Based on XGBoost+SHAP: Revealing the driving factors of ecological vulnerability in Sichuan Province and its ecological protection assessment

Objectives This study employs the single indicator of Net Primary Productivity (NPP) combined with machine learning methods to overcome the subjective limitations inherent in traditional comprehensive indicator systems. It aims to quantitatively analyze driving mechanisms, thereby providing a scientific basis for zoning control, ecological protection and restoration, and ecological protection assessment within the territorial space of Sichuan Province.

Methods  Based on the IPCC definition of ecological vulnerability, this study utilizes vegetation Net Primary Productivity (NPP) from 2001 to 2023 in Sichuan Province as a single evaluation indicator. It integrates spatial autocorrelation, hotspot analysis, and the XGBoost+SHAP machine learning model to systematically reveal the spatial pattern of ecological vulnerability and the magnitude of influence of driving factors in Sichuan Province.

Results  ① The overall ecological vulnerability in Sichuan Province is relatively high, with approximately 70% of the region experiencing moderate or higher levels of vulnerability. The spatial distribution exhibits a "high in the west, low in the east" pattern, with extremely vulnerable areas concentrated along the basin margins and the Hengduan Mountains. ② Ecological vulnerability is controlled by the interactive effects of natural and human activity factors. Pearson correlation analysis identifies elevation, mean temperature, land surface temperature, and rainfall as key natural driving factors, while SHAP values quantitatively indicate that land use intensity (highest contribution), land surface temperature, and mean temperature are the core drivers. ③ Spatial agglomeration of ecological vulnerability is significant, with hotspot areas concentrated in the Western Sichuan Plateau and basin margins.

Conclusions  Ecological vulnerability demonstrates a significant positive spatial correlation. The degree of spatial agglomeration is highest for ecological sensitivity, followed by vulnerability, with adaptability being the lowest. Land use intensity, land surface temperature, and mean temperature are the primary driving factors influencing changes in the ecological vulnerability index.