The PAGES2K article to be published tomorrow will show eight regional reconstructions, which are plotted below. In today’s post, I’ll try to briefly summarize what, if anything, is new about them.
Antarctica: This is a composite of 11 isotope series (mostly d18O). It includes some new data (e.g. Steig’s new WAIS series) and some long unavailable data (Ellen Mosley-Thompson’s Plateau Remote). It shows a long-term decline with nothing exceptional in the 20th century. Steig has recently characterized the recent portion of Antarctic isotope as “unusual”, but this is really stretching the facts to the point of disinformation. I’ll post separately on this.
Arctic: This is a somewhat expanded version of the Kaufman data, unsurprising since Kaufman seems to have been the leader of the program. It shows an increase from 1800 to 1950, with leveling off since 1950. Its modern values are higher than medieval values. It is heavy on varvology (22 varve series), but, like Kaufman et al, also has ice cores (16) and tree rings (13, including Briffa’s Yamal) plus a few others. They use Korttajarvi, but Kaufman has issued one correction on this already in 2009 and avoided use of the contaminated portion. We’ve discussed Arctic d18O values from time to time, observing that their 20th century values are rather unexceptional. My surmise is that the varve data, which, as discussed in other CA threads, is highly problematic, is the main contributor to the modern-medieval differential in the PAGES reconstruction.
Asia: This reconstruction is based entirely on tree rings (229 series), all, interestingly, used in a positive orientation. 20th century values are elevated but the reconstruction lacks the distinctive blade of, for example, the Gergis stick. The majority of the tree ring data is unarchived: chronologies have been included in the PAGES2K data, but the underlying measurement remains unarchived.
Australia: this is the Gergis reconstruction. There are only two long series (both tree ring). As is well known, Gergis picked data according to ex post correlation to temperature (contrary to the representation in the disappeared article). The present network is little changed from the network in the disappeared article, with the precise differences remaining to be explained. The network is about half tree ring data and about half is short coral (nearly all O18) data. The blade in the Gergis stick comes almost entirely from coral O18 data – for which corresponding medieval information is lacking. The reconstruction is thus a sort-of splice of low-amplitude tree ring data with high amplitude coral O18. Coral specialist literature nearly always uses Sr data as a measure of temperature. The 20th century increase in coral Sr data is much less than O18 data: however, Gergis screened out the Sr data and almost exclusively used coral O18 data.
Europe: The network is 10 tree ring series and one documentary. I don’t know at present how the series were chosen. Most of the increase in the reconstruction took place prior to 1950. Late 20th century values equal and then exceed mid-century values. It will be interesting to see whether sustained ring widths will be maintained with these particular chronologies during warmer temperatures.
North America. There are two North American reconstructions. A reconstruction using pollen is at 30 year intervals and ends in 1950. It shows elevated temperatures in the late first millennium that exceed the most recent values in the series. The other reconstruction uses tree rings. It includes many series from the MBH98 dataset, including the Graybill bristlecone chronologies. Although the tree ring data is accurately dated, the reconstruction is only reported at 10-year intervals. Although the data set includes new data reaching into the present century, the reconstruction is shown only to 1974.
South America: This network is particularly hard to understand. It shows particularly low medieval values relative to the modern period – a point that is relevant to assertions on medieval-modern differential. The network also uses intrumental data. It has two long ice core series from Quelccaya, which, as previously noted, appear (according to the SI) to have been inverted, a decision which, if correct, would rather detract from conclusions about modern-medieval differential drawn from this reconstruction, given that the medieval portion of the reconstruction only has a few contributors, of which Quelccaya is prominent.
Climate Audit 
48 Comments
RC has a suppressed post on the PAGES2K reconstruction preserved in a CA comment https://climateaudit.org/2013/04/20/pages2k-south-america/#comment-414634
The RC post has been liberated.
I was wrong here: the Antarctica reconstruction is indeed the full recontruction not the combined West and East reconstruction as I suggested. But they have cut off the 2-166AD part from it!
Real Climate has unsettled minds
For you’ve put them in previous binds
Their fears have increased
So that post’s not released
They’re watching, awaiting your finds
You’d think that the Team would now know
That the mud from Tiljander won’t go
These folks have some nerve
Using those varves with verve
Plus the odd trees to put on this show
And results are of course overrated:
Between data that’s oddly re-dated
Using odd, delayed starts
Dropping “off-message” parts
While unhappy late bits get truncated
===|==============/ Keith DeHavelle
Not much to say except thanks for the breakdown. The Asia data is pretty big, it might be fun when/if it is released.
