Groundwater Storage Dynamics and Lagged Responses to Precipitation in the Qinling Mountains (2004–2021)

Abstract

Addressing the challenges of sparse hydrological stations and limited groundwater monitoring in the complex terrain of the Qinling Mountains, this study integrated multi-source datasets from 2004 to 2021, including GRACE satellite observations and ERA5 meteorological reanalysis. An extreme gradient boosting (XGBoost) machine learning model was developed to estimate groundwater storage (GWS), followed by a systematic investigation of its spatiotemporal evolution and lagged responses to precipitation. The results demonstrate that: ① The XGBoost model achieved high retrieval accuracy on the test dataset (R^2=0.9549, RMSE=15.07" mm" ); ② Over the past 18 years, GWS across the Qinling Mountains exhibited a significant declining trend (-33.8" mm/a" ), with the rate of depletion following the spatial pattern of Central (-3.34" mm/a" ) > Southern (-2.98" mm/a" ) > Northern (-2.00" mm/a" ); ③ The response of GWS to precipitation revealed pronounced spatial heterogeneity and lag effects—the southern region responded rapidly with a 0-month lag, whereas the northern and central regions, buffered by thick soils and human disturbance, exhibited lag times of 8 and 9 months, respectively. These findings elucidate the modulation of groundwater recharge mechanisms by complex mountainous topography and anthropogenic activities.