Forecasting Models for Some Water Quality Parameters of Shatt Al-Hilla River, Iraq


This paper provides Artificial Neural Networks model versions for forecasting the monthly averages of some chemical water quality parameters of Shatt Al-Hilla River, which is located at Hilla City, south of Iraq. The water quality parameters investigated were Sulphate, Magnesium, Calcium, Alkalinity, and Total Hardness. Results indicate that for Sulphate and Calcium high correlation coefficients models were observed to be (0.9 and 0.88), while for Magnesium, Alkalinity and Hardness low correlation coefficients model were observed to be (0.48,0.58, and 0.51) respectively. Serial correlation behavior of these variables indicate at that high lag time correlations sequences are observed for the first two variables and low ones for the last three water quality parameters. A serial correlation coefficient analysis was done and indicates that as the variable exhibited weak lag correlation structure, then a successful ANN forecasting model could not be obtained even if many trials were done to enhance it's performance, such as increasing the number of nodes, the lagged input variables, and/or changing the learning rate and the momentum term values, or the use of different types of activation functions. On the other hand, those variables that have a strong lag correlation structure can easily fit successful ANN forecasting models