Analysis of the Influence of Driving Factors on Vegetation Changes Based on the Optimal-Parameter-Based Geographical Detector Model in the Yima Mining Area

Abstract

The growth of vegetation directly maintains the ecological security of coal mining areas. It is of great significance to monitor the dynamic changes in vegetation in mining areas and study the driving factors of vegetation spatial division. This study focuses on the Yima mining area in Henan Province. Utilizing MODIS and multi-dimensional explanatory variable data, the Theil–Sen Median + Mann–Kendall trend analysis, variation index, Hurst index, and optimal-parameter-based geographical detector model (OPGD) are employed to analyze the spatiotemporal changes and future trends in the EVI (enhanced vegetation index) from 2000 to 2020. This study further investigates the underlying factors that contribute to the spatial variation in vegetation. The results indicate the following: (1) During the period studied, the Yima mining area was primarily characterized by a moderate-to-low vegetation cover. The area exhibited significant spatial variation, with a notable pattern of “western improvement and eastern degradation”. This pattern indicated that the areas that experienced improvement greatly outnumbered the areas that underwent degradation. Moreover, there was an inclination towards a deterioration in vegetation in the future. (2) Based on the optimal parameter geographic detector, it was found that 2 km was the optimal spatial scale for the analysis of the driving factors of vegetation change in this area. The optimal parameter combination was determined by employing five spatial data discretization methods and selecting an interval classification range of 5–10. This approach effectively addresses the subjective bias in spatial scales and data discretization, leading to enhanced accuracy in vegetation change analysis and the identification of its driving factors. (3) The spatial heterogeneity of vegetation is influenced by various factors, such as topography, socio-economic conditions, climate, etc. Among these factors, population density and mean annual temperature were the primary driving forces in the study area, with Q > 0.29 and elevation being the strongest explanatory factor (Q = 0.326). The interaction between temperature and night light was the most powerful explanation (Q = 0.541), and the average Q value of the interaction between the average annual temperature and other driving factors was 0.478, which was the strongest cofactor among the interactions. The interactions between any two factors enhanced their impact on the vegetation’s spatial changes, and each driving factor had its suitable range for affecting vegetative growth within this region. This research provides scientific support for conserving vegetation and restoring the ecological system.

1. Introduction

The large-scale and high-intensity mining of coal resources has led to the degradation of vegetation in mining areas, which has led to a series of environmental problems [1]. This seriously affects and restricts the ecological environmental protection work and development of mining areas. With its immense carbon sequestration rate and potential in multiple areas, vegetation plays an indispensable role in mining area ecosystems. The growth of vegetation plays a crucial role in ensuring the ecological integrity of mining areas [2] and serves as a significant measure for restoring ecosystems and conserving the environment [3].

Remote sensing technology has been widely used in vegetation monitoring in different areas and various conditions due to its unique advantages in monitoring and analyzing surface vegetation changes [4]. The vegetation index is a key parameter describing the growth trend in vegetation, which can effectively reflect the changes in vegetation ecosystems [5]. At present, domestic and foreign scholars have conducted a lot of research on vegetation change monitoring using MODIS-NDVI [6,7], GIMMS-NDVI [8,9], and Landsat-NDVI [10,11] datasets. At the same time, the vegetation index also plays a key role in the study of vegetation changes in mining areas [12,13,14,15,16,17]. However, in areas with a dense vegetation cover, due to NDVI saturation, the accuracy of studies is adversely affected [18]. The EVI has the advantages of anti-atmospheric interference, high vegetation coverage adaptability, and the ability to handle high reflectance from the surface and discriminate bare land. Due to the introduction of a blue light band for calculation, its evaluation of vegetation coverage and growth status has become more accurate [19], and it has been actively applied to monitor long time-series changes in vegetation [20,21,22,23].

Vegetation change is comprehensively affected by various natural factors and human factors. Owing to variations in geographical regions, the influencing factors of vegetation change exhibit regional disparities [24]. Many researchers have developed various methods to study the driving forces of vegetation change, among which correlation analysis plays an important role in studying the effects of climate factors on vegetation change [25,26,27,28]. However, most past studies regarded natural and human factors as independent and ignored their interaction in changing vegetation cover [29]. The geographic detector is an important model created by Wang Jinfeng et al. to explore spatial differences and their driving factors [30]. This approach can assess the non-linear correlation between vegetation and its influencing factors and uncover the synergistic impact of different influencing factors, thereby addressing the limitations of conventional analysis methods. However, the traditional geodetector model was limited by the spatial scale, spatial data discretization method, and hierarchical number. On this basis, Song et al. [31] developed a new geodetector model (OPGD) with optimal parameters, which resolves the data discretization and subjective selectivity of spatial scale issues and evaluates the research object more scientifically [32]. At present, the OPGD model has been applied to many research fields, such as climate change [33], ecology [34], land utilization [35], urban environments [36], and so on.

