Downscaling of Groundwater Reserves in Henan Province Based on Xgboost

Abstract

Addressing the challenge posed by the relatively low spatial resolution of gravity satellite data, which hampers the fine monitoring of groundwater in provincial regions, this study focuses on Henan Province and develops a method utilizing eXtreme Gradient Boosting (XGBoost) for statistical downscaling of Groundwater Storage Anomaly (GWSA). Initially, the land water storage anomaly was derived from the GRACE/Grace-FO Mascon product provided by CSR. This was combined with surface water storage components, including soil moisture, snow water equivalent, and canopy water, as modeled in the GLDAS Noah land surface model, to calculate the groundwater storage anomaly. Subsequently, multiple source factors—such as precipitation, air temperature, relative humidity, surface temperature, evapotranspiration, normalized vegetation index, and elevation—were standardized to a 0.01° grid. This facilitated the establishment of a nonlinear mapping relationship between the coarse-resolution GWSA and high-resolution environmental factors, ultimately generating groundwater reserve anomaly products at a 0.01° resolution for Henan Province. The model results were validated through coarse-scale aggregation consistency tests, time series comparisons of monitoring wells, and error index evaluations. The findings indicate that XGBoost effectively preserves the spatiotemporal variation characteristics of the original GRACE results. After aggregation to 0.25°, a strong correlation with the original GWSA is observed. The downscaling results generally align with the variation trends observed in the monitoring wells. This research offers a high spatial resolution data foundation for the dynamic monitoring of groundwater resources, agricultural irrigation management, and regional water security assessments in Henan Province.