RUNOFF ESTIMATION USING ARTIFICIAL NEURAL NETWORK

Abstract

The rainfall-runoff process is believed to be highly nonlinear, time-varying and spatially distributed. Therefore, sometimes it may not be possible to accurately predict the runoff using simple models. Recently, significant progress has been made in rainfall-runoff modeling using Artificial Neural Network (ANN). It has the ability to model the rainfall-runoff process using relatively small amount of data which is highly desirable for any basin in general and the basins of developing countries like India, in particular, for which scarcity of data is a major problem. ANN also provides a more systematic approach and shortens the time spent in model development. At the same time, it also provides better prediction accuracy and flexibility. In the present study, the performance of ANN has been evaluated in terms of prediction accuracy and efficiency. Standard Soil Conservation Service (SCS) model has been used to evaluate its performance. Various neural network architectures have been evaluated to assess their effect on the predication accuracy of runoff estimation. From these, an optimal neural network model has been identified. Besides, the use of ANN for runoff estimation, a neural network model has also been developed for the prediction of Curve Number (CN) value which is an important factor in SCS model for runoff estimation. The findings of the present work show that the ANN models have immense potential to predict more accurate runoff than SCS model. From the results, it has been found that ANN derived runoff values are more close to actual values in high discharge period than SCS model. Improvement in prediction accuracy has also been observed when the data size is iii increased, Similarly, CN values predicted by ANN model have also been found to be very close to experimental values. The model for CN value prediction has been used to estimate the CN values in Roorkee and surrounding areas by considering a land cover map produced from IRS-1B LISS II image. These CN values are then used to predict the runoff in the area selected. The results are found to be comparable with those obtained by SCS model.