A couple of days ago, I reported on the comparison of Jones et al 1990 data to the TR055 data archived at NCDC, noting that the data was the same for all but one series. I’ve now been comparing the TR055 versions to GHCN v2 and HadCRU3 and noticing some puzzling aspects to both GHCNv2 and HadCRU3.
In a few cases, I’ve been able to closely replicate the HadCRU3 gridcell version from an individual station series (rather as I’d been able to do with some Russian gridcells.) Here’s one such example where the two series are virtually identical from 1950 on. You can tell this in two ways from this graphic: the GHCN version is plotted in black and the HadCRU3 version in red. If the two series match, then the later series overwrites the earlier – that’s why you don’t see the black GHCN series except for a little interval at the start. In order to show this, I then re-plotted th GHCN with points in large black points – their locations coincide with the HAdCRU3 version, so you can tell that the two series match over their recent history. The HadCRU3 version includes some early history not recorded at the station here.
Haerbin is another case where the HadCRU3 gridcell value closely matches an individual station – in this case the match extends to the pre-WW2 portion as well. Shenyang is another close match.
Black – GHCN; red HadCRU3.
While a few GHCN versions of Jones et al stations match through to recent values, I noticed that many, if not most, of the Jones et al stations were not recorded at GHCN after 1990, but HadCRU3 values in the gridcell continued to the present. Here’s one such example, where the GHCN station data for Qitai end in 1990, but the gridcell continues to the present. (In this and other cases, this is due to another station kicking in, but examining this phenomenon proved very surprising.)
Yesterday I showed the large seeming inhomogeneity at Chongqing station. Today I’m going to show this station together with a station in an adjacent gridcell to illustrate the complications of post-hoc adjustments. On the left is Kunming station; on the right, Chongqing. The adjacent gridcells have quite different Hadcru3 histories (red) and one cannot avoid feeling that some amount of the difference is due as much to different adjustment strategies as to different histories in the two adjacent gridcells.
Another interesting case involving an entirely different adjustment strategy came in the gridcells 27.%N, 117.5-122.5E. Here are 4 GHCN versions compared to the HadCRU3 gridcell in red. For these two cells, the HadCRU3 version contains much lower early 20th century values than the GHCN versions for the Jones et al 1990 selected stations.
What accounts for the big discrepancy between GHCN station values and HadCRU3 gridcell values for these cells (and, since I’ve been able to more or less replicate some cells, my guess is that the difference lies in an adjustment to the data rather than in a failure to replicate some unreported and obscure aspect of the methodology.) More on this in the coming week. Also you’ll notice that several of the station series ended in 1990 (black dot versions). More on this peculiar phenomenon as well – it turns out to be pervasive in the Chinese data as archived at GHCN).