The PAGES2K (2013) Arctic reconstruction of Kaufman et al has attracted considerable attention as a non-Mannian hockey stick. However, it’s been fraught with problems since day one, including a major re-statement of results in August 2014 (McKay and Kaufman, 2014 pdf), in which Kaufman conceded (without direct acknowledgement) Climate Audit criticism that their results had been impacted by the use of contaminated data and upside-down data. But there’s a lot more.
In March 2013, almost exactly contemporaneous with PAGES2K, Hanhijarvi et al, pdf here, the originators of the paico method, published their own Arctic reconstruction, which has undeservedly received almost no publicity. (In this post, I will use “PAGES2K” to refer to the PAGES2K Arctic reconstruction; the full PAGES2K study includes other areas, including Gergis’ Australian reconstruction.) But unlike PAGES2K, its medieval reconstruction has higher values than its modern reconstruction – a finding that has received negligible coverage. Because its methodology matches the PAGES2K methodology, the difference necessarily arises from proxies, not from method.
Nor is the issue merely “regional” coverage though Hanhijarvi et al’s Arctic reconstruction is based on North Atlantic proxies though it would be puzzling even as a “regional” result. These proxies from a very large subset of the PAGES2K Arctic data (27 of 59 series, using no other data). With such a large subset, one can only obtain the PAGES2K Arctic results if there is a superstick for the rest of the data (non-H13 proxies). As a regional result, specialists would have to explain the physics of a medieval warm period in the North Atlantic concurrent with extreme cold in the rest of the Arctic, if one were to take these results at face value.
But before attempting such a complicated solution, it is important to note that Kaufman’s proxies are fraught with defects. Kaufman has already acknowledged that one of his supersticks (Igaliku) was contaminated by modern agriculture; and that another non-H13 series (Hvitarvatn) was used upside down. Several series, thought to be temperature proxies as recently as 2013, were removed in August as no longer “temperature proxies”. For inexplicable reasons, Kaufman failed to remove all the contamination from the Igaliku series and his inversion of the Hvitarvatn points to major inconsistencies with other series. Further, although Kaufman has acknowledged multiple errors in the PAGES2K Arctic reconstruction, he has not issued a corrigendum, thereby permitting the erroneous series to continue in circulation, while, oddly, thus far not providing a digital version of the amended reconstruction.
The Paico Reconstructions
First, here are the H13 (blue) and PAGES-2013 (black) reconstructions, overplotted on the New York Times version of Kaufman et al 2009 (see Andy Revkin here). The PAGES-2013 reconstruction was similar in shape to the Kaufman-2009 reconstruction, but had nearly double the amplitude. Its medieval period was much cooler than the late 20th century -indeed, it was about the same temperature as the Little Ice Age of Kaufman et al, only a few years earlier. While the wiggles of the H13 reconstruction have much in common with the wiggles of PAGES-2013, the net result is completely different:its medieval warm period is over 1 degree warmer than PAGES-2013 and warmer than the 20th century portion of the reconstruction. Once again, given that the method is identical and that the H13 data is a very large subset of the PAGES-2013 data, the difference is really quite extraordinary.
Figure 1. Kaufman et al 2009 reconstruction from New York Times, overplotted by H13 reconstruction. 40-year gaussian smooth. Both centered to 1961-1990 reference.
PAGES2K Figure S3a
None of the PAGES2K graphics are very clear. The clearest depiction of their arctic reconstruction is their Figure S3a, which, however, is in standard deviation units. In the diagram below, I’ve plotted smoothed versions of the PAGES-2013 (yellow) and H13 reconstructions (red), both in SD units for consistency. Unlike the previous diagram where the series are aligned at the closing reference period (1961-1990) and diverge in earlier periods, when plotted in this perspective, the divergence occurs in the HS-blade at the end. This information is identical to the previous plot, but the perspective is different. In this perspective, the difference seems to arise from the amplitude of the blade.
The PAGES2K non-corrigendum corrigendum did not show a direct comparison of their new reconstruction to the 2013 version, but did show the corrections in Figure 2 as shown below. I’ve overplotted the difference between the H13 and PAGES-2013 reconstructions (in blue), showing that the corrections have already gone about halfway towards H13. Unfortunately, Kaufman did not archive the new version. I’ve written to Nature asking them to require Kaufman to archive the results, but they instead told me to get the results from the authors. I argued that the corrected results should be available officially and not merely in a grey version. I’ll see how this goes.
