Anthony Watts posted his take on the USHCN adjustments. Here is my take on his take. Anthony’s graph shows my method in red.
The spike in 2014 is due to USHCN fabricating temperatures in their adjusted version. The adjusted version includes a lot of May temperatures from stations which USHCN doesn’t have any data for. May is a warmer month than January-April, so the USHCN final average gets elevated relative to the raw average, because the final version includes more (imaginary) May station data. If USHCN didn’t fabricate temperature data, this spike wouldn’t exist.
My method takes the average of all monthly readings at all stations in a given year, and subtracts the average of the raw data from the average of the final data. I assume a Monte Carlo distribution of missing data, which is probably a safe assumption for a large data set. This shows the total amount of adjustment for all of the USHCN temperature gymnastics. Zeke wants me to take the average of all of the individual station deltas. My approach is probably more meaningful because it also shows the effects of station loss.
Note in Anthony’s plot, that my approach and his approaches are roughly parallel from 1930 to 1995. After 1995, my approach shows a sharp increase in slope while his tails off. Not coincidentally, since 1995 the number of stations has dropped dramatically. It is possible that stations which showed less warming (or more cooling) disappeared selectively after 1995. Whatever it is, this an important piece of information and should not be hidden by the methodology. I don’t know exactly why the methodologies diverge after 1995, but the fact that the break is coincident with a drop off in stations strongly hints at a possible correlation.
Gridding would provide a better average for the whole country than what I am doing, but that isn’t my intent. I’m just showing the average adjustment across all stations.
Zeke says “everybody uses anomalies” but that isn’t correct. NCDC publishes absolute temperatures for their US temperature data, and most of my comparisons are HCN raw vs. NCDC published.
As I showed yesterday, my approach to raw data (blue below) is pretty close to Hansen 1999, except for the USHCN V1 adjustments – which were much smaller than the V2 adjustments.
No matter how you look at it, USHCN and NCDC are cooling the past – when they should actually be cooling the present due to UHI.