No abstract

Abstract. Seeking more accuracy and reliability, the hydrometeorological community has developed several tools to decipher the different sources of uncertainty in relevant modeling processes. Among them, the ensemble Kalman filter (EnKF), multimodel approaches and meteorological ensemble forecasting proved to have the capability to improve upon deterministic hydrological forecast. This study aims to untangle the sources of uncertainty by studying the combination of these tools and assessing their respective contribution to the overall forecast quality. Each of these components is able to capture a certain aspect of the total uncertainty and improve the forecast at different stages in the forecasting process by using different means. Their combination outperforms any of the tools used solely. The EnKF is shown to contribute largely to the ensemble accuracy and dispersion, indicating that the initial conditions uncertainty is dominant. However, it fails to maintain the required dispersion throughout the entire forecast horizon and needs to be supported by a multimodel approach to take into account structural uncertainty. Moreover, the multimodel approach contributes to improving the general forecasting performance and prevents this performance from falling into the model selection pitfall since models differ strongly in their ability. Finally, the use of probabilistic meteorological forcing was found to contribute mostly to long lead time reliability. Particular attention needs to be paid to the combination of the tools, especially in the EnKF tuning to avoid overlapping in error deciphering.

No abstract

Abstract. Seeking for more accuracy and reliability, the hydrometeorological community has developed several tools to decipher the different sources of uncertainty in relevant modeling processes. Among them, the Ensemble Kalman Filter, multimodel approaches and meteorological ensemble forecasting proved to have the capability to improve upon deterministic hydrological forecast. This study aims at untangling the sources of uncertainty by studying the combination of these tools and assessing their contribution to the overall forecast quality. Each of these components is able to capture a certain aspect of the total uncertainty and improve the forecast at different stage in the forecasting process by using different means. Their combination outperforms any of the tool used solely. The EnKF is shown to contribute largely to the ensemble accuracy and dispersion, indicating that the initial condition uncertainty is dominant. However, it fails to maintain the required dispersion throughout the entire forecast horizon and needs to be supported by a multimodel approach to take into account structural uncertainty. Moreover, the multimodel approach contributes to improve the general forecasting performance and prevents from falling into the model selection pitfall since models differ strongly in their ability. Finally, the use of probabilistic meteorological forcing was found to contribute mostly to long lead time reliability. Particular attention needs to be paid to the combination of the tools, especially in the Ensemble Kalman Filter tuning to avoid overlapping in error deciphering.

The increasing demand for uncertainty assessment in streamflow forecasts has drawn the hydrological community's interest toward ensemble forecasting techniques. The widespread deterministic hydrological forecasting point of view focuses to a great extent on the search for a hydrological model that would come as close as possible to "perfection" (i.e. the aim is to implement a model that produces a point forecast that is as close as possible as the observed outcome). On the other hand, ensemble forecasting departs from the deterministic point of view by avoiding the assumption that the "perfect" model exists and instead focuses on issuing a type of forecast that accounts explicitly for the uncertainty inherent to the forecasting process as a whole. In this paper, one-day-ahead hydrological ensemble forecasts obtained by stacked neural networks are presented and analysed. To do so, three simple performance assessment criteria are presented. Those criteria were originally developed in the meteorological and statistical communities to accommodate the need for a quality assessment methodology that is coherent with the probabilistic nature of ensemble weather forecasts. It will be shown that, even though the ensemble forecasts suffer from underdispersion, they outperform point forecasts.