Simulation of Rainfall-Runoff Hydrograph Considering the Rainfall Temporal Pattern Using Artificial Neural Network in Kasilian Reagent Basin

Rainfall-runoff process is a non-linearity and complex phenomenon in hydrology. Comprehensive models are widely implemented for rainfall-runoff modeling. However, these models require a large number of detail information and their application are limited to just field scale studies. In case of lack of detailed data, black boxes model like artificial neural networks can be implemented to model the complex and nonlinearity relationships. To simulate the rainfall-runoff hydrograph in the kasilian basin a multi layer perceptron design of artificial neural network with architecture 9-10-7 was implemented. First precipitation data was divided into four groups to consider the temporal pattern of rainfall. For each group, rainfall distribution in different time quartiles, base flow of hydrograph, total depth of rainfall, and rainfall depth until time of concentration, rainfall duration and antecedent precipitation index was derived and inserted to the ANN model as an input parameters. The output of ANN model consists of peak discharge and its occurrence time, hydrograph base time, time of the 50 and 75 percent of peak discharge occurrence and hydrograph widths corresponding to these discharges. ANN model was executed for the different groups of data using the various activation functions in hidden and output layers. The results indicated that there is a strong correlation between model outputs and measured data. Correlations varies from 0.9107 (RMSE= 0.0882) for the first group to 0.99 (RMSE=0.0678) for the fourth group. These strong correlations confirm that in case of lack of detailed data, ANN model can be used for simulation of hydrological parameters.