In summary, the purpose of this study mainly includes two aspects. Firstly, we reveal the spatial heterogeneity of vegetation in the Yima mining area from 2000 to 2020 and its long-term change process and change trend in the future. Secondly, on the basis of optimizing spatial scales and the related parameters, we objectively analyze the explanatory power of various driving factors causing spatial differentiation in vegetation. Through this study, we hypothesize that the spatial heterogeneity of vegetation is significant, and its change trend is improved within this time range. Simultaneously, the development of vegetation is influenced by a multitude of factors, including natural conditions and human activities, which will be quantified and elucidated in this analysis. The research results can provide scientific support for ecological environmental protection in mining areas.

2. Materials and Methods

2.1. Study Area

The Yima mining area (China) is located between the coordinates 34°17′10″–35°02′37″ N and 111°08′57″–112°29′29″ E (Figure 1a). It extends from the Sanmenxia City Development Zone in the west to Mengjin County in the east, and from Minggao Town in Yichuan County in the south to the south bank of the Yellow River in the north (Figure 1c). The study area spans 2212.5 km² and is characterized by a typical temperate monsoon climate. The Yima mining area is characterized by a low mountain and hilly landform. The topography exhibits an elevation gradient, with higher terrain in the western and southern regions and lower terrain in the eastern and northern regions (Figure 1b). The topography of individual areas is complex within the mining area, and cropland is the predominant land type (Figure 1d). Croplands constitute 78.4% of the area, followed by construction land and forestland, which account for 10.5% and 7.2%, respectively. The smallest and yet negligible proportion belongs to shrub land. The expansion of the mining sector has stimulated economic growth in the surrounding region, yet it has also had a significant detrimental effect on the local vegetation [37].

2.2. Data Sources and Preprocessing

The EVI data used in this study came from the MODIS13Q1 dataset provided by NASA. This dataset has a temporal resolution of 16 days and a spatial resolution of 250 m. Firstly, the GEE platform was used to preprocess the original images by splicing, format conversion, and resampling to obtain the monthly EVI time series. Then, based on the maximum value synthesis method (MVC), the interference of clouds and atmosphere was eliminated to obtain the annual maximum EVI time series. This represents the optimal condition of the vegetation growth of the year and also eliminates the impact of missing EVI data in January 2000 in the MODIS13Q1 product dataset [38].

In this study, climate, terrain, social, land-use type, and soil type data were selected as the explanatory variables for explaining the spatial differences in vegetation cover in the Yima mining area. The data sources are shown in

Table 1. Data sources.

Data Type	Driving Factors	Code	Resolution	Temporal Extent (Year)	Data Sources
Climatic data	Temperature	X1	1 km	2000–2020	Resources and Environmental Science and Data Center, Chinese Academy of Sciences https://www.resdc.cn/ (accessed on 25 February 2024).
	Precipitation	X2	1 km	2000–2020
	Evaporation	X3	1 km	2000–2020
	Sunny hours	X4	1 km	2000–2020
	Relative humidity	X5	1 km	2000–2020
Social data	GDP	X6	1 km	2000–2020
	Population Density	X7	1 km	2000–2020
	Night light	X8	1 km	2000–2020	China Tibetan Plateau Scientific Data Center https://data.tpdc.ac.cn/ (accessed on 2 March 2024).
Topographic data	Elevation	X9	30 m	2020	USGS https://www.usgs.gov/ (accessed on 25 February 2024).
	Slope	X10	30 m	2020
Land and soil data	Land-use type	X11	30 m	2000, 2020	Resources and Environmental Science and Data Center, Chinese Academy of Sciences https://www.resdc.cn/ (accessed on 5 March 2024).
	Soil type	X12	1 km	2020

. All datasets are uniformly projected and resampled to a 1 km resolution based on Arcmap10.2.

2.3. Methods

2.3.1. Theil–Sen Median + Mann–Kendall Trend Analysis

The Theil–Sen Median trend analysis is a non-parametric statistical method [39]. This method is insensitive to outliers, has high computational efficiency, is suitable for the trend analysis of long time-series data, and can provide robust trend estimation [40]. In this study, the Theil–Sen Median trend analysis was used to determine the trend of EVI changes in the Yima mining area from 2000 to 2020. The calculation principle is as follows:

$$\mathrm{S}\mathrm{l}\mathrm{o}\mathrm{p}\mathrm{e}=\mathrm{M}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{n}\left({\displaystyle \frac{\mathrm{x}\mathrm{j}-\mathrm{x}\mathrm{i}}{\mathrm{j}-1}}\right),\mathrm{i}<\mathrm{j}$$

(1)

where xi and xj are the measured pixel values in year i and the year j, respectively. When the slope is positive, the time series of the measured data shows an upward trend; when the slope is negative, the time series of the measured data shows a downward trend; and when the slope is equal to 0, the time series of the measured data has no obvious trend.

