Communication of Uncertainties in Practice – an example from Bavaria

Contributed by Alfons Vogelbacher

Hochwasser_Passau_2013-06-03Bavaria has been hit by extreme flooding in June 2013 with the highest water levels recorded in the border town of Passau  since 1501 . The Bavarian Flood Information service  began its life in 1902 and is now a collaborative venture involving four types of organisation: the Flood Information Centre at the Bavarian Environment Agency in Munich; five regional flood forecast centres; seventeen state offices for water management; and the local authorities. Only the local authorities are actually tasked with warning the public, whereas all  organisations contribute to information which is published on the internet, phone services and teletext.

Example of Probabilistic Flood Forecast issued by the flood Flood information Information centre Centre at the Bavarian Environment Agency in Munich

Bavaria began  to display uncertainties to the public in 2007. The uncertainty bands are determined for each gauging station by statistical analysis of the errors in previous forecasts. The output is an estimate of the 90%- and 10%-quantile of the expected flows and water levels at the forecast station. The operational execution of the forecasts is managed using bespoke software (HUGO) which also enables re-forecasting and scientific development. The complete ensemble  of flood forecast simulation results are made available to the regional water authorities, regional governments and ministries by a JavaClient application (FLIPPER) within the data network of the Bavarian water management administration.

The Flood Information Centre recently conducted a survey, investigating the response of the regional forecasters to uncertainty bands. Such inter expert communication is one of the key corner stones in enabling effective communication to the wider public and other end users. Overall the experts found the uncertainty bands helpful to their decision making:94% of all experts judge the uncertainty bounds as useful. Only 6% thought them to be not very helpful, but 15% felt that the uncertainty bounds created insecurity. 88% thought that the uncertainty bounds are particularly helpful in estimating the exceedance of possible flood thresholds, whilst 9% complained that the observations are often outside the uncertainty band.

As part of this study the Flood Information Centre also investigated exactly when regional flood forecasters would issue a warning based on the variety of information (deterministic/ensemble forecasts) available. Forecasters generally are still relying on the deterministic forecasts rather than making full use of the ensemble:  87 % only issue a pre-warning if the deterministic forecast (red line in fig. 2) exceeds the flood alert level. 13 % issue a pre-warning, if the upper uncertainty bound exceeds the flood alert level. However, their decision making also depends on the exact features of the case at hand, such as the gradient of the rising limb of the hydrograph: when confronted with figure 2, 63 % of all forecasters would issue a pre-warning.

Figure 2

Although it is acknowledged that there is further work to do in communication and decision making with probabistilic flood forecasts, overall the introduction of the uncertainty bounds to forecasts issued by the Flood Information Centre can be judged a success. Future development will concentrate on developing the decision support system further.

For more details on operational flood forecasting in Bavaria, please visit here.



  1. Could you say a few more words about figure 2. Are you saying that forecasters would react differently between the upper left panel and the upper right? Intuitively, I can see how the first (in which the hydrograph is rapidly rising) would cause more alarm than the second (where conditions are flat)– perhaps the eye is extending the trend of the forecast (e.g. on 2.9.12, surely the red line will be above the yellow threshold). But from a verification/forecast evaluation point of view, the two should be no different.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.