Uncertainty estimation of hydrological models using bayesian inference methods

Abstract

Intensive investigations of hydrologic model calibration during the last two decades have resulted in a reasonably good understanding of the issues involved in the process of estimating the numerous parameters employed by these codes. Nevertheless, these classical "batch" calibration approaches require substantial amounts of data to be stable, and the subsequent model forecasts do not usually represent the various imbedded uncertainties. Especially in the light of thousands of uncalibrated catchments in need of model simulations for streamflow predictions, a parameter estimation approach is required that is able to simultaneously perform model calibration and prediction without neglecting the substantial uncertainties in the computed forecasts. This thesis introduces the Bayesian Recursive Estimation scheme (BaRE), a method derived from Bayesian probability computation and adapted for the use in "on-line" hydrologic model calibration. The results of preliminary case studies are presented to illustrate the practicality of this simple and efficient approach.