Interactions and Driving Force of Land Cover and Ecosystem Service Before and After the Earthquake in Wenchuan County

Abstract

The Wenchuan earthquake, an unexpected magnitude 8.0 mega-earthquake that struck on 12 May 2008, significantly changed land cover (LC), particularly affecting vegetation and rock cover. However, the long-term effects of LC changes on ecosystem services (ESs) remain unclear in earthquake-affected regions, especially across different spatial scales. This study, focusing on Wenchuan County, employs a multi-model framework that integrates fractional vegetation coverage (FVC), rock exposure rate (FR), and ecosystem services (ESs), combining correlation analysis, geographically weighted regression (GWR), Self-organizing map (SOM) clustering, and XGBoost-SHAP model, to analyze the spatiotemporal dynamics, interrelationships, and driving mechanisms of land cover (LC) and ESs before and after the earthquake. Results show that: (1) From 2000 to 2020, FVC and FR fluctuated markedly under earthquake influence, with slight declines in habitat quality (HQ) and carbon storage (CS) and notable improvements in soil conservation (SC) and water yield (WY). (2) With increasing elevation, the FVC–CS–SC group exhibited a downward trend and synergy, while the FR–HQ–WY group increased and also showed synergy; trade-offs and synergies became more pronounced at larger scales, displaying strong spatiotemporal heterogeneity. (3) Elevation (explaining 10–60% of variance) was the main driver for LC and ESs, with land use, slope, human activities, climate, and geological conditions significantly impacting individual indicators. At the same time, the existing geological hazard points are mainly concentrated along both sides of the river valleys, which may be associated with intensified human–land conflicts. These findings offer valuable insights into ecological restoration and sustainable development in earthquake-affected regions.

1. Introduction

Earthquakes directly alter the surface morphology and vegetation distribution through secondary disasters such as landslides and debris flows, which have profound impacts on the structure and function of ecosystems [1,2]. Taking the 2008 Wenchuan Earthquake as an example, the secondary disasters triggered by the earthquake destroyed approximately 1249.50 km² of vegetation, leading to an imbalance in the supply and demand of key ESs such as water conservation, soil preservation, and carbon storage, resulting in severe ecological degradation in the affected area [3,4]. In recent decades, climate change and human activities have significantly influenced the self-restoration process of ecosystems [5], making certain areas more vulnerable. As the carrier of terrestrial ecosystems, land type, pattern, and intensity of use often undergo significant changes after an earthquake, subsequently affecting the provision of ESs [6,7]. Therefore, to understand the long-term impact of earthquakes on ecosystems, there is an urgent need to strengthen research on the interactions between LC and ESs in earthquake-affected areas (abbreviations shown in

Table A4. Table of Abbreviations.

Full Form	Abbreviation	Full Form	Abbreviation
Land cover	LC	Ecosystem service	ES
Fractional vegetation coverage	FVC	Rock exposure rate	FR
Habitat quality	HQ	Carbon storage	CS
Soil conservation	SC	Water yield	WY
Normalized difference vegetation index	NDVI	Normalized difference rock index	NDRI
Geographically weighted regression	GWR	Self-organizing map	SOM
Extreme gradient boosting	XGBoost	Shapley additive explanations	SHAP

).

There is a close interaction between ESs and LC [8,9]. For example, forests usually enhance CS and SC, whereas urbanization typically degrades HQ [10,11,12]. Natural disturbances such as earthquakes can alter land cover and spatial configurations (e.g., causing forest fragmentation and grassland degradation) [13], thereby driving fluctuations in ES supply. FVC and FR, calculated via a pixel-based binary method, can quantitatively capture these changes and their feedback on LC [14,15,16,17,18]. Meanwhile, integration of the InVEST model with other models (e.g., the PLUS and FLUS model) has become a powerful tool for assessing ES under multiple scenarios, including the combined effects of climate and land use change [19,20]. However, its core assumptions of static, independent grid cells and linear, additive relationships limit its ability to represent dynamic feedback and non-linear, cross-scale interactions [21,22,23]. By integrating multi-temporal datasets into InVEST and coupling its outputs with FVC/FR by varied methods across multi-scale, it may overcome many of these assumption-based limitations and provide a robust, quantitative framework for evaluating ecosystem recovery following major disturbances.

Understanding the interactions between LC and ESs in earthquake-affected areas is crucial. The driving effects of factors such as elevation, slope, rainfall, lithology, and others on LC and ESs have been widely studied [14,24,25]. However, discussions on FVC, FR, and ESs in earthquake-affected areas are relatively isolated. Studies on vegetation primarily focus on spatiotemporal evolution and driving mechanisms, while research on rock exposure is more focused on geological disasters such as landslides and debris flows [1,2,17,18,26]. Additionally, earthquakes often occur in mountainous areas with complex topography, where factors like elevation and slope significantly influence vegetation distribution and rock exposure [27]. However, the impact of vegetation and rock exposure on ESs in mountainous areas has not been thoroughly studied [6,7]. Therefore, investigating the interactions between FVC, FR, and ESs before and after an earthquake is not only key to understanding the ecological effects triggered by earthquakes but also forms the scientific basis for optimizing ecological restoration strategies.

Trade-offs/synergies are often used to study the interactions between two or more ESs. A trade-off refers to the negative interaction between ESs, while a synergy refers to the positive interaction between ESs [28]. Combining correlation analysis with GWR can reveal the relative numerical relationships and spatial heterogeneity of ecosystem service trade-offs/synergies [29]. Meanwhile, SOM clustering can reflect both the spatial aggregation characteristics of different ESs and the trade-offs/synergies between multiple ESs [30]. Methods for identifying driving forces include spatial autocorrelation analysis, geographical detectors, multi-scale geographically weighted regression, and machine learning algorithms. Among these, machine learning algorithms combined with SHAP can more accurately and intuitively reflect the driving effects and are widely used in ES driving mechanism studies [31,32,33]. However, previous studies on ESs have mostly used coverage indicators (such as NDVI NDRI) as qualitative driving factors, which makes it difficult to provide spatial guidance for improving ESs. Furthermore, there has been insufficient attention to the spatial heterogeneity of mountainous areas, making it difficult to avoid scale effects [34]. Therefore, there is an urgent need to incorporate coverage indicators and ESs into the same analytical framework in post-earthquake ecological restoration processes and fully consider scale effects.