Given the size of the changes arising from corrections to date, it will be interesting to see what happens when contamination in the Igaliku is fully removed (the incomplete removal being unacceptable) and when the orientation of the Big Round Lake series matched the corrected Hitarvatn series.
Figure 3. Differences to PAGES-2013 version. Black – PAGES-2014; blue- Hanhijarvi et al 2013.
My 2006 AGU Union Presentation
As one further comparison, the next figure compares the Hanhijarvi reconstruction to a figure in McIntyre 2006, The Impact of NRC Recommendations on Climate Reconstructions, a presentation to the 2006 AGU Union session. I present this example, in part, as one further rebuttal of Nick Stokes’ fabricated claim that I have supposedly been reluctant to show the effect of criticisms on proxy reconstructions or that such graphics are in any way “inconvenient”. One of the original themes of Climate Audit has been that the canonical reconstructions relied on small networks and that the same data was used over and over again, and that slight and plausible changes in proxy selection, replacing bristlecones and/or Yamal with equally or more plausible proxies led to very different looking reconstructions. To support this assertion, the closing slide in my presentation showed the effect of slight changes in proxy networks on several canonical reconstructions (Mann, Jones, Esper, Moberg.) Presentation of this figure caused Hughes to splutter in rage, claiming that this was one of the most despicable things that he had ever seen at a scientific conference. I’ve overplotted the H13 reconstruction on this figure, showing that it is, if anything, an even more extreme example of the variations in my 2006 presentation.
Figure 4. Hanhijarvi reconstruction (black) plotted on figure from McIntyre 2006, The Impacts of NAS Recommendations on Reconstructions, AGU Union Session, 2006. For each reconstruction, I had made plausible variations in proxy selection, yielding different long-term appearance, but relatively similar modern “verification”. The Hanhijarvi reconstruction has a similar pattern but greater amplitude.
The Proxy Networks
I’ve already observed that the H13 network is a very large subset of the PAGES2K network (27 of 59 proxies.)
H13 focused on North Atlantic and northern Europe. Between 45 west (in Greenland) and 30.5 east (Finland), PAGES2K used 29 proxies, 27 of which were in the H13 network. The only additional PAGES2K proxies were Hvitarvatn – which Kaufman used upside down; and a Finnish chironomid series (Pieni-Kauro) that did not reach back to the MWP.
PAGES2K used on only five series east of 31 east, only three of which reached the MWP. All three are familiar to CA readers: the Yamal superstick of Briffa 2008 (since downsized in Briffa et al 2013); Avam-Taimyr of Briffa 2008 and Esper’s Polar Urals. None of these break new ground obviously.
25 series come from west of 45 west. There are 16 lake sediment series, many of which have been discussed, mostly critically at CA; there are 4 D’Arrigo tree ring series and 5 ice core series from western Greenland and the Canadian archipelago.
By Proxy Type
Ice core O18: Both H13 and PAGES2K contain substantial subsets of ice core O18, an important proxy class without a material HS. H13 uses 11 of 16 PAGES2K ice core series (Dye3, Renland, Crete, GISP2, GRIP, B16, NGRIP1, B18, Lomonosovfonna, B21 and Austfonna); the five additions in PAGES2K are from west Greenland and the Canadian islands: Camp Century, Agassiz, Devon, Penny (O18 and melt). None of these five are material to the HS. H13 used the longer Lomonosovfonna version, adopted in the 2014 PAGES non-corrigendum.
Ocean: All six ocean cores used in PAGES2K are used in H13: P1003, MD99-2275_diatom, MD99-2275_alkenone, MD95-2011_diatom, MD95-2011_alkenone and MSM5_5-712.
Speleothem: the only PAGES2K speleothem (Okshola) is used in H13.
Documentary: the only PAGES2K documentary series (Bergthorsson Iceland sea ice) is used in H13.
Tree rings: H13 used four tree ring series out of 13 PAGES series: Jamtland, Tornetrask (Grudd), Forfjorddalen-2, Finnish Lapland. PAGES2K added nine series, five from Siberia (Briffa’s Avam-Taimyr and Yamal series; Esper’s Polar Urals series and two other series :Indigirka STD, Lower Lena ARS) and four Jacoby/D’Arrigo series from North America (Seward Peninsula, Gulf of Alaska, Yukon and Coppermine series). PAGES2K-2014 changed to Briffa’s Tornetrask version in 2014.
