Unveiling kazakhstan's ecosystem service puzzle: Spatiotemporal shifts and drivers of supply and demand through multi-model integration and machine learning methods

Kazakhstan, a key arid and semi-arid region in Central Asia, faces growing mismatches between ecosystem service (ES) supply and demand under complex ecological and socio-economic pressures. This study quantifies the spatiotemporal dynamics of water yield (WY), carbon storage (CS), and food production (FP) from 1995 to 2022 using InVEST and Google Earth Engine(GEE), and identifies dominant drivers of the ecosystem service supply–demand ratio (ESDR) through Random Forest(RF) and SHAP analysis. Normalized Partial Dependence Plots (PDPs) reveal thresholds and nonlinear responses of these drivers, while Geographically and Temporally Weighted Regression (GTWR) captures their spatial heterogeneity. Key findings include:(1)Temporal changes: WY fluctuated over 25 %, CS declined by 18.6 %, and FP increased by 42.3 % over the study period.(2)Spatial patterns: WY was concentrated in eastern and northern regions, CS in central and southern regions, and FP in northern and western regions, with natural barriers limiting ES supply in the south and east.(3)ESDR dynamics: WY surplus or balanced areas reached 64.8 %, with severe deficits decreasing from 20.7 % to 16.5 %. CS high-capacity areas declined from 19.3 % to 14.2 %, while low-capacity areas rose from 44.7 % to 51.1 %, indicating reduced carbon sequestration. FP high-capacity areas grew from 13.3 % to 20.8 %, but deficit areas also increased from 17.3 % to 21.7 %, highlighting emerging food security risks. (4) Mechanistic drivers: Population growth, climate variability, and energy consumption collectively explain over 60 % of ESDR variance, illustrating how demographic and anthropogenic pressures drive ES mismatches. These findings provide actionable, quantitative guidance for region-specific resource management and sustainable development in arid Central Asia.