The Wenchuan Earthquake, one of the most destructive earthquakes in recent years, provides a unique case for studying the interaction between LC and ESs. Previous studies have revealed significant post-earthquake declines in vegetation coverage, increased bedrock exposure and soil erosion, as well as degradation of ecosystem service functions [4,13,17,18]. However, these studies have mostly focused on post-disaster assessments at a single time point, lacking a systematic analysis of the long-term dynamic relationship between LC and ESs. Therefore, this study takes Wenchuan County, the epicenter of the Wenchuan Earthquake, as the study area. Using remote sensing and the InVEST model, we quantified FVC, FR, and four ESs from 2000 to 2020. At both the grid and sub-watershed scales, we explore their interaction mechanisms and driving forces using correlation analysis, GWR, SOM clustering and XGBoost-SHAP model. The study also thoroughly examines the spatiotemporal distribution characteristics of different ecological indicators. The study aims to (1) analyze the spatiotemporal characteristics of FVC, FR, and ESs before and after the earthquake in Wenchuan County; (2) determine the interactions between LC and ESs before and after the earthquake; and (3) investigate the driving mechanisms of LC and ESs. This study provides scientific support for optimizing ecological restoration policies in earthquake-affected areas.

2. Materials and Methods

2.1. Study Area Overview

Wenchuan County is located on the eastern edge of the Tibetan Plateau. The terrain slopes from northwest to southeast, alternating between mountains and valleys, and is characterized by karst topography. The western part of the region, Mount Siguniang, has an elevation exceeding 6000 m, while the elevation of Xuan’kou Town at the mouth of the Min River in the southwest is only 780 m. The climate in the study area is predominantly temperate monsoon, gradually changing from southeast to northwest with the terrain. The distribution of water, light, and heat is uneven: the southeastern part receives abundant rainfall, while the northern Min River valleys are arid. The geological structure is complex, with the central fault and the rear mountain fault of the Longmenshan Fault Zone extending from northeast to southwest. The epicenter of the 2008 Wenchuan Earthquake was located in Yingxiu Town. The earthquake triggered landslides and debris flows due to the loosening of mountain slopes. Following the earthquake, large-scale ecological restoration projects were implemented to restore vegetation and improve habitat quality. The location is shown in Figure 1.

2.2. Data and Preprocessing

This study selected six ecological indicators, including two LC indicators and four ESs, covering multiple aspects of the ecological environment. The ecological indicators were selected based on the following principles: (1) the feedback on the disaster and ecological environment before and after the Wenchuan Earthquake; (2) the inclusion in the “General Plan for Post-Wenchuan Earthquake Reconstruction” and other ecological protection objectives. For the LC indicators, the study chose FVC to monitor the destruction and recovery of vegetation before and after the earthquake and FR to monitor the impact of earthquake-induced secondary disasters and ecological restoration on rock desertification. In terms of ESs, HQ can reflect fluctuations in the habitat quality of key species, such as the giant panda; CS and SC are crucial for mitigating forest damage and soil erosion caused by landslides, and debris flows; WY was selected because the Min River is an important tributary of the Yangtze River and a vital water source for downstream cities.

The acquisition and preprocessing of the data were conducted based on the principles of authenticity, accessibility, and general applicability. The data involved were categorized into two types: the first type was used for calculating the six ecological indicators, and the second type was used for calculating 11 driving factors, including elevation, slope, and soil types. To ensure comparability before and after the earthquake while considering data availability, this study selected 2000, 2005, 2010, 2015, and 2020 for analysis, using a five-year cycle. All raster data were resampled to a spatial resolution of 30 m and projected using the WGS_1984_UTM_Zone_48N coordinate system to ensure consistency in spatial accuracy across all datasets.

Table 1. Data source.

Data type	Format	Resolution	Data Source
Administrative boundaries	Shpfile	/	National Platform for Common Geospatial Information Services (www.tianditu.gov.cn)
Major road data	Shpfile	/	Open Street Map (www.openstreetmap.org)
Land use data	Raster	30 m	Resource and Environment Science and Data Center(www.resdc.cn)
DEM	Raster	30 m	Geospatial Data Cloud Platform (www.gscloud.cn)
Precipitation	Raster	1000 m	National Tibetan Plateau Science Data Center (https://data.tpdc.ac.cn)
Evapotranspiration	Raster	1000 m	National Tibetan Plateau Science Data Center (https://data.tpdc.ac.cn)
Soli type	Raster	250 m	Harmonized Worldwide Soil Data base: (HWSD) (http://www.fao.org)
Geological data	Shpfile	/	Geoscientific Data and Discovery Publishing System (http://dcc.ngac.org.cn)

lists the types, formats, resolutions, and sources of the primary data required for this study.

2.3. Methods

The analytical framework of this study consists of four parts: (1) Calculating the ecological indicators for the period 2000–2020 using remote sensing data and the InVEST model and analyzing the trend of changes; (2) Studying the trade-off and synergy relationships between ecological indicators at both the grid and sub-watershed scales using correlation analysis and GWR, revealing the spatial heterogeneity of interactions between ecological indicators; (3) Identifying ecological indicator bundles at both the grid and sub-watershed scales and exploring the spatial clustering characteristics of ecological indicators; (4) Analyzing the driving effects of ecological indicators based on the XGBoost-SHAP model, investigating the impact of the earthquake on ecological indicators. The technical route is shown in Figure 2.

