EGU General Assembly edition:2012 location:Vienna, Austria date:22-27 April 2012
Reliable flood forecasts are the most important non-structural measures to reduce the impact of floods. However
flood forecasting systems are subject to uncertainty originating from the input data, model structure and model
parameters of the different hydraulic and hydrological submodels. To quantify this uncertainty a non-parametric
data-based approach has been developed. This approach analyses the historical forecast residuals (differences
between the predictions and the observations at river gauging stations) without using a predefined statistical error
distribution. Because the residuals are correlated with the value of the forecasted water level and the lead time,
the residuals are split up into discrete classes of simulated water levels and lead times. For each class, percentile
values are calculated of the model residuals and stored in a ‘three dimensional error’ matrix. By 3D interpolation
in this error matrix, the uncertainty in new forecasted water levels can be quantified.
In addition to the quantification of the uncertainty, the communication of this uncertainty is equally important.
The communication has to be done in a consistent way, reducing the chance of misinterpretation. Also, the
communication needs to be adapted to the audience; the majority of the larger public is not interested in in-depth
information on the uncertainty on the predicted water levels, but only is interested in information on the likelihood
of exceedance of certain alarm levels. Water managers need more information, e.g. time dependent uncertainty
information, because they rely on this information to undertake the appropriate flood mitigation action. There are
various ways in presenting uncertainty information (numerical, linguistic, graphical, time (in)dependent, etc.) each
with their advantages and disadvantages for a specific audience. A useful method to communicate uncertainty of
flood forecasts is by probabilistic flood mapping. These maps give a representation of the probability of flooding
of a certain area, based on the uncertainty assessment of the flood forecasts. By using this type of maps, water
managers can focus their attention on the areas with the highest flood probability. Also the larger public can
consult these maps for information on the probability of flooding for their specific location, such that they can take
pro-active measures to reduce the personal damage.
The method of quantifying the uncertainty was implemented in the operational flood forecasting system for the
navigable rivers in the Flanders region of Belgium. The method has shown clear benefits during the floods of the
last two years.