One of the draft AR5 figures (the one reworked due to baseline errors, mind you) showed HadCRUT4 variability for the 2 - sigma bounds around the model mean rather than
model variability.
As often the observed and
modelled variability is pointed out it doesn stop stupid remarks such as this citation stemming apparently from David Whitehouse.
Collins 2012, part a:
model variability itself; part b: model variability (precision) represented as physical uncertainty (accuracy).
This will allow us to assess the quality of the models and to provide information on the model - to -
model variability, in particular with respect to our predictions.
An ensemble of 24 forecasts were made to provide estimates of mean and
model variability.
Bear in mind that red line only represents the average model expectation, not the range of
model variability.