Artificial Neural Network models for disaggregation of monsoon season runoff series for a hilly watershed

Abstract

This paper presents the application of artificial neural networks for disaggregating of monsoon season comprising of June, July, August and September (JJAS) months of the year runoff series and is illustrated by an application to model the river flow of Naula watershed of Ramganga river in Uttrakhand state, India. Two different seasonal ANN models were developed. In the first ANN model neural network was trained with monsoon season comprising of June, July August and September (JJAS) months of the year runoff as input and runoff (corresponding to 4 months, JJAS) of June, July, August and September month output. The ANN architecture 1-1-1-4 was found best in training and testing. In the second ANN model most active monsoon months (July and August) total runoff and relatively less active months (June and September) total runoff was used as input and runoff of July and August; June and September runoff, respectively as output. The ANN architecture 1-4-4-4-2 for June-September total and 1-4-4-2 for July-August total was found best in training and testing. Lower value of RMSE and higher value of CE was obtained for all the months for the selected architecture. The correlation coefficient between measured and simulated data series are 0.92 and 0.75 for training and testing, respectively for model 1 whereas for model 2, the values are 0.99 and 0.87 in July-August and 0.97 and 0.89 in June-September for training and testing, respectively.