Performance Evaluation of Artificial Neural Network in Predicting Dissolved Oxygen and Total Phosphorus Concentrations in the Ilam Dam Catchment

In this research, a multi-layer perceptron neural network (MLP-NN) model was used to simulate dissolved oxygen and total phosphorus in the Ilam Dam catchment. The NN model was developed using experimental data from three sub- catchments of Ilam Dam during the 2009-2010 period. Input variables of NN model including water PH, electricital conductivity, total suspended solids, temperature, total phosphorus, sulfate, ammonia, iron and total nitrogen were employed for the dissolved oxygen modeling. Input variables for total phosphorus modeling were temperature and phosphate, which have been measured at one station on the reservoir vicinity at different depths. The performance of the models was assessed by the coefficient of determination, relative error, and sum of squared errors (SSE) indices. The neural network results indicated that all the input variables affected modeling of the dissolved oxygen concentration, while the most effective parameter was the total suspended solids. Phosphate presented the highest impact on the total phosphorus modeling than the temperature. The cofficient of determination values for modeling dissolved oxygen and total phosphorus were 0.813 and 0.940, respectively. The results of NN models were then compared with the two-dimensional, laterally averaged CE-QUAL-W2 model. Based on the results, the multi-layer perceptron model showed more accurate responses than the numerical model in predicting water quality variables. Results also showed that the NN was able to predict the eutrophication process with acceptable accuracy and could be used as a valuable tool for qualitative management of reservoir water.