Uncertainty propagation within a water level ensemble prediction system

Abstract

Reliability of a water level ensemble prediction system (WL-EPS) is directly related to the quality of the available data and the processes integrated into each predictive layer. In this project, we evaluate a hydrometeorological ensemble prediction system (H-EPS) coupled to a hydraulic module (HM) developed on the Chaudière River watershed in Quebec, Canada. Ensemble streamflow forecasts account for the three main sources of uncertainty: meteorology (ensemble forecasts), initial conditions (through data assimilation), and model structure (multimodel). The HM consists of a 1D river model that is informed at the boundary conditions by the H-EPS outputs. In order to evaluate the forecasting system in an operational context, a sampling procedure is implemented on the hydrometeorological ensemble forecasts in order to limit the simulation time of the hydraulic component without altering the description of the uncertainty provided beforehand. Forecasts reliability and accuracy are evaluated, for events of various magnitude covering a wide range of flows, based on performance metrics such as the reliability diagram and the continuous rank probability score (CRPS). Overall, results show that the performances of the ensemble flow and water level forecasts provided by the proposed system are quite reliable over the range of forecast horizons evaluated. Water level forecasts are dependent on the quality of the streamflow forecast, which is coherent with the assumption that streamflow is the main source of uncertainty of the HM. However, at control point where flow and water level forecasts are available, a slight deterioration in the forecast reliability is noted from flow to water level; nonetheless, the reliability of the forecasts remains comparable, thus confirming the strong link between the H-EPS and HM uncertainties. Consideration of the uncertainty related to the roughness coefficient uncertainty turned out to have a negligible effect on the spread of the water level forecasts, confirming once more the domination of the hydrologic uncertainty.