Prediction of qualitative parameters concentration in the groundwater resources using the Bayesian approach

Abstract

The instruments that help decision-makers for evaluating the quality of groundwater resources are necessary. This study introduces a Bayesian network to simulate and predict the nitrate and TDS concentrations as two quality parameters affecting water consumption. For this purpose, various quantitative and qualitative predictor variables were evaluated using different scenarios. For simulation the TDS, the predictor variables were determined with the correlation coefficient (R2) of 0.86 which are sodium chloride, chloride, drawdown, return flow, and precipitation. Nitrate was simulated using concentrations of sulfate, calcium, hardness, electric conductivity, drawdown, aquifer recharge, return flow from consumption, groundwater exploitation, and precipitation according to an R2 of 0.81. Considering the role of land use in the quality of groundwater resources, two zoning methods were used to evaluate the accuracy of zoning. IDW and Kriging methods were used to estimate the prediction results using Bayesian network. Results showed that the Kriging method was a suitable zoning technique for both TDS and nitrate with R2 values of 0.81 and 0.76, and mean squared errors of 43.1 and 2.7, respectively. Combination of the prediction results with qualitative zoning can significantly contribute to the evaluation of land development and improvement.