Downscaling of global climate models for flood frequency analysis: where are we now?

The issues of downscaling the results from global climate models (GCMs) to a scale relevant for hydrological impact studies are examined. GCM outputs, typically at a spatial resolution of around 3° latitude and 4° longitude, are currently not considered reliable at time scales shorter than 1 month. Continuous rainfall-runoff modelling for flood regime assessment requires input at the daily or even hourly time-step. A review of the different methodologies suggested in the literature to downscale GCM results at smaller spatial and temporal resolutions is presented. The methods, from simple interpolation to more sophisticated dynamical modelling, through multiple regression and weather generators, are, however, mostly based directly on GCM outputs, sometimes at daily time-step. The approach adopted is a simple, empirical methodology based on modelled monthly changes from the HadCM2 greenhouse gases experiment for the time horizon 2050s. Three daily rainfall scenarios are derived from the same set of monthly changes, representing different possible changes in the rainfall regime. The first scenario represents an increase of the occurrence of frontal systems, corresponding to a decrease in the rainfall intensity; the second corresponds to an increase in convective storm-type rainfall, characterized by extreme events with higher intensity; the third one assumes an increase in the monthly rainfall without any change in rainfall variability. A continuous daily rainfall-runoff model, calibrated for the Severn catchment, was used to generate daily flow series for the 1961–90 baseline period and the 2050s, and a peaks-over-threshold analysis was undertaken to produce flood frequency distributions for the two time horizons. Though the three scenarios lead to an increase in the magnitude and the frequency of the extreme flood events, the impact is strongly influenced by the type of daily rainfall scenario applied. We conclude that if the next generation of GCMs produce more reliable rainfall variance estimates, then more appropriate ways of deriving rainfall scenarios could be developed using weather generators rather than empirical methods. Copyright  2002 John Wiley & Sons, Ltd.