Weighted frequency distributions express modelling uncertainties in the ENSEMBLES regional climate experiments

Abstract

Fourteen regional climate models (RCMs) were driven by general circulation models (GCMs) in FP6-ENSEMBLES to provide 17 fine-scale (25 km) climate change scenarios for the period 2021-2050. In a preliminary exercise, these RCMs were driven by gridded observations (ERA40 reanalysis) to simulate as accurately as possible the 1961-2000 period. The quality of this reproduction was used to calculate a weight for each model. Each individual model climate had an uncertainty due to the finite sampling (30 yr). These spreads were combined by those weights to produce an ensemble uncertainty. We illustrate here the daily and climatological frequency distributions for winter and summer temperature and precipitation in 3 European cities (Budapest, Dublin and Lisbon). The distribution obtained by ENSEMBLES weights was compared with a distribution using equal weights, distributions using random weights and distributions based on a single model. As far as the reproduction of the observed distribution (1961-1990) is concerned, there is no evidence that the ENSEMBLES weight system provides results closer to observation than equal weights or weights drawn at random. A single model taken at random yields a quality score not better than ENSEMBLES in the case of precipitation, and worse than ENSEMBLES in the case of temperature. As far as climate change for 2021-2050 is concerned, the use of ENSEMBLES weights instead of equal weights also leads to a similar response at daily as well as 30 yr mean time scales.