The Euro Hockey Team (EHT) present a variety of results using CVM methods on proxies from 5 canonical Team studies: Jones Briffa 1998; MBH99; Esper 2002; Moberg 2005 and Hegerl 2006. CVM (take the average of scaled proxies and re-scale the average to match the instrumental record) seems to me to be a less dangerous method than inverse regression. However, the supposedly "robust" results are all from very small subsets, which have been "data snooped" in advance. I’ve replicated 4 of the CVM results so far – the non-MBH ones.
The EHT purport to test for robustness – but their idea of robustness is to test the impact of removing one proxy after they’ve already ensured that there are, say, two foxtails plus Yamal. On numerous occasions, I’ve stated that medieval-modern levels in Team studies are highly dependent on the presence/absence of a few stereotyped series: bristlecones/foxtails, the Yamal substitution, Thompson’s Dunde ice core. The EHT present 5 CVM results and these provide a nice template for showing this one more time. Today I’ll do quick robustness analyses of the CVM-versions of Esper 2002 and Jones 1998 and do the others later.
Here is their key spaghetti graph of CVM reconstructions for the 5 studies plus the Union (All-Star?) reconstruction.
Euro Team Fig. 4. Reconstruction back to AD 1000, calibrated on 1856 to 1980 northern hemisphere temperature, using CVM, for a variety of different data collections. The MBH1999 and HPS2000 NH reconstructions and the Jones et al. (1998) instrumental data are shown for comparison. Graphs have been smoothed with a 21-year running mean and centred on 1866 to 1970. The maximum of the “Union” reconstruction in the pre-industrial period (0.25 K, 1091 AD) is shown by a cyan cross, the maximum of the instrumental record (0.841 K, 1998 AD) is shown as a red.
The figure below isolates the Esper composite, which consists of the 5 Esper series available to 1000 – two foxtail series, a Tornetask version, the Polar Urals Update (only used by Esper) and Taymir. It’s surprising that they didn’t include Mongolia in their Esper composite as it theoretically goes back to 1000 and has strong 20th century growth. There is something weird going on with Esper’s Mongolia version. This is the only one (out of 14 chronologies) that Esper didn’t produce when requested by Science. Hegerl et al, who rely on Esper versions, mention in passing that Mongolia was unavailable and performed a re-collation of 9 sites – Sol Dav/Tarvagatny Pass (mong003) will be one of them, but the others are unknown at present. It looks like Esper might have misplaced his Mongolia chronology as used. If not, the fact that it continues to be Missing In Action is really quite remarkable.
First, I’d like to demonstrate that I’ve replicated their result sufficiently to permit a sensitivity analyis. In the figure below, the archived EHT composite was plotted in black, with my emulation in red overlying it. The difference is plotted in blue. Because the emulation is very close, you can barely see the black. The correlation between the two versions is good to more than three 9’s. An obvious observation when this series is disentangled from the spaghetti is that, by itself, I don’t think that anyone would call it a "Hockey Stick". 20th century values are somewhat elevated, but not in an exceptional range – and this is with 2 foxtail series.
Figure 2. Replication of EHT Esper composite. Black – Archived by EHT; red – emulation; blue – difference.
Fig. 4. Reconstruction back to AD 1000, calibrated on 1856 to 1980 northern hemisphere temperature, using CVM, for a variety of different data collections. The MBH1999 and HPS2000 NH reconstructions and the Jones et al. (1998) instrumental data are shown for comparison. Graphs have been smoothed with a 21-year running mean and centred on 1866 to 1970. The maximum of the “Union” reconstruction in the pre-industrial period (0.25 K, 1091 AD) is shown by a cyan cross, the maximum of the instrumental record (0.841 K, 1998 AD) is shown as a red cross.
The obvious question is what happens if you don’t use the foxtails – after all, these results are supposed to be fantastically "robust". The figure below shows the EHT Esper composite on the left and a Esper variation on the right without the two foxtail series. As you see, without the foxtails, 11th century values are higher than modern values and, indeed, the series ends up at about its long-term average. So the amount of 20th century elevation is very small even in the EHT Esper version and even this limited 20th century "warmth" vanishes without the foxtails. The 1856-1980 correlation of the EHT Esper composite to their instrumental temperature record is 0.59 – higher than the level of 0.43 for the non-foxtail composite. However both results would be "significant" under the statistical methods presented by the EHT.
Figure 2. Left – EHT Esper composite (same as previous figure); Right – variation without foxtails.
Jones et al 1998 Version
Jones et al 1998 is shown in red in the EHT spaghetti graph (their Figure 4 excerpted above.) The EHT Jones composite uses 6 series – three NH and 3 SH: Polar Urals (the Briffa 195 version); Tornetrask (the Briffa 1992 version – which has an amusing "adjustment" discussed last year – see the Jones et al category); West Greenland ice core (a slightly different version than MBH); Cook’s Tasmania tree ring chronology and the Lara-Villalba tree ring chronology from Rio Alerce, Argentina.
Once again, I was able to make a nearly exact replication of the EHT Jones composite as shown below (using the same color scheme as above). The appearance of this composite is obviously significantly different than the Esper composite – with more high-frequency. Again one would not be inclined to label this paticular series as a Hockey Stick.
EHT Jones composite. Black -archived version; red- emulation; blue difference.
In this case, I’ve done a routine sensitivity analysis in which I’ve replaced the West Greenland stack used in Jones et al 1998 with the slightly later version used in MBH99 (I’ve verified that the Jones version is an earlier version with the originator of the data.) In the Euro All-Star Team, in their unaccountable policy of using older data, they use the Jones version. There’s not much difference between the two series, but I’ve used the later series. I’ve also used the updated Polar Urals version (from Esper) instead of the Briffa Polar Urals series (there are other problems with this series in the 11th century which I wrote about last year. I’m convinced that some of the tree cores in the Briffa 1995 study have been misdated.) Anyway, merely by using the updated Polar Urals version, the relative medieval-modern levels are changed. In this case, the 1856-1980 correlations for the variation (0.37) are a very slight improvement on the correlations for the Juckes version (0.36) – so any "significance" attributed to the Juckes version is shared equally by the version with a higher MWP.
EHT Jones composite – Left – As archived; Right – variant in which Polar Urals update substituted for 1995 Briffa version.