COMPARISON OF DAILY STREAMFLOW ESTIMATION TECHNIQUES

The aim of the first part of our research, described in this paper, was to compare daily streamflow estimation techniques and models. A general application software named DebEst was developed for the purpose. The Saint-Francois River basin was used as a physical test area because of the availability of several hydrometric stations in this region. All techniques and models used gave good results. However, principal component analysis and multiple regression applied to deterministic models gave better results than ARIMA models. The least square recursive algorithm was more flexible than the other techniques, although discrepancies sometimes appeared because of incorrect weighting of measuring and modelling noise. Results improved significantly when seasonal models were used and when the variation of parameters was taken into account as a function of flow. All techniques described yielded autocorrelated residuals, at least for the first three time lags. The amplitude of the residual autocorrelation function was reduced by seasonal models although it still remained high. In the second part of our research, the Kalman filter technique will be used in conjunction with the methods described above to extract residual information and generate truly independent residuals.