Selecting components in a probabilistic hydrological forecasting chain: the benefits of an integrated evaluation
Joseph Bellier, Guillaume Bontron, Isabella Zin
Abstract
Probabilistic streamflow forecasts are commonly issued by propagating meteorological ensemble forecasts into a hydrological model. However, a variety of components can be added to the chain to improve forecast skill. For operational centres, it may be easier to follow a modular strategy, and develop these components incrementally. However, only an integrated evaluation, based on the final streamflow forecasts, can ensure all components remain beneficial whatever the chain configuration. Also, it provides helpful guidelines about which developments to prioritise. Here, we consider two ensemble-based extensions, a meteorological grand ensemble and a hydrological multi-model, and two statistical corrections, a pre-processing and a post-processing. These components are opposed to their simpler alternative, and all configurations are evaluated on a common testbed comprising six catchments in the French upper Rhone River. Results show that all four components remain systematically beneficial, thereby validating the modular development strategy. Also, unlike other studies, neither the effect of the pre-processing nor that of the grand ensemble is found to vanish with hydrological modelling, which disputes the idea that all limitations in the input meteorological ensemble forcing can be rectified by a post-processing. However, only the post-processing can ensure the reliability of the streamflow forecasts; hence, it should be systematically prioritised.