Climate Audit

Keigwin’s Sargasso Sea temperature reconstruction, used in Moberg et al 2005, was de-selected by Juckes in what he represented to be the Moberg CVM composite and in making the Union composite (although he managed to use Tornetrask twice ?!? under different names). Although he de-selected the Sargasso Sea temperature reconstruction, he used Moberg’s Arabian Sea percentage G bulloides – a HS shaped proxy discussed here on a number of occasions – in both the Moberg CVM and the “Union” composite.

Previously, I reported that a Moberg CVM using the Sargasso Sea proxy instead of the Arabian Sea G bulloides series and Indigirka instead of Yamal led to a very different result – higher MWP than modern.

Juckes did not report that he had de-selected any Moberg proxies and did not identify which proxies were de-selected. When confronted with the Sargasso Sea de-selection, he replied that the proxy ended in 1925, a point which he said that he confirmed with the originator (Lloyd Keigwin). It sort of sounds like a plausible reason, but I thought that I’d follow up.

There are a couple of issues that I had in mind: (1) on its face, Keigwin’s Sargasso Sea proxy is a vastly more valid proxy for SST than G bulloides percentage – which is said by the authors to be a proxy for wind speed; is highly non-normal in distribution with G bulloides counter-intuitively being a cold water foraminifera, demonstrating coldwater upwelling rather than warm SSTs; (2) since both are box cores, I wondered why there would be material differences in the end dates of the series – is this an artifact? (3) are there modern benchmarks for the Keigwin series?

Core Intervals
Both cores are box cores. Bermuda core BC-004 was taken 3 years after Arabian Sea core RC2730 – 1989 versus 1986. Sedimentation rates in Bermuda are higher (~19 cm over the last 1000 years) than at the Arabian Sea site (~11 cm over the last 1000 years.) The sampling interval in Bermuda (1 cm) was wider than the sampling interval for RC2730 (2 mm). Both cores were dated with AMS radiocarbon. The Arabian Sea core was dated through a linear fit to the samples with the intercept removed as a reservoir effect. The Bermuda core was in effect fitted with a pretty flexible curve so that imputed sedimentation rates were very rapid at some points and very slow at other points. Juckes “evaluation” of reconstructions did not evaluate this issue. Intuitively, the dating method used for RC2730 seems better and, for the two cores to be used in some kind of composite, one would have thought that Moberg, at least, would have addressed this issue.

The average bin period for Arabian Sea RC2730 is therefore about 18 years as compared to 52 years for Bermuda BC004. This would imply that the top sample in the ARabian Sea core had a range of 1968-1986 (midpoint 1977) as compared to a Bermuda range of 1937-1989, midpoint 1963. Attributed dates are, however, 1986, and 1925 respectively.

Bermuda: Hudson89-038-BC004

Notes:
(1) Anderson et al 2002; (2) SI to Anderson et al 2002 http://www.sciencemag.org/cgi/data/297/5581/596/DC1/1; (3) ftp://ftp.ncdc.noaa.gov/pub/data/paleo/contributions_by_author/anderson2002/rc2730.txt (4) ftp://ftp.ncdc.noaa.gov/pub/data/paleo/contributions_by_author/keigwin1996/fig4bc-004d_ruber_d18-o (5) Juckes at CA

Keigwin Correspondence
When I noticed this, I wrote to Keigwin to ask about whether there were any narrower samples:

Dear Dr Keigwin,
in your original Sargasso Sea study, it is my understanding that the top part of the core was analyzed in 1 cm intervals. I was wondering if any subsequent analysis of this or nearby cores had been done in narrower (say 2 mm) intervals? Thanks, Steve McIntyre

He promptly wrote back:

Steve, no we never sampled closer than 1-cm. The forams are probably too rare to get enough for dating in such a thin slice, and I doubt we’d be able to extract additional information. Consider the % carbonate curve. It looks pretty smooth at 1 cm spacing, so I doubt there would be higher frequency variability hidden there. Lloyd

It makes one wonder about the Arabian Sea data. Anyway I wrote back to ask:

Thanks. Are there any modern G ruber samples in the same or nearby locations that provide relevant benchmarks? Steve

Keigwin replied almost immediately:

Steve, the best modern samples would be from Werner Deuser’s sediment traps in the Sargasso Sea near Bermuda (cf his 1987? paper I cited in mine). I used his isotope results on those specimens to correct for disequilibrium effects when calculating the paleotemperatures in my data.