The Mann–Kendall significance test is a widely used non-parametric test [41]. It adopts a null hypothesis that the data are independent and randomly distributed. This study uses this method to determine whether the EVI trend in the Yima mining area is significant. The calculation principle is as follows:

$$Z\left\{\begin{array}{c}\begin{array}{c} \begin{array}{c}{\displaystyle \frac{S-1}{\sqrt{Var\left(S\right)}}},S>0\\ 0,S=0\\ {\displaystyle \frac{S-1}{\sqrt{Var\left(S\right)}}},S<0\end{array} \end{array}\end{array}\right.$$

(2)

$$S={\sum }_{i=1}^{n-1}{\sum }_{j=i+1}^{n}sign(xi-x j)$$

(3)

$$sign\left(xi-x j\right)=\left\{\begin{array}{c}+1,xi-x j>0\\ 0,xi-x j=0\\ -1,xi-x j<0\end{array}\right.$$

(4)

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

(5)

where xi and xj are the measured pixel values in year i and the year j, respectively, and n is the length of the time series. In this study, according to the value of |Z|, the change significance of the measured data is divided into the following two categories: if |Z| ≥ 1.96, the change trend of the measured data is significant; if 0 ≤ |Z| ≤ 1.96, the trend of the measured data is slightly significant.

2.3.2. Coefficient of Variation

The coefficient of variation measures the degree of variation of a data sequence and reflects the degree of spatial data variations over time so as to evaluate the stability of the time series [42]. In this study, this method is used to characterize the stability of EVI data at pixel scale in the Yima mining area. The calculation principle is as follows:

$$CV={\displaystyle \frac{\sqrt{\sum _{i=1}^n (X_i-\overline{X}{)}^2/(n-1)}}{\overline{X}}}$$

(6)

where X_i is the EVI pixel value in year i; $\overline{X}$ is the multi-year average value of the EVI pixel; and n is the length of the time series (n = 21).

2.3.3. Hurst Index

Time series with self-similarity and long-range dependence are ubiquitous, and the Hurst index is one of the main methods to quantitatively describe long-range dependencies [43]. The Hurst index can be estimated by a variety of methods, such as R/S analysis, wavelet analysis, and the Whittle method. The most commonly used method is the R/S analysis. In this study, the R/S analysis is used to calculate the Hurst index to describe whether the EVI’s future trend is self-similar to the past trend, so as to predict the future changes. The calculation principle is as follows.

For any given time series (t), t 1, 2, 3...... at any time 1, the mean sequence is constructed as

$${ \beta }_{\lambda }={\displaystyle \frac{1}{\lambda }}{\sum }_{i=1}^{\lambda } \beta \left(t\right) ,\lambda =\mathrm{1,2},3\dots \dots$$

(7)

Then, the cumulative deviation sequence is formed as

$$\gamma \left(t,\lambda \right)=\sum \left(\beta \left(u\right)-{\beta }_{\lambda }\right),1\le t\le \lambda$$

(8)

The computational range is

$$R\left(\lambda \right)=\gamma (t,\lambda {)}_{max}-\gamma (t,\lambda {)}_{min},\lambda =\mathrm{1,2},3\dots \dots$$

(9)

The standard deviation is

$$S\left(\lambda \right)={\left[{\displaystyle \frac{1}{\lambda }}{\sum }_{i=1}^{\lambda } {\left(\beta \left(t\right)-{\beta }_{\lambda }\right)}^2\right]}^{{\displaystyle \frac{1}{2}}},\lambda =\mathrm{1,2},3\dots \dots$$

(10)

The Hurst index is then defined as

$$H={\displaystyle \frac{{\mathit{log}}_{\lambda }\left(R\left(\lambda \right)\right)}{c\times S\left(\lambda \right)}},0<H<1$$

(11)

where c is a constant. When H = 0.5, it indicates that the time series is an independent random series with limited variance. When 0 < H < 0.5, it indicates that the time series has long-term dependence and behaves as anti-persistence, that is, the future trend is opposite to the past trend. Closer H values to zero indicate a stronger anti-persistence. When 0.5 < H < 1, it indicates that the time series has a long-term dependence and shows persistence, that is, the future trend is consistent with the past trend. Closer H values to 1 indicate a stronger persistence. The Theil–Sen Median trend analysis chart was superimposed on the Hurst index chart to obtain the vegetation future trend chart [44] (

Table 2. Classification of the EVI variation in the future.

