On Marotzke and Forster - WAG Index?

There is considerable discussion ongoing over the Marotzke and Forster paper published this month in Nature Magazine.  Nic Lewis noted that the paper was based on circular reasoning in a post on Climate Audit.  That is based on a previous Forster paper that estimated change in forcing using estimated surface temperature then Marotzke using that forcing compared to surface temperature to estimate model "natural" variability.

Prior to all this I had been looking at what should be the "actual" surface temperature versus what the models assume to be the actual surface temperature.  Climate Explorer has a good deal of the model runs archived so that you can mask regions.  The tropics are my main region of concern primarily because of the huge latent heat generation which is a source for the majority of clouds.  Water vapor response, clouds radiant forcing and deep convection are all very uncertain climate factors.  So I have prepared and may continue to prepare comparisons of model "actual" temperatures to observational temperatures.  This isn't really as simple as it should be.  Models use TAS, Temperature Air Surface which isn't really something measured.  Land surface temperature is measured roughly 2 meters above the surface located at the surface station and the elevation of those surface stations varies considerably.  A "potential" surface air temperature is one way to approach this issue, but over the ocean no "surface" air temperature is reliably measured only a sea "surface" temperature which is often really a sea "sub-surface" temperature.  That makes "what surface?" an important question that really doesn't have a good answer.

So when I started putting this together I used ERSSTv4  as a reference and whenever possible masked the model tas for just ocean data points.  In the tropics, the tas should be close to the measured SST less roughly 1 C degrees.  I have the observed SST in red with a thicker line so it stands out and the models I selected to show a rough range of the model temperatures for the same area.  Not all of the models are set up for the sea point masking so GISS-E is just 30S-30N.  Unless I made an error, the rest are masked just for sea points.  There is roughly a 2.5 C range of temperatures in the models for what should be a "known" value to at least a half degree by my estimation.  The two Chinese model runs bbc-csm-1 and -1m start the closest to what should be the "real" temperature.  GISS which also produces a temperature product comes in third with one of the GFDL estimates a close forth.

Initially, it looked like the models with the more lower than observed temperature had the greatest "sensitivity", but this collection implies that there is not much correlation between getting surface temperature right and sensitivity.  It looks more like a bunch of wild assed guesses.

One of Forster's conclusions was that the spread in the wild assed guesses indicated that the modelers were not "adjusting" their models to match observation.  I heartily agree with that conclusion.  Another was the present and future forcing and feedbacks was the source of the variability in the model "forecasts".  Well, if you "adjust" the model runs to a 1860 to 1999 baseline anomaly or any other past period anomaly, then the majority of the variability is in the future.  If you "adjust" the model runs to a future baseline for anomaly then most of the variation is in the past and would be due to wild assed guessing.

Now if you used the Marotzke and Forster method on this you could come up with an estimate of"Wild Assed Guess Variability".  Perhaps we could create a WAG-Index for future model generations?