2.3.1. Ecological Indicators Assessment

Remote sensing can extract land cover information through different band combinations and is widely used in ecological assessments and geological environmental monitoring [15,35]. The NDVI and NDRI for the study area were calculated as the average values using Landsat satellite data from the Google Earth Engine platform. Before calculation, cloud cover and the impact of vegetation growth stages on data quality were excluded, and missing data were filled in using data from adjacent years to improve accuracy [16]. Furthermore, NDVI and NDRI were used to generate FVC and FR through the pixel-based binary method (as shown in the following formulas:). At the same time, the accuracy and reliability of FVC/FR calculations were ensured through methods such as visual inspection, cross-validation, and temporal consistency analysis.

$$\mathrm{F}\mathrm{V}\mathrm{C}={\displaystyle \frac{\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}-{\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}}_{\mathrm{m}\mathrm{i}\mathrm{n}}}{{\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}}_{\mathrm{m}\mathrm{a}\mathrm{x}}-{\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}}_{\mathrm{m}\mathrm{i}\mathrm{n}}}}$$

(1)

NDVI: $\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}$ of the current pixel; ${\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}}_{\mathrm{m}\mathrm{a}\mathrm{x}}$ and ${\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}}_{\mathrm{m}\mathrm{i}\mathrm{n}}$ are the maximum and minimum $\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}$ in the study area.

$$\mathrm{F}\mathrm{R}={\displaystyle \frac{\mathrm{N}\mathrm{D}\mathrm{R}\mathrm{I}-{\mathrm{N}\mathrm{D}\mathrm{R}\mathrm{I}}_{\mathrm{m}\mathrm{i}\mathrm{n}}}{{\mathrm{N}\mathrm{D}\mathrm{R}\mathrm{I}}_{\mathrm{m}\mathrm{a}\mathrm{x}}-{\mathrm{N}\mathrm{D}\mathrm{R}\mathrm{I}}_{\mathrm{m}\mathrm{i}\mathrm{n}}}}$$

(2)

NDRI: $\mathrm{N}\mathrm{D}\mathrm{R}\mathrm{I}$ of the current pixel; ${\mathrm{N}\mathrm{D}\mathrm{R}\mathrm{I}}_{\mathrm{m}\mathrm{a}\mathrm{x}}$ and ${\mathrm{N}\mathrm{D}\mathrm{R}\mathrm{I}}_{\mathrm{m}\mathrm{i}\mathrm{n}}$ are the maximum and minimum $\mathrm{N}\mathrm{D}\mathrm{R}\mathrm{I}$ in the study area.

The ecosystem services and Trade-off Comprehensive Assessment (InVEST) Model, which can quantify and visualize multiple ESs and has been scientifically validated [36], was used in this study to assess HQ, CS, SC, and WY. The formulas are shown in Appendix A,

Table A1. Model formulas.

Type		Formula
Coverage Indicators	NDVI	$\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}={\displaystyle \frac{\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{R}\mathrm{e}\mathrm{d}}{\mathrm{N}\mathrm{I}\mathrm{R}+\mathrm{R}\mathrm{e}\mathrm{d}}}$
		Where $\mathrm{R}\mathrm{e}\mathrm{d}$ and $\mathrm{N}\mathrm{I}\mathrm{R}$ are the red band and the near-infrared band of remote sensing images (Landsat TM/OLI), respectively.
	NDRI	$NDRI={\displaystyle \frac{SWIR1-NIR}{SWIR1+NIR}}$
		where $SWIR1$ and $NIR$ are the short-wave infrared band-1 and the near-infrared band of remote sensing images (Landsat TM/OLI), respectively.
Ecosystem Service	Habitat quality	$Q_{xj}=H_j\left[1-\left({\displaystyle \frac{D_{xj}^z}{D_{xj}^z+k^z}}\right)\right]$
		Here, Qxj represents the habitat quality of a grid within habitat type j, where Dxj indicates the disturbance level experienced by grid $x$ in habitat type $j$. The constant $k$ is the half-saturation value, typically set at half of the maximum value obtained from $Q_{xj}$ during a preliminary trial, and $H$, denotes the suitability of habitat type $j$.
	Carbon storage	$CS=C_{\mathrm{above}}+C_{\mathrm{below}}+C_{\mathrm{dead}}+C_{\mathrm{soil}}$
		In this equation, $C_{\mathrm{above}}$ denotes above-ground organic matter, $C_{\mathrm{below}}$ denotes below-ground organic matter, $C_{\mathrm{dead}}$ represents dead organic matter, and $C_{\mathrm{soil}}$ represents soil organic matter.
Ecosystem Service	Soil conservation	$SC=S{C}_{pot}-S{C}_{act}$ $=R\times K\times LS\times \left(1-C\times P\right)$
		In this equation, $SC$ represents soil conservation capacity, while $S{C}_{pot}$, $S{C}_{act}$ denote potential and actual soil erosion, respectively. $R$, $K$, $LS$, $C$, and $P$ represent rainfall erosivity, soil erodibility, slope length, vegetation cover management, and soil conservation practices.
	Water yield	$Y_x=\left(1-{\displaystyle \frac{AE{T}_x}{P_Y}}\right)\times P_x$
		In this equation, $Y_x$ represents the water yield for pixel $x$; $P_x$: the annual precipitation for pixel $x$; and $AE{T}_x$ represents the annual evapotranspiration for pixel $x$.
Spearman’s correlation analysis		$R_{XY}=1-{\displaystyle \frac{6\sum _{i=1}^n{\left(X_i-Y_i\right)}^2}{n\left(n^2-1\right)}}$
		where $R_{XY}$ is the correlation coefficient between $X$ and $Y$; $X_i$ and $Y_i$ are the $i$ sample values of $X$ and $Y$, respectively; and $n$ is the number of samples.
Geographically Weighted Regression (GWR)		${\gamma }_i={\beta }_0\left(u_i,v_i\right)+{\displaystyle \sum _{k=1}^m}{\beta }_k\left(u_i,v_i\right)x_{ik}+{\mathsf{\epsilon }}_i$
		where ${\gamma }_i$ is the sample values at position $i$, $x_{ik}$ $(k=1,2,\dots .,m)$ is the remained sample values at position $i$; $\left(u_i,v_i\right)$ are the spatial coordinates of sampling point $i$; ${\beta }_0\left(u_i,v_i\right)$ is the intercept term; ${\beta }_k\left(u_i,v_i\right)$ is the regression coefficient and ${\mathsf{\epsilon }}_i$ is the error term.
Extreme Gradient Boosting model (XGBoost model)		${\mathrm{O}\mathrm{b}\mathrm{j}}^{\left(t\right)}\approx {\displaystyle \sum _{i=1}^n}\left[L\left(y_i,{\widehat{y}}_i^{\left(t-1\right)}\right)+g_i{f}_t\left(x_i\right)+{\displaystyle \frac{1}{2}}{h}_i{f}_t{\left(x_i\right)}^2\right]+{\displaystyle \sum _{k=1}^t}\Omega \left(f_k\right)$
		where $L\left(y_i,{\widehat{y}}_i\right)$ represents the loss function,$\Omega \left(f_k\right)$ indicates the regularization term that suppresses the overfitting of the base learner at each iteration, and it is not involved in the integration of the final model. $g_i$ corresponds to the first-order derivative of the loss function, $h_i$ refers to the second-order derderivative of the loss function, and $t$ denotes the iteration number.
Shapley Additive Explanations (SHAP)		${\varphi }_i={\displaystyle \sum _{S\subseteq N\setminus i}} {\displaystyle \frac{\left|S\right|!\times \left(\left|N\right|-\left|S\right|-1\right)!}{\left|N\right|!}}\times \left[f\left(S\cup i\right)-f\left(S\right)\right]$
		where ${\varphi }_i$ is the SHAP value for a specific feature $i$, indicating its influence on the model’s prediction. This value is computed by evaluating the contributions of feature $i$ across all potential subsets $S$ of the other features $N$ (excluding $i$).

