Seamless Prediction @BafG: Probabilistic medium-range and seasonal hydrological forecast products for the German Waterways

Contributed by Bastian Klein and Dennis Meissner

Since the 1980s the Federal Institute of Hydrology (BfG) develops hydrological forecasting systems (WAVOS = Water-Level Forecasting System) and offers operational forecasting services focussing on navigation on the Federal Waterways in Germany.

Besides BfG provides and maintains flood forecasting models used operationally by German states being in charge for flood protection and risk management.

By setting up the Research & Development project “Seamless Prediction @BafG” in 2012 the BfG meets the challenge to provide their forecast-customers with improved and enhanced real-time products containing information on predictive uncertainties.

Furthermore BfG intends to act as a link between research and practical use by showing that lots of promising methods related to probabilistic forecasting could be implemented successfully in operational systems.

Project partners are the Institute of Hydrology, Water Resources Management and Environmental Engineering at the Ruhr-University Bochum (Task Hydrological Model Uncertainty) and the Institute of Applied Mathematics at the University of Heidelberg (Task Probabilistic Quantification of Streamflow Ensembles). The project is structured in five work packages (see Fig.1).

The aim of the project is the:

• application and development of advanced methods in operational forecasting,
• optimization and extension of operational models,
• stepwise closing the gap between short / medium range forecasts and climate projections by developing seasonal streamflow forecasts.

Dealing with probability and uncertainty in general becomes even more essential when aiming at seasonal hydrologic predictions in addition to the current short to medium range forecasts. Therefore the “seamless idea” of the project refers to the methods as well as to the forecast lead times.

Synergetic effects aren’t yet utilized. Dealing with forecast lead times from one day up to the seasonal scale of six months all available information and methods have to be used to generate reliable probabilistic forecasts. Hence in the analysis framework of the project (see Figure 2)  aspects of ensemble techniques, data assimilation and statistical post processing are considered.

To quantify the sources of meteorological uncertainty different available meteorological ensemble products with different lead times (e.g. SRNWP-PEPS, COSMO-LEPS, ECMWF-VarEPS) are used, analyzed and applied.

Data assimilation, in this context, refers to incorporating all available information to improve the forecast quality. E.g. real-time observations (e.g. runoff, snow, soil moisture) are used to derive the optimal initial state variables of the hydrologic model at the beginning of the forecast range. Hydrologic models in this forecast framework covers different types of deterministic models applied by hydrologists, mainly rainfall-runoff and hydrodynamic models.

Statistical post-processing of hydrological ensemble forecasts is required to obtain probabilistic forecasts of the parameter of interest (e.g. runoff, water-level). Operational forecasts, even if the models are parameterized and updated as good as possible, often show errors of unknown source that degrade the quality of the forecasts. Statistical Error Correction serves as a last chance to improve forecast accuracy.

In the next step the application of probabilistic methods such as Bayesian Model Averaging or Ensemble Model Output Statistics result in the probability distribution of the predictive uncertainty.

The resulting forecasts and their uncertainty have to be communicated to the different end-user groups in a proper manner. This essential step in the operational forecast chain, often neglected in hydrological studies, needs an intense communication and exchange with potential end-user groups as well as other scientific disciplines such as social studies, media research and communication sciences.

The study areas of the project are the German federal waterways (see Figure 3) with special focus on the River Rhine basin being one of the most important inland waterways in Europe.

In the context of HEPEX several experiments in the fields Data Assimilation, and Ensemble Post-processing are planned in the next year.