CR 22:189-204 (2002)  -  doi:10.3354/cr022189

ABSTRACT: Near-surface air temperature and precipitation are frequently presumed to be indicators of anthropogenic climate change; in this study, they are compared with each other in terms of their simulated signal-to-noise ratio. Special emphasis is given to the characteristic time scales of temperature and rainfall signals. By means of analysis of variance, based on an ensemble of 4 greenhouse-gas (GHG) induced, long-term, coupled climate model experiments, the contributions of internal noise and external GHG forcing to total variability are quantified. The part of the variance accounted for externally is a valuable measure for the predictability of a meteorological variable. Further, the sensitivity to different climate models is evaluated, based on multi-model ensembles. With regard to temperature, the GHG forcing accounts for more than 90% of the total variance in the tropics and 50 to 70% in the extra-tropics, implicating a warming trend all over the globe. On the other hand, precipitation is largely induced by internal variability. The GHG scenario accounts for at best 30% of the variance in rainfall. The corresponding trend pattern is more differentiated, predicting dryer conditions in the subtropical regions and rising precipitation elsewhere. The temperature signal, arising from 1970 onward, is evident on time scales of 30 yr and longer. For precipitation, there is a weak but statistically significant low-frequency signal on a regional scale, which is apparent for 60 yr time slices but not at shorter time scales. At present, precipitation, in contrast to temperature, represents an unfavourable detection variable for GHG-induced climate change. The super-ensemble approach reveals that intermodel variations contribute the major part to total variability in high latitudes, whereas the GHG impact on temperature is still evident in the tropics. Large model uncertainties occur in the regions of sea ice margins, mountains and over Antarctica.

KEY WORDS: Climate change · Detection variable · Signal-to-noise ratio · Multi-model ensemble · ANOVA