. Additionally, based on the geomorphological characteristics of the study area, the mean values of ecological indicators were calculated at different elevations and watersheds to reveal their spatiotemporal distribution patterns. Further studies were conducted at the 1000-m grid scale and sub-watershed scale to explore scale effects.

2.3.2. Trade-Off/Synergy Analysis of Ecological Indicators

Correlation Analysis

Spearman’s correlation analysis is a non-parametric method that is robust to noise, supports multiple scales and data types, and is easy to interpret. It is an effective tool for quantitatively assessing the relationships between ESs [37]. In this study, the “corrplot” package in R (version 4.4.1) was used to quantitatively analyze the trade-off and synergy relationships of the six ecological indicators for the years 2000, 2005, 2010, 2015, and 2020, and significance testing was performed. The absolute value of the correlation coefficient indicates the strength of the trade-off or synergy relationship: negative values indicate trade-offs, while positive values indicate synergies (Appendix A Table A1).

Geographically Weighted Regression (GWR)

The trade-offs and synergies between ecological indicators exhibit significant spatial heterogeneity [38]. GWR is capable of revealing this spatial heterogeneity and is easily visualized, making it an important complement to Spearman’s correlation analysis [39]. In this study, the “GWR” package in R was used to analyze and visualize the trade-off and synergy relationships between the six ecological indicators in pairs.

2.3.3. Identification of Ecological Indicator Bundles

Ecological indicator bundles represent the aggregation of multiple ecological indicators within a specific spatiotemporal range, revealing the spatial associations and distribution patterns between different ecological indicators. SOM performs unsupervised clustering of multiple ecological indicators and can identify bundles of different spatial units within a region, providing strong objectivity and accuracy [40]. In this study, SOM clustering was performed using the “kohonen” package in R, and the area change trends of different bundles was analyzed. The optimal number of bundles was dynamically adjusted based on the spatial distribution’s rationality and readability at different scales, with the optimal number of bundles determined to be five.

2.3.4. Driving Force Analysis

The combination of Machine Learning and SHAP has been widely used in ecosystem service driving force analysis. The XGBoost model captures the complex relationships in non-linear, high-dimensional data, while SHAP, based on game theory principles, identifies the contributions and influence directions of different factors on ecological indicators [41]. The combination of both models allows for the identification of key driving factors of ecological indicators and provides a clear visualization of the importance and mechanisms of different driving factors. Positive SHAP values represent driving effects, while negative SHAP values represent suppressive effects.

3. Results

3.1. Spatiotemporal Distribution Characteristics

3.1.1. Spatial Distribution Characteristics

At different elevations, all ecological indicators exhibited a clear “gradient effect”. FVC, CS, and SC gradually decrease with increasing elevation, while FR, HQ, and WY gradually increase with elevation. Across the watersheds, FVC, CS, and SC also decrease with elevation. The values are higher in watersheds E to G and lower in the higher elevation watersheds C and D, showing a “high in the east, low in the west” distribution pattern. In contrast, FR, HQ, and WY increase with elevation. They have lower values in watersheds E to G and higher values in watersheds C and D, showing a “high in the west, low in the east” distribution pattern (Figure 3 and Figure 4).

3.1.2. Temporal Variation Characteristics

Affected by the Wenchuan Earthquake, FVC showed a clear “V” shape fluctuation in different elevations and watersheds, with 2010 as the turning point. In contrast, FR exhibited an inverted “V” shape fluctuation, both indicating a distinct “recovery period”. After the earthquake, FVC in the medium and low elevation areas decreased by more than 0.05, while FR increased by more than 0.05. Among the watersheds, the most significant changes occurred in watersheds F and G, where FVC decreased by more than 0.15 and FR increased by more than 0.15. Overall, vegetation and rock exposure in most areas had recovered to pre-earthquake levels by 2015. However, in areas more severely affected by the earthquake, recovery was slower, such as in watersheds F and G. At the same time, HQ showed a slight decline after the earthquake, and CS also decreased to some extent in the medium and low elevation areas. During the period from 2000 to 2020, SC and WY exhibited a fluctuating upward trend with similar trends. More details are shown in Figure 3 and Figure 4.

