Spatiotemporal dynamics and nonlinear drivers of ecosystem service synergies and trade-offs in Shaanxi, China.

Spatiotemporal interactions among ecosystem services (ES) and their drivers are essential for sustainable landscape management. This study employs the Dynamic Ecosystem Trade-off Synergy Index (DETSI), local indicators of spatial association (LISA) and interpretable machine learning (XGBoost-SHAP) to assess ES synergies and trade-offs, examine nonlinear driver-response patterns, and identify threshold effects in Shaanxi Province, China, from 2000 to 2022. Six ES indicators, net primary productivity (NPP), habitat quality (HQ), water yield (WY), soil retention (SR), cultural landscape (CL) and food supply (FS), were analysed to characterise spatial patterns and temporal dynamics. Results show persistent south-north gradients and stable high-value clusters in ecological core zones. Regulating service pairs exhibit strong synergies, whereas cultural-provisioning and provisioning-regulating pairs show persistent trade-offs. From 15 ES pairs, five representative pairs (CL-WY, FS-WY, NPP-SR, NPP-WY and SR-WY) were examined using SHAP, achieving accuracy of 0.8112-0.9696 and macro-F1 of 0.8110-0.9603. Eighteen key drivers, including NDVI, precipitation, elevation, road density and points of interest, show nonlinear and threshold-like responses. Moderate conditions support synergies, while extremes drive decoupling. The Qinling-Daba corridor emerges as a stable synergy zone, with northern and southeastern Shaanxi identified as recovery and rupture areas. This framework provides a robust transferable approach for spatially differentiated ecological assessment and planning.