The gravity recovery and climate experiment (GRACE) satellite mission revolutionized remote measurement of total water storage (TWS) anomalies, providing comprehensive insights at regional to continental scales. With a wealth of applications, GRACE data has been leveraged for monitoring ice sheets and glaciers, uncovering anthropogenic groundwater depletion, tracking droughts, and predicting floods, among others. Its legacy continues with the launch of GRACE Follow-On (GRACE-FO) on 22 May 2018. However, the time span of GRACE(-FO) observations (covers so far), limited to 20 years of monthly data with a year gap between GRACE and GRACE-FO, hinders capturing long-term climate projections. A short record may not accurately reflect the changes in climate and its impacts on various aspects of the environment, such as water resources, agriculture, and ecosystems. Additionally, the shortage of data from the last decades of the 20th century may lead to inaccurate results (underestimation or overestimation) in various GRACE(-FO) applications, such as characterizing droughts. To accurately assess global and regional climates, continuous observations for at least 30 years (more favorably 60 years) are highly recommended.
To overcome the limitations of the short record of GRACE data, we have employed a hindcasting approach using state-of-the-art datasets from Global Hydrological Models (GHMs), Land Surface Models (LSMs), and atmospheric reanalysis models. The intersected period between GRACE and these models was divided into training and validation periods. A Gaussian Process Regression (GPR) was trained using the model’s total water storage anomalies (TWSA) as features and GRACE TWSA as target values, at a global land scale, for each grid cell. The trained model then hindcasted TWSA back to 1980. The results before 2002 were compared with key hydrological fluxes, such as precipitation, evapotranspiration, and runoff, at both the pixel and basin scales. The newly obtained long-term dataset of TWSA can be utilized for climate-related research and to enhance extremes monitoring within the GRACE period. This approach offers a valuable solution to the problem of the limited record of GRACE data, allowing for more accurate and comprehensive analysis of global and regional climates.
Figure 1 presents the results from the hindcasted total water storage anomalies (GRACE-H) over four selected basins. The performance of the GPR is compared against the ensemble mean of the models. Our proposed method has significantly improved the results in terms of the Nash-Sutcliffe Efficiency (NSE), with an increase from 12 % in the Yangtze basin to 30 % in the Amazonas basin.The Normalized Root-Mean-Squared-Error (NRMSE) also demonstrates significant improvement, ranging from 50 % in the Yangtze basin to 70 % in the Amazonas basin. The same evaluation conducted at the global scale indicated general enhancement in the reconstruction of GRACE TWSA from the models, with improvements ranging from 5 to 90 % for NSE and 10 to 100 % for NRMSE.