Spatial Modeling of Groundwater Level in Bangladesh Using Physio-Climatic Variables: Machine Learning and Statistical Approaches

Abstract

Abstract Groundwater monitoring is essential for sustainable groundwater resource management in a country like Bangladesh, where this precious resource is gradually declining due overextraction. Acquiring groundwater level (GWL) over a large area is time-consuming and expensive. This study proposes an alternative approach to groundwater monitoring using freely available daily groundwater storage (GWS) gridded data of the Global Land Data Assimilation System (GLDAS) with other freely available data, including population, rainfall, temperature, irrigation, elevation for modeling GWL data of Bangladesh with a spatial resolution of 0.25o × 0.25o. This was accomplished by employing multiple linear regression (MLR) and artificial neural networks (ANN), using weekly in-situ GWL data at 844 locations distributed over Bangladesh. The results showed the inability of GWS data to estimate the country's groundwater spatial variability and trend. The relative performance of MLR and ANN models revealed a higher capability of ANN in estimating GWL from GWS and other data with an overall correlation coefficient (R) of 0.95 and mean squared error (MSE) of 0.64. The study revealed population and rainfall have the most decisive influence in determining GWL. The model developed using ANN can be used to estimate GWL at locations where observation data are unavailable and thus monitor GWL for sustainable groundwater management.