PAGES 2017: Arctic Lake Sediments

Arctic lake sediment series have been an important component of recent multiproxy studies.  These series have been discussed on many occasions at Climate Audit (tag), mostly very critical.  PAGES 2017 (and related Werner et al 2017) made some interesting changes to the Arctic lake sediment inventory of PAGES 2013, which I’ll discuss today.
Some prior Climate Audit criticisms have resulted in withdrawal or major changes or lingering controversy, including:

  • the controversy over Mann’s use of Korttajarvi (Finland) varve thickness and XRD proxies (tag) due to two factors:  recent disturbance of varves by agricultural runoff and road construction; use of the data upside-down to the interpretation of the author. While Mann denied any error and refused to correct, Kaufman et al 2009, who had followed Mann’s incorrect orientation, grudgingly conceded the error after it had been pointed out at Climate Audit and issued a corrigendum inverting the orientation of the series (also reversing the orientation of four other series (of 23) in the corrigendum;
  • the orientation of the Hvitarvatn varve thickness series in PAGES 2013 was immediately criticized at Climate Audit in April 2013 (here). Subsequently, McKay and Kaufman (2014) conceded the error and issued an amended version of their Arctic reconstruction, but, like Mann, refused to issue a corrigendum to the original article. [Correction July 29, 2017: on October 7, 2014, immediately after publication of McKay and Kaufman 2014, I wrote to Nature, pointing out that McKay and Kaufman 2014 primarily addressed errors in PAGES 2013 (not new information) and urged that they issue a proper Corrigendum to PAGES 2013.  In November 2015, over a year later, PAGES 2013 issued a belated Corrigendum.  Neither Nature nor the authors notified me of this and I was unaware of this until a comment in this thread.]
  •  almost immediately after publication of PAGES 2013, I pointed out that their Igaliku series (which had a huge hockey stick) was contaminated in its modern portion by agricultural runoff and erosion. I recall that Nick Stokes vigorously denied that this constituted an error.  McKay and Kaufman 2014 partially responded to the problem by truncating a few data points towards the end (but not fully extinguishing the problem.)
  • I observed that the Kepler Lake (Alaska) d18O series was structurally similar to the Mount Logan (Alaska) ice core d18O series, about which I had criticized its ex post exclusion by PAGES 2013 on spurious grounds – the series went the “wrong” way and was excluded due to “regional” effects.  Instead of re-instating Mount Logan (according to principled ex ante criteria), PAGES (McKay and Kaufman) instead excluded Kepler Lake.

Other prior CA criticisms of Arctic sediment proxies  included:

  • inhomogeneity of the Iceberg Lake varve thickness series (tag) due to changing physical configuration of the moraine-bounded lake;
  • arbitrary exclusion of the early (pre-730) portion of the Blue Lake varve thickness series (tag), where the alleged inhomogeneity was simply that the climate was believed to be “warm with precipitation inferred to be higher than during the twentieth century”, not a physical disturbance.

Some of these issues had material impact on underlying reconstructions e.g. the notorious Mann et al 2008 “no-dendro” series or Kaufman et al 2009 (here).

McKay and Kaufman 2014 (PAGES)

McKay and Kaufman 2014 discontinued the following three series used in PAGES 2013 on the  grounds that there was “insufficient evidence that they were sensitive to temperature”

  • the Kepler Lake (Alaska) d18O series mentioned above;
  • a particle-size series from East Lake, Southampton Island in the Canadian Arctic Archipelago;
  • a diatom (Fragilariopsis cylindrus) series from Holsteinborg Dyb in west Greenland.

They grudgingly inverted the Hvitarvatn series.
Sections of the following two sediment records were truncated on the grounds that the record was not “sensitive to temperature” during these intervals:

  • Blue Lake varve thickness prior to AD730;
  • Igaliku pollen accumulation rate after AD1920

A 50-year dating error in the Lone Spruce (Alaska) BSi series was fixed.
PAGES 2017
Four more PAGES 2013 Arctic sediment series were rejected in PAGES 2017, while two series used prior to PAGES 2013 (but not in PAGES 2013) were re-instated.
The following four PAGES 2014 Arctic sediment series were rejected in PAGES 2017:

  • the Iceberg Lake (Alaska) varve series (previously criticized at CA tag) because of its “unclear relation to temperature”
  • the truncated Korttajarvi X-ray density (XRD, a form of grey scale) see CA tag, originally used in Mann et al 2008 and Kaufman et al 2009, now also said to have an “unclear relation to temperature”
  • the previously truncated Igaliku series (CA tag)now removed in its entirety due to its  “unclear relation to temperature”
  • the Lone Spruce (Alaska) BSi series (introduced in PAGES 2013, but only mentioned in passing in CA posts) was also removed due to its “unclear relation to temperature”.

