Combining physics-based modeling and machine learning for GRACE satellite data fusion and reconstruction

Abstract:

Global hydrological models and land surface models are increasingly being used to simulate spatial and temporal patterns of total water storage. Missing processes and/or uncertain parameters in these models may introduce significant bias and uncertainty in model predictions, hampering the timely application of the model results in regional water management and forecasting activities. The Gravity Recovery and Climate Experiment (GRACE) satellite mission and its follow-on, GRACE-FO, have provided unprecedented opportunities to quantify the impact of climate extremes and human activities on total water storage at large scales. There is strong interest in fusing GRACE data to improve the fidelity of global hydrological model simulations. Like many other satellite missions, GRACE missions have data gaps, in particular the approximately one-year gap between the two GRACE missions needs to be imputed to maintain data continuity and maximize mission benefits. This presentation summarizes our recent work in two related areas, namely, application of deep learning techniques to learn the model-GRACE mismatch patterns, and use of automated machine learning for reconstructing GRACE-like data. We show that blending machine learning with hydrological model outputs may provide an effective way for combining the best of both worlds.