The recent revisions to PAGES2K included a dramatic flipping of the Hvitarvatn varve series to the opposite orientation used in the 2013 version used in IPCC AR5. Together with other changes (such as a partial – but still incomplete – removal of contaminated sediments from the Igaliku series), this unwound most of the previous difference between medieval and modern periods. While Kaufman and coauthors are to be commended for actually fixing errors, contamination still remains in the Igaliku series. In addition, the revised Hvitatvatn orientation is now inconsistent with the Big Round Lake (Baffin Island) varve series, which is now almost a mirror image.
The revised Arctic2K removed three series as no longer viewed as being temperature proxies. Each of these deserves to be looked at, as to whether this is simply post-hoc. One of these was a O18 series from Kepler Lake, Alaska. I’ll discuss this in a separate post. Obviously, O18 is a workhorse proxy and it is disquieting that an O18 series can be removed post-hoc with the following footnote:
Omitted (not temperature sensitive).
Removing some O18 series , while keeping other series of the same proxy class that go the “right way”, obviously introduces potential bias since there is an obvious possibility that some of the “right way” examples are overshooting?.
There were 18 O18 series in the original PAGeS2K Arctic network, 15 of which were ice core with one lake sediment series (Kepler Lake – now removed), one ocean sediment series ( P1003, Serjup 2011) and one speleothem series (Okshola, Norway). The Okshola O18 series was inverted in PAGES2K (as in Ljungqvist) to the orientation of polar ice core O18 series (datasets that are the backbone of paleoclimate.)
In the figure below, I’ve compared the Okshola speleothem O18 series (bottom panel) to the Renland, Greenland O18 series (top panel) over the Holocene. As I’ve mentioned from time to time, when one is uncertain of the correct orientation of a series, it seems to me to be a good idea to look for Holocene analogues. Out of the various ice core O18 series as possible comparisons, I chose Vinther’s Renland series, because, as an editorial comment, I think that Vinther’s analysis of uplift and height changes in Greenland ice cores is very convincing and, when I’m looking at data offline, I frequently use it as a comparison benchmark, as I’m doing today. The Renland O18 series (top panel) shows a peak around 9-10000 BP during the Holocene Optimum, with a gradual (Milankowitch) decline.
Although PAGES2K show only a short Okshola version, I noticed a longer version in the SI to Hanhijarvi et al 2013 back to nearly 8000 BP and I’ve plotted it (bottom panel). It shows a decreasing trend through the later Holocene, though with considerably more medium-frequency variability. On its face, there doesn’t seem to be any reason – let alone persuasive reason – for inverting the orientation of the Okshola series, relative to the Greenland ice cores, and yet this was done in PAGES2K and, before them, in Ljungqvist et al 2012.
Figure 1. Top – Renland, Greenland O18 series (Vinther, 2010); bottom – Okshola, Norway speleothem O18 (Linge 2009: data from Hanhijarvi et al 2013).
In addition, although the Okshola series is being included in a high-resolution reconstruction, it only has three values after 1800(?!) – not enough to provide an independent calibration. Nonetheless, the downtick of the Okshola series in the 20th century does seem to give support for this inverted orientation. But, before doing so, it’s also worthwhile to look at Renland in high resolution. While there is much to endorse in Vinther’s analysis, his 20-year series ends in 1960, when there is actually high resolution data to 1985. In the figure below, I’ve shown three windows on the Renland data: look first at the middle panel, where I’ve shown the data in 5-year resolution, one of the available resolutions: it has a downtick not shown in the 20-year data. I’ve plotted this point in black in the left panel. On the right, I’ve plotted annual resolution data in light grey, 5 -year data in black and 20-year data in red (a consistent color scheme below).
This shows support for relatively elevated O18 values in the 1930s, decreasing into the 1980s. On this record, Greenland O18 measurements as of the 1980s were well within the Milankovitch variability (so to speak), and actually somewhat on the “cold” side of the Renland series.
Figure 2. Vinther’s Renland data (adjusted for uplift). Red- 20-year resolution; black – 5-year resolution; grey – annual.
I’d love to see some Greenland data more recent than the early 1990s, but (to my knowledge, and I’ve looked hard), there hasn’t been a single published O18 measurement for the years after the early 1990s. (One of my earliest posts was “Bring the Proxies Up to Date”). To further reassure any reader who thinks that he can see something in the closing squiggle of the Okshola series, here is its modern portion on an expanded scale to facilitate comparison to the Renland right panel above.
Figure 3. Okshola O18, 1800 on.
