Abstract:
In the present study, an effort has been made to develop statistical and ANN models for estimation of sodium concentration in pre-monsoon and post-monsoon seasons using routinely monitored water quality parameters of ground water wells in Jaipur district, Rajasthan (India). The Best Subset procedure based on R2 (coefficient of determination) and F (Fisher's test) values was used in model dissemination. It was found that electrical
conductivity, hardness, chloride, and sulphate could be used as surrogate parameters for the prediction of sodium. The model values of Na when compared with actual values (validation) showed a reasonably good matching. Further it is was noticed that there was not a single model which could be used to predict the Na levels. It is primarily attributed to the fact that sodium concentration not only varies from site to site but also varies from
season to season. Secondly, Principal component analysis was used to
__—
predict the dominating water quality constituents and it was revealed that four
principaLcomponents...are-accounted for the total chemical variability in the ground water quality for pre-monsoon season and three principal components for post-monsoon season, respectively. The common factors conductivity, fluoride, nitrate, alkalinity, and phosphate have perceptible influence on the quality of groundwater of Jaipur district, Rajasthan. Finally, Back Propagation, two layer feed forward ANN models for both pre-monsoon and post-monsoon season was developed for estimation of sodium using the steepest descent optimization technique. ANN models were developed considering a fixed number of iterations as 1000 and these were verified on the data not considered in calibration. The input variables considered for different model structures were identified through correlation analysis. Based on the statistical performance evaluation criteria such as root mean square error (RMSE), correlation coefficient (CC), and coefficient of efficiency (CE), the results indicated satisfactory performance of ANN based model.
