Fowler, H.J. and Ekström, M. 2009. Multi-model ensemble estimates of climate change impacts on UK seasonal precipitation extremes. International Journal of Climatology, 29(3), 385-416.
Thirteen regional climate model (RCM) integrations from the PRUDENCE ensemble are used together with extreme value analysis to assess changes to seasonal precipitation extremes in nine UK rainfall regions by 2070-2100 under the SRES A2 emissions scenario. Model weights are based on similarities between observed and modelled UK extreme precipitation calculated for a combination of (i) spatial characteristics: the semi-variogram parameters sill and range, and (ii) the discrepancy in the regional median seasonal maxima. These weights are used to combine individual regional RCM bootstrap samples to provide multi-model ensemble estimates of percent change in the return value magnitudes of regional extremes. The contribution of global climate model (GCM) and RCM combinations to model structural uncertainty is also investigated.
The multi-model ensemble estimates project increases across the UK in winter, spring and autumn extreme precipitation; although there is uncertainty in the absolute magnitude of increases, these range from +5 to +30% dependent upon region and season. In summer, model predictions span the zero change line although there is low confidence due to poor model performance. RCM performance is shown to be highly variable; extremes are well simulated in winter and most poorly simulated in summer. The ensemble distributions are wider (projections are more uncertain) for shorter duration extremes (e.g. 1 day) and higher return periods (e.g. 25 year). There are rather limited differences in the weighted and unweighted multi-model ensembles; perhaps a consequence of the lack of model independence between ensemble members. The largest contribution to uncertainty in the multi-model ensembles comes from the lateral boundary conditions used by RCMs included in the ensemble. Therefore, the uncertainty bounds shown here are conservative despite the relatively large number of RCMs contributing to the multi-model ensemble distribution.