Predictive Modeling of Terrestrial Water Storage Anomalies in Kalimantan Basins: Bridging the GRACE and GRACE-FO Data Gap with Extreme Gradient Boosting

Abstract

Terrestrial water storage (TWS) anomaly has been a robust indicator in predicting and monitoring hydrometeorological hazards and sustainable water resources management to comprehend the water dynamics on Earth. The Gravity Recovery and Climate Experiment (GRACE) satellite identifies this change by heeding the Earth’s mass anomalies since 2002. However, due to an 11-month data gap before the operation of GRACE-FO, continuous investigation using GRACE has been challenging. This study employed an extreme gradient boosting (XGBoost) algorithm to reconstruct GRACE TWS anomaly by integrating the hydroclimatic variables from Noah surface models over a span of approximately 20 years, focusing on five Kalimantan basins. The testing set was evaluated using three statistical metrics, resulting in a correlation coefficient (CC) of 0.943, Nash–Sutcliffe efficiency (NSE) of 0.887, and scaled root-mean-square error (RMSE*) of 0.337. This approach effectively addresses the research gap in utilizing the GRACE product in an archipelago state such as Indonesia and offers an efficient method for reconstructing TWS anomalies for various hydrological systems at the local scale.