Spatially Heterogeneous and Nonlinear Responses of Ecosystem Service Value to Human Activity Intensity in an Arid Mountain–Oasis–Desert Urban Agglomeration

ABSTRACT Rapid urbanization and agricultural expansion are reshaping human–environment relationships in arid regions; however, the spatially heterogeneous and nonlinear relationships between human activities and ecosystem services remain insufficiently understood. This study investigated the urban agglomeration on the northern slope of the Tianshan Mountains (UANSTM), Xinjiang, China, from 2000 to 2020. Ecosystem service value (ESV) was estimated using a regionally adjusted equivalent‐factor approach, while human activity intensity (HAI) was assessed. Using a watershed‐based “scale dimensionality reduction–response decoupling” analytical framework, the relationships between HAI and ESV were examined through coupling coordination analysis, spatial association analysis, and an extended geographically Gaussian process regression (GGPR) model interpreted using GeoShapley. The results showed that land‐use transitions initially generated net ESV gains, mainly through the conversion of bare land to grassland and cropland; however, these gains declined after 2010 as urban expansion increasingly converted cropland, grassland, and bare land into built‐up land. High‐ESV areas were consistently concentrated in mountainous regions, whereas high HAI values were mainly distributed in oasis and urbanized areas. Although the coupling relationship between HAI and ecosystem services remained relatively stable, the coupling coordination degree ( D ) was generally low. Provisioning services showed stronger coordination with HAI than regulating, supporting, and cultural services. In most cases, the GGPR model showed better predictive performance than conventional Gaussian process regression, random forest, and extreme gradient boosting models. Moreover, ESV and several individual ecosystem services exhibited predominantly nonlinear, inverted U‐shaped associations with HAI, with positive or stable model‐derived marginal contributions at moderate HAI levels but increasingly negative marginal contributions under high HAI conditions.