Remote sensing-based statistical modeling for shallow groundwater mapping using a machine learning technique: the case of Borkena sub-basin, Awash River basin, Ethiopia

Insufficient comprehension of the prevalence of shallow groundwater impairs the accuracy of energy and water cycle modeling. This study aims to evaluate the occurrence of shallow groundwater in the Borkena sub-basin, Awash River basin, Ethiopia, using statistical modeling based on ANN-MLP. This was achieved by employing groundwater depth (GWD) as a dependent variable and the following predictor variables: latent energy (LE) flux and its coefficient of variation (CV), elevation (ELV), nighttime land surface temperature (NLST) – CV, normalized difference vegetation index (NDVI) and its CV, and reflectance intensity of shortwave infrared radiation (SWIR). The results indicated that shallow groundwater resources occur in about 16.5% of the Borkena sub-basin. According to multivariate linear regression, LE-flux and its CV, NDVI and its CV, and NLST-CV rise when GWD decreases; however, ELV and reflectance intensity for SWIR decrease. Additionally, while looking at the occurrence of shallow groundwater, LE-flux was found to be more important than other variables. Growing interest in remote sensing observation-based methods to identify shallow groundwater at local to regional scales is met by an ANN-MLP-based statistical model, which could help enhance models of the land surface.