Varved lake sediments: H13 used 2 Finnish varve series (Korttajarvi, Lehmilampi) in which modern contamination had been removed. PAGES2K added one Iceland series (Hvitarvatn) and five North American series (Blue Lake, Lower Murray, C2, Hvitarvatn, Iceberg, Donard and Big Round), all of which had been previously used in Kaufman et al 2009 and repeatedly used thereafter in recent multiproxy studies e.g. Ljungqvist. The Kaufman series are all problematic.
Chironomids: H13 used one chironomid series (Hamptrask from Luoto); PAGES2K added another Luoto series from Finland and four series from North America (Moose, Hudson, Screaming Lynx and Lake 4). All are low resolution and none look like they would tip the calculation from H13 results to PAGES2K results.
Other lake sediments: PAGES2K also used seven singleton lake sediment series (“singletons” are proxies that only occur once in the network). I think that multiproxy studies should avoid singletons in favor of proxies with enough population to see what might be going on. H13 included one of these series (the truncated Nautajarvi series). Three of the six PAGES2K singletons were withdrawn in PAGES-2014 as no longer “temperature sensitive”: Kepler, Alaska (O18); East Lake, Melville Island (particle size), Holsteinborg, Greenland (Fragilariopsis cylindrus). The closing dates of another series (Lone Spruce, Alaska) were amended by 50 years in PAGES-2014. The Igaliku, Greenland (pollen accumulation) series is severely contaminated by modern agriculture and has a huge non-climatic HS. PAGES-2014 acknowledged the contamination, but only removed the contaminated portion after 1970, leaving prior contamination (enormous) still in the network. The other singleton is the Braya So (alkenone) series.
First, a little backstory, then some discussion.
Hanhijarvi’s background appears to be mathematics, rather than birds and bees and forests, and his publications include actual mathematics. His two coauthors were Martin Tingley, a young Canadian statistician now active in paleoclimate, and Atte Korhola, a prominent Finnish paleolimnologist, both of whom have been mentioned at CA.
Tingley was praised by David Appell in 2009 for the hockey-stickness of his reconstruction with Peter Huybers (then in draft). They(presumably inadvertently) used Mann’s version of the contaminated Korttajarvi sediments – upside-down as well. Although the problem had been well publicized at Climate Audit, Mann’s failure to issue a corrigendum left this contaminated series still in play. In October 2009, a few weeks before Climategate, I criticized their use of Mann’s contaminated data. Their 2009 draft article was not published until 2013. The final article, Tingley and Huybers 2013, still used Mann’s contaminated data, an unwise decision that detracted from the significance of their particular HS. I was very critical of this carelessness in my CA comment at the time.
In the same month (October 2009) that Tingley’s use of upside-down Tiljander was first publicized here, the third coauthor, Atte Korhola, severely criticized Mann’s use of contaminated data, using language and terms that Steyn could directly quote in court (below is Jean S’ translation):
the Proxies have been included selectively, they have been digested, manipulated, filtered, and combined, for example, data collected from Finland in the past by my own colleagues has even been turned upside down such that the warm periods become cold and vice versa. Normally, this would be considered as a scientific forgery, which has serious consequences.
Korhola’s involvement in the present project is salutary as Korhola, unlike Tingley’s coauthor in Tingley and Huybers, is familiar enough with the data to know when it is pointing up or pointing down and not to use data that has been agriculturally contaminated. Keeping proxies in the correct orientation seems little to ask of multiproxy authors, but it seems to be an almost insuperable challenge for many supposed specialists.
I’m not presently in a position to comment on technical merits of paico other than briefly. I haven’t handled the code. (I don’t have any present plans to try to translate from Matlab into R and the authors have thus far only made the package available in Matlab – no complaint on that in terms of documentation, but it does restrict usability for people who speak R.) Like most CPS algorithms, paico presumes that the orientation of each proxy is known ex ante – a protocol that has long been advocated at Climate Audit. This avoids flipping, as can happen in some multivariate methods (including principal components). It also avoids the many scaling and rescaling operations that litter most proxy reconstruction methods: such rescaling is assumed by practitioner to be cost-free, but this is not necessarily the case (but this is outside the scope of today’s post).