I’m puzzled by the fact that the core top foram data do not approach the warming of 1000 yrs ago. In fact, I’ve been keeping a keen eye for other data sets that might show this, and I have seen none. This includes a beautiful but unpublished record from a core in a basin in the Gulf of Mexico. It has much better resolution than the Bermuda Rise record, but the MWP still seems warmer than today.
Lloyd

Benchmarking Ocean Sediment Proxies
The reference to Deuser 1987 in Keigwin (Science 1996) was as follows:

For isotope analysis, I chose the planktonic foraminfera G ruber (white variety 150 to 230 microns) The white variety of this species lives year-round in the upper 25 m of the northern Sargasso Sea and has a relatively constant annual mass flux and shell flux (18- W.G. Deuser, 1987. J Foraminiferal Res 17, 14.). Thus of all planktonic foraminifera in this location this species is most appropriate for reconstructing annual average SSTs (18).

This raises an interesting point connecting back to my concerns about percentage G bulloides as a supposed proxy for (presumably) SST. Keigwin, an authority in the field, could have chosen percentage G bulloides as a proxy, but didn’t. After specifically turning his mind to what was an appropriate proxy, he used G ruber white variety 150-230 microns. In contrast, Moberg et al (followed by Juckes et al) simply grabbed a foraminiferal series that had never been calibrated to temperature – with highly non-normal distributions and inserted it by brute force in small subset composites. Ironically, as I observed before, the same proxy was recently used in Nature (Treydte et al 2006) as a proxy for precipitation in Asia.

Googling Deuser, I re-discovered an Oceanus 1996 article which I’d commented on before, but not in precisely this this context. It stated:

Since 1978, Scientist Emeritus Werner Deuser has collected a nearly continuous suite of deep sediment trap samples at the Ocean Flux Program site near Station S. The Ocean Flux Program traps are shown following recovery aboard the Bermuda Biological Station vessel Weatherbird II. The traps were deployed along a bottom tethered mooring at 500, 1,500 and 3,200 meters depths to intercept particles sinking through the water column.

Importantly, it contained the following graphic in which Keigwin’s reconstruction is explicitly linked to modern SST.

Caption: Estimated sea surface temperature from Station S annual averages and from Globigerinoides ruber shell oxygen isotopes averaged at 50-year intervals. Note that the range of sea surface temperature variability on longer time scales is much larger than what has been observed since 1954 at Station S.

The Oceanus 1996 article went on to show the following graphic for Bermuda Station S linking SST to estimated dO18:

Caption: Bermuda Station S hydrography shows the oxygen isotope ratio that a foram would have if it deposited its shell in equilibrium with the annual average sea surface temperature and salinity observed since 1954 at Station S near Bermuda. The large decrease in sea surface temperature and increase in salinity in the late 1960s was caused by unusually unpleasant weather those years. (Temperature and salinity data provided by Terry Joyce)

Conclusion:
In the Keigwin reconstruction, there is a bit of splicing of modern sediment trap information with core information – indeed Keigwin refers to this in his calibration. But is this a strength or a weakness? And is it done in other chronologies? Think about tree ring chronologies. Collections of living trees typically do not yield millennial chronologies. These are obtained by splicing modern samples with subfossil information. I would submit that there are many and greate inhomogeneities in the splice of living chronologies to subfossil chronologies (e.g. changing altitudes, latitudes; modern sample bias; other topics discussed here) than in Keigwin’s linking of Deuser’s sediment trap information to his box core information.

It would be very difficult to develop objective criteria for excluding this linkage while at the sametime permitting use of Briffa’s version of the Yamal tree ring chronology. If there is, Juckes sure hasn’t provided. Indeed, Juckes did not provide any objective criteria for rejection of this series in the article. Indeed, he didn’t even report the de-selection, which was only identified through detective work here. He provided an ad hoc reason after the issue was raised.

As to the calibre of the data: Keigwin identified G ruber as a SST proxy. Where is the evidence that percentage G bulloides is a SST proxy? It was not calibrated to temperature in the original article, where the rationalization for its inclusion was weak to say the least. Juckes says that he’s “evaluating” reconstructions – how can he do that without evaluating the quality of the proxies in small subsets?

Finally, Keigwin says that there’s a “beautiful” series at higher resolution from the Gulf of Mexico waiting in the wings. we know that if you take cherry-picked and data-snooped series – Yamal, bristlecones/foxtails, and Dunde – mix in some noisy “proxies” with no common signal, you can get a HS series in different ways. Whoop-de-doo. It’s new samples that haven’t been data snooped should be the life blood of multiproxy studies. Indigirka. Millar et al 2006’s ecological niche study. Thompson’s unreported Bona-Churchill dO18. Hughes’ unreported Sheep Mountain bristlecones. The new Gulf of Mexico series.