Slope	Hurst
	0 < H < 0.5	H = 0.5	H > 0.5
Slope < 0	From degradation to growth	Uncertain	Sustainable degradation
Slope > 0	From growth to degradation	Uncertain	Sustainable growth

).

2.3.4. Optimal-Parameter-Based Geographical Detector Model (OPGD)

The OPGD model includes five components: parameter optimization, factor detection, interactive detection, risk detection, and ecological detection. Parameter optimization refers to the discrete optimization of the spatial scale and the data. Scale optimization is used to compare the changes of Q values of the explanatory variables at different spatial scales. When 90% quantiles of Q values of all explanatory variables at the same spatial scale reach the maximum, the spatial scale is optimal. Discrete optimization involves employing various discrete techniques to identify the division of continuous variables. This is carried out by utilizing the Q value obtained from the factor detector to determine the optimal combination of parameters. It also provides a collection of discretization methods and breakpoints for each continuous variable, allowing for the calculation of their respective Q values [45]. The explanatory power of each driving factor X to Y as well as the heterogeneity of the geographical dependent variable Y are examined using the factor detector. The calculation formula is as follows:

$$Q=1-{\displaystyle \frac{h=1}{N{\delta }^2}}=1-{\displaystyle \frac{SSW}{SST }}$$

(12)

$$\begin{array}{r}SSW=\sum _{h=1}^L{N}_h{\delta }_h^2,SST=N{\delta }^2\end{array}$$

(13)

where Q is the factor’s explanatory power ranging from zero to one; h is the classification interval of the independent variable X; Y is the dependent variable; N is the number of partitions in each layer or the whole domain; ${\delta }_h^2$ and ${\delta }^2$ are the variances of X and Y classification intervals, respectively. SSW and SST are the total variance of the intra-layer and the whole region, respectively.

The interaction detector is used to determine whether the synergy’s explanatory power for Y differentiation is improved, weakened, or unaffected if any two driving factors, X1 and X2, interact. The core concept involves the following procedure: firstly, determining the explanatory impact of factors X1 and X2 on Y as q(X1) and q(X2), respectively; secondly, calculating the combined explanatory power q(X1∩X2) of the of X1 and X2 intersection on Y; and finally, comparing q(X1), q(X2), and q(X1∩X2) to complete the classification of the explanatory power (

Table 3. The underlying principle of determining interaction.

Type of Interaction	Judgment Basis
Non-linear reduction	Q(X1∩X2) < Min[Q(X1), Q(X2)]
Single-factor non-linear reduction	Min[Q(X1), Q(X2)] < Q(X1∩X2) < Max[Q(X1), Q(X2)]
Two-factor enhancement	Q(X1∩X2) > Max[Q(X1), Q(X2)]
Non-linear enhancement	Q(X1∩X2) = Q(X1) + Q(X2)
Independent	Q(X1∩X2) > Q(X1) + Q(X2)

Note: X1 and X2 represent two different independent variable factors, and Q is the explanatory power of the independent variable factor X to the spatial differentiation of the dependent variable factor Y.

).

Risk detection is used to determine whether there are any significant differences between the average attribute values among the subregions of the two factors. Risk detection was performed by calculating the EVI average of the subregions for the influencing factors, which was followed by statistical significance testing. The subregion of the influencing factors for vegetation growth becomes more suitable as the average EVI increases in each region. This can be used to identify the appropriate range or type of each impact factor and to find areas with high vegetation coverage. Risk detection using t statistics is calculated as follows:

$$t={\displaystyle \frac{{\bar{Y}}_{h=1}-{\bar{Y}}_{h=2}}{{\left[\frac{Var\left(Y_{h=1}\right)}{n_{h=1}}+\frac{Var\left(Y_{h=2}\right)}{n_{h=2}}\right]}^{\frac{1 }{2}}}}$$

(14)

where ${\bar{Y}}_{h }$ is the mean of the dependent variable of subregion h; n_h is the sample size in subregion h; and var is the variance.

Ecological detection is used to detect the significance of driving factors on the spatial distribution of vegetation [46] measured by the F statistic:

$$F={\displaystyle \frac{N_{x1}{N}_{x2}-1SS{W}_{x1}}{N_{x2}{N}_{x2}-1SS{W}_{x2}}}$$

(15)

$${SSW}_{x1}={\sum }_{h=1}^{L1} N_h{\sigma }_h^2,{SSW}_{x2}={\sum }_{h=1}^{L2} N_h{\sigma }_h^2$$

(16)

where N_x1 and N_x2 represent the sample size for each factor, respectively; SSW_x1 and SSW_x2 represent the sum of the layers formed by the factors, respectively; and L₁ and L₂ represent the number of layers for each factor. Here, the null hypothesis is H₀: SSW_x1 = SSW_x2. If H₀ is rejected at the significance level of α, it means that there is a significant difference between the two factors in explaining the spatial differentiation of vegetation cover.

