Deep Learning-aided Temporal Downscaling of Satellite GravimetryTerrestrial Water Storage Anomalies Across the Contiguous United States (CONUS)

Abstract

Gravity Recovery and Climate Experiment (GRACE) and GRACE-FollowOn (GFO) satellites can monitor the global spatio-temporal changes in terrestrial water storage anomalies (TWSA) with monthly temporal and ~300 km spatial resolutions. Since these native resolutions may not be adequate for various studies requiring better localization of TWSA signal both in spatial and temporal domains, in recent years, considerable efforts have been devoted to downscaling TWSA to higher resolutions. However, the majority of these studies have focused on spatial downscaling; only a few studies attempted to improve the temporal resolution. Here, we utilized an in-house developed Deep Learning (DL) based model to downscale the monthly GRACE/GFO Mass Concentration (Mascon) TWSA to daily resolution across the Contiguous United States (CONUS). The simulative performance of the DL algorithm is tested by comparing the simulations to independent (non-GRACE) dataset and the land hydrology models. In addition, we assessed the potential of our daily simulations to detect long- and short-term variations in TWSA. The validation results show that our DL-aided simulations do not overestimate or underestimate GRACE/GFO TWSA and can monitor variations in the water cycle at a higher temporal resolution.