Spatial non-stationarity and causal pathways of hydro-thermal responses in alpine grasslands intertwined with climate warming and surface disturbances

Against a backdrop of climate warming and increasing humidity, alpine grasslands along the northeastern margin of the Qinghai–Tibet Plateau have exhibited a widespread “greening” trend. However, the mechanisms underpinning this apparent recovery are characterized by pronounced spatial non-stationarity. Conventional approaches based on global assumptions are insufficient to disentangle the complex interplay between climatic benefits and anthropogenic disturbances in heterogeneous terrain. Focusing on Tianzhu County in the eastern Qilian Mountains, this study integrates long-term (2001−2023) net primary productivity (NPP) data with multi-source environmental variables, employing spatial clustering, geographically weighted regression (GWR), and structural equation modeling (SEM) to systematically unravel the nonlinear drivers and causal pathways of alpine grassland dynamics. The results indicate that: (1) NPP increased significantly across 98.45% of the study area, although the quality of recovery exhibits marked spatial heterogeneity. (2) Multidimensional clustering identifies a representative “low-elevation ecological trap” (Zone 2), which, despite possessing the highest moisture-driven potential in the region ( Coef f Pre = 0.28 ), is highly exposed to anthropogenic disturbances due to the absence of topographic barriers, resulting in substantial interannual variability ( CV = 0.20 ). In contrast, mid to high-elevation areas (Zone 4) demonstrate compensatory growth following the release of environmental pressures. (3) SEM analysis confirms a causal feedback loop characterized by “topographic constraint–hydrothermal offset–disturbance interception”: elevation imposes a metabolic ceiling on ecosystem development ( β = − 0.413 , p < 0.001 ), while warming induces a pronounced hydrothermal offset effect ( r = − 0.36 ). Critically, in areas lacking topographic protection, human activities bypass natural mediating processes and exert a direct suppressive impact on ecosystem functioning ( β = − 0.185 , p < 0.001 , Partial R 2 = 3.99 % ). These findings demonstrate that grassland recovery in arid and semi-arid regions is not governed solely by climatic benefits, but rather by the dynamic interplay between natural potential and anthropogenic interception. The proposed mechanistic framework of a “structural ecological trap” provides a novel basis for differentiated, precision-oriented management and adaptive governance in ecologically fragile alpine regions.