Techniques for Extreme Rainfall and Flood Runoff Estimation
The EURO-FRIEND flood project covers both real time forecasting and simulation for design purposes (frequency estimation of peak flows and flood inundation) and scientific purposes (understanding runoff generation). On most of the studied catchments not only rainfall but also snow is important. A central point is an effort to estimate uncertainty in the predictions.
Current activities follow to a large extent two important recent developments (Beven, 2006; 2007):
In the Manifesto a new development of the GLUE (Generalised Likelihood Uncertainty Estimation) methodology is introduced. It relies more on prior evaluations of model acceptability relative to observations and less on likelihood measures based on model residuals after a model has been run. This focuses, for example, on incommensurability errors (differences between the nature of observed and predicted variables due to scale and heterogeneity effects) and the effects of input errors, both of which are often ignored. Simulations (the parameter sets) which lie outside the range of the effective observational error are rejected.
With the ever increasing computer power and software (middleware) possibilities we will soon have computer models of everywhere, i.e. all places of interest will be represented. The data to constrain our models of particular places will assume greater importance than particular model structures. Models should be applied within a learning framework. Models, i.e. not only non-behavioural parameter sets but also the model structures which for a certain place are not in agreement with the observed data will be rejected.
Uncertainty in environmental modelling in general is summarised in a new book Beven, K J, (2009), Environmental Modelling: An Uncertain Future?, Routledge: London.