Conclusion
While PAGES has conceded that they used Hvitarvatn upside down and have commendably addressed their error, there are numerous remaining problems, some of which I’ve discussed already. It seems to me that their justification for inverting the orientation of the Okshola speleothem O18 series is unconvincing. It seems to me that flipping the series makes it inconsistent with better-quality O18 series (as Renland above). Nor do I understand the benefit of including a series with only three post-1800 values in a high-resolution reconstruction.
15 Comments
The last sentence, starting “Nor do I understand …) seems to have been cut off. On my screen it ends: “…and can picture circumstances under”
yes, same here, post may have gone up prematurely?
Same for me.
first image not coming through for me…. I’m on Windows 7 with Google Chrome… anyone else having a problem?
(please delete this comment)
checked from my iPad and also from Windows 8 PC, still can’t see first image
Linge et al.(2009) is ambiguous about the suitability of Okshola δ18O as a temperature proxy:
If there isn’t a consistent relationship between temperature and the proxy, doesn’t that eliminate its usefulness for quantitative reconstructions?
I do not see how a proxy with such issues can give results that inspire confidence. IMO dO18 is the most reliable proxy but cave deposits have complex and imponderable origins.
Are these changes based on new material in the specialist literature… or is it just “because we now say so” or “because it seems more correct this way” or “because our algorithm or PCA has identified this as the more robust way”..?
I find an “omitted – not temperature sensitive” reversal quite troubling. It makes me think the science hasn’t really progressed or isn’t ‘that good’ in the first place if they’re willing to come to such a grand conclusion (as you know, finding that one proxy of a million trees or ice cores that happens to be a temperature proxy would therefore be a diamond-in-the-ruff big deal when apparently so few are declared extant) and then completely wave it off without anything new added to the specialist literature…
it seems disturbingly ad hoc, especially when they remove portions of series during periods when they are supposedly “not temperature sensitive”. However, to compound the confusion, it appears to me that they’ve mis-stated the reasons for not using the early parts of the three tree ring chronologies. The deleted portion of the Coppermine tree ring chronology had fewer than 10 cores – 10 being a rule of thumb in the trade for minimum requirements for a chronology. So the truncation is from usual quality control, not ad hockery, though one wonders why they didn’t do this in the first instance.
Note that Nick Stokes has staunchly opposed this sort of quality control in arguing for retention of Gaspe in the AD1400 network – there were only 1-2 cores in the early portion of Gaspe. Mann claimed in Mann et al 1998 that such quality control had been carried out as one of the supposed strengths of his study, but Gaspe did not meet the claimed QC standards, which therefore seem to have been applied inconsistently.
Proxy nonsense. As an engineer, if I were to record data related to my design’s performance, throw away the ones I didn’t care for, and use others with inverted signs, I would be fired and shamed, and rightly so.
In a comment on the previous post I wondered why the revised reconstruction showed early twentieth century (c.1911-1944) Arctic warming as being twice the size of late twentieth century (c.1976-1978) warming, when global temperature anomalies show the later period as being greater in size. At least part of the answer may lie in the inclusion of the Okshola, Norway speleothem O18 and Renland, Greenland O18 series, which both show a downturn at the end of the twentieth century.
Steve: you’re forgetting that PAGeS2k used Okshola in inverse orientation. It contributes an uptick. I should have noted that PAGES2K uses the 5-year Renland version ending in 1983.
“…one lake sediment series (Kepler Lake – now removed)”
Now that’s odd. Kaufman edited a major synthesis effort looking at NH lake sediments, all under the heading of climate change, in 2012. Kepler Lake has been studied for temperature signal for decades. I wonder what happened.
Last year, I wrote about PAGES2K and Kepler Lake, observing that it closely matched Mt Logan, which PAGES2K had excluded because it went the “wrong way”. I observed:
Having considerable experience of Scandinavian Holocene climate history I can say with very high confidence that any proxy that does not show the early-middle Holocene as being warmer than the present is either upside-down or not a temperature proxy at all.
This is shown by a wealth of data, palynology, plant macrofossils, animal micro- and macrfofossils, treeline changes, lake-level changes, and current distribution of taxa with relict ranges. The postglacial climatic optimum was definitely considerably warmer than the present, both winter and summer.
There is plainly no “potential” with respect to the bias of retaining some d-O18 series but discarding others. The bias is explicit.
2 Trackbacks
[…] temperature anomalies show the later period as being greater in size. Steve McIntyre’s latest post shows that at least part of the answer may lie in the inclusion of the Okshola, Norway speleothem […]
[…] the history of this particular set of discussion I direct you now to: Revisions to Pages2K Arctic / Okshola: which way is up? / PAGES2K: More Upside Down? and PAGES2K vs the Hanhijarvi Reconstruction and which a the time […]