Step-ahead forecasting of GRACE-derived terrestrial water storage spatial downscaling in Saudi Arabia using Elman recurrent learning and support vector regression

Abstract

The growth and sustainability of human development rely on water availability. Proper management of Groundwater as the main source of one-third of freshwater resources is compulsory because of its renewable sources and to safeguard food sanctuary and water security. The steps ahead modeling of GRACE Terrestrial Water Storage (TWS) was employed in this study using the GRACE TWS data from 2007 to 2017 and TWS spherical harmonic solution obtained from the University of Texas (UT) center for space research.  The data was used to create the artificial intelligence-based models viz: Elman neural network (ENN) and Support vector regression (SVR) based on several input variables, including t-12, t-24, t-36, t-48, and TWS as the output variable. The models were evaluated using mean absolute error (MAE), root means square error (RMSE), mean absolute percentage error (MAPE), correlation coefficient (CC), and Nash–Sutcliffe efficiency (NSE). The estimation outcomes depicted that only SVR-M1 (NSE=0.993, MAE0.0346) generated promising results, with ENN-M3 (NSE=0.6586, MAE=0.6895) as the second-best model. All other models’ combinations were within the range of good to marginal accuracies, which are unreliable for decision-making.