Just read the paper
I have just read[*] the Karl et al paper in Science (this is a full text link, the Reg article links to the abstract). Not that there is much detail, but what I did find there looks rather bizarre. Consider:
1. They deal with sea surface temperature (SST). Historically, these were taken by ships - throw buckets into water, raise to deck, stick a thermometer in. Lots of biases (e.g., ships are not everywhere, measurements were irregular, etc).
2. At some point (starting before WWII) ships got fitted with engine intake sensors - these gotta give you higher readings - they are next to the engines, after all.
3. Recently (it is not clear to me how recently - the paper does not say) buoy measurements have become wide-spread. These do not suffer from (as) many biases as ship-based measurements (see above). But buoy-based measurements do not go nearly as far back as ships, so if you want long-term historical data...
4. Buoys systematically show slightly lower temperatures than ships in close-by locations.
5. So, let's compute the somehow averaged difference between co-located ships and buoys and adjust all (i.e., not just co-located) buoy-based data upwards.What the...?!?!?
6. Now, since buoys are nominally more accurate let's give them higher weights. This would normally be correct, but it's blatantly inconsistent with the previous adjustment - why did you choose to "correct" the more accurate data?!?
7. Let's extrapolate the results to polar regions where very little or no real data exist. [**]
As a result, the more recent observations (buoys) are adjusted upwards without any justification as far as I can see, AND given higher weights, AND extrapolated way beyond the sample that exhibited the bias that caused the adjustment in the first place.
Frankly, I think I can bend just about any trend in any direction using methods like this. Now, can I get a government grant? On second thought, a research grant would probably not be a generous enough compensation for the lost ability to look at myself in the mirror.
[*] I've been meaning to do it fora few days now, ever since it made waves in mainstream media. Always go to the source - got to justify my Reg handle...
[**] Aside: many numerical models only cover polar regions, for various reasons - I fully expect some influence on model calibration in future publications, though it remains to be seen, of course.