A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering, School of Environment, Resources and Development

Thesis (M.Eng.) - Asian Institute of Technology, 1993

The present study explores the applicability of Back propagation (Neural network) model to the problem of daily flow forecasting which is of great importance from the viewpoint of water resources development and planning. Discharge and rainfall data for Nan River Basin, in the periods of 1985-1988 and 1989 were used in calibration and verification respectively. Multiple Linear Regression model is also used for the same data set for comparison purposes. From a careful comparative study, it was found that back propagation results in better forecasts compared to the multiple linear regression model. Several conclusions can be drawn from this study. Firstly the rainfall data alone could not be adequate in forecasting river flow in the Nan River Basin. The forecasts were significantly improved when discharge data at the station under consideration were included in the rainfall data. It was found that discharge at the station under consideration is a function of past discharges at that station, discharges at the upstream stations and rainfall at the upstream stations. The regression model is simple while the back propagation model is more complicated and time consuming.