I’ve collated the data from the Excel spreadsheet to
Proxy.tab
recon.csv
info_pages2k.csv
at http://www.climateaudit.info/data/multiproxy/pages2k
Proxy.tab is an R-list of 8 items for the 8 regions, each collated to a time series.
Proxy[[“asia”]] is the Asia network of chronologies. You need to use download.file (…mode=”wb”) to download R-objects.
Buaah!! This is turning from a tragedy to a comedy and the paper has not even published yet! Darrell Kaufman:
“Similar”, only in Climate Science! I plotted Hanhijärvi’s recons (based on the same data as the “official” recons plotted by Steve above). Results:
Australian (Gergis) Hockey Stick — poof — gone
European Hockey Stick — poof – gone
South American Hockey Stick — poof — gone
Asia (no hockey stick even in P2k version) — poof — much elevated MWP
Only Arctic hockey stick remaining. Pretty funny as Hanhijärvi did not find any hockey stick in his own recent Arctic Atlantic area reconstruction:
Sami, I think you may have a communication problem with these people. I’m pretty sure you didn’t say that your PaiCo reconstructions are similar to those plotted by Steve as they seem to differ in the crucial places (MWP – modern period). Maybe next time you should use some good old Finnish adjectives like “perkele” to get your real message across. 🙂
Jean S, if the paper as published also makes this claim, rather than just a redacted RC comment from one of the apparent authors, then you should take this evidence straight to the journal and demand a correction or retraction.
Same as with the Gergis portion if the new paper makes the same proxy selection claim as the one that was withdrawn when shown not true.
Should be an interesting week, as we will soon know whether the actual paper contains such provably wrong statements, or is silent, or is nuanced.
I bet silence on Gergis, and nuanced claim without support on PaiCo. That is a bet that these folks are reasonably smart, and moderately evasive. Risky bet, because it might well be they are not that smart, as Marcott proved thanks to you and Steve M. Hope I lose the bet, and that they can be nailed on more bad science erroneously represented.
Regards
Re: Rud Istvan (Apr 21 15:36),
I guess you have not been in this “game” long enough … I just read the SI (the paper is now released) where they make the comparision, and indeed they look at least somewhat similar … then I looked the units: sd. So instead of directly comparing the two reconstructions, which both are supposingly in temperature anomalies, they converted them to zero mean and unit variance (over the whole series length) and then compared! Does it make any sense? No, but they look somewhat similar when you plot them on top of each other 😉
But now I have to go to sleep. You guys have a nice evening with the paper, I’m sure there are plenty of interesting things to find! For instance, it seems to contain the new Gergis et al. selection criterion. Also Antarctic REs are reported there (in a figure!) … apparently didn’t bother any reviewers.
If it makes any sense to correlate these proxies, Spearman rank correlations are generally significant (p<0.05) but low, ranging from 0.07 to about 0.3.
https://dl.dropboxusercontent.com/u/75831381/PAGES%202K%20stats%20%26%20Spearmans.docx
Lance, I too looked at the table. In natural world, correlations above 0.9 are rare, between about 0.6 and 1 are potentially useful and the rest is really no correlation at all, despite the significance test.
These data are so noisy that you have to (weary old phrase) torture them to get something to write home about.
Seems odd that six proxies have >1000 observations each, but North America has only two (pollen and trees), each with <100 observations. Why so few?
I have posted an active viewer for the Pages2K proxies.
Nick, your graphic is a great way to look at a huge and comprehensive number of proxies all in one place and one setting.
Thanks Kenneth,
I must say that Pages2k provided a very well organized spreadsheet, which made it a lot easier.
Thank you, Nick. This adds clarity quickly. Geoff.
This probably explains a lot of what we see in climate science …
“For many young people who aspire to be scientists, the great bugbear is mathematics. Without advanced math, how can you do serious work in the sciences? Well, I have a professional secret to share: Many of the most successful scientists in the world today are mathematically no more than semiliterate.”
http://online.wsj.com/article/SB10001424127887323611604578398943650327184.html
Print news in Australia today, 20th century ‘hottest in 1400 years’. No mention of Gergis or the controversy.
The Real Climate post is back up and I noticed some superficial things that didn’t make sense. They say there is no evidence of a worldwide Mideval Warm Period or Little Ice Age. But looking at their graph, 800-1000AM looks rather warm in all the reconstructions that extend back that far. There is also a rather cool period also in virtually all proxies in the 18th or 19th century. There is the usual bit about “the present appears warmer than any period in the last 1400 years. Are they still trying to justify that one?
