Estimation of the recharging rate of groundwater using random forest technique

Accurate knowledge of the recharging rate is essential for several groundwater-related studies and projects mainly in the water scarcity regions. In this study, a comparison between different methods of soft computing-based models was obtained in order to evaluate and select the most suitable and accurate method for predicting the recharging rate of groundwater, as the natural recharging rate of the groundwater is important in efficient groundwater resource management and aquifer recharge. Experimental data have been used to investigate the improved performance of Gaussian process (GP), M5P and random forest (RF)-based regression method and evaluate the potential of these techniques in the prediction of natural recharging rate. The study also compares the prediction of recharging rate to empirical (Kostiakov model, multilinear regression, multi-nonlinear regression) equations. The RF method was selected for the recharging rate prediction and was compared with the M5P tree, GP and also empirical models. While GP, M5P tree and empirical models provide good quality of prediction performance, RF model showed superiority among them with coefficient of correlation (R) values as 0.98 and 0.91 for training and testing, respectively. Out of 106 observations collected from laboratory experiments, 73 were used for developing different models, whereas rest 33 observations were used for the assessment of the models’ performance. Sensitivity analysis recommends that time parameter (t) is the main influencing parameter, which is crucial for the prediction of the recharging rate. RF-based model is suitable for accurate prediction of recharging rate of groundwater.