Predictions of the GRACE-derived terrestrial water storage anomalies generated by a Bayesian convolutional neural network

This dataset provides predictions of the GRACE-derived terrestrial water storage anomalies (TWSAs) to fill the approximately one-year gap (July 2017-May 2018) between GRACE and its successor GRACE Follow-On. The predictions were obtained by a Bayesian convolutional neural network driven by the ERA5-land data (monthly precipitation, temperature, TWSAs, and cumulative water storage changes). The training period is April 2002-March 2014, the test periods are April 2014-June 2017 and June 2018-August 2020.