A thesis submitted in partial fulfillment of the requirement for the degree of Master of Science, School of Engineering and Technology

Thesis (M.Sc.) - Asian Institute of Technology, 1988

In this thesis a system is developed for daily flow forecasting consisting o f five regression methods: - The Least Squares Method, - The Ridge Regression , - The Principal Component Regression, - The Stepwise Regression, a nd - The regression with minimum s um of absolute errors. The system was used to evaluate the performances of the regression techniques and of the following models: - The - The - The - The model involving discharge, rainfall and evaporation, mode l using only discharge and rainfall, Linear Perturbation Mode l, a nd Extended Linear Perturbation Model. From applications to data on two catchment areas , in Laos and Th ail and , it was found that: - The performances of the least squares method, the stepwise regression and the regression with minimum sum of absolute errors are almost the same. But the regression with mini mum sum of absolute errors requires mu c h more executing time. - The performance of the ridge regression is not as good as that of the least squares method in the case of nonsingular matrix . The ridge regression is recommended for use only when the least squares method cannot give the results because of the multicollinearity present among the predictor variables. - The performance of the principal component regression is a l so not as good as that of the least squares method. - The performances of the model involving discharge, rainfall and evaporation , the model using only discharge and rain fall, and the Extended Linear Perturbation Model are almost t he same. - The performance of the model using only discharge and rainfall is better than that of the Linear Perturbation Model.