The Artificial Neural Network (ANN) has been used as a predictive tool for the estimation of certain parameters of Subernarekha, an important river in the Jharkhand state in India. The network used two algorithms for this purpose and was sufficiently accurate in predicting the most economically heavy and time-consuming set of data. The Levenberg-Marquardt Backpropagation Network (trainlm) and the Resilient Backpropagation Network (trainrp) were the two algorithms used for the estimation of metallic species and physicochemical parameters of the river. The MAPE for metallic species were found to be 0.71 for cadmium, 0.182 for copper and 0.771 for chromium, while physicochemical parameters were 16.645 for alkalinity, 5.883 for dissolved oxygen (DO) and 23.28 for chemical oxygen demand (COD). Both the algorithms were used with different sets of hidden layers i.e., one for trainlm with 5 neurons and three for trainrp with 5, 4 and 5 neurons, and these were determined using the trial and error method. This method is not only economically favorable but time-adaptive as well, as it depends on the amount and span of data available and it can predict values to a very high degree of accuracy. This method can successfully be employed for the prediction of parameters of any river system with confidence.
©2011 Walter de Gruyter GmbH & Co. KG, Berlin/Boston
