First step towards a multi-model probabilistic flood forecasting system for Europe

Contributed by Anna Kauffeldt, Fredrik Wetterhall, Florian Pappenberger

The European Flood Awareness System (EFAS) provides flood information to national authorities around Europe. Warnings are based on a fully operational probabilistic forecasting system which addresses uncertainties in weather forecasts by using multiple weather forecasts (deterministic and ensembles) from several centres. These forecasts are used to force a hydrological model (LISFLOOD) and discharge predictions are compared to return flows determined from long-terms runs with observed data to issue alerts if certain threshold levels are reached.

This modelling framework allows for addressing some of the uncertainties in the weather forcing, but the use of one single hydrological model means that uncertainties introduced by model structure and parameters is neglected and therefore underestimated. Our use of models to describe the highly non-linear responses and complex interaction in the hydrosphere, whether simple conceptual or more complex process-based ones, inherently entail uncertainties in modelled storages and fluxes.

EFAS is now looking to expand the modelling framework to include an ensemble of hydrological models to address part of these uncertainties. However, it is not a straight-forward task to build such an ensemble and the first step is to find potential models that fit into such a framework. Therefore, a comprehensive review of large-scale models in the context of suitability of Pan-European operational hydrological forecasting is currently being undertaken. A total of 23 models were initially listed as potential candidates (see list below). These models were included because of their specific aims towards large-scale applications, since the model need to be applicable at continental scale for the EFAS project.

The suitability of the models for operational forecasting in Europe is assessed e.g. in terms of:

• process and flux descriptions important for flood modelling
• input data requirements and the availability on European scale
• spatial and temporal resolution
• availability of source code for tailoring to specific needs of the project

The work is currently being conducted at ECMWF through literature studies and communication with model developers. The list of models covers a wide range of levels of complexity and includes models that were originally developed in meteorological and the hydrological communities.

The study will render a short-list of models which fulfil the requirements of EFAS and can be used within its modelling framework. Ultimately, it is hoped that a multi-model approach will add value to the forecasting system by reducing the overconfidence that a one model approach might cause. We welcome suggestions of models that may have been overlooked or any other feedback in the comments, so please help us in our (impossible?) quest to create a full probabilistic hydrological forecasting system!