Evaluating Ecological Vulnerability and Its Driving Mechanisms in the Dongting Lake Region from a Multi-Method Integrated Perspective: Based on Geodetector and Explainable Machine Learning
This study focuses on the Dongting Lake region in China and evaluates ecological vulnerability using the Sensitivity–Resilience–Pressure (SRP) framework, integrated with Spatial Principal Component Analysis (SPCA) to calculate the Ecological Vulnerability Index (EVI). The EVI values were classified...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-07-01
|
| Series: | Land |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2073-445X/14/7/1435 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study focuses on the Dongting Lake region in China and evaluates ecological vulnerability using the Sensitivity–Resilience–Pressure (SRP) framework, integrated with Spatial Principal Component Analysis (SPCA) to calculate the Ecological Vulnerability Index (EVI). The EVI values were classified into five levels using the Natural Breaks (Jenks) method, and spatial autocorrelation analysis was applied to reveal spatial differentiation patterns. The Geodetector model was used to analyze the driving mechanisms of natural and socioeconomic factors on EVI, identifying key influencing variables. Furthermore, the LightGBM algorithm was used for feature optimization, followed by the construction of six machine learning models—Multilayer Perceptron (MLP), Extremely Randomized Trees (ET), Decision Tree (DT), Random Forest (RF), LightGBM, and K-Nearest Neighbors (KNN)—to conduct multi-class classification of ecological vulnerability. Model performance was assessed using ROC–AUC, accuracy, recall, confusion matrix, and Kappa coefficient, and the best-performing model was interpreted using SHAP (SHapley Additive exPlanations). The results indicate that: ① ecological vulnerability increased progressively from the core wetlands and riparian corridors to the transitional zones in the surrounding hills and mountains; ② a significant spatial clustering of ecological vulnerability was observed, with a Moran’s I index of 0.78; ③ Geodetector analysis identified the interaction between NPP (<i>q</i> = 0.329) and precipitation (PRE, <i>q</i> = 0.268) as the dominant factor (<i>q</i> = 0.50) influencing spatial variation of EVI; ④ the Random Forest model achieved the best classification performance (AUC = 0.954, <i>F</i>1 score = 0.78), and SHAP analysis showed that NPP and PRE made the most significant contributions to model predictions. This study proposes a multi-method integrated decision support framework for assessing ecological vulnerability in lake wetland ecosystems. |
|---|---|
| ISSN: | 2073-445X |