3. Results

3.1. Parameter Optimization

The data used in this study are all multi-year averages from 2000 to 2020. This study takes 1 km, 2 km, 3 km, and 4 km as different classification standards [31]. We found that 2 km is the optimal spatial scale since the 90% quantile of the Q value of the explanatory variable reached its maximum at this time (Figure 2). In addition, the discretization interval of 10 continuous explanatory variables was set to 5–10 categories, and five methods of natural discontinuity, standard deviation discontinuity, geometric discontinuity, quantile discontinuity, and equal interval discontinuity were used to discretize them. The OPGD model automatically selects the best discretization method and the number of intervals for each continuous variable according to the Q value calculated by the factor detector. The results show that the optimal parameter combination of X1 and X8 is divided into eight intervals by the standard deviation discontinuity method. The optimal parameter combination of X2, X4, X6, and X10 is divided into 10 intervals by the quantile discontinuity method. The optimal parameter combination of X3 is divided into nine intervals by the natural discontinuity method. The optimal parameter combination of X5 is divided into 10 intervals by the natural discontinuity method. The optimal parameter combination of X7 is divided into eight intervals by the standard deviation discontinuity method. The optimal parameter combination of X9 is divided into 10 intervals by the standard deviation discontinuity method (Figure 3).

3.2. Analysis of the Vegetation Spatial Heterogeneity

The spatial pattern of vegetation cover is depicted by calculating the multi-year EVI average at the pixel scale from 2000 to 2020 (Figure 4a). It can be seen that the spatial heterogeneity of the multi-year EVI average in the Yima mining area is obvious, and its value increases from northwest to southeast. The EVI index of the Yima mining area is segment into five grades: 0–0.2, 0.2–0.4, 0.4–0.6, 0.6–0.8, and 0.8–1 [47] (Figure 4b), representing extremely low, low, medium, high, and extremely high vegetation coverage. The Yima mining area was dominated by medium–low vegetation cover. The area of the EVI ≤ 0.4 in 2000 and 2020 accounted for 88.74% and 90.92%, respectively. The coverage area of the high and extremely high vegetation cover only accounted for 9.07% of the total area, which is mainly concentrated in Yanzhen Township in the middle of the Yima mining area.

In 2000, the Yima mining area was dominated by a low vegetation coverage, which was mainly distributed in the west of the area (Figure 5a). In 2020, the Yima mining area was mainly covered by moderate vegetation, which was distributed in other areas except the urban center and its surrounding areas, the northern Yellow River basin, and the western open-pit mines (Figure 5b). The magnitude of moderate vegetation coverage in the Yima mining area increased from 62.51% to 80.76%, and the area of low and high vegetation coverage decreased from 25.94% and 11.13% to 9.24% and 9.06%, respectively. The changes in the medium and the low vegetation cover classes are the main type of vegetation cover transfer in the mining area. The change in vegetation coverage in the mining area was primarily caused by the shift of low vegetation in the western region and high vegetation in the southeastern region to the medium vegetation class (Figure 5c). This is related to the implementation of ecological restoration projects in large open-pit coal mines in the western mining area in recent years as well as the disturbance of vegetation in the cropland in the southeast by hydrothermal conditions.

3.3. Analysis of Vegetation Spatiotemporal Variations

3.3.1. Temporal Variations of EVI

The annual EVI average is between 0.399 and 0.537, rising from 0.472 in 2000 to 0.502 in 2020. The mean EVI value is 0.484 with a standard deviation of 0.037, suggesting that the annual average EVI values are closely clustered. There are no substantial changes in the EVI in the area, although there is a considerable fluctuation range during the period. The changes in EVI can be roughly divided into three stages (Figure 6).

3.3.2. Spatial Variation Characteristics of EVI

The slope of the EVI changes in the area from 2000 to 2020 is between −0.023/a and 0.0166/a, which is spatially characterized by “improvement in the west and degradation in the east” (Figure 7a). The areas with low EVI change values are primarily found in the urban areas of XinAn County, Yiyang County, Yima City, Mianchi County, Yadi Village, and part of the southeast Yima mining area. These regions are mostly flat plains and urban and suburban areas. The high-value areas are mainly concentrated in the western mountains, southern hills, and the Yellow River Basin in the north of the area.

