Cascade Correlation Artificial Neural Networks for Estimating Missing Monthly Values of Water Quality Parameters in Rivers

Abstract

Three layer cascade correlation artificial neural network (CCANN) models have been developed for the prediction of monthly values of some water quality parameters in rivers by using monthly values of other existing water quality parameters as input variables. The monthly data of some water quality parameters and discharge, for the time period 1980–1994, of Axios river, at a station near the Greek – FYROM borders and for the time period 1980–1990, of Strymon river, at a station near the Greek – Bulgarian borders, were selected for this study. The training of CCANN models was achieved by the cascade correlation algorithm which is a feed-forward and supervised algorithm. Kalman's learning rule was used to modify the artificial neural network weights. The choice of the input variables introduced to the input layer was based on the stepwise approach. The number of nodes in the hidden layer was determined based on the maximum value of the correlation coefficient. The final network arhitecture and geometry were tested to avoid over-fitting. The selected CCANN models gave very good results for both rivers and seem promising to be applicable for the estimation of missing monthly values of water quality parameters in rivers.