HadCRUT4.6 has released their annual temperature result for 2019 as 0.74C. They grid the station data (CRUTEM4) with SST data (HADSST3) in 5×5 degree bins and perform an area weighted: global average. CRUTEM4 has 7680 stations which contribute to calculating this average. Although GHCN-V4 has more stations (17280) it is not clear to me that the coverage is really all that much better. CRUTEM4 is similar to the coverage of V3 but with some additional stations.
I downloaded the CRUTEM4 station data and calculated the global average with HadSST3 using my 3D averaging method (Spherical Triangulation). This is exactly the same data as that used by HadCRUT4. Both results are compared below together with those using GHCN-V4 and HadSST3..
The largest difference is that between spatial integration techniques. Exactly the same data produces a difference of 0.07C in the result for 2019. The reason for this is simply because spherical triangulation makes an implicit interpolation over both poles, whereas the traditional 5×5 lat/lon grid averages over just the occupied cells. This difference will depend from year by year on just how much warmer high latitude stations warm as compared to those at lower latitudes. So in 2004 and 2015 there was little difference between the two as compared to say 2016.
Here is the 3D grid used to calculate the anomaly for December 2019.
This shows how the triangulation connects together all station locations covering all the earth’s surface and as a result interpolates the average temperature from the 3 vertices across each triangular area. The coloured triangles show the relative increase in temperature anomalies relative to 1961-1990..
This procedure also gives very similar results to those of Cowtan & Way who instead use a kriging technique to interpolate into polar regions.