3.2. Trade-Offs and Synergies Between Ecological Indicators

3.2.1. Results of Correlation Analysis

The results show that most of the results passed the significance test at the grid scale, while at the sub-watershed scale, most results for 2000 and 2005 passed the significance test. However, the number of significant results for 2010, 2015, and 2020 decreased, indicating that the ecological impacts of the earthquake on the sub-watersheds have not fully recovered.

According to the trade-off and synergy effects, the six ecological indicators were divided into two groups. The first group included FVC, SC, and CS, and the second group included FR, HQ, and WY. Ecological indicators within the same group showed synergy effects, while ecological indicators between different groups exhibited trade-off effects. Specifically, at the grid scale, FVC-FR showed a strong trade-off relationship, while in other combinations, FVC-HQ and FVC-WY show trade-off relationships, and FR-HQ and FR-WY show synergy relationships. FR with CS and SC shows trade-off relationships, while FVC with CS and SC shows synergy relationships. HQ-WY and CS-SC exhibit synergy relationships. Affected by the earthquake, the absolute value of the correlation coefficient between 2000 and 2020 decreased and then increased with 2010 as the turning point, as shown in Figure 5a. At the sub-watershed scale, the trade-off and synergy relationships of the same ecological indicator combinations are similar to those at the grid scale, but the absolute values of the correlation coefficients are higher, indicating a certain scale effect, as shown in Figure 5b.

3.2.2. Results of GWR

The trade-offs or synergies between ecological indicators gradually strengthen as the scale increases from the grid scale to the sub-watershed scale. For example, the synergy between HQ-WY at the sub-watershed scale is stronger than the grid scale. At the same time, the scale effect amplifies the impact of geomorphological features on the trade-off and synergy effects. At the grid scale, the spatial distribution of trade-off and synergy relationships is fragmented, and the spatial distribution pattern is more complex, revealing more details. In contrast, at the sub-watershed scale, the influence of elevation is more prominent.

At the grid scale, the proportion of trade-off areas was far greater than the proportion of synergy areas for FVC-FR, FVC-WY, FR-CS, HQ-CS, CS-WY, and SC-WY, indicating that these ecological indicators primarily exhibited trade-off relationships in their spatial distribution. For FVC-CS, FR-WY, and HQ-WY, the proportion of synergy areas was much greater than the proportion of trade-off areas, indicating that these ecological indicators primarily showed synergy relationships in their spatial distribution. For FVC-HQ, FVC-SC, FR-HQ, FR-SC, HQ-SC, and CS-SC, the area proportions of trade-off and synergy relationships were relatively similar, but the spatial distribution of these indicator combinations was more influenced by geomorphological features compared to combinations with larger proportions of trade-off or synergy areas. This influence was more evident in valleys and high-altitude areas (Figure 6).

At the sub-watershed scale, except for FVC-FR, CS-WY, and SC-WY, which showed strong trade-off effects, the trade-off and synergy effects of the other ecological indicator combinations were influenced by elevation. As the elevation increased, the synergy between FVC-HQ, FVC-WY, FR-CS, FR-SC, HQ-CS, and HQ-WY weakened from the northeastern to the southwestern part of the study area, while the trade-off effect strengthened. Conversely, for FVC-CS, FVC-SC, FR-HQ, FR-WY, HQ-SC, and CS-SC, the trade-off effect weakened, and the synergy effect strengthened as elevation increased from the northeastern to the southwestern part of the study area (Figure 7).

3.3. Identification of Ecological Indicator Bundles

This study identified five ecological indicator bundles at both the grid and sub-watershed scales using the SOM algorithm and calculated the dynamic changes in the area of the bundles at different scales. At the grid scale, the spatial distribution of each bundle was fragmented and influenced by multiple factors. At the sub-watershed scale, the spatial distribution of the bundles was influenced by elevation, and the results were cross-validated with the GWR results.

At the grid scale, Bundle 1 was dominated by the synergy between FR-HQ-WY and was mainly distributed in the high-altitude areas at the edges of the study area. Its area fluctuated little and occupied 15% to 20% of the total study area. Bundle 2, influenced by human activity, was characterized by the trade-off between FVC HQ and WY. It was mainly found in the transitional zones between forests and alpine meadows along the valleys, occupying less than 5% of the study area. Bundle 3 was dominated by the synergy between FVC-SC-CS-HQ, while Bundle 4 was dominated by the synergy between FVC-CS-HQ. Bundles 3 and 4 were distributed in areas with denser vegetation in the study area. In years with higher SC, the area proportion of Bundle 3 was larger, and in years with lower SC, Bundle 4 occupied a larger proportion. As a result, the area of Bundles 3 and 4 fluctuated greatly in different years, but the combined area of Bundle 3 and Bundle 4 remained relatively stable, occupying about 60% of the study area. Bundle 5 was a synergy bundle dominated by HQ, mainly distributed in valley areas with less human disturbance. Its area remained relatively stable, occupying about 20% of the study area (Figure 8).

At the sub-watershed scale, the spatial distribution characteristics and area of the bundles were more stable. Except for Bundle 5, which was a synergy bundle, the other bundles were related to the spatial distribution of FVC or FR. Bundle 2 was concentrated in sub-watersheds with higher FR in alpine meadows, dominated by the synergy between FR-HQ-SC-WY, occupying about 20%. Bundles 1, 3, and 4 were distributed in low-elevation areas with denser vegetation, with their total areas ranging from 50% to 55%. Bundle 5, a synergy bundle dominated by SC, was distributed in high-altitude sub-watersheds with denser vegetation, with its area ranging from 20% to 25% (Figure 9).

3.4. Results of Driving Force Analysis

This study used the XGBoost-SHAP model to analyze the relative importance and explainability of 11 driving factors for six ecological indicators at both the grid scale and sub-watershed scale from 2000 to 2020. The results showed that at the grid scale, the main driving factors of different ecological indicators varied significantly, revealing more details, while at the sub-watershed scale, elevation was the primary driving factor for the ecological indicators.

