A novel XGBoost-based approach for reconstruction terrestrial water storage variations with GNSS in the Northeastern Tibetan Plateau

Abstract

Accurately estimating terrestrial water storage (TWS) variations is essential for ensuring the sustainable management of global water resources. The Global Navigation Satellite System (GNSS) offers a promising approach for monitoring TWS changes with high spatial and temporal resolution. However, its application is significantly constrained by the sparse and uneven distribution of GNSS stations. In this study, we build upon traditional GNSS inversion techniques by employing the Extreme Gradient Boosting Machine Learning (XGBML) model to simulate crustal deformation caused by hydrological loading. The simulation is conducted on a $0.5^{°}\times 0.5^{°}$ grid across the Northeastern Tibetan Plateau (NETP). This study compared TWS variations derived from the XGBML simulations and traditional inversion methods with data from the Gravity Recovery and Climate Experiment (GRACE) satellite and the Global Land Data Assimilation System (GLDAS). The Pearson Correlation Coefficients (PCC) between TWS changes derived from the XGBML inversion technique and those from GRACE and GLDAS data were 0.72 and 0.50, respectively, representing improvements of 8.82 % and 11.10 % compared to the conventional inversion approach. Furthermore, GNSS-DSI, GRACE-DSI, and SPEI were integrated to analyze hydrological drought events in the study area, revealing that precipitation and temperature are important drivers of hydrological drought in the NETP. These findings highlight the effectiveness of the XGBML model in simulating GNSS vertical displacements induced by hydrological loading and demonstrate its potential as a novel tool for identifying water storage variations in regions with uneven GNSS station distribution.