Regressing CMIP3 SRES A1B against HADCRUT4 over the entire
hindcast interval of 1900 - 2000 gives a scaling factor of 0.89 (model is too hot).
Thus, reliability over
a hindcast interval is not necessarily a sufficient condition to demonstrate that the model forecasts are good (Y12).
Now, it doesn't matter how you select / constrain over
the hindcast interval, the range of forecast warming still has no supremum because even if B is bounded, d is unbounded due to being a gaussian.
For a very long
hindcast interval, you can't fit many instances into the data (and they all overlap, so are not independent).
Not exact matches
We don't even have the data needed to intelligently initialize the models we have got, and those models almost certainly have a completely inadequate spatiotemporal resolution on an insanely stupid, non-rescalable gridding of a sphere... the ongoing failure of the GCMs to actually predict or
hindcast anything at all particularly accurately outside of the reference
interval.»
Nic, considering the first part of your comment, let's write the response of a model over the
hindcast and forecast periods as something like (A + e, B + d) where A and B are the forced response over the two
intervals (which depends on the parameter choices) and e and d are gaussian deviates due to internal variability (which depends on random initial conditions).
You state exactly that which you know within an acceptable confidence
interval that is the norm for that discipline, but not by
hindcasting, but by making predictions of key variables whose values were not determined in the past, to show the validity of your theory / model.