Determination of groundwater potential using ensemble machine learning models in GIS (Case Study: Birjand plain)

Predicting the potential of groundwater is very important for the systematic development and planning of water resources. The main purpose of this study was to develop ensemble machine learning models including random forest (RF), logistic regression (LR) and Naïve Bayes (NB) by random subspace Classifier (RS) algorithm to predict groundwater potential areas in Birjand plain. Therefore, for implementation, geo-hydrological data of 37 groundwater wells (Number of wells, location of wells and groundwater level or Water table) and 17 hydrology, topographic, geological and environmental criteria were used. The least squares support vector machine (LSSVM) feature selection method used to determine the effective criteria to increase the performance of machine learning algorithms. Finally, groundwater potential prediction maps were prepared using RF-RS, LR-RS and NB-RS models. The performance of these models evaluated using the area under the curve (AUC) and other statistical indicators. The results showed that the RF-RS hybrid model (AUC = 0.867) has a very high predictability for groundwater potential in the study area. It was also found that the elevation criterion is most important in predicting the groundwater potential in the study area. The results of the present study can be useful for making appropriate decisions and planning regarding the optimal use of groundwater resources.