National flood forecasting for Germany

A new online flood-forecasting system service for Germany came into operation in January 2008. At present, flow forecasts are available at about 200 locations on the rivers Danube, Elbe, Rhine, Oder and Weser as well as their major tributaries.

The forecasts are on a regional level but can be extended to cover local areas to obtain more detailed water level and discharge information.

Forecasts are provided around the clock for the following 48 hours, and are updated each hour based on readily available metrological and hydrological data. This provides more accurate and reliable forecasts than before.

The forecasts use DHI’s suite of MIKE 11 river-modelling tools. Metrological data in the form of precipitation and temperature is provided by the more than 1,500 observation stations and from a 4km grid-based model operated by Meteomedia.

Other hydrological data is retrieved by a WebCrawler, which scans the various sources of data, such as water level and discharge from other publicly available sources. The data is processed and compiled into time-series’, which feed the forecast model. About one gigabyte of data is sampled hourly to make the forecasts.

Sampling data from so many resources is quite a challenge, and this has been met through close cooperation with the German metrological company Meteomedia.

The MIKE 11 model set-up is one of the largest and most complex river models developed for this purpose. The river model set-up covers 15,000km of river with 100 branches in the entire river basin areas and 585 sub-catchments.

The catchments typically cover 500-1,000km2. This is the first time a complete regional model of the main German rivers, including the upstream catchment areas outside Germany, has been developed.

It takes 20-25 minutes to gather and process the necessary data, while the actual forecast on average requires 15 minutes of computational time. A further 10 minutes is required to disseminate the forecasts.

Forecasts are disseminated via websites, with public access, subscription access or as dedicated and tailor-made sites, specifically developed for the end-users’ requirements. Forecasts are presented as time-series’ at selected locations on the rivers and as a coloured river status map showing the flood status of the river branches.

The entire data-sampling and forecasting system is fully automatic and will prepare for the next forecast to be made shortly after a forecast has been made available.

In the forecasting period, the model uses hourly input from a

4km grid-based meteorological model. Furthermore to obtain the best performance at time of forecast, the model uses a data assimilation routine, where available real-time flow information is used by hydrodynamic model to ensure high accuracy.

Part of the MIKE river model is the rainfall run-off module, which is a lumped, conceptual rainfall run-off model. This simulates the overland flow, the interflow and the base flow components based on input of rainfall, temperature and evaporation.

The module can either be applied independently or used to represent one or more contributing catchments that generate lateral inflows to a river network. In this manner, a single catchment or a large basin is schematised as an ensemble of many catchments connected with a complex network of rivers and channels.

The model describes the behaviour of each individual component in the hydrological cycle, at catchment level, by using a group of interconnected conceptual elements. In particular, it simulates the precipitation run-off process by continuously accounting the moisture content in four storages. These represent physical elements of the catchment: snow storage, surface storage, root zone storage, and groundwater storage.

Based on the meteorological input data, the model produces catchment run-off as well as other components of the hydrological cycle, such as actual evapotranspiration, soil moisture content, and groundwater level. The resulting catchment run-off is split conceptually into overland flow, interflow and base flow components.

Combining this with the forecasting module, the model provides real-time data management, allowing automatic updating of the analysis and correcting for differences in observed and computed hydrographs. Therefore, the module serves two purposes, i.e. forecasting and updating.

The flood-forecasting component predicts the variation in discharges and water levels in a river system as a result of catchment rainfall and inflow/outflow through boundaries in the river system. Updating consists of conditioning the model predictions to the observed data until the time of forecast.

The standard updating procedure provided is an error correction routine capable of distinguishing between phase error and amplitude error.

By adopting this approach, accurate and reliable forecasts can be made based on the available data. This is further achieved by updating the forecasts on an hourly basis. As with any forecast, the accuracy is highest the closer to the time of prediction.

The new forecasting service not only provides forecasts of water levels and discharges but also enables a number of important parameters to be forecasted, such as those used for decision support systems. The forecasting service now available has received considerable interest and is expected to be exploited for a number of different purposes in the near future.