We have some firm sightings of Waldo in Siberia, as Warwick Hughes has long told us. There are very remarkable differences between temperature series depending on the site in Siberia. When Gavin Schmidt or James Hansen encounter differences of this order of magnitude between the U.S. and the ROW, they ascribe it to “regional” climate change – a pattern of temperature change in the 20th century which has left temperatures in the U.S. with relatively little change since the 1930s, while there has been much larger increases in the ROW.
There is a remarkable microcosm of this pattern in Siberia, where cities like Irkutsk and Bratsk have experienced sharp increases, while other sites have experienced relatively little change – a pattern no doubt ascribed by Schmidt and Hansen to “regionalization”. Here are a few plots :
For a little less muddy version, click on the image. These are all dset=0 versions before Hansen combining and adjusting. If you think about these series in statistical terms purely as time series – forget about the temperature history: think of these plots as different realizations of regional climate average plus “error”. What kind of error process could yield such diverse realizations?
If you posit that these time series are realizations of a gridcell plus noise: what kind of noise would be involved? You can’t add white noise or low order red noise to some sort of “true” gridcell composite and get Irkutsk on the one hand and Skovorindo on the other. An econometrician looking at the residuals from such an enterprise would observe highly autocorrelated residuals and conclude that the process had been mis-specified or perhaps that some of the data was bad. In this case, the cities (Irkutsk, Bratsk) would be the obvious candidates for exclusion.
Of course, Hansen says that he’s adjusted for urbanization. At this stage, I’ve not examined his adjustments in detail. It’s possible that this particular step is well described and will replicate easily. So far, we’ve had extreme difficulty even figuring out how Hansen combines nearly identical series, so who knows what perils lurk ahead.
To give a flavor of the Hansen adjustment, here’s the Hansen gridcell containing Irkutsk. You can see that the adjustment in this case removes the trend up to the 1980s and leaves it intact after the 1980s. At some point i the future, we’ll try to figure out why.