Rethinking pathways of ecological management: Insights from climate-driven ecosystem health

Translating multidimensional ecosystem-health assessment into climate-sensitive management remains challenging in arid regions. Taking Xinjiang, China, as a representative desert-oasis-mountain system, this study developed a research framework comprising separate modules for VORS-based Ecosystem Health Index (EHI) assessment, LSTM time-series projection, climate-EHI LSTM-SHAP analysis, and management zoning. EHI was assessed at five-year intervals from 1990 to 2020 and projected to 2035 using historical EHI sequences. Model selection employed temporally blocked validation, with 2020 retained as an independent test set. Across five runs, LSTM outperformed linear regression and Random Forest. Spatially, relatively high EHI values were concentrated in the Altay, Tianshan, and Kunlun Mountains, whereas the Junggar and Tarim basins were dominated by low-to-moderate health levels. Historical changes were generally small and spatially heterogeneous, with localized improvement and degradation coexisting across Xinjiang. A separate climate-EHI LSTM-SHAP analysis identified precipitation of the wettest quarter, precipitation of the warmest quarter, and annual precipitation as the most influential climatic variables and revealed nonlinear, variable-specific response ranges. The nine selected climatic variables were used to construct the Ecosystem Climate Sensitivity Index (ECS), which was coupled with EHI to delineate five mutually exclusive management zones. The resulting zoning distinguishes priorities for ecological intervention, climate-sensitive regulation, early warning, functional maintenance, and coordinated conservation and enhancement.