The SI to PAGES 2017 also contained the following lengthy quality-control comment on Hvitarvatn, which had been a source of PAGES 2013 embarrassment:

Based on studies of glacier mass balance and glaciology in Iceland (e.g. Bjornsson; Flowers), Icelandic glacier fluctuations are dominantly controlled by changes in melt season temperature. Glacier fluctuations influence the production and transport of eroded material and the eventual deposition of this sediment in a downstream basin (i.e. a proglacial lake). .. On short timescales (seasonal, annual, inter-annual), changes in sediment accumulation can be driven by many factors and we can all agree that identifying individual controls is messy. But on longer timescales (for example, centennial timescales, … I would argue strongly that changes in sediment accumulation are driven by changes in glacier size. This is laid out in Larsen et al., 2011 QSR. We subsequently expanded on this initial study to: 1) include the whole Holocene (Larsen et al., 2012 QSR attached, which demonstrates a clear “8.2ka-event” signal and subsequent Neoglacial onset), and 2) by measuring varve thickness in multiple cores along a lake transect and tying the core data to seismic stratigraphy (Larsen et al. 2013 EPSL attached). This latter work demonstrates that the trends in sediment accumulation are consistent and observed throughout the lake basin. Given the available data, I feel comfortable summarizing as follows: Icelandic glacier fluctuations are dominantly controlled by summer temperature. On longer timescales, fluctuations of the Langjokull ice cap can be reconstructed from changes in mean varve thickness at glacial lake Hvitarvatn.
Previous comment: QC failed: article states “sediment flux to Hvítárvatn is dominantly controlled by the integrated rate of sediment production by erosion beneath Langjökull, modulated on annual to decadal timescales by the efficiency of the subglacial fluvial sediment delivery system.”, variability function of proximity, absolute values function of sediment availability. This is _not_ temperature!; QQ by PF not passed

The two paragraphs seem inconsistent.  They did not discuss or attempt to explain the similarity of Hvitarvatn and Big Round varve thickness series.
The following two series were re-instated:

  • Soper Lake, a varve thickness series used in Mann et al 2008. It was too short for either Kaufman et al 2009 (1000 years required) or PAGES 2013 (minimum AD1500 start).  However, it has a pronounced Hockey Stick shape over the shorter period.  I think that the policy of Kaufman et al (restricting to long series) is by far the better way to achieve interpretable results and regard this re-instatement as quality deterioration.
  • Hallett Lake, which like Lone Spruce, was an Alaska BSi series (one with slightly lower resolution). It had been used in Kaufman et al 2009 but dropped in PAGES 2013. It’sfar from obvious why one Alaskan BSi series is now held to be temperature sensitive and not the other – especially when the opposite was concluded in PAGES 2013.

The Hallett Lake BSi series extends back to the mid-Holocene and is shown below to keep changes in the last millennium in perspective. The Hallett BSi series does indeed have somewhat of a Hockey-Stick shape (which might have explained why it was preferred to Lone Spruce BSi), but on a Holocene scale, the blade of the stick is inconsequential – a point that can be missed when multiproxy techniques first convert series to SD units.

PAGES 2017 introduced two new low-resolution alkenone series, both by D’Andrea:

  • Lake E, a Greenland Lake adjacent to the Braya So alkenone series already in PAGES2013. (The original author even combined the two series in a version).
  • Kongressvatnet in Svalbard (16.1 years)

Werner et al (CPD 2017)

Werner et al (CPD 2017) is an Arctic reconstruction by PAGES2017 authors.  Strangely, it rejects three more Arctic sediment series, each of which had been involved in previous controversy:

  • it rejected Hvitarvatn (a non-HS series) on the grounds that its “annual and centennial signal inconsistent”. It didn’t say how it arrived at this conclusion.
  • it rejected Blue Lake (which had very elevated first millennium values) on the grounds that it had a “very nonlinear response, short overlap with instrumental, unclear interpretation”.
  • it rejected Lehmilampi on the grounds that “exact interpretation unclear from original article”. Yet it retained Nautajarvi from the same authors, even though its “darksum” series is, if anything, harder to interpret.

In their reconstruction, they elected not to use 9 series on the grounds that they lacked annual resolution: six chironomid/midge reconstructions (Hamptrask, Lake 4, Pieni-Kauro, Hudson, Moose, Screaming Lynx), three alkenone series (Braya So, Kongressvatnet, Lake E) and the one remaining BSi series (Hallett Lake).
The net result is that the sediment series in Werner et al 2009 reverted back to five series from Kaufman et al 2009 (Donard, Big Round from Baffin Island; C2 and Lower Murray from Ellesmere Island; truncated Nautajarvi, Finland) plus the short Soper Lake (Ellesmere Island) series to help with the HS.

Taking Inventory

The inventory flows of Arctic sediment proxies are summarized below. 32 “different” (non-isomorphic) series were introduced in the four studies as “temperature sensitive”. (For the purposes of this inventory, flipped versions are treated as one series.)  16 of the 32 series were rejected in a subsequent study as not being “temperature sensitive” after all.  This is a very high casualty rate given original assurances on the supposed carefulness of the original study. The casualty rate tended to be particularly high for series which had a high medieval or early portion (e.g. Haukadalsvatn, Blue Lake).

Not only is the number of surviving series (16) a discouraging proportion of the opening inventory, but even the opening inventory was itself culled from a much larger population of lake sediment series.   Making matters worse, because the inventory of proxies changes only slightly from study to study, the networks of sediments in each study are not  independent, as opposed to slight variations.  Because data snooping and ex post selection are endemic, little to no credence can be given to the (very slight) HS displayed by any composite. (This is not to say that the “true” answer is a non-HS, only that the answer is tainted.)
In trying to get a more informed understanding of Arctic proxies, I’ve found it helpful to examine Arctic lake sediments over a Holocene perspective and to examine multiple series (e.g. varve thickness, magnetic susceptibility, BSi) from a single site at the same time.  I think that it is most helpful to work with proxies which are available at multiple sites (PAGES too often uses singletons). I also think that it is best to work out from the best understood sites – a principle used in the mineral exploration industry. I have some notes on some sites that have particularly interested me (Hvitarvatn, Big Round in particular) and will try to write them up some time.
 
 

Source