Machine learning and GIS-RS-based algorithms for mapping the groundwater potentiality in the Bundelkhand region, India

Abstract

Mapping the groundwater potential zones with high accuracy is always a difficult task. The combination of Geographic Information System (GIS) and Remote Sensing (RS) with machine learning techniques provide a reliable method to map the groundwater prospective areas. This research used support vector machine (SVM) learning and random forest (RF) regression algorithms to predict the groundwater potential areas in the Bundelkhand craton region. All the parameters that affect the groundwater occurrence in this region, such as altitude, aspect, distance to drainage, distance to road, distance to faults, geomorphology, distance to river, lithology, normalized difference vegetation index (NDVI), rainfall, slope, soil, drainage density, land use land cover (LULC), topographic wetness index (TWI), lineament density and curvatures have been prepared by the remote sensing data as well as data taken from different departments and organizations. The training and testing dataset is generated from the groundwater potential zones map prepared through the frequency ratio (FR) technique. A total of 23,917-pixel locations have been selected in the research region. These locations contain the location of groundwater points and non-groundwater points equally. These points were randomly portioned into 70:30 for training and testing the model, respectively. The 2417 unknown points were taken from the study area and given to the trained model to predict the groundwater prospective areas. The maps of groundwater potential zones obtained using machine learning models were categorized into five classes: very low, low, moderate, high, and very high. The outcome of the employed algorithm is validated through the well discharge data and area under the receiver operating characteristics curve (AUC-ROC) method. The developed model's outcome gives valuable information regarding the effective management of groundwater in a particular region to government agencies and private sectors.