Water Quality Assessment Using Artificial Intelligence Techniques: SOM and ANN-A Case Study of Melen River Turkey

Abstract

Artificial intelligence methods have been employed with regard to 26 sets of physical and chemical pollution data obtained from the Melen River by the Turkish State Hydraulic Works during the period of 1995-2006. Water-quality data are divided into two parts relating to the high- and low-flow periods for the 1 KMP, 2 BMP, and 3 BMA stations. The self organizing map-artificial neural networks (SOM-ANNs) is employed to evaluate the high-low flow period correlations in terms of water-quality parameters. This is done in order to extract the most important parameters in assessing high-low flow period variations in terms of river water quality. The map size chosen is 9 x 9 in order to ensure that the maximum number of groups would be obtained from the training data. The groups explaining the pollution sources are identified as being responsible for the data structure at each dataset. The SOM, supported by ANN, is applied to provide a nonlinear relationship between input variables and output variables in order to determine the most significant parameters in each group. The multilayer feed-forward NN is chosen for this study. The most crucial parameters are determined, and the groups are conditionally named as mineral structure; soil structure and erosion; domestic, municipal, and industrial effluents; agricultural activity waste-disposal sites; and seasonal effects factors. Based on the explanation of the parameters, we can have an opinion about other parameters which can lead to cost and time savings. The aim of this study is to illustrate the usefulness of artificial intelligence for the evaluation of complex data in river- and water-quality assessment identification, and pollution sources, for effective water-quality management.