Groundwater potential mapping using multi-criteria decision, bivariate statistic and machine learning algorithms: evidence from Chota Nagpur Plateau, India

Abstract

Abstract Increased consumption of water resource due to rapid growth of population has certainly reduced the groundwater storage beneath the earth which leads certain challenges to human being in recent time. For optimal management of this vital resource, exploration of groundwater potential zone (GWPZ) has become essential. We have applied Analytical Hierarchy Process (AHP), Frequency Ratio (FR) and two machine learning techniques specifically Random Forest (RF) and Naïve Bayes (NB) here to delineate GWPZ in Gandheswari River Basin in Chota Nagpur Plateau, India. To achieve the goal of the study, twelve factors that determine occurrence of groundwater have been selected for inter-thematic correlations and overlaid with location of wells. These factors include elevation, drainage density, slope, lithology, geomorphology, topographical wetness index (TWI), distance from the river, rainfall, lineament density, Normalized Difference Vegetation Index (NDVI), soil, and Land use and Land cover (LULC). A total 170 points including 85 in well site and 85 in non-well site have been selected randomly and allocated into two parts: training and testing at the share of 70:30. The implemented methods have significantly provided five GWPZs specifically Very Good (VG), Good (G), Moderate (M), Poor (P) and Very Poor (VP) with high and acceptable accuracy. The study also finds that geomorphology, slope, rainfall and elevation have greater importance in shaping GWPZs than LULC, NDVI, etc. Model performance has been tested with receiver operator characteristics (ROC), Accuracy (ACC), Kappa Coefficient, MAE, RMSE, etc., methods. Area under curve (AUC) in ROC curve has revealed that accuracy level of AHP, FR, RF and NB is 78.8%, 81%, 85.3% and 85.5, respectively. The machine learning techniques coupled with AHP and FR unveil effective delineation of groundwater potential area in said river basin which by genetically offers low primary porosity due to lithological constrains. Therefore, the study can be helpful in watershed management and identifying appropriate location wells in future.

Keywords

• Analytical Hierarchy Process
• Frequency Ratio
• Machine learning
• Random Forest
• Naïve Bayes
• ROC curve