The usage of artificial neural networks in microbial water quality modeling: A case study from the Lake İznik

Abstract

The aim of this study was to develop faecal pollution model structures with artificial neural networks (ANNs) for cost-effective lake water quality management studies. In this study 5 artificial neural networks model structures were applied to predict the Faecal coliform concentrations for 4 different coast areas "Golluce, Inciralti, Darka, Orhangazi" and all data of the coasts in Lake Iznik-Turkey. The Levenberg-Marquardt and backpropagation algorithm was proposed for feed-forward neural networks training. According to performance functions root mean squared error (RMSE), neural network model structures provided acceptable results. Correlation values (R) were found between 0.590 and 0.999. Increasing the number of hidden layer in the model structures was not raised the model efficiency in each trial. Type and number of input parameters were more effective for some model efficiency. Increasing the number of hidden layer and inputs in the model structures did not raise the model efficiency in each trial. Because depending on the numbers and chemical compositions of the substrates in the lake water microorganism's metabolism and their growth rates could be influenced differently and the larger error values of the modeling results determined in Golluce and Orhangazi Coasts which influenced by pollution sources. Water quality modeling studies and increasing of monitoring would provide more productive results for protection and management of coastal.