I’ve now written a number of posts on PAGES2K (see tag). I had noticed almost immediately after publication that they had used a sediment series with a huge HS due to recent agricultural contamination (Igaliku) – an error eerily similar to Mann’s upside-down Tiljander. In addition, I reported that another of the major contributors to their HS (the Hvitarvatn, Iceland varves) had been used upside-down (also here) and that PAGES2K had inexplicably used a short version of the Lomonosovfonna, Svalbard ice core, even though one of the PAGES2K coauthors had earlier data with a higher MWP. I criticized PAGES’ post hoc exclusion of the Mount Logan ice core O18 series because it went the “wrong way” due to “regional” fluctuations, while keeping the upticked Austfonna series, the uptick of which could just easily be “regional”. I later observed that PAGES’ Kepler lake sediment O18 series was similar to the Mount Logan ice core series and that it was impossible to explain the exclusion of one and not the other (I disagreed with ex post exclusion of Mount Logan). Late last year, I wrote a lengthy review of Kaufman’s varve thickness series that have become common to nearly all recent reconstructions, though even the orientation of these series is sufficiently unclear to confuse specialists and lead to contradictions.
All of these criticisms were adopted in the PAGES-2014 non-corrigendum without specific acknowledgement. Part (but not all) of the Igaliku contamination was removed; the Hvitarvatn series was inverted; the longer Lomonosovfonna series was deployed. Rather than include Mount Logan, they removed Kepler. In addition, they removed the early portion of several tree ring series (Coppermine, Lower Lena, Forfjorddalen) where there were insufficient cores (incorrectly explaining the change as “restricted to temperature sensitive section”) and replaced a couple of tree ring series with different versions (replacing a Swedish version of Tornetrask with a more HS-shaped version from the University of East Anglia and another in Alaska).
Although the above errors warrant a corrigendum, Kaufman did not post a corrigendum at the PAGES2K, here apparently adopting Mann’s technique of not reporting corrigenda at the original article (thereby permitting continued use of the impacted results), but quietly disclosing the errors in a different location as CYA.
In my initial response to the non-corrigendum corrigendum, I observed that there were other shoes that might still drop: Kaufman only removed post-1970 contamination from the Igaliku record, which remains seriously contaminated in the mid-20th century. Some of Kaufman’s other series remain very problematic: the Big Round varve series has extraordinary similarity to the Hvitatvatn series, but they are now in opposite orientation. Other Kaufman muds are a mess: in the amendments, Kaufman removed the early portion of the Blue Lake, Alaska series as not being “temperature sensitive”, but the removal seems completely ad hoc and post hoc; meanwhile, other series with clearly identified inhomogeneities e.g. Iceberg Lake (but HS contributors) are left untouched.
The very large differences between H13 and PAGES2K Arctic arise not from methodology, but from proxy selection, with the diagnosis being sharpened by the fact that H13 is an exact subset of PAGES2K. It would be very interesting to a paico run on the non-H13 subset of PAGES2K – does it have an even more pronounced superstick? This sort of inconsistency between proxies with a medieval warm period and HS-shaped proxies is, of course, familiar to CA readers.
In addition, there are still important remaining issues in the PAGES2K, including unremoved contamination in the Igaliku superstick and the puzzling mirror orientation of Hvitatvan and Big Round. Such changes would take PAGES2K even further to the direction of Hanhijarvi et al.
A few days ago, Jean S extracted their reconstruction, which I’ve overplotted on their Figure 7 to prove the match. If you squint, you can see the dotted black line (digital) overplotting the blue H13 reconstruction. The H13 reconstruction goes from 0 to 2000. H13 Figure 7 showed decadal averages for 0-9 years. The closing uptick in the graphic comes from the single 2000 value. Any other smoothing yields a closing value that is not only below the medieval values, but the 1930s values. The other series in the spaghetti graph are Mann et al 2008 (CPS and EIV), Moberg and Kaufman et al 2009 (red).
Hanhijarvi et al,, 2013 (Clim Dyn.) Pairwise comparisons to reconstruct mean temperature in the Arctic Atlantic Region over the last 2,000 years.
McKay and Kaufman, 2014 (Scientific Data). An extended Arctic proxy temperature database for the past 2,000 years
PAGES2K Consortium, 2013 (Nature), Continental-scale temperature variability during the past two millennia.