Guest Post by Willis Eschenbach
The CERES dataset is satellite data that is based on radiation measurements made from low earth orbit. The CERES data has two parts. The first part is observational data, measurements of downwelling and upwelling solar radiation and of upwelling longwave radiation. It is usually referred as the CERES “top-of-atmosphere” data. The official name is “CERES EBAF-TOA”, and it is available here.
However, the second part of the CERES is not top-of-atmosphere observations from the satellite. Instead, it is calculated surface data based on the CERES TOA observations along with other satellite observations. It’s called the “CERES EBAF-Surface” dataset, and is available from the same location.
As a result, I’ve always been concerned about the accuracy of the CERES surface data. After all, it’s just calculations, it’s not actual observations. So I got to thinking that I could “ground-truth” the CERES surface observations by using the TAO buoy data. It’s not a comprehensive test by any means, but the TAO buoys cover a region of great interest to me, the tropical Pacific. The TAO buoy data is available here.
I started out by seeing how well the CERES surface longwave radiation data agreed with the TAO sea surface temperature (SST) data. Now, the CERES dataset doesn’t have SST data, but we can convert the CERES surface radiation data into temperature by using the Stefan-Boltzmann relationship. First I looked at a string of TAO buoys that run along the Equator. I used the location of each TAO buoy, and compared it with the CERES surface calculated result for that location. Figure 1 shows that comparison for the eight TAO buoys along the Equator which have SST data.
I was pleasantly surprised by this result. The greatest bias is ± two tenths of a degree, and the correlation is very high, 0.97 to 0.99.
Having looked at an east-west line of buoys, I then looked at a north-south line of buoys. These are all at 165°E, in the warmest area in the Pacific.
Again the correlations are good, although there is one of the seven down at a correlation of 0.92. And the bias is slightly larger, -0.3 to -0.4°C. In this region all of the CERES data is slightly below the TAO buoy data.
Overall, however, if the bias errors are only on the order of a few tenths of a degree and the correlation is on the order of 0.97 or better, I’m more than happy to say that the sea surface temperature is extremely well represented by the CERES surface calculations.
However, that’s the easiest of the variables. Next I looked at something much harder to estimate—the available solar radiation at the surface after atmospheric reflection, absorption, and scattering. Figure 3 shows the available solar measurements from buoys along the Equator.
This one surprised me quite a bit. I wouldn’t have guessed that the satellite calculations would come this close. Not only do they get the annual cycles right, but they also get the occasional departures from the annual cycles. Yes, the correlation of some of them is lower, but they still do a good job. And the bias in all cases is less than 2%, a respectable showing.
I then looked at the same group of north-south buoys I’d used above. Here are those results.
Again, a very good showing, with correlations from 0.87 to 0.97, and bias of less than 2%.
There is one more overlapping dataset between CERES and TAO, that of downwelling longwave radiation (DLR). Unfortunately, there are only five TAO buoys with DLR data, and one record is very short … but we use what we have. Here is that group of buoys.
Of all of the results, I was most surprised by this one. I would say that this one would be the hardest to calculate from the satellite data. But despite that, if we set aside the very short (5-month) dataset in the first panel of Figure 5, the other four have good correlations (0.82 to 0.97), and the three longer datasets (panels 2, 3, and 4) have a bias of well under 1%.
Conclusions? Well, as I said, this is far from a comprehensive test … but I am greatly encouraged nonetheless. The CERES surface dataset, despite being calculated rather than observed, is a very good match to the TAO buoy data in all available respects. Makes me feel much better about using it.
My regards to you all,
PS-If you disagree with someone please have the courtesy to quote the exact words you disagree with. This lets all of us understand the exact nature of what you think is incorrect.