To supplement the scary heat target map of Oz in the Herald Sun and the interview with Phipps in the Age, there are some lovely quotes from Overpeck responding to Kelly’s cue-card question on Oz ABC RN Breakfast Radio:
http://www.abc.net.au/radionational/programs/breakfast/global-study-reconstructs-temperatures-over-2000-years/4642898
Michel Mann is the man who manages to isolate one spike out of these irregular “sawtooth” graphs and pronounces it the Medieval Climate “Anomaly”. Maybe he finally saw the light? He does not label any other anomalies.
/sarc off
The PAGES 2K paper is now published online.
I can’t see Gergis et al in the article references but the SI Note 52 reads:
52. Gergis, J.et al.
Evidence of unusual late 20th century warming from an Australasian
temperature reconstruction spanning the last millennium.
J. Clim.(in review).
Every time I check in to climateaudit, I amazed at the amount of info that’s posted. How you yeomen and/or yeowomen find the time for such thoughtful analysis is beyond me. Keep up the good work and Thanks.
AAP (Australian Associated Press) already has this piece of propaganda out about PAGES 2K:
http://www.theaustralian.com.au/news/breaking-news/late-20thc-was-warmest-in-1400-years/story-fn3dxix6-1226625537082
I’m puzzled by the Australasia bit. You say it is just the withdrawn Gergis construction. And they say in the SI that “all data were linearly detrended over the 1921-1990 period”, same wording as in the Gergis ex-paper. But of course this was the claim made in the Gergis paper that turned out to be untrue leading to the paper being withdrawn. ??
I didn’t say that it was “just the withdrawn” reconstruction. I said that the network is mostly identical – they used 21 of the 27 proxies of the disappeared paper and even kept the numbering in their proxylist as little changed as possible – and that the reconstructions are substantially the same. At the time of my first post, I did not have access to the SI which appears to say that they used detrended correlations. However, a straightforward application of detrended correlations would yield the results previously reported by Jean S and myself.
Here’s a longer excerpt from the SI (which was not available at the time of my post):
So you have to watch the pea. Originally they checked for correlation with the Australian temperature that they were trying to reconstruct. This was changed to correlation with some gridcell somewhere. Note that their algorithm checks numerous gridcells but they do not appear to have changed their “significance” benchmark to reflect the search of numerous gridcells.
It seems odd that this changed criterion would result in a list substantially similar to the original criterion. My guess is that they would have to experiment quite a bit before finding this particular criterion. A methodological report according to SImonsohn’s anti-fraud criteria would require them to report all the criteria that they had examined before settling on the rather strange criteria ultimately adopted.
They also write, “To account for proxies with different seasonal definitions than our
target SONDJF season (for example calendar year averages) we calculate the
correlations after lagging the proxies for -1, 0 and 1 years.” And all without (apparently) modifying the p-value threshold to account for multiple testing.
That’s a new one, as far as I’ve heard here. There could be four gridcells within a 500km radius of a proxy. If the three lags are used with the gridcell picking at the same time, this is not Pick Two Keno but rather “Pick Three to Twelve Daily Keno” depending on where the proxy is located.
Re: HaroldW (Apr 22 13:38),
I have hard time following the rationale for the lag +1 (I’m reading “lag” as something subtracted from the value).
They supposingly dated their target summer season instrumental according to the SOND year (since there are four months). Let’s call that year n. Now lagging a proxy record by -1 means that they correlate the target instrumental of the year n with the proxy measurement from the year n+1. That makes some sense, since the target value from the year n includes JF from the year n+1.
But the lag +1? Now they correlate the instrumental value from the year n (consisting of the average of SOND from the year n and JF from n+1) with a proxy value from the year n-1??? What kind of proxies are those in Australia that can predict temperatures?
I also wish that if any of the series were included based on a non-zero lag, they also remembered to do the same lagging when they constructed the reconstruction.
If the proxies are very good and the data is not autocorrelated, the -1 wouldn’t matter much, because it’d only very rarely be chosen (as the correct lags, most likely lag 0 but possibly lag +1, will correlate better).
However, if the proxies are noisy or autocorrelated (or both, as is the reality), the “wrong” lag (-1 or arguably +1) could get chosen pretty frequently. Even if we say that lag 0 and lag +1 are equally preferable, you still gave yourself some materially increased chance to pick a better correlation. If the gridcell procedure wasn’t adjusted for in benchmarking, this probably wasn’t either. I do again wonder if the checking against multiple gridcells and multiple lags were performed in the same step, which would multiply (rather than add) the number of unaccounted-for daily keno picking.
JasonScando: “There could be four gridcells within a 500km radius of a proxy.”
