Radiative Imbalance in CMIP3 Models

While reading over Roger Pielke Sr. blog, he made the point with respect to Loeb et. al (2012) that it might be interesting to compare the measured TOA radiative imbalance to those projected in models over this period.  Through PCMDI, I downloaded the first run net top-of-atmosphere (TOA) flux (rtmt) from the more popular models and calculated a global average, thinking I’d generally see between ~0.25 W/m^2 to 1 W/m^2 over the relevant period, with the more sensitive models showing a larger net imbalance (since they would generally equilibrate slower with respect to temperature changes).  However, I am quite perplexed with what I’ve found:

 

 

The above charts are for the 1 and 10 year averages of net radiative flux at TOA.  As you notice, these are all over the place (even though there is little variation per model in 10 year averages).  Furthermore, several of the models seem to show a negative (!) flux imbalance, which makes no sense because that would mean the Earth is *losing* energy, not gaining, and thus we would not expect to see the rising temperatures or ocean heat content that appears in the model runs.

My code is rather simple, and it’s always possible I’ve screwed something up.  However, when I performed the analysis
for the 20th century GISS runs, the trend/fluctuations look right (compare to figure 1C from Hansen et. al 2005):

The main difference from Hansen et al. (2005) seems to be that the flux imbalance is shifted up the Y axis in his figure.  While one may argue whether the incorrect values for absolute surface temperatures are relevant versus temperature changes, you certainly couldn’t  argue the absolute flux imbalance is irrelevant in GCMs, since the heat accumulated is the integral of this net TOA flux (and a model that increases heat content with a negative TOA flux would violate the 1st law of thermodynamics).

So I’m left a bit confused here.  IPCC AR4 chapter 8 seems to show the zonal RMSE of outgoing longwave and shortwave flux, but I couldn’t find anywhere that shows the net global TOA radiative imbalance of models versus observations.  The net radiative imbalance must be expressed in absolute terms here, right?   What am I missing?