GIS-based groundwater potential mapping using machine learning methods in Busan, South Korea

Groundwater is a finite natural resource, and is a vital resource of drinking water, irrigated agriculture, and many ecosystems. It is essential to establish systematic plans for using groundwater. Groundwater potential map (GPM), designed to evaluate groundwater potential, is an efficient tool to maintain and manage groundwater. The goal of this study is mapping GPM in Busan, South Korea using geographic information system (GIS)-based datasets and two machine learning methods: random forest (RF) and support vector machine (SVM). The datasets consist of two parts; first datasets were constructed for groundwater well datasets as dependent variable, second datasets were constructed for groundwater influence factors as independent variable. Groundwater well datasets were collected 308 springs from extensive field surveys and governmental reports, 214 springs (70%) were used for model training, 94 springs (30%) were used for model validation. Groundwater influence factors consisted of 14 factors; 4 topographical factors (altitude, slope degree, slope aspect, and curvature), 4 hydrological factors (topographic wetness index, stream power index, distance from drainage, and drainage density), 5 geological factors (lithology, distance from fault, fault density, distance from lineament, and lineament density), and land cover factor. GPMs were constructed using RF and SVM models. The predicted results from RF and SVM models were validated using the receiver operating characteristics (ROC) curve. The accuracy assessment of GPMs for RF and SVM models was calculated using the area under the curve (AUC) and the AUC for RF and SVM models were calculated as 0.7441 and 0.7138. Therefore, RF model presents the higher prediction result compared to SVM model.