1Assistant Professor, Soil & Water Engineering, FAE, IGKV, Raipur; Email: irapsugarcane@rediffmail.com
2Dean, Faculty of Agricultural Engineering, IGKV, Raipur
3Scientist ‘F’, National Institute of Hydrology, Roorkee
4Professor, Soil & Water Engineering, FAE, IGKV, Raipur
Modelling rainfall-runoff transformation is essential for several hydrological and water management studies. A rainfall-runoff model was developed for the Upper Kharun catchment (2511 km2) in Chhatisgarh state, based on multi-layer perceptron (MLP) artificial neural network (ANN) trained with Baysian Regularization back propagation algorithm. The daily data for the years 1990–2009 were divided into two sets for model training (1990–2004) and for testing (2005–2009). The best geometry of the ANN rainfall-runoff model was identified in terms of number of hidden nodes through a performance evaluation in both training and testing dataset. The mean areal precipitation over the catchment estimated through Thiessen polygons was a main input to the model. The results of the MLP model were compared with multiple linear regression (MLR) model for the catchment. The results showed that the MLP ANN technique has great potential in simulating the rainfall-runoff transformation process.
Multi layer perceptron, artificial neural networks, rainfall-runoff models, hidden layers, model training
