Scale effects and time-lag responses of groundwater hydrochemistry to land use change: A spatiotemporal analysis of a rapidly urbanizing typical agricultural area in the Lower Yellow River alluvial plain

Study region Typical agricultural areas along the lower Yellow River in China Study focus Traditional assessments often overlook hydrochemical spatial heterogeneity and spatiotemporal land-use lags. Utilizing datasets (1985–2010), this study reconstructs groundwater evolutionary trajectories. We developed an improved SOM-PMF (Self-Organizing Map and Positive Matrix Factorization) model for refined source apportionment, and a SOM-RF-SHAP framework to evaluate scale-dependent and lagged associations between land-use predictors and groundwater chemistry. New hydrological insights for the region Groundwater hydrochemical evolution followed a non-linear Natural Baseline-Pulse Disturbance-System Recovery trajectory. High-intensity human activities (1991–2000) triggered sharp increases in sulfate and total dissolved solids (TDS), shifting groundwater toward sulfate-enriched Ca-Mg/Na hydrochemical facies, with pronounced recovery signals emerging only after 2006. By integrating SOM-based hydrochemical classification with PMF source apportionment, the SOM-PMF model captured multivariate hydrochemical patterns and revealed clear spatial heterogeneity. This approach identified three interpretable source factors-carbonate weathering, cation exchange, and anthropogenic mixed inputs. Furthermore, SOM-RF-SHAP revealed distinct scale-dependent and lagged land-use responses. High-mineralization areas showed strong local, long-lag responses, with cropland most strongly associated with TDS (SHAP = 599.7; 500 m; 7 years), consistent with cumulative agricultural salinization. Carbonate weathering–cation exchange mixed areas showed a broader, shorter-lag sulfate response to impervious surfaces (SHAP = 67.5; 1 km; 4 years), whereas low-mineralization areas showed weak land-use explanatory power.