Climate Audit

Jones et al [1998]] did not archive their proxy set as used. I requested this information from Jones this year and was refused, although he did send me two series (Lenca, Law Dome) not locatable anywhere in a public archive about a year ago. Here is some background on the proxies in Jones et al [1998] – 10 NH series and 7 SH series . I doubt that any of this information will be directly usable in my review of this study which is underway and so I’m posting it up.

Of the 17 proxies, 13 are used directly in MBH98-99, which does have an archive, and for one (Jacoby northern treeline composite), the 10 constituent series are used individually in MBH. Not used in MBH98-99 are 1 NH series (Jasper) and 2 SH series (Lenca, Law Dome) . So the two studies are remarkably non-independent in proxies. However, J98 does not use principal component methods and so any hockey-stick-ness of J98 is not a direct result of an oddity like PC methodology. I’m not going to get to methodology in this post.

Eleven (7 NH; 4 SH) of the 17 proxies are also used in Jones and Mann [2004] : C England; Tornetrask (Fenno); Polar Urals; Western USA (Briffa density); Jasper; Jacoby Northern Treeline; Crete (Fisher’s Greenland dO18); GBR5 (Great Barrier Reef); Galapagos; New Caledonia and Lake Johnston (Tasmania). The standardizations were different, but, on a correlation basis, 7 series were identical (correlation>0.999): Jasper, C Eng, Briffa WUSA, Jacoby treeline, Fisher dO18, GBR5 and Galapagos; one was a new version similar to the older one: Tasmania (correlation : 0.978); one was a different version with moderate similarity: New Caledonia (correlation -0.84) and two were new versions with little similarity to the older versions: Tornetrask (correlation – 0.29) and Polar Urals (correlation- 0.19). I’ve not yet tracked the reasons for the big changes in the Tornetrask and Polar Urals editions. However, because of the obvious differences in versions, it made more sense to use the MBH archive as a base for replicating Jones et al [1998] data, than the Jones and Mann [2004] archive.

Of the 4 series not at the MBH98-99 archive, 2 series (Jasper, the Jacoby composite) are publicly archived at Briffa’s website. Two series (Lenca, Law Dome) could not be located in the public domain. I obtained versions of these two series from Jones over a year ago.

Although there is no data archive for J98, there are plots of all the proxy series, which are shown in Figure 1 below. In addition, the WDCP archive for J98 lists NH and SH counts by years, which provides an additional control. Several differences arose in the above implementation.

The versions of the C England and C European series differ from the MBH versions, which were JJA series, which were truncated to start dates of 1730 (not 1659) and 1550 (not 1525) respectively. Mann et al. [Corrigendum 2004 ] attributed the truncation to Bradley and Jones [1993], where indeed truncated versions of the series are plotted (although the truncation is not reported in Bradley and Jones [1993] either). Non-truncated versions were at the MBH98 FTP site, but these were annual versions, which did not match when plotted up. There are original sources for both series. JJA versions were collated from original sources and ended up with a plausible match in both cases.

Some values for the Law Dome series are missing in the 19th century. I don’t know the background for this. The data file provided by Jones on Law Dome had monthly readings. Some 19th century values are missing; a conversion to annual data using na.rm=TRUE yielded more 19th century annual values than observed in the archived count series. The NA values were manually adjusted to match the counts.

The SH count for 1990-1991 in the archives was 2, while the count in the dataset constructed here is only 1. The only possibility that I can think of is that J98 used a slighltydifferent Tasmania dataset than MBH98 (which ends in 1989). The NH count for 1983 in the archives is one higher than observed in this dataset.

Figure 1 compares the J98 graphs to my emulations. In general, the graphs of series from the MBH archive match pretty much exactly (other than the C England truncation), as do the Lenca and Law Dome versions obtained directly. These versions do NOT necessarily match archived versions (as discussed in MM03 for MBH98). For example, the three coral series used here are "grey" versions that differ from the archived version. I inquired on this in connection with MBH98 and the originator of the Great Barrier Reef series (Lough) said that she did not provide the data to Mann and surmised that it was a version provided to Ed Cook, presumably passed along to Jones and Mann.

The Jasper series from the Briffa archive looks like it may be a problem: it matches closely, but it does not appear to match exactly especially in the MWP. As will be seen, there are some replication problems in the MWP period of Jones et al [1998] and this may be a factor.

Figure 1. Left. Proxies from Jones et al [1998]; right – versions in replication dataset

When you look at the individual proxies, always a useful exercise, it’s remarkable how few have hockey stick shapes and instructive to see which ones do. The most prominent hockey stick shapes are: the Jacoby treeline composite; Kameda melt; New Caledonia dO18; and to a small extent, Tasmania. Of these, only Tasmania goes into the MWP.

The key series for achieving a low MWP in the eventual reconstruction is the Polar Urals, where its distinctive very cold MWP drags down the J98 NH reconstruction in the MWP (which only has 3 series in the early 11th century). I’ve posted up a lot about Jacoby, where data mining issues ("a few good men") need to be closely examined. Tasmania is another Hockey Team data set (Cook).

In MM03, we talked a lot about the use of "grey" (usually obsolete) data versions in MBH98. Because of the commonality of data versions in J98, there is the same problem here. While most of my attention has been to the NH series, in passing, here are 3 versions of the New Caledonia dO18 series: the archived WDCP version (inverted); the MBH98 version (inverted) and the Jones and Mann [2004] version. You will see that the downspike in the archived version is earlier than the downspike in the MBH98 or JM04 versions. In this case, there is a difference between MBH98 and JM04 versions (which is not the case with the other coral series). The JM04 and MBH98 versions are identical for the other coral series, but which differ from the archived versions. The 18th century differences in these versions is disquieting.

What I can show right now and am working up in a presentation is that the hockey-stickness of J98 depends on the extension of the Polar Urals data beyond its range of confidence (and then relying on three incorrectly dated cores which I discussed earlier.) I’m trying to figure out how the impenetrable J98 confidence interval calculations were done in order to tie up this report.