GRACE-Based Analysis of the Spatiotemporal Evolution and Driving Factors of Groundwater Storage in the Heilongjiang (Amur) River Basin

Abstract

Accurately characterizing groundwater storage variations and identifying their dominant drivers are essential for regional groundwater assessment and sustainable management. This study focuses on the Heilongjiang (Amur) River Basin and constructs a monthly groundwater storage anomaly (GWSA) series for 2003–2022 using GRACE/GRACE-FO, GLDAS, and multiple environmental datasets. The spatiotemporal evolution of GWSA was analyzed, and groundwater sustainability was further evaluated. On this basis, an XGBoost model with five-fold year-grouped cross-validation was developed, and SHAP and lagged correlation analyses were combined to quantify the model-based importance of the main environmental factors and characterize their nonlinear predictive relationships with GWSA, interaction patterns, and lagged responses. The results showed that basin-averaged GWSA exhibited a significant declining trend during the study period, with groundwater deficits becoming markedly more pronounced after 2017, while substantial spatial heterogeneity was observed among the sub-basins. The groundwater system exhibited relatively low overall sustainability and limited recovery capacity following groundwater deficits. The XGBoost model showed relatively stable performance in characterizing GWSA variations, with a correlation coefficient of 0.660 and an RMSE of 48.54 mm between the pooled predictions from the five test folds and the GRACE-derived GWSA. TreeSHAP analysis showed that precipitation had the highest relative SHAP importance (33.15%), while evapotranspiration, runoff, air temperature, land use, and snowmelt also exhibited varying degrees of SHAP importance and pronounced nonlinear patterns and interaction characteristics. Snowmelt and GWSA exhibited a seasonal timing offset, with a pronounced time–frequency association around the annual scale. GWSA remained elevated after spring thaw, and the annual frozen fraction decreased significantly during 2003–2022. The results provide scientific support for sustainable groundwater management and engineering planning in cold regions.