From 2000 to 2020, the area of EVI improvement in the Yima mining site was greater than the area of degradation (Figure 7b). The EVI showed an upward trend in 67% of the regions, of which 28% showed a significant increase. They were predominantly located in the Jinping Mountain area of Yiyang County, the surrounding areas of the Yellow River basin, and the low hills and ridges in the western part of the Yima mining area. A trend of degradation was found in 33% of the regional EVI, with 28% showing slight significant degradation in the central and southeastern part of the Yima mining area and 5% showing significant degradation in urban centers, their surrounding areas, and the open-pit bauxite mining area in the northwest.

3.3.3. Temporal EVI Stability

In this study, the fluctuation in vegetation change was divided into four grades: mild, moderate, severe, and extremely severe, separated with a CV value of 0.05 [48]. Over the past 21 years, the vegetation changes in the Yima mining area have been relatively unstable (Figure 8), with moderate fluctuations covering 61% of the total area. The low-fluctuation area accounts for 10%, mainly distributed in the north and east of the mining area. The areas with high and extremely high fluctuations account for 29%, which are mainly distributed in the city centers and the suburbs, as well as the rural areas in the west and southeast.

3.3.4. Prediction of Future EVI Trend

The Hurst value of the EVI in the mining area is between 0.13 and 0.85 (Figure 9a). The area with a Hurst < 0.5 accounts for 83% of the total, meaning that the future development trend of the EVI in these areas is not maintainable. The area with a Hurst > 0.5 accounts for about 17% of the total, implying that the development trend of the EVI in these areas is supportable in the future. These areas are mainly located in the northern, western, and southern parts of the Yima Mining Area.

In order to predict the positive and negative persistence of EVI future changes in the area, the Hurst index raster data and the trend analysis results of the vegetation EVI over the 21 years were superimposed (Figure 9b). The area proportion of the future change trend of the EVI in the Yima mining area has the following order: from growth to degradation (56%) > from degradation to growth (27%) > continuous growth (11%) > continuous degradation (6%). This shows that the vegetation in the Yima mining area will have a degradation trend in the future. These areas are mainly located in the western part of the Yima mining area, the surrounding area of the Xiaolangdi Reservoir in the north, and the western part of Yiyang County. The area with an improving vegetation condition in the future is mainly concentrated in the central and southeastern part of the Yima mining area.

3.4. Driving Factors of Vegetation Spatial Differences

3.4.1. Factor Detection Analysis

The Q value was used to measure the explanatory power of each driving factor of the overall pattern of EVI spatial distribution, which was obtained as the following order: elevation (X9) > population density (X7) > annual average temperature (X1) > GDP (X6) > annual average sunny hours (X4) > night light (X8) > annual average evaporation (X3) > land-use type (X11) > annual average precipitation (X2) > relative humidity (X5) > soil type (X12) > slope (X10) (Figure 10). Among these, elevation, population density, and temperature, with Q values above 0.29, were the main driving factors of EVI change in the Yima mining area, while soil type and slope, with Q values less than 0.052, had little influence on EVI changes. The Q values of the sunny hours, GDP, night light, evaporation, land-use type, precipitation, and relative humidity were between 0.122 and 0.214. This indicates that the vegetation in the Yima mining area is affected by topography, the social economy, and climate. Different from other regions, the vegetation change in the Yima mining area is greatly affected by elevation.

3.4.2. Interactive Detection Analysis

Interaction detection can further analyze the interaction between the driving factors, as well as identify their explanatory power. The Q values of the interaction between the driving factors are higher than that of any single factor (Figure 11). The interaction between factors demonstrates the correlation between non-linear enhancement and enhancement resulting from the combination of two factors. There is no factor that works independently, and there are obvious differences in the interaction between factors.

The primary factors in the single-factor detection were elevation and population density. The average Q values for the interaction detection with other factors were 0.470 and 0.450. However, the interaction between temperature and other factors was the strongest, with a Q value greater than 0.39 and an average Q value of 0.478. Temperature ∩ nighttime light (Q = 0.541) had the highest explanatory power. Nighttime light images can reflect the intensity of human activities at the macro-level. From a micro-perspective, the nighttime light environment also has a direct impact on plant photosynthesis and carbon metabolism [49]. This implies that the combination of temperature and nighttime light could have a more pronounced impact on the spatial distribution of vegetation in the region. It was also found that the explanatory power of soil type ∩ slope was the lowest (Q = 0.148).

