Hydrological post-processing of streamflow forecasts issued from multimodel ensemble prediction systems

• Published in: Journal of hydrology (Amsterdam) Vol. 578; p. 124002
• Main Authors: Xu, Jing, Anctil, François, Boucher, Marie-Amélie
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2019
• Subjects: Bayesian model averaging (BMA); Bayesian theory; hydrologic models; Hydrological post-processing; Hydrometeorological ensemble forecasts; prediction; stream flow; Uncertainty; watersheds; Uncertainty; Bayesian model averaging (BMA); Hydrometeorological ensemble forecasts; Hydrological post-processing
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2019.124002

Main flowchart of the study. [Display omitted] •Using Bayesian model averaging to post-process multimodel hydrological forecasts.•BMA improves over raw multimode forecasts for longer forecasting horizons.•Maintain performance and reliability when partial sources of uncertainties included.•Compensate for lack of forecast reliability caused by biased weather forecasts.•Provide users with flexible forecasting tools choices building a multimodel HEPS. Hydrological simulations and forecasts are subject to various sources of uncertainties. Thiboult et al. (2016) constructed a 50,000-member great ensemble that ultimately accounts for meteorological forcing uncertainty, initial condition uncertainty, and structural uncertainty. This large 50,000-member ensemble can also be separated into sub-components to untangle the three main sources of uncertainties mentioned above. However, in Thiboult et al. (2016) model outputs were simply pooled together, considering equiprobable members. This paper studies the use of Bayesian model averaging (BMA) to post-process multimodel hydrological forecasts. BMA assigns multiple sets of weights on different models and may then generate more skillful and reliable probabilistic forecasts. BMA weights explicitly quantify the level of confidence one can have regarding each candidate hydrological model and lead to a predictive probabilistic density function (PDF) containing information about uncertainty. The BMA scheme improves the overall quality of forecasts mainly by maintaining the ensemble dispersion with the lead time. It also has the ability to improve the reliability and skill of multimodel systems that only include two sources of uncertainties that the 50,000-member great ensemble using all forecasting tools (i.e., multimodel, EnKF, and meteorological ensemble forcing) could predict jointly. Furthermore, Thiboult et al. (2016) showed that the meteorological forecasts they used were somehow biased and unreliable on some catchments. The BMA scheme is capable to improve the accuracy and reliability of the hydrological forecasts in that case as well.