Climate Forcings

Stephen E. Schwartz of Brookhaven and co-authors have a Nature comment.
I got the link from Lubos Motl who explained that it had something to do with the GDP of Cuba and the relative length of Gavin's (Schmidt?) hands, but I had trouble following the advanced math, so I was forced to read the original article - which seemed rather clear.

Schwartz and co-authors note an oddity: The estimated uncertainty in anthropogenic forcing has a much larger range (0.6 to 2.4 W/m^2) than the uncertainty in the model predictions of temperature range as a result of anthropogenic forcings (0.5C to 1.0C). This, he says, looks strange.

First, how could this happen? Schwartz does not say so explicitly, but he seems to think that the forcings are an input parameter for the models. I’m no modeler, but I doubt that that is the case – the input parameters, I suspect, are the things that drive the forcings: CO2 concentration, aerosol, etc. Physically, the forcings *are* the things that drive the climate, but in the models, the things that can be measured easily are not the forcings, but the drivers of the forcings. The forcings are deduced, I suspect, from models that capture some elements of the global climate models. In effect, the forcings become sort of a second order product of the models, calculated from model output differences. This has the effect of amplifying the uncertainties, in a way vaguely analogous to the way that errors in position measurements are compounded when one uses those measurements to compute velocities.

In other words, I think that the models get surface temperatures from drivers (CO2, aerosol, etc) and then a sort of reverse model gets forcings from something like surface temperatures. If the process introduces the same error at each stage, the error for forcings should be larger - maybe twice as large - as the temperature error.

Of course if it's this simple, why did Nature bother to publish the comment? Maybe Eli or Belette can explain.