A Comprehensive Evaluation of GRACE‐Like Terrestrial Water Storage (TWS) Reconstruction Products at an Interannual Scale During 1981–2019

Given the success of the Gravity Recovery and Climate Experiment (GRACE) mission in mapping terrestrial water storage (TWS) since 2002, recent reconstructions of long‐term TWS rely on the use of statistical machine learning to apply GRACE‐derived information to past decades. Evaluating the interannual accuracy during nonobservational periods is a key challenge. This study develops a comprehensive framework to discuss the interannual accuracy of three different TWS reconstructions during 1981–2019, including (a) global‐scale accuracy assessment using GRACE and satellite laser ranging data; (b) regional‐scale accuracy testing across various underlying surfaces (i.e., rivers, lakes, and glaciers); and (c) investigation of relevant evidence from other Earth subsystems (i.e., historic climate events, sea level budget, and polar motion). Among the three reconstructions, the one that additionally corrects glacial TWS changes (REC2) detects a breaking point in the 1990s and further closes the interannual sea level budget with an absolute difference reduction to 5.13 mm; the reconstruction that is forced by local meteorological conditions (REC1), accounting for 54% of the GRACE‐derived signal energy, underestimates glacial TWS variability but outperforms the other reconstructions in reproducing lake levels, basin‐scale water balances, and climate events at the interannual scale, while the others consider 95%–99% of the GRACE‐derived signal energy. The relatively high accuracy of REC1 (and REC2) in reflecting interannual changes in nonglacial (and glacial) regions is further confirmed by explaining the χ2‐ (and χ1‐) component polar motion. Ten to 20% of the interannual polar motion remains unexplained, indicating room for improvement in interannual TWS reconstruction. Key Points We investigate multisource evidence among Earth subsystems to develop a comprehensive evaluation framework for terrestrial water storage (TWS) reconstructions There are some drawbacks at the interannual scale when using statistical machine learning to generate Gravity Recovery and Climate Experiment‐like TWS reconstructions Polar motion evaluates TWS reconstructions from multiple perspectives and reinforces the reliability of results using other metrics