COMPARISON OF 2 PROCEDURES FOR IMPROVING SHORT-TERM FORECASTS OF THE NATURAL FLOWS IN A DETERMINISTIC MODEL

The parameter values of a conceptual rainfall-runoff model must be est imated so as to represent, as faithfully as possible, a drainage basin under study, to reproduce its natural flows. The parameters are usual ly evaluated to reproduce the mean flow conditions of the basin. On a daily basis, however, the conditions of a basin at any given time are rarely equivalent to the basin's mean conditions, which introduces a b ias not taken into account by the parameters. This bias affects, in pa rticular, the quality of the short-term natural flow predictions. Two procedures, the objective of which is to improve the forecasts of a co nceptual deterministic model, are therefore described and compared in this work. The first proposes using the Kalman filter to update some p arameters of the conceptual deterministic model on a daily basis as a function of the most recent natural flow observed on a basin. As the m ain element developed in this paper, this procedure allows better cons ideration of the random component in modeling natural flows. The secon d procedure, currently used by Alcan, corrects the forecast results of the deterministic model. Lac-Saint-Jean (Quebec, Canada) basin data a re used to test these two procedures. The results indicate that the co rrection procedure used by Alcan performs better than the one using th e Kalman filter. Nevertheless, the correction procedure using the Kalm an filter behaves coherently in that it varies the parameters of the m odel at periods during which they are normally required to. Such behav ior confirms that the principle of combining deterministic tools with others of a probabilistic nature remains a good one, and would be wort h exploring in greater depth for modeling natural flows in future deve lopments.