I am often surprised at how different climate models can be. Shown below is the annual mean, global mean surface air temperature from control runs of 21 of the AR4 GCMs, along with the observations (HadCRUT3). The spread in the magnitude of variability is enormous, as seen by eye, but also the timeseries have very different ‘memory’ characteristics. The second figure attempts to quantify these differences for the AR4 models by showing the total standard deviation against the lag-1 auto-correlation of the time series. [The timeseries are not all well represented by an AR(1) process, due to ENSO variability presumably, but it is still good enough for this purpose.] Anyone else find the range in behaviours in this global mean property surprising and/or interesting?
[Click for higher resolution versions]