local responses during mining disturbance and ecological restoration. Stage-averaged precipitation and temperature stably represented the hydrothermal background, shifting interpretation from a single topographic or anthropogenic driver toward an integrated framework of climatic constraint, topographic modulation, and amplification by human activity. High-intensity mining altered local ecological processes but did not constitute the sole dominant mechanism reshaping regional ecological quality. In the context of an arid mining area, EEQ remained jointly constrained by climate, topographic setting, and the intensity of human activity [39].

4. Discussion

4.1. Spatiotemporal Evolution of RSEI

RSEI in the Shendong Mining Area exhibited marked interannual fluctuation and a weak downward trend from 1999 to 2024. EEQ neither deteriorated nor improved continuously in a single direction, and annual mean RSEI ranged from 0.4249 to 0.6756. The time-series trajectory was closely coupled with changes in mining practices and formed two phases with distinctly different fluctuation amplitudes. Mean RSEI was higher but more variable during 1999–2010; during 2011–2024, the mean declined and the amplitude narrowed.

Pronounced RSEI fluctuations during 2005–2006 and 2009–2011 resulted not only from simultaneous changes in greenness, wetness, dryness, and heat but also from mining development stages, ecological governance projects, and interannual climatic conditions. Around 2005, mine construction and production capacity were expanding continuously; higher NDBSI and LST and lower NDVI jointly produced low RSEI. During 2006–2009, vegetation and moisture conditions gradually recovered as surface drying and thermal pressure weakened. RSEI increased overall and reached the study-period maximum in 2009. This improvement may have reflected the cumulative effects of the Three-North Shelterbelt Program, conversion of cropland to forest and grassland, soil and water conservation, and mine-land reclamation, which partly offset mining disturbance. Around 2010, large-scale expansion and the rapid formation of major mine clusters increased NDBSI and reduced NDVI, causing a marked RSEI decline; unfavorable regional moisture conditions may have further amplified the disturbance. After 2011, high-intensity mechanized mining proceeded alongside ecological governance and restoration. RSEI fluctuated around a lower mean and gradually exhibited narrowing amplitude. Changes in key years therefore represented stage-dependent ecological responses to the combined effects of mining disturbance, climatic conditions, artificial restoration, and natural recovery rather than Persistent decrease driven by a single factor. High-intensity mining remained a negative influence on regional EEQ, but its apparent effect narrowed under continued governance and restoration.

Compared with the mining-modernization and capacity-expansion stage, rapid extraction from ultra-large longwalls in Shendong produced a pronounced zonation of ground fissures. Dynamic ground fissures were less extensive and closed rapidly, forming predominantly reticulate patterns (Figure 11a). Their development cycle was only 18 days, indicating strong self-healing capacity [5]. Edge ground fissures formed an O-shaped ring above the mining boundary and were dominated by step-like cracks (Figure 11b). The fissure angle was nearly vertical, the distribution range was markedly reduced, total surface displacement and subsidence remained unchanged, the subsidence rate increased substantially, and the duration of mining influence on the surface was shortened by approximately 50%. The relatively uniform subsidence zone reached 78% in a single episode, facilitating the post-mining recovery of pre-mining topography [6]. This reduction in land damage suggests that high-intensity extraction did not intensify all manifestations of surface damage; some were weakened or showed self-healing capacity [7]. Coal-seam occurrence and mining technology are relatively consistent across Shendong, producing common patterns of surface movement and deformation, with a subsidence coefficient of 0.5–0.6 [40], and self-healing is widespread.

Figure 11. Distribution characteristics of ground fissures in the Shendong Mining Area: (a) dynamic ground fissures (Red ellipses indicate representative ground fissures); (b) edge ground fissures.

Nevertheless, subsidence and associated fissures may damage root-zone soil structure and accelerate soil-water loss, reducing NDVI and WET. Increased bare land and altered surface hydrothermal conditions may raise NDBSI and LST and ultimately reduce RSEI. These effects are spatially heterogeneous. Previous work projected only weak future vegetation degradation in Shendong [41]. The NDVI time series exhibited an overall phenological cycle, and the effect of mining subsidence on vegetation cover was not uniformly negative; areas with declining vegetation cover accounted for 27.65% of the central mining area [42]. These findings support the observed weak decline, converging amplitude, and dynamic disturbance–recovery–redisturbance adjustment. After 2016, intensified geological environment governance and green mine development strengthened the offsetting effects of subsidence treatment, land reclamation, and vegetation reconstruction. RSEI recovered after several low-value years and gradually shifted to narrower fluctuations. After 2021, ecological governance entered a more systematic implementation and consolidation stage. Mining and governance continued in parallel; although RSEI still fluctuated interannually, the amplitude narrowed further, indicating a transition from strong disturbance responses toward dynamic regulation by mining disturbance, engineering governance, natural recovery, and climatic variability.