The performance of the driving force analysis model was evaluated using R², MSE, and MAE. At the grid scale, all evaluation metrics were within a reasonable range, indicating strong robustness and fitting ability of the model (

Table 2. Average values of model performance evaluation metrics at the grid scale.

	FVC	FR	HQ	CS	SC	WY
R2	0.95	0.92	0.75	0.81	0.89	0.96
MAE	0.04	0.05	0.04	0.04	0.03	0.04
RMSE	0.06	0.07	0.05	0.06	0.04	0.05

and

Table A2. Grid scale.

	2000			2005			2010			2015			2020
	R2	MAE	RMSE	R2	MAE	RMSE	R2	MAE	RMSE	R2	MAE	RMSE	R2	MAE	RMSE
FVC	0.96	0.04	0.05	0.96	0.03	0.05	0.92	0.05	0.06	0.94	0.04	0.06	0.94	0.04	0.06
FR	0.94	0.05	0.07	0.94	0.04	0.06	0.89	0.05	0.08	0.91	0.05	0.07	0.91	0.05	0.07
HQ	0.69	0.04	0.05	0.69	0.04	0.05	0.78	0.04	0.05	0.78	0.04	0.05	0.78	0.03	0.05
CS	0.82	0.03	0.05	0.82	0.03	0.05	0.81	0.04	0.06	0.80	0.04	0.06	0.81	0.04	0.06
SC	0.89	0.03	0.04	0.89	0.03	0.04	0.88	0.03	0.04	0.90	0.02	0.03	0.89	0.03	0.04
WY	0.96	0.04	0.05	0.96	0.04	0.05	0.95	0.04	0.05	0.95	0.04	0.05	0.96	0.04	0.05

). However, at the sub-watershed scale, limited by the sample size, the performance evaluation metrics indicated that some ecological indicators had a poor fit with the XGBoost-SHAP model (

Table A3. Sub-watershed scale.

	2000			2005			2010			2015			2020
	R2	MAE	RMSE	R2	MAE	RMSE	R2	MAE	RMSE	R2	MAE	RMSE	R2	MAE	RMSE
FVC	0.8	0.14	0.16	0.7	0.14	0.18	×	0.16	0.21	0.69	0.14	0.18	0.76	0.13	0.17
FR	0.77	0.18	0.19	0.7	0.15	0.19	0.52	0.18	0.21	0.65	0.14	0.18	0.61	0.17	0.20
HQ	×	0.13	0.19	×	0.15	0.21	×	0.14	0.19	×	0.15	0.19	×	0.14	0.18
CS	×	0.25	0.28	×	0.26	0.3	×	0.28	0.34	×	0.28	0.35	×	0.26	0.33
SC	×	0.15	0.27	×	0.15	0.26	×	0.15	0.27	×	0.16	0.28	×	0.15	0.28
WY	0.85	0.11	0.14	0.82	0.12	0.15	0.83	0.13	0.14	0.81	0.13	0.15	0.84	0.12	0.13

Note: × indicates R² < 0.5.

). Therefore, after synthesizing the results from both the grid scale and sub-watershed scale, it was found that elevation was the main driving force for FVC and FR (explainability > 40%). FVC was negatively correlated with elevation, while FR was positively correlated with elevation. Additionally, due to the earthquake in 2010, lithology strengthened the driving effect on FVC and FR (explainability increased by >5%), indicating significant differences in vegetation damage and rock exposure across different lithologies caused by the earthquake. HQ was mainly driven by land use types (explainability > 20%) and was related to the distance from major roads and elevation (explainability > 15%). Higher HQ values were observed in areas with dense vegetation, such as forests, while human activities along major roads had a suppressive effect on HQ. CS was mainly driven by land use types (explainability > 40%) and was influenced by elevation and evapotranspiration (explainability > 10%). Vegetation provided higher carbon storage, and as elevation increased, vegetation decreased, leading to lower CS. SC was mainly driven by slope (explainability > 35%) and was also related to rainfall, elevation, and soil types (explainability > 10%). Areas with steeper slopes had stronger soil conservation capabilities, and rainfall promoted SC, while different soil types had a significant impact on soil conservation. WY was driven by evapotranspiration (explainability > 25%), elevation, rainfall, and land use types. More details are shown in Figure 10 and Figure 11.

4. Discussion

4.1. Spatiotemporal Evolution of LC and ESs

The results indicate that the spatial pattern of LC in the study area was significantly affected by the Wenchuan Earthquake, but it was still primarily dominated by elevation. The continuous, gradual changes in precipitation, temperature, and evapotranspiration that shift with increasing elevation, which in turn drive continuous, progressive shifts in the spatial distribution of vegetation cover and rock exposure, showing the “gradient effect”. The large-scale landslides and collapses post-earthquake were the main reasons for the sharp decline in vegetation coverage, and the spatial distribution of landslides and collapses was closely related to factors such as elevation and lithology. This is consistent with previous studies [3,42,43]. However, unlike previous studies, this study found that the rate of vegetation recovery varied significantly under different terrain conditions. Vegetation in high-altitude areas was less affected by the earthquake and recovered more quickly, while recovery in low-altitude and steep-slope areas was slower. This may be related to vegetation types, soil erosion, and the spatial distribution of secondary disasters [44,45]. Moreover, FR significantly increased post-earthquake, especially in watersheds F and G, where landslides were more concentrated. This was consistent with the decline in vegetation coverage. However, vegetation recovery and rock exposure recovery did not occur synchronously. By 2020, the FVC in watersheds F and G had recovered to pre-earthquake levels, but the FR remained higher than before the earthquake. This may be related to extreme weather events that triggered landslides and debris flows again in areas with high landslide density post-earthquake [46,47].

