Abstract:
Daily river flow forecasts are essential for water resources planning and management. The purpose of the study is to demonstrate that Neural Networks can be successfully applied to flood forecasting. The ANNs are recognized as being universal approximators and are capable of extracting the underlying relationship between any input and its subsequent output. The bootstrap method is a computational procedure that uses intensive resampling with replacement to reduce uncertainties. The present study makes an attempt to improve the credibility of the data-driven models among researchers and practitioners. The Study Area, Kulsi River Basin, is a part of the Brahmaputra sub-basin is situated on the south bank of the mighty River Brahmaputra. The basin was identified as Pilot Basin for the Centre for Flood Management Studies, Guwahati under XII Plan.This sub-basin spreads in the Kamrup District of Assam as well as west Khasi hills and Ribhoi district of Meghalaya. The river Kulsi drains out a total area of 2806 sq. km. The present research is undertaken with the following objectives: (i) To explore the potential of flood forecasting using ANN in Kulsi River Basin &; (ii) To quantify the uncertainty in flood forecasts using bootstrap technique.
The results of the study indicate that BANN models were found to replicate the rise and fall of observed discharge with considerable accuracy. These results show the applicability of BANN models in forecasting floods when limited data is available. Best performance was observed for model with one day lagged rainfall values as input which shows the significance of adding lagged rainfall as input in the BANN model and finally RMSE and MAE quantified the uncertainty in model predictions
