Impact of Ensemble Size on the Assessment of Model Climate Signal

Abstract

Climate simulations realized with an atmospheric General Circulation Model (GCM) are mainly a boundary-forced problem, the forcing being sea surface temperature (SST) and sea ice distribution. However, recent results obtained by, among others, Harzallah and Sadourny (1995), Stern and Miyakoda (1995), and Barnett (1995) demonstrate that model internal variability can be large and it obscures sometimes the good detection of climate signal. Although the cause of this internal variability is not fully determined, it seems that ensemble approach can overcome the difficulty, since we can then enhance the signal/noise ratio. In the scientific community, ensemble approach is more and more employed although the need on computing resource is largely increased. One important question is thus the following: what is the minimum number of runs necessary to overcome the model internal variability and to study, with confidence, the model’s response to external forcing ? It is clear that the answer depends upon the signal to study and the method used to detect the signal.