Spatial mapping of groundwater potentiality applying ensemble of computational intelligence and machine learning approaches

Abstract

Groundwater plays a significant role in satisfying the human need, and it regulates the earth's hydrological system. Due to increase in population, the stress on groundwater has been successively increased. First step for managing the groundwater is the identification of groundwater potential areas with sound methods. Malda district has high agricultural dependence and in turn huge amount of groundwater has been extracted for irrigation and drinking. As a result the problems of water scarcity for drinking and other purposes has aggravated in this study area. Main aim of this research article is to prepare spatial groundwater potential maps (GWPMs) of Malda district of India using hybrid ensemble of computational intelligence and machine learning approaches including Bagging, artificial neural network (ANN)-Bagging, random forest (RF) and support vector machine (SVM). Total 93 well locations and 17 groundwater potentiality determining factors (GWPDFs) were selected. GWPDFs were selected based on the rank correlation and multi-collinearity analysis. The contribution of GWPDFs was assessed by the RF and sensitivity analysis. As per these methods distance from the river, rainfall, lithology are the most important factors in determining the potentiality of groundwater. The robustness and accuracy of the four groundwater potential maps were evaluated by the nine statistical methods. GWPMs showed that only 23.41% area of this district is very highly potential for groundwater. All the statistical methods used for testing the accurateness of the GWPMs including receiver operating characteristics (ROC), accuracy, etc. acknowledged the significant results of the groundwater potential models. Among the selected models ANN-Bagging ensemble model achieved the highest accuracy (area under curve = 0.96) followed by Bagging, RF and SVM. Thus, the present study ensured that ANN-Bagging ensemble method might be used as an important tool for identifying groundwater potential area that could be helpful to land use planners and local people for groundwater management.