87th AMS Annual Meeting

Monday, 15 January 2007
Comparison of observed and modeled trends in annual extremes of temperature and precipitation
Exhibit Hall C (Henry B. Gonzalez Convention Center)
Dmitry Kiktev, Hydrometeorological Centre of Russia, Moscow, Russia; and J. Caesar and L. V. Alexander
The potential for extreme climatic events, such as heat waves, floods and droughts, to occur more frequently in the future is of great concern. To understand the capabilities and limitations of climate change projections it is important to compare climate model output with historical observations.

Our analysis focuses on temporal trends in annual extremes indices derived from daily maximum and minimum temperature and daily precipitation series. A new dataset of gridded extremes indices (HadEX), covering the latter half of the 20th century, is used to form an observational benchmark for the assessment of the climate models. We utilise data from five coupled atmosphere-ocean IPCC AR4 models run with historical anthropogenic forcing. The availability of multiple ensemble members allows us to assess the benefits of ensemble modeling for inter-decadal climate change simulations.

We use a bootstrapping approach to estimate trend uncertainty, and also to estimate the uncertainty in similarity between observed and modeled trend patterns. We then compare the frequency distributions of trend pattern similarity of individual members and multi-model ensemble means with the observed trend patterns.

The climate models are more successful at reproducing trends in upper temperature percentiles compared with lower percentiles. Trends in precipitation extremes were found to be reproduced with lower skill, irrespective of whether the global mean trend was included in the calculation. Whilst there can be wide variation in skill, even between individual runs from the same model, it appears that much of the skill can be explained by the globally averaged trend that is likely to be associated with the effect of increases in well mixed greenhouse gas concentrations.

Multi-model ensembles perform well if there is reasonable skill in the individual contributing members, in many cases performing at the level of the best ensemble members, and in some cases even better.

Supplementary URL: