The effect of internal variability on anthropogenic climate projections

Abstract

A single model ensemble of five members, with a CO2 increase of 1% per year, has been used to investigate the spread in climate change estimates. It is furthermore explored how this spread is related to internal variability of the climate system and the strength of the CO2 forcing. As expected, the fraction of the globe where a statistically significant climate change could be detected increased with the strength of the CO2 forcing. For temperature this fraction increased rapidly with integration time, while the increase of areas with statistically significant precipitation change was much slower. It is shown that increasing the averaging periods for estimating climate change from 20 to 40 yr reduced the spread due to internal variability by 40% in the Arctic regions. This is due to better sampling of decadal internal variability. In order to reduce sampling uncertainty below a fixed threshold, the number of ensemble members needed is much larger in the polar region than in the lower latitudes. This should be taken into account when designing modelling strategies (high-resolution modelling versus larger number of ensemble members) to reduce uncertainty in future polar climate change. At the time of doubling of CO2, the variance of the climate change estimates, based on 20 yr averages from a single model ensemble, constitutes 13% and 42% of corresponding multimodel ensemble variances for temperature and precipitation, respectively. In the case of precipitation, this indicates that some caution should be used in attributing different climate change estimates to differences in model formulations.