Prediction of Water Quality Parameters for Irrigation in Konkan Region using Artificial Neural Network Technique

Abstract

In the present study, artificial neural networks (ANN) were used to derive and develop models for prediction Kelly’s ratio (KR ), percent sodium (Percent Na), permeability index (PI ), residual sodium carbonate (RSC), sodium absorption ratio (SAR) and soluble sodium percentage (SSP) as groundwater quality parameters of Ratnagiri district by using post monsoon season groundwater quality parameters (Na, Mg, K, CaCO₃ and HCO₃) collected for time period 1999-2014 from Groundwater Surveys and Development Agency, Navi Mumbai as input variables. The ANN model was developed with multilayer feed forward back propagation (MLFBP) with sigmoid transfer function. While developing ANN model for different input parameters, three steps were followed as identification of model structures, to evaluate the performance and adopting model for forecasting. The model development data set was further divided into three subsets; training set, cross validation set and testing set in 70:15:15 proportions. The ANN models were developed for prediction KR, Percent Na, PI, RSC, SAR and SSP using neurosolutions. Performance of model was evaluated by statistical criteria which included correlation coefficient, root mean square error, index of agreement and mean bias error. The analysis revealed that the selected ANN based models had correlation coefficient > 0.95, RMSE < 0.3438, IA > 0.91 and MBE <-0.2450 for Model 3-2-1 to predict KR, Model 4-2-1 to predict percent Na, Model 4-6-1 to predict PI, Model 4-4-1 to predict RSC and Model 3-6-1 to predict SAR and SSP during post monsoon season. The results confirmed that developed ANN models were found suitable for prediction of water quality indicators used for irrigation purpose. It was recommended that two numbers of nodes in hidden layer can be used for modelling of KR and percent Na, four number of nodes for RSC and six numbers of nodes can be used for SAR, SSP and PI under limited data availability.