Ecosystem Service Value Dynamics Under Land-Use Transition in Henan Province, China

Rapid land-use transition and urbanization are increasingly associated with changes in the spatial and temporal patterns of ecosystem services in agricultural regions, yet the nonlinear and spatially varying relationships involved remain insufficiently understood. This study integrated ecosystem service assessment, spatial statistical analysis, interpretable machine learning, and geographically weighted regression to investigate ecosystem service value (ESV) dynamics and their spatially varying associations with human and natural factors in Henan Province, China, from 1990 to 2020. ESV was estimated using the equivalent factor method, with the standard equivalent value adjusted using regional grain production and price data. Multi-source topographic, climatic, vegetation, socioeconomic, infrastructure, and land-use intensity variables were further integrated to examine factors associated with ESV variation. Spatial statistical analyses were used to characterize temporal changes and spatial differentiation, XGBoost combined with SHapley Additive exPlanations (SHAP) was applied to identify nonlinear associations and relative predictor contributions, and geographically weighted regression (GWR) was used to examine spatial heterogeneity in predictor–ESV relationships. Total ESV showed a decline–recovery–slow growth trajectory, decreasing from 1807.99 billion CNY in 1990 to 1725.90 billion CNY in 2000 and subsequently increasing to 1818.07 billion CNY in 2020. High-ESV areas were persistently concentrated in the western and southwestern mountainous regions, whereas low-ESV areas were mainly distributed in the eastern and northern plains. Land-use intensity, carbon emissions, nighttime light intensity, population density, and GDP density showed relatively high predictive importance, although their associations with ESV were nonlinear and context-dependent. GWR further revealed pronounced spatial variation in these relationships. These findings highlight the importance of considering nonlinear human–natural associations and spatial heterogeneity when assessing ecosystem service dynamics and developing differentiated land-management strategies.