Key points are not available for this paper at this time.
It is crucial to monitor water resources and acquire knowledge on water-related natural disasters to fully understand the Earths climate system and guarantee its long-term sustainability. The Gravity Recovery And Climate Experiment (GRACE) and GRACE-FollowOn (GRACE(-FO)) satellites have contributed significantly to our knowledge of variations in Earths Total Water Storage (TWS) throughout the last twenty years. Nevertheless, the ability to detect hydrological activities is hindered by restrictions in spatial and/or temporal resolutions. This study introduces a sophisticated Deep Learning (DL) paradigm that is specifically developed for the spatial downscaling of GRACE TWS anomalies (TWSA). The technique employs soft constrained knowledge by developing a novel loss function to obtain higher spatial resolutions while preserving the integrity of the GRACE TWSA signal. The time series of Mass Conservation (Mascon) TWSA from Jet Propulsion Laboratories (JPLM) are downscaled in spatial resolution from 300 km to 50 km using spatiotemporal correlations of TWSA derived from the WaterGAP Hydrology Model (WGHM). The TWSA simulations consist of monthly time series spanning from April 2002 to December 2022. These downscaled TWSA time series were evaluated not only internally using statistical metrics but also externally comparing with the non-GRACE dataset that are related to the glacier mass loss signals, the interannual TWSA signal variations that are triggered by El Nino-Southern Oscillation (ENSO), InSAR subsidence rates and the altimetry-derived water levels in major rivers. Furthermore, the DL methodology has been examined in both drought and flood events, where it was able to effectively bridge the gap between the GRACE and GRACE(-FO) satellite missions. These evaluations reveal the suggested DL strategy preserves the GRACE/-FO TWSA signal while also achieving an evolution to higher spatial resolution.
Uz et al. (Mon,) studied this question.