It is important to note that elevation has a dominant effect on the spatial distribution of ESs [10,48], and the impact of the earthquake on ESs is relatively small [49]. CS and HQ showed a decrease in the low-elevation areas along the valley sides, which may be associated with land exposure caused by landslides and increased human-environment conflicts due to disaster reconstruction [50]. In contrast, SC and WY showed relatively stable spatial distributions between 2000 and 2020, with fluctuations upwards under the influence of climate change and limited impact from the earthquake. The stable supply of ESs may be closely related to ecological restoration efforts. For example, vegetation in most areas recovered to pre-earthquake levels within 10 years, reducing the long-term damage to ESs caused by secondary disasters, surface runoff, and sediment transport [18,51].

4.2. Interactions Between LC and ESs

LC has a constraining effect on ESs. Vegetation growth typically promotes ESs, while increased land exposure often triggers ecological risks [9,52]. Previous studies have shown that in mountainous ecosystems, FVC and the provision of ESs follow a “low at both ends, high in the middle” unimodal pattern with elevation changes. Low and high-elevation areas are limited by human activities or natural conditions, with sparse vegetation and exposed land, while mid-elevation areas, with appropriate moisture and soil conditions, become the core regions for ecosystem service provision [53,54,55,56]. Considering the spatial distribution differences in the river valleys of the study area, this pattern aligns with the findings of this study: FVC, SC, and CS showed a synergistic relationship and gradually decreased with elevation; FR and WY showed a synergistic relationship and gradually increase with elevation. In contrast to other studies [57], this study found that HQ and FVC exhibited a trade-off relationship, but there were regional differences influenced by multiple factors, such as the distribution of vegetation in mountain ecosystems, human activities, and regional protection policies [58]. In low-elevation river valleys with dense vegetation, conflicts between humans and land have become more prominent due to infrastructure development and human activities. National Highway 213 and 317, as well as the Rongchang Expressway, have led to a significant decline in HQ along the sides of the river valleys. Meanwhile, high-altitude regions in the western part of the study area, including the Wolong Nature Reserve and the Caopo Nature Reserve, have effective management measures within the protected areas that may have improved HQ [59].

Additionally, the trade-offs and synergies between ecological indicators show significant heterogeneity across different regions and times [57]. Under the influence of the earthquake, land fragmentation may trigger differences in the fluctuations of LC and ESs, thus altering the original trade-offs and synergies between ecological indicators. This led to a decrease and then an increase in the absolute value of the correlation coefficient. Meanwhile, Existing studies indicate that trade-offs/synergies are more pronounced at larger scales, consistent with the findings of this study [60]. This phenomenon may result from accumulated local differences enhancing spatial heterogeneity at larger scales and reduction of local random disturbances, allowing overall trade-offs/synergies to be more systematically revealed. Furthermore, in areas with frequent secondary disasters after the earthquake, there are also significant spatial differences in the trade-offs and synergies between ecological indicators. For example, in areas with landslide debris accumulation, SC and CS showed a significant trade-off relationship, while in the areas where land was converted back to forest, CS and WY showed a trade-off relationship.

4.3. Analysis of Driving Mechanisms

The spatial distribution and dynamic changes of LC and ESs are complex processes driven by both natural and human factors [39,40]. Similar to the scale effects in trade-off and synergy relationships, data aggregation at different spatial scales also influences the analysis of driving mechanisms. At the larger watershed scale, elevation emerges as the most significant driver, whereas at the finer grid scale, a more diverse array of driving factors is apparent. Among these, elevation is a key factor that significantly influences climate (e.g., rainfall, evapotranspiration), topography, and human activities, directly or indirectly dominating LC and Ess [56]. In low-elevation river valleys, the climate and topography are favorable for agricultural development and urban expansion, leading to high land use intensity and prominent human-land conflicts. This results in vegetation destruction, land degradation, and habitat fragmentation, with relatively low HQ and CS in the region [10]. In the mid-elevation areas, as soil and water conditions improve and human activities decrease, a land use pattern dominated by forests and supplemented by agriculture and livestock farming is formed. FVC increases, and CS and SC are higher in the region [48]. In high-elevation areas, harsh natural conditions (e.g., low temperatures and steep slopes) limit human activities. In addition, the long-term influence of gravity has led to widespread alpine meadows, barren lands, and glaciers, with rock exposure, sparse vegetation, and disordered runoff, resulting in high WY [61].

LC and ESs are not only dominated by elevation but are also influenced by factors such as lithology, climate, and human activities [6]. For example, After the Wenchuan earthquake, the explanatory power of lithology increased by more than 5% for FVC and FR in 2010; this may be attributable to fluctuations in FVC and FR were more pronounced in areas with exposed igneous rocks. The influence of lithology on FVC and FR significantly increased, but its effect gradually weakened over time while the impact of elevation continued. At the same time, geological hazards are more densely distributed in valley areas with higher human activity intensity, which may also contribute to lower FVC and higher FR at lower elevations. Climate factors (e.g., precipitation and evapotranspiration) significantly affect SC and WY, but their spatial distribution patterns are ultimately constrained by the elevation gradient [10]. Human activities, especially land use policies and post-earthquake ecological restoration projects, have improved local LC and ESs to some extent. However, their effects exhibited significant spatial heterogeneity and are generally weaker than the gradient effects dominated by elevation. In low-elevation gentle slope areas, human activities have promoted vegetation recovery and enhanced ESs, while in high-elevation and landslide-prone areas, the effects of policies are limited due to terrain constraints and ecological vulnerability.

4.4. Significance of the Study and Policy Recommendations

Accurately identifying the spatial distribution and driving factors of ESs is crucial for formulating appropriate ecological protection strategies [62,63]. SOM could identify the spatial aggregation characteristics of multiple ESs [29,30], and afforestation and desertification control are important methods for managing Ess [64,65]. Therefore, this study performed a clustering analysis of FVC and FR with ESs through SOM, providing more direct methods and spatial guidance for ecosystem service management. Finally, this study proposed zonal management strategies to enhance the resilience of ecosystems in responding to geological disasters, soil erosion, and climate change based on the different bundles at the grid scale in 2020. The specific policy recommendations are as follows:

Bundle 1: The FR-HQ-WY synergy bundle is mainly distributed in the high-altitude regions at the edges of the study area. The region has widespread alpine meadows and barren land, with a relatively fragile ecological environment. This area requires enhanced ecological monitoring, regulation of tourism activities at Mount Balang and Mount Siguniang, and prevention of water and soil erosion risks caused by the degradation of alpine meadows.

