GRAiCE: reconstructing terrestrial water storage anomalies with recurrent neural networks

Abstract

Abstract The Gravity Recovery and Climate Experiment (GRACE) and its follow-on (GRACE-FO) missions have provided estimates of Terrestrial Water Storage Anomalies (TWSA) since 2002, enabling the monitoring of global hydrological changes. However, temporal gaps within these datasets and the lack of TWSA observations prior to 2002 limit our understanding of long-term freshwater variability. In this study, we develop GRAiCE, a set of four global monthly TWSA reconstructions from 1984 to 2021 at 0.5° spatial resolution, using Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM) neural networks. Our models accurately reproduce GRACE/GRACE-FO observations at the global scale and effectively capture the impacts of climate extremes. Overall, GRAiCE outperforms a previous reference TWSA reconstruction in predicting observed TWSA and provides reliable water budget estimates at the river basin scale. By generating long-term continuous TWSA time series, GRAiCE will offer valuable insights into the impacts of climate variability and change on freshwater resources.