3.4.3. Risk Detection Analysis

In order to analyze the range or type of driving factors of vegetation growth, this study used the risk detector in the optimal parameter geographic detector. The findings indicate that there are notable variations in the impact of driving factors on the EVI across different classification intervals. Among the climatic factors, the average annual temperature was 14.9 °C–15.2 °C, the average annual evaporation was 1430 mm–1450 mm, the average annual sunshine hours were 1985 h-1998 h, the average annual relative humidity was 62.4%–62.6%, and the average annual precipitation was 662 mm-688 mm. When vegetation growth is more robust, the average EVI can exceed 0.5. This suggests that optimal hydrothermal and light conditions are more conducive to vegetation growth. Among the terrain factors, the average EVI reached a maximum of >0.51 in the fourth range (elevation 360 m–431 m, slope 4°–5°). Among the socio-economic factors, when the population density was 377–525 people/km², the per capita GDP was 2590–2630 CNY/km², and the night light was 364–714 cd/km², the average EVI value was the largest, which is most conducive to vegetation growth. Forestland with red clay soil exhibited a higher average EVI and superior vegetation growth (Figure 12). This shows that social and economic factors have a significant impact on the EVI in the Yima mining area. The vegetation growth is improved when the level of human disturbance is reduced [50].

3.4.4. Ecological Detection Analysis

There is no significant difference among the following driving factors: population density (X7) and average annual temperature (X1); elevation (X9) and population density (X7); average annual precipitation (X2) and average annual relative humidity (X5); average annual precipitation (X2) and land-use type (X11); annual evaporation (X3) and land-use type (X11); and annual sunshine duration (X4) and night light (X8). There were significant differences between each pair of factors except those above (Figure 13).

4. Discussions

4.1. Spatiotemporal Variations of Vegetation

The EVI grades in the cropland and forestland areas of the Yima mining area are predominantly categorized as medium and high. Human factors play a significant role in the agro-ecosystem. In addition to suitable climatic conditions, crop growth is mainly managed and regulated by human activities. Forests possess a robust inherent capacity to resist external disturbances during their natural development. Forests generally maintain a stable growth state unless they are seriously disrupted by human activities like deforestation and urban expansion or by extreme adverse climatic conditions. As a result, vegetation growth is favorable in both areas. Compared with the first two land-use types, the grassland ecosystem is relatively fragile. The utilization and degradation of grasslands have formed a complex and close relationship with the natural environment. Thus, a specific portion of the grassland area is characterized by low EVI values (Figure 14).

The land-use type of the Yima mining area is mainly cropland. Limited by the temperate monsoon climate and topographic conditions, the agricultural system mainly includes cropping winter wheat(Triticum aestivum L.) and performing summer maize(Zea mays L.) rotations [51,52]. The growth of crops is strongly influenced by climate change. Winter wheat and summer corn in particular require higher temperatures [53,54]. Additionally, rain-fed agriculture is prevalent in most of the region [55], which further increases the uncertainty surrounding the impact of climate on crop growth. Thus, the EVI change exhibits greater temporal volatility. However, due to the regularity of the cropping cycle, the changes in the EVI on the whole are not obvious in 2000–2020 (Figure 6). The annual average of the EVI continued to decrease from 0.472 in 2000 to 0.399 in 2002, marking its lowest value in 21 years. On the one hand, following the improvements in coal sale conditions in 2000, intensified coal extraction activities in the Yima mining area led to a proliferation of dumping sites, including open-pit coal gangue hills and cinder piles, resulting in direct adverse impacts on vegetation. On the other hand, there was a decline in the average annual temperature and precipitation (Figure 15c), reaching its lowest level during 2000–2002, which created unfavorable hydrothermal conditions that hindered vegetation growth. However, the improved vegetation area was much larger than the degraded area (Figure 7), which was consistent with the results of Wang H et al. [37]. The areas that exhibited improvement were primarily situated in the western and southern regions of the Yima mining area. These areas have relatively higher elevations and are mostly covered with forests. These forests enjoy stable growth, which makes them resistant to disturbances [56]. At the same time, in recent years, the ecological restoration projects in the abandoned open-pit mines have also improved vegetation conditions [57]. The degraded areas were primarily found in the urban centers and their surrounding regions, as well as in the agricultural areas in the southeast and the open-pit bauxite mining areas in the northwest. These areas exhibited significant fluctuations. The study area’s level topography and dense population make it an ideal site for urban expansion and agricultural activities. The ecological environment of the mining site is inherently delicate, which has a direct impact on the growth of vegetation, aligning with previous research findings [58,59]. Open-pit mining has had a great influence on the disturbance and destruction of the ecological environment [60]. Furthermore, the vegetation in the degraded areas continues to exhibit a deteriorating trend for the future. Therefore, it is necessary to promote vegetation restoration projects and the ecological environment of the area [61].

