Guest Post by Willis Eschenbach
We have gotten three more years of data for the CERES dataset, which is good, more data is always welcome. However, one of the sad things about the CERES dataset is that we can’t use it for net top-of-atmosphere (TOA) radiation trends. Net TOA radiation is what comes in (downwelling solar) minus what goes out (upwelling longwave and reflected solar). The difference between the two is the energy that is being stored, primarily in the ocean.
The problem is that according to the raw, unadjusted CERES data, there’s an average net TOA radiation imbalance of ~ 5 W/m2 … and that amount of imbalance would have fried the planet long ago. That means that there is some kind of systematic error between the three datasets (solar, reflected solar, and longwave).
So, the CERES folks have gone for second best. They have adjusted the CERES imbalance to match the Levitus ocean heat content (OHC) data. And not just any interpretation of the Levitus data. They used the 0.85 W/m2 imbalance from James Hansen’s 2004 “smoking gun” paper. Now to me, starting by assuming that there is a major imbalance in the system seems odd. In any case, since the adjustment is arbitrary, the CERES trends in net TOA radiation are arbitrary as well. Having said that, here’s a comparison of what the Levitus ocean heat content (OHC) data says, with what the CERES data says.
Figure 1. CERES and Levitus ocean heat content data compared. The CERES data was arbitrarily set to an average imbalance of +0.85 W/m2 (warming).
I must admit, I don’t understand the logic behind setting the imbalance to +0.85 W/m2. If you were going to set it to something, why not set to the actual trend over the period of the CERES data? My guess is that it was decided early on, say in 2006, when the trend was much closer to +0.85 W/m2 and people still believed James Hansen. In any case, the way they’ve set it doesn’t tell us much. Let’s see what else we can learn from the two datasets. First lets take a look at the full Levitus dataset, and its associated error estimates.
I gotta say, I’m simply not buying those errors. Why would the error in 2005 be the same as the error in 1955?
In any case, we’re interested in the period during which the CERES and the Levitus datasets overlap, which is March 2000 to February 2013. To compare the two, we can adjust the CERES trend to match the Levitus data. Figure 3 shows that relationship. I’ve included the error data (light black lines.
Now, I’m sure that you all can see the problems. In the CERES data, the change from quarter to quarter is always quite small. And this makes sense. The ocean has huge thermal mass. But according to the Levitus data, in a single quarter the ocean takes huge jumps. These lead to excursions that are much larger than the error bars.
To visualize this, we can plot up the quarter-to-quarter changes in ocean heat content. Figure 4 shows that relationship.
Figure 4. Quarterly changes in the ocean heat content. Note that this shows the quarterly change in OHC, so the units are different from those in Figures 1 and 3. Standard errors of the quarterly change are larger than those of the quarterly data, because two errors are involved in the distance between the two points.
As Figure 4 highlights, the disagreements between the Levitus and the CERES data are profound. For some 60% of the Levitus data, the error bars do not intersect the CERES data …
Conclusions? Well, my first conclusion is that I put much more stock in the CERES data than I do in the Levitus data. This is because of the very tight grouping of the CERES data in Figures 3 and 4. Here are the boxplots of the data shown in Figure 4:
Remember that the tight grouping of the CERES data is the net of three different datasets—solar, reflected solar, and longwave. If you can get that tight a group from three datasets, it indicates that even though their accuracy is not all that hot, their precision is quite good. It is for that reason that I put much more weight on the CERES data than the Levitus data.
And as a result, all that this does is reinforce my previous statements about the error bars of the Levitus data. I’ve held that they are way too small … and both Figures 3 & 4 show that the error bars should be at least twice as large.
Next, the CERES data doesn’t vary a lot from a straight line. In particular, it doesn’t show the change in trend between the early and the later part of the Levitus record.
Finally, the CERES data provides a very precise measurement of the quarterly changes in OHC. Not only is their overall variation quite small, but they are highly autocorrelated. In no case are they greater than 0.5e+22 joules.
So for me, until the Levitus quarter-to-quarter changes get down to well under 1e+22 joules, I’m not going to put a whole lot of weight on the Levitus data.
NOTE: see my previous post for the data and code.