Mapping stability and instability hotspots in Jiangsu's vegetation: an explainable machine learning approach to climatic and anthropogenic drivers.

Understanding vegetation stability is essential for evaluating ecosystem resilience and informing adaptive land management under changing climatic conditions. This study investigated the spatiotemporal patterns and climatic drivers of vegetation stability across Jiangsu Province, China. Vegetation productivity was assessed using the annual maximum kernel normalized difference vegetation index, while stability was quantified through two indicators: proportional variability (PV) and lag-one autocorrelation (AR). Results revealed that 15.77% of the province experienced increases in PV and AR, indicating growing vegetation instability, particularly in the south-central and southeastern regions. In contrast, 84.23% of the area showed declining PV and AR trends, suggesting enhanced stability, mainly in the southwestern, northern, and central regions. Spatially, high AR values were observed in western and southern Jiangsu, while high PV values were concentrated along the eastern coast and near Lake Taihu. More stable areas-characterized by low PV and AR-were primarily located in the central and northwestern regions. An interpretable machine learning model identified background solar radiation and its temporal variability as the dominant drivers of vegetation stability, followed by vapor pressure deficit (VPD). Precipitation variability had minimal influence. SHAP dependence plots revealed nonlinear responses: moderate radiation and higher soil moisture promoted stability, while elevated VPD and radiation variability reduced it. Most regions were in favorable ecological condition, although ~20% were classified as poor and another ~20% remained uncertain. These findings highlight the key roles of radiation and moisture in regulating vegetation stability and offer insights for climate-resilient land management in intensively cultivated landscapes.