Overall, EEQ in the Shendong Mining Area followed a two-phase, wave-like trajectory. The high-intensity mining stage was associated with a lower mean RSEI, but no persistent and irreversible regional decline was observed during the study period. Instead, narrowing RSEI amplitude and stage-specific recovery in the later period indicate that the arid mining-area ecosystem retained a degree of recovery potential under the combined effects of engineering restoration and natural recovery.

4.2. Spatial Explanatory Power of Drivers and Mining-Area Responses

The drivers of EEQ change in Shendong were both stage-dependent and compound. The study area lies in the transition zone between the northern Loess Plateau and the Mu Us Sandy Land, where low precipitation, strong evaporation, and sparse baseline vegetation make the ecosystem highly sensitive to hydrothermal conditions. Studies of ecological risk in arid regions similarly show that EEQ and risk patterns are commonly shaped by the joint effects of hydrothermal conditions, surface-cover change, and human activity intensity, producing pronounced spatiotemporal heterogeneity [43]. Geographical detector results showed that stage-averaged temperature had the strongest single-factor explanatory power during 1999–2004, precipitation and temperature were jointly strong during 2011–2015, and temperature retained a relative advantage during 2016–2020. Climatic constraints were therefore not merely background noise but fundamental controls on stage-dependent RSEI fluctuation. In particular, the strong interactions of precipitation with temperature and slope during 2011–2015 indicate that ecological fluctuation in the initial high-intensity mining stage was especially sensitive to stage-specific hydrothermal conditions.

Elevation, slope, and aspect also explained variation in every stage, indicating that topographic setting modulated the spatial response of RSEI to climatic fluctuation and mining disturbance by affecting water redistribution, soil-erosion intensity, and conditions for vegetation recovery. Elevation already had a relatively high q value during 1999–2004 and became the strongest single factor during 2021–2024. These results show that ecological quality change was not spatially homogeneous but remained constrained by landform units and topographic gradients. Topography is therefore better understood as a spatial modulator of ecological response, with a more prominent role when climatic fluctuation weakens or during stable post-restoration stages.

Human activity effects were most evident during high-intensity mining and engineering restoration. The single-factor explanatory power of population-density change, nighttime light change, and land use was generally lower than that of climate and topography. However, their interactions with climatic and topographic factors increased substantially in several stages, reflecting the spatial targeting of mine construction, goaf treatment, waste-dump reclamation, and green mine projects. During 2021–2024, the q value of population-density change increased to 0.0917, second only to elevation, indicating that residual human disturbance remained important during the stable post-restoration stage. Collectively, single-factor and interaction results showed stage-dependent spatial associations of RSEI with the spatial stratification of hydrothermal conditions, topographic setting, and human activity, and multifactor combinations generally explained more variation than individual factors. The geographical detector reveals spatial associations and explanatory power; it does not establish causal mechanisms.

The RSEI time series did not show a persistent, irreversible overall decline from 1999 to 2024. This temporal pattern cannot be explained solely by the geographical detector results and may reflect the combined influence of regional climatic variability, mining disturbance, natural recovery, and artificial restoration.

4.3. Limitations

This study analyzed internal spatiotemporal variation in RSEI within the Shendong Mining Area and its statistical relationships with climatic, topographic, and human activity factors. No nearby non-mining control area matched for climate, topography, and initial land cover was included. The observed correspondence between mining stages, individual mine units, industrial mining concentrations, and RSEI change should therefore be interpreted as a spatial and temporal association with mining-related ecological responses rather than a strict causal-effect estimate. Although the geographical detector showed that hydrothermal conditions, topographic setting, and human activity jointly explained spatial differentiation in RSEI, it could not fully separate the independent contributions of regional climatic variation, widespread land-cover change, and mining. Future studies should incorporate matched non-mining reference areas and before-after-control-impact or difference-based designs to quantify the net ecological effect of mining more rigorously.

