Assessment of Intelligent models for Estimating the Electrical Conductivity in Groundwater (Case study: Mazandaran plain)

Groundwater resources along with surface water supply the needs for municipal, industrial and agriculture uses, and their quantity and quality should be investigated. Salinity is one of the most important parameters in assessing the quality of groundwater. In this study, application of artificial neural networks and Bayesian network in predicting the electrical conductivity in 8 observation wells in Mazandaran plain was investigated. For this purpose, hydrogen carbonate, chloride, sulfate, calcium and magnesium were selected as input and output parameters for electrical conductivity at monthly a scale during 2003-2013. The criteria of correlation coefficient, mean absolute error and Nash Sutcliff coefficient were used to evaluate the performance of the model. The results showed that artificial neural network model has the highest correlation coefficient (0.989), the lowest mean absolute error (0.019 ds/m) and the highest standard of Nash Sutcliffe (0.970) ranked the first priority in the validation phase. The results indicate acceptable capability of artificial neural network models to estimate the electrical conductivity of groundwater.