Application of Artificial Neural Network to Improve DRASTIC-Based Groundwater Vulnerability Assessment

Groundwater vulnerability assessment systems have been developed to achieve a suitable method for protecting groundwater resources from future contaminants. The DRASTIC model is the most well-known approach in groundwater vulnerability assessment. This model calculates the vulnerability index of the aquifer based on effective parameters of depth to the water level, net recharge, aquifer media, soil media, topography, unsaturated zone characteristics, and hydraulic conductivity. After ranking and integrating these seven layers, the groundwater vulnerability map was prepared. This study evaluated the susceptibility of the Ardabil plain aquifer using the DRASTIC method in GIS software. Also, the Artificial Neural Network (ANN) modeling has been developed to improve the results of the DRASTIC model and to predict the nitrate concentrations in plain as an indicator of pollution. The inputs and outputs of the ANN were seven parameters of the DRASTIC model and nitrate concentrations, respectively. To calibrate the ANN model, the inputs and output data were divided into two datasets for training and test purposes (75% for Training and 25% for test). This study provided both point and class predictions to assess groundwater vulnerability. Finally, used models were evaluated using the Root Mean Square Error (RMSE), Determination Coefficient (DC), and correlation coefficient (r) criteria (for point predictions) and Total Accuracy (TA) and Heidke Skill Score (HSS) criteria (for class predictions). According to the obtained results, the proposed ANN model improved the results of the conventional DRASTIC model and provided more accurate pollution predictions.