Runoff forecasting artificial neural network for an idealized catchment

• Main Author: Muhammad Shafi Bueari.
• Other Authors: Chua Hock Chye Lloyd
• Format: Final Year Project
• Language: English
• Published: 2010
• Subjects: DRNTU::Engineering::Civil engineering::Water resources
• Online Access: http://hdl.handle.net/10356/40940
• Institution: Nanyang Technological University

A study investigating the forecast of runoff for an overland flow using the artificial neural network was carried out. Data from an experiment which includes rainfall data collected from 6 October 2002, 14 October 2002, 27 October 2002, 3 November 2002, 13 November 2002, 17 November 2002, 18 November 2002, 22 November 2002 and 5 December 2002, 7 December 2002 was analyzed. In this study, a total of six ANN models were used. They are Rt_Rt-8, Rt_Rt-8Qt, Rt-4_Rt-6, Rt-4_Rt-6Qt, Qt and Qt- 1_Qt. Feed forward neural network with back-propagation algorithm was selected as the modeling tool. MATLAB will be used to run the artificial neural network. Different sets of ANN models were used and their accuracy was based on the comparison of Nash‐Sutcliffe efficiency (NS), R2, mean absolute error(MAE) and root mean square error (RMSE). Results from the ANN model were compared with Constant model and Autoregressive moving average (ARMA). It has been found that among all the models used in this study, ANN models show a good generalization of rainfall-runoff relationship and is better than the other two models that have been used in this study. Upon comparing the NS, R2, MAE, RMSE and the number of shift, Rt_Rt-8 and Rt_Rt-8Qt shows that it is the best rainfall runoff model for this study. Rt_Rt-8 will be useful to forecast early lead time and Rt_Rt-8Qt will be useful to forecast at lead time greater than Qt+4. ANN models are still preferred in the forecasting of rainfall runoff in this study due to its ability to train and learn the data inputs much efficiently as compared to ARMA and Constant models.