资源环境科技发展态势分析平台

Launched in May 2018, the Gravity Recovery and Climate Experiment Follow-On mission (GRACE-FO)-the successor of the erstwhile GRACE mission-monitors changes in total water storage, which is a critical state variable of the regional and global hydrologic cycles. However, the gap between data of the two missions is breaking the continuity of the observations and limiting its further application. In this study, we used three learning-based models, that is, deep neural network, multiple linear regression (MLR), and seasonal autoregressive integrated moving average with exogenous variables, and six GRACE solutions (i.e., Jet Propulsion Laboratory spherical harmonics (JPL-SH), Center for Space Research SH (CSR-SH), GeoforschungsZentrum Potsdam SH (GFZ-SH), JPL mass concentration blocks (mascons) (JPL-M), CSR mascons (CSR-M), and Goddard Space Flight Center mascons (GSFC-M)) to reconstruct the missing monthly data at a grid cell scale. Evaluation showed that the three learning-based models were reliable for the reconstruction of GRACE data in areas with humid and no/low human interventions. The deep neural network models slightly outperformed the seasonal autoregressive integrated moving average with exogenous variables models and significantly outperformed the multiple linear regression models in most of 60 basins studied. The three GRACE mascon data sets performed better than the SH data sets at the basin scale. The models with SH solutions showed similar performance, but the models with the mascon solutions varied markedly in some basins. Results of this study are expected to provide a reference for bridging the data gaps between the GRACE and GRACE-FO satellites and for selecting suitable GRACE solutions for regional hydrologic studies.

Key Points

Gaps between the GRACE and GRACE-FO data sets are filled across global land areas Performance of reconstructing total water storage anomalies using three learning-based models and six GRACE solutions is evaluated Findings of this study improve understanding of global and regional hydrologic cycles, various GRACE products, and learning models