I see that Mr. Goddard's article has gotten quite a discussion going. I've emailed The editors at The Register, but haven't heard back, and just recently Mr. Goddard directly.
Mr. Goddard's approach to counting pixels is simply not the correct approach. First, let me clarify a couple things.
1. The satellite data doesn't directly measure sea ice area or extent. It measures brightness temperature - a measure of the amount of energy emitted by the ice. This is converted to area or extent using an algorithm.
2. There are several different algorithms and they can yield different results in terms of absolute numbers. However, their trends and change from year to year show similar magnitudes. The Bremen AMSR data is from a different algorithm - hence it looks different.
3. Both UIUC and NSIDC use the same brightness temperature data.
4. Both UIUC and NSIDC use the same algorithm, but with some differences in the specifics, so the numbers aren't perfectly matched, but there is very good overall agreement and they yield the same conclusions about changes in Arctic sea ice.
5. People have talked a lot about "pixels", but one needs to understand what one is talking about. There are two types of "pixels". One is "data pixels"; this is a function of the spatial resolution of the sensor (i.e., how small of an area the sensor can resolve). For the data UIUC and NSIDC uses, the data pixels are about 25 x 25 km. The other is "image pixels", which describes the qualities of the image.
6. The data has to be gridded onto a projection, which yields a gridded resolution, which is also about 25 x 25 km, but varies depending on the type of projection and where the grid cell within the projection. The input data for both UIUC and NSIDC is on a 25 x 25 km grid. The UIUC grid that Mr. Goddard analyzes has been interpolated onto a different grid. I do not know the specifics of that grid, but such interpolation will change how the data looks when viewed.
7. The data can then be conveyed in an image. The image has an "image pixel" resolution. This is generally given in dpi or dots per inch. Higher dpi means a sharper image. However it does NOT change the fundamental resolution of the data.
8. An image is simply a way to convey data; it is not data itself. Therefor it is not proper to do analysis on the image. You need to use the data.
9. The gridded data, when analyzed, must account for the projection in terms of the area of the grid cells. You have to sum the ice, weighted by the correct area for each grid cell. NSIDC uses a polar stereographic projection with a true latitude of 70 N. Other than at 70 N there will be distortion that needs to be corrected for, as NSIDC does.
10. NSIDC freely distributes all the data, tools to work with the data, and the grid cell area files. So anyone can do their own analysis.
11. NSIDC's methods have been around for over 20 years, have been thoroughly vetted in peer-reviewed science journals, and confirmed numerous times over by independent scientists conducting the proper method.
12. Finally, Mr. Goddard need not have wasted his time doing his image pixel counting. He could've simply referred to the UIUC site, which actually counts the pixels properly and creates a timeseries plot:
http://arctic.atmos.uiuc.edu/cryosphere/IMAGES/current.365.jpg
If you look at that plot, you'll that because it is area instead of extent, the raw numbers are lower. However, while it's a bit hard to make out the values real accurately, you see that this year on Aug. 11, it was ~4 million sq km, while last year on Aug. 12, it was ~3.6 million sq km (actually, since it's a one-year sliding window, Aug. 12, 2007 is no longer visible, but that's what it was on that data and the current range shows a similar difference). That's a bit more than an 11% difference. So Mr. Goddard's analysis of UIUC's data doesn't even agree with UIUC's analysis.
Hopefully Mr. Goddard will make a corrections soon.
Walt Meier
Research Scientist
National Snow and Ice Data Center
University of Colorado at Boulder
