A METHOD OF ESTIMATING OPTIMAL CATCHMENT MODEL PARAMETERS

A review of a calibration method developed earlier (Ibrahim and Liong, 1992) is presented. The method generates optimal values for single ev ents. It entails randomizing the calibration parameters over bounds su ch that a system response under consideration is bounded. Within the b ounds, which are narrow and generated automatically, explicit response surface representation of the response is obtained using experimental design techniques and regression analysis. The optimal values are obt ained by searching on the response surface for a point at which the pr edicted response is equal to the measured response and the value of th e joint probability density function at that point in a transformed sp ace is the highest. The method is demonstrated on a catchment in Singa pore. The issue of global optimal values is addressed by applying the method on wider bounds. The results indicate that the optimal values a rising from the narrow set of bounds are, indeed, global. Improvements which are designed to achieve comparably accurate estimates but with less expense are introduced. A linear response surface model is used. Two approximations of the model are studied. The first is to fit the m odel using data points generated from simple Monte Carlo simulation; t he second is to approximate the model by a Taylor series expansion. Ve ry good results are obtained from both approximations. Two methods of obtaining a single estimate from the individual event's estimates of t he parameters are presented. The simulated and measured hydrographs of four verification storms using these estimates compare quite well.