prediction of groundwater quality parameters in golestan province using response surface method, decision tree and neural network

Objective: groundwater quality is a main issue in most of the plains in iran. therefore, quality management and monitoring of water resources are of great importance. material and methods: in this study, water quality parameters, including sodium adsorption ratio (sar), total dissolved solids ratio (tds), and electrical conductivity (ec), were predicted using an artificial neural network (mlp type), decision tree model (m5tree), and response surface method (rsm). the quality data acquired from 96 observation wells located in golestan province were used for model inputs, including sodium, water ph, chloride, sulfate, calcium, and magnesium. models were evaluated utilizing three criteria: root mean square error (rmse), determination coefficient (r2), and mean absolute error (mae). three different input combinations were considered to predict ec, sar, and tds. results and discussion: the results of this study showed that the parameters na and cl have the greatest effect on the accuracy of the models. according to the results, the decision tree model (m5tree) was found to have the highest accuracy in predicting ec (r2 =0.997, rmse=23.351, mae=13.607) followed by the rsm (r2 =0.98, rmse=56.871, mae=33.0428) and ann (r2 =0.991, rmse=37.1073, mae=17.279) models. however, the rsm model has a higher efficiency than the other models in predicting the sar and tds. according to the results obtained, the rsm generally predicts the groundwater quality parameters with relatively better accuracy.conclusions: the results obtained from this research showed that the models having all input parameters offered the best accuracy in predicting ec, sar, and tds. also, the results showed that the na and cl parameters have the greatest effect on the accuracy of the prediction results for all three methods. groundwater quality is a main issue in most of the plains in iran. therefore, quality management and monitoring of water resources is of great importance. in this study, water quality parameters including sodium adsorption ratio (sar), total soluble solids ratio (tds) and electrical conductivity (ec) were predicted using artificial neural network (mlp), decision tree model (m5tree), and response surface method (rsm). the quality data acquired from 96 observation wells located in golestan province were used for model inputs are sodium, water ph, chloride, sulfate, calcium and magnesium. models were evaluated utilizing three criteria of root mean square error (rmse), detemination coefficient (r2) and mean absolute error (mae) were used. three different input combinations were considered to predict ec, sar and tds. the results of this study showed that the parameters na and cl have the greatest effect on the accuracy of the models. according to the results, the decision tree model (m5tree) was found to have the highest accuracy in predicting ec followed by the rsm and ann models. however, the rsm model has a higher efficiency than the other models in predicting the sar and tds. according to the obtained results, it can be said that the rsm in general predicts the groundwater quality parameters with relatively better accuracy. conclusions: the results obtained from this research showed that the models having all input parameters offered the best accuracy in predicting ec, sar, and tds. also, the results showed that the na and cl parameters have the greatest effect on the accuracy of the prediction results for all three methods.