4.4. Implications for Ecological Restoration in Desert Mining Areas

The long-term RSEI changes in the Shendong Mining Area indicate that ecological governance in desert mining areas should not adopt a uniform-intensity engineering restoration strategy. For areas with low restoration potential caused by the combined effects of elevation, slope, and aspect, soil and water conservation measures and drought-resistant vegetation restoration should be prioritized. For mining core zones with strong changes in nighttime light intensity and population density, disturbance source control and land reclamation quality assessment should be strengthened. For peripheral buffer zones and areas with favorable natural recovery conditions, excessive engineering intervention should be reduced, and ecological stability should be improved through enclosure protection and natural succession.

5. Conclusions

Using GEE and Landsat imagery from 1999 to 2024, we constructed RSEI for the Shendong Mining Area, a representative arid mining area in China. Theil–Sen median trend analysis, the Mann–Kendall test, the Hurst exponent, and the geographical detector were integrated to characterize EEQ at the mining-area and individual-mine scales and to examine its trajectories and drivers. The principal conclusions are as follows:

(1)

From 1999 to 2024, RSEI in the Shendong Mining Area exhibited pronounced interannual fluctuations and a weak downward trend, following a two-phase, wave-like pattern corresponding to shifts in mining practices. Mean RSEI was 0.5669 during 1999–2010 and 0.5182 during 2011–2024, with substantially narrower fluctuations in the latter period. Overall, RSEI shifted from a state of high variability and relatively high values to one of lower variability and a lower mean, without exhibiting continuous deterioration.

(2)

Eco-environmental quality (EEQ) in the Shendong Mining Area was characterized by pronounced spatial patchiness and heterogeneity. The moderate and good grades together accounted for 66.16% of the study area and constituted the dominant ecological quality classes. Low-RSEI areas were concentrated mainly in mining core zones and around industrial sites, construction land, and transportation corridors, whereas high-RSEI areas occurred primarily along the mining-area periphery and in areas with relatively dense vegetation cover. The overall spatial pattern was therefore characterized by lower EEQ in mining core zones and higher EEQ in peripheral buffer zones.

(3)

RSEI remained at a moderate level across the 13 representative mines, although substantial differences were observed among individual mines. Buertai and Cuncaota No. 1 were relatively stable, whereas Liuta exhibited the greatest fluctuations and was therefore more sensitive to mining disturbance. During 2011–2024, mean RSEI decreased slightly at most mines, while the magnitude of fluctuation narrowed. These results indicate that mining disturbance continued to adversely affect mine-scale EEQ, although no pronounced retrogressive ecological succession was observed at the regional scale.

(4)

Slight degradation was the dominant type of EEQ change in the study area. Improved and degraded areas accounted for 24.34% and 74.87%, respectively, producing an overall pattern dominated by slight degradation but accompanied by pronounced localized improvement. The Hurst analysis revealed marked spatial differences in future ecological trajectories. Areas projected to undergo Persistent decrease accounted for 39.01%, indicating that some mining core zones remain at risk of continued ecological degradation.

(5)

Based on the temporal evolution of RSEI, shifts in mining practices, and the progression of ecological governance, this study established an analytical framework comprising two major phases and five substages. The spatial differentiation of RSEI in the Shendong Mining Area was jointly influenced by climate, topography, and human activity. The dominant controls shifted from temperature in the early period to the combined effects of precipitation and temperature during the initial high-intensity mining stage, and subsequently to elevation and changes in population density during the restoration and stabilization stage. Topographic conditions modulated spatial differences in ecological responses, whereas human activities amplified localized responses during mining disturbance and engineering restoration. Multifactor interactions were generally stronger than individual-factor effects, demonstrating that changes in mining-area EEQ resulted from the coupled influences of hydrothermal conditions, topographic setting, mining disturbance, and ecological restoration.

Overall, the study period was characterized by a weak decline in mean RSEI and a progressive narrowing of interannual variability. High-intensity mining was spatially and temporally associated with lower eco-environmental quality, whereas natural recovery and ecological restoration partly offset disturbance-related effects in some areas. Nevertheless, Persistent decrease risks remained in several mining core zones. Because no matched non-mining control area was included, these findings should be interpreted as integrated ecological responses within the mining area rather than as causal estimates of the net ecological effect of mining.