Investigating the Nonlinear Response of Ecosystem Health in China’s Karst Region Using Explainable Machine Learning

Karst regions, characterized by environmental fragility, are of great value for ecological research into regional ecological security and sustainable development. Ecosystem health (EH) serves as a critical indicator of environmental sustainability, effectively delineating the impacts of both natural variability and human interventions. Yet, a systematic understanding of its nonlinear responses to complex interacting drivers across China’s Karst region remains lacking. This study integrated multi-source datasets, including core remote-sensing datasets (NDVI, nighttime light, DEM, slope) and socioeconomic land-use datasets (administrative boundary, land-use, population-relevant socioeconomic data) to assess the spatiotemporal dynamics of EH from 2000 to 2020 while also utilizing the XGBoost–SHAP model to quantify the influence of key drivers and their relationships with EH. The results revealed that: (a) EH and its components in China’s Karst region exhibited a distinctive spatial heterogeneity, with high values concentrated in South China and low values across the Tibetan Plateau; (b)The Tibetan Plateau (TP) shows a pattern of “high in southeast and low in northwest”. The forest ecosystem in the southeast Hengduan Mountains is in good health due to its integrity, whereas the environment in the northwest Plateau is weakened due to cold and drought stress. (c) SHAP analysis revealed vegetation and precipitation as key drivers of EH, with stronger anthropogenic influence in South China and climate-driven patterns in the north and Tibetan Plateau. Our findings provide a robust scientific foundation for ecological management in karst regions, underscoring the critical role of scale-dependent dynamics and multi-factor interactions in effective ecosystem monitoring and conservation.