4.2. Spatial Scale Optimization and Spatial Data Discretization

The findings indicated that the 90th percentile Q value of the 10 explanatory variables is highest when measured at the 2 km spatial scale. This spatial scale is considered optimal for investigating the spatial and temporal patterns of the EVI in the Yima mining area. The OPGD model is suitable for both continuous variables and categorical variables. However, continuous variables need to be optimized by discrete partitions. Five discretization methods were employed to optimize the 10 parameters, thereby elucidating the most promising mechanism underlying the spatial variations in vegetation and the driving factors. The results of some scholars also verify the effectiveness and reliability of the OPGD model parameter optimization [31,62,63]. Compared with previous studies, the main role of the OPGD model is to optimize the combination of the spatial scale parameters, discretization methods, and classifications in an objective way and to judge the explanatory power of the driving factors of vegetation spatial differences. This method is more scientific and accurate in revealing the characteristics of the spatial data and providing support for exploring the spatial pattern and heterogeneity of geographic information.

4.3. Vegetation Spatial Differences’ Driving Factors

Vegetation growth and change are influenced by a variety of factors [64]. Elevation, population density, and average annual temperature have an absolute advantage in explaining the spatial differentiation of vegetation. Different from many studies, elevation has the strongest single-factor explanatory power (Figure 10), and vegetation growth is the best when elevation is 360 m–431 m (Figure 12). On the one hand, the Yima mining area has a rugged topography, and vegetation types, hydrothermal conditions, and population density change with elevation. Therefore, elevation directly affects the distribution pattern of vegetation, which is supported by previous studies [50,65,66]. On the other hand, more cropland is distributed between the elevations 360 m and 431 m (Figure 1), and crops are more sensitive to hydrothermal conditions [67]. The average annual temperature change rate (Figure 15a) and the average annual precipitation change rate (Figure 15b) in this elevation zone mostly approach 0, indicating a stable condition suitable for the growth of crops. Cui Y et al. [68] also showed that changes in temperature and precipitation can cause differences in crop growth, while stable hydrothermal conditions play an important role in sustainable crop growth.

Of all the climate factors, the annual mean temperature had a stronger explanatory power than the annual mean precipitation. However, the highest explanatory power was not solely attributed to the annual mean temperature, but rather to its interaction with other important factors. The Yima mining area experiences a temperate monsoon climate that gradually transitions from east to west, shifting from a plain to a hilly and mountainous climate. The average annual precipitation is generally moderate, so vegetation in most areas has a low response to precipitation. However, under the influence of the regional geographical environment, in areas with obvious terrain fluctuations, air temperature becomes the key climate factor affecting vegetation growth, and this is consistent with the findings of Peng W F et al. [69] and Lamchin M et al. [70]. The results of risk detection show that vegetation grows well under favorable moisture, heat, and light conditions and with less human intervention, which is consistent with the results of Wang Y et al. [29]. Most of the driving factors have significantly different impacts on vegetation in the Yima mining area. In general, it is clear that different driving factors have different action mechanisms and degrees of influence on vegetation’s spatial differentiation. In order to effectively protect the vegetation resources in this area, ecological management strategies need to fully consider the unique role of each driving factor.

5. Conclusions

Based on the MODIS13Q1 data of the Yima mining area from 2000 to 2020 obtained on the GEE platform, this study used the Theil–Sen + Mann–Kendall trend analysis, the coefficient of variation, and the Hurst index to clarify the spatiotemporal change trend, stability, and future trend of the EVI. The optimal spatial scale and data discretization parameters were determined using the OPGD model, and the driving factors behind the spatial differences of vegetation in the Yima mining area were revealed. The research results can provide reference for future vegetation ecological management strategies and ecological environment protection and restoration in mining areas. The main conclusions are as follows:

(1)

Temporally, the average annual EVI of the area did not change significantly from 2000 to 2020, but it fluctuated greatly during the period. Spatially, the Yima mining area is dominated by low and medium vegetation cover, with an obvious spatial heterogeneity. The vegetation change is moderately fluctuating, showing a pattern of “improvement in the west and degradation in the east”. The area showing an improvement trend is much larger than the area showing a degradation trend. The areal proportion of vegetation change trend in the future was in the following order: from growth to degradation (56%) > from degradation to growth (27%) > sustained growth (11%) > sustained degradation (6%).

(2)

The analysis of the optimal parameter geographic detector model shows that the 90% quantile of the driving factor Q is the largest on the spatial scale of 2 km, which is the best scale by which to study the vegetation spatiotemporal pattern in the Yima mining area. The study employed five spatial data discretization methods and 5–10 categories to objectively select the optimal parameter combination. This approach effectively addressed the subjective selectivity of the spatial scale and data discretization, leading to an improved accuracy in studying vegetation change and its driving factors.

(3)

In factor detection, the single-factor explanation power of elevation is the strongest (Q = 0.326). However, the average annual temperature is affected by topography and exerts an important influence as the strongest auxiliary factor in the interaction. The interaction between any two factors was found to be non-linear and bivariate enhancements, which intensifies their influence on the spatial changes in vegetation. It is important to consider the appropriate vegetation range for each driving factor in future ecological protection efforts in mining areas.