Bundle 2: The FVC-FR and WY trade-off bundle is mainly found in the transition zones between alpine meadows and forests, as well as in river valleys disturbed by human activities. This area requires enhanced monitoring of tree growth, forest protection to prevent deforestation, and regulation of grazing activities to reduce land desertification risks.

Bundle 3 is the FVC-CS-SC synergy bundle, corresponding to areas with strong ecosystem service provision capacities but also areas with significant vegetation destruction and slope instability. These regions are characterized by dense vegetation and steep slopes. This area needs appropriate land use restrictions to promote natural vegetation recovery, strengthen slope management, and enhance soil conservation and carbon storage capabilities in the region.

Bundle 4 is the FVC-CS synergy bundle, also an area with strong ecosystem service provision capabilities, where vegetation is dense but has been severely impacted by the Wenchuan Earthquake. This region has relatively gentle slopes, and vegetation recovery can be promoted through human interventions to maintain carbon storage capabilities.

Bundle 5 is the HQ-dominated synergy bundle, mainly distributed in river valley areas with minimal human activity. This area needs to avoid uncontrolled human expansion and strengthen the regulation of infrastructure development.

Furthermore, drawing on management experience from Balang Mountain and Siguniang Mountain nature reserves and lessons learned from debris-flow control in Banzigou, this study proposes an implementation pathway that prioritizes natural restoration, supplements it with targeted intervention, and emphasizes prevention over remediation. It underscores the necessity of strictly limiting development intensity in steep watersheds with extensive igneous-rock exposures and regulating mountain tourism to reduce human-environment conflicts. Furthermore, we recommend establishing tiered management standards and ecological compensation mechanisms with clearly defined responsibilities at each governance level to effectively alleviate human-environment tensions and maximize ecological recovery outcomes.

4.5. Limitations and Prospects

In this study, we treated post-earthquake ecological restoration and climate change impacts on ecosystem services as separate factors; however, in real ecosystems, these drivers interact and cannot be fully disentangled. Therefore, future research should explicitly examine the coupled effects of climate change and human interventions to reveal their synergistic and offsetting effects. Moreover, the datasets used here differ markedly in spatial resolution (e.g., 30 m for land use vs. 1000 m for precipitation), which may introduce inconsistencies in spatial precision. Subsequent work should systematically evaluate how data resolution influences the robustness of ecosystem service assessments.

It is also important to recognize that land use conflicts and ecological risks are highly spatially coupled, particularly in rapidly urbanizing areas or ecologically vulnerable where zones of intense conflict frequently coincide with areas of heightened ecological vulnerability, resource competition, and environmental degradation [19,21]. Although the InVEST model provides valuable insights into dynamic changes in ESs, its core assumptions of static, independent grid cells and linear responses limit its ability to capture socio-ecological feedback and land use tensions. At the same time, remote sensing monitoring still entails certain uncertainties. Future research should ensure the validity of InVEST model parameters and the accuracy of remote sensing monitoring through field surveys and other methods. Furthermore, integrating higher temporal and spatial resolution socio-economic and policy datasets (e.g., The “Grain-for-Green” program and post-earthquake ecological restoration projects) will enable quantitative assessment of the contributions of anthropogenic interventions to vegetation recovery, thereby fully elucidating the coupled mechanisms between natural and human drivers in ecosystem recovery following major disturbances.

5. Conclusions

This study quantitatively assessed the spatiotemporal distribution of FVC), FR, and four typical ESs in Wenchuan County from 2000 to 2020. It revealed the interactions, driving mechanisms, and scale effects of LC and ESs before and after the earthquake, providing the scientific basis and spatial guidance for ecosystem service management in earthquake-affected areas. The main conclusions are as follows:

(1)

The spatial distribution of ecological indicators exhibited a significant “gradient effect”: FVC, CS, and SC decreased with increasing elevation, showing a “high in the east, low in the west” distribution pattern; FR, HQ, and WY increased with increasing elevation, showing a “high in the west, low in the east” distribution pattern.

(2)

The earthquake had a significant but periodic impact on LC. Post-earthquake, FVC showed a “V”-shaped fluctuation, while FR showed an inverted “V”-shaped fluctuation. However, in most areas, vegetation and rock exposure had recovered to pre-earthquake levels by 2015, indicating a clear “recovery period”. In contrast, ESs were less affected by the earthquake. HQ and CS decreased slightly after the earthquake, while SC and WY showed an upward trend driven by climate change from 2000 to 2020.

(3)

Ecological indicators could be divided into two groups based on trade-off and synergy relationships. One group consists of FVC, CS, and SC, while the other group consists of FR, HQ, and WY. Ecological indicators within the same group exhibited synergy, while those between different groups exhibited trade-offs. The trade-off and synergy relationships of ecological indicators were similar across different scales but were more pronounced at larger scales and showed strong spatial heterogeneity.

(4)

Ecological indicators were driven by both natural and human factors, with elevation being a key driving factor. Elevation (explainability > 45%) was the primary driver of FVC and FR, and the earthquake temporarily enhanced the driving effect of lithology. The spatial distribution and variation of ESs are driven by multiple factors, including topography, climate, and human activities, all of which are closely related to the elevation gradient.

The terrain of the study area is complex, and with the intensification of extreme weather events and regional economic development, ESs may be further weakened, especially in river valleys and high-altitude areas. Therefore, formulating scientific ecological restoration strategies, promoting vegetation recovery, and combating desertification while balancing economic development and ecological protection are crucial to ensuring the rapid recovery and sustainable development of ecosystems in earthquake-affected areas. This study provides scientific support for post-earthquake ecological restoration and promotes the sustainable use and protection of ESs in the affected region.