A 5×5-degree gridcell is about 550km in extent in the latitude-direction, and smaller than that by the ratio of cos(lat) in the longitude-direction. Assuming that what the authors meant by “grid-cell within a search radius of 500 km from the proxy site” is that the gridcell *center* is within that distance, then (at the equator) sites more than about 50 km from a gridcell center in lat and long will have four such gridcells (2×2) qualify. Roughly speaking, about 80% x 80% = 64% of locations will be in this category. ~32% will have two such gridcells, and only ~4% will be tested against one gridcell. Further from the equator, it is possible to involve more gridcells. E.g. in Tasmania (lat 42S; cell size ~400 km in longitude) either two or three longitude cells fit the description.
On the other hand, if the criterion is that some part of the gridcell is within 500 km of the proxy site [which seems a little unlikely, to be sure], then up to nine gridcells (3×3) would qualify, and four (2×2) would be a minimum, at the equator.
Steve: I’m not sure why you regard this possibility as “unlikely”. Regardless, it’s the sort of thing that is easy enough to state clearly and they ought to have done so.
Jean S (Apr 22, 2013 at 3:17 PM) –
I don’t see any need for the authors to involve lag for the instrumental record; all that’s necessary is to (say) define “seasonal year X” as SOND of calendar year X and JF of X+1, and compute the instrumental average. However, there may be uncertainty about the year reported by the proxy compilers; perhaps they called that season “year X+1” rather than X. Obviously, one should try to identify the choice made in the proxy report, if this is possible then there is no need for alternative correlations. But at most there can be two comparisons.
And yes, I agree that if the correlation passed with a lag, then the proxy values should also be lagged in the reconstruction. As usual, words alone leave ambiguities; only the code is definitive.
Steve: I’m not sure why you regard this possibility as “unlikely”.
Mainly because it requires extra code to compute whether any part of a grid cell is within 500 km of a proxy site, compared to computing whether the grid cell center is within range. To extend the number of comparisons, it would be easier to use the simpler method but increase the free parameter of distance.
Is there a valid rationale for classifying equatorial trends seasonally?
Mann et al 2008, that laboratory of statistical horrors, also picked the better of adjacent gridcells, a point discussed at the time. The following text was in a draft post that for some reason wasn’t published at the time, but which shows the effect of pick-two in increasing the yield of “significant” correlations. Gergis multi-pick would obviously excerbate the effect,
So far we’ve described how M08 deflated the benchmark from arguably 22% to 13%. The other leg of their strategy was grossing up the proportion of “passing” proxies from 22% to 28%. This was done by the use of what I’ve called “Pick Two Daily Keno” to measure correlation, while using Pick One benchmarks (as above).
The Pick Two methodology is stated as follows:
Through inspection of source code, we’ve been able to determine that Mann chose the correlation with the highest absolute value. This has the effect of “hollowing out” the middle portion of the distribution – a point that we observed last fall and illustrated below. In the dendro network, the yield above 0.1063 increases from 22.1% to 28%.
Figure 4. M08 Dendro Network: Pick Two versus Pick One
A similar increase in yield will occur in any red noise situation. If the two comparanda gridcells were uncorrelated, then the expected yield would nearly double (a little less due to overlap). Because adjacent gridcells are correlated, the expected increase is noticeably less than a double. What would be the expected increase? Who knows? My guess, based on this network, is that it would be to about 28% 🙂 Obviously, Mann should have accounted for pick two in his benchmarks.
But aside from that, using pick two is objectionable because it is pointless and weird. Has anyone ever seen this sort of test used in a relevant case? (If so, it should have been cited to show the purpose of this test.) Absent any justification, it has the appearance of being introduced for no purpose other than to gross up the yield of passing proxies – an appearance made all the more disquieting by the failure to account for pick two in the benchmarks.
I don’t know of any cases where it was used, but I do get the logic of it. The boundaries of grid cells are arbitrary. That a proxy falls within a grid cell doesn’t mean the stations within it are the ones that represent it best. Pick-two (or more) methods could theoretically improve the skill of a reconstruction.
There’s no excuse for not changing the benchmarks though. Or for that matter, not testing your results with a pick-one method.
Already featured on SkS: http://www.skepticalscience.com/pages2k-confirms-hockey-stick.html – and Mann is more than happy to retweet the announcement.
It took that long?
Since the paper is behind a paywall and they presumably have access to it – maybe someone would want to ask if it includes a physical explanation for their use of some of the South American proxies?
Specifically, why CAN Composite 4 (tree rings) has a negative correlation to temperature and CAN Composite 31 (more tree rings, about 8 KM away from CAN Comp 4) has a positive correlation with temperature? I would, but can’t bring myself to either read or post at SkS anymore given their past shenanigans…
Assuming the coordinates are accurate, one looks a lot drier today than the other and about 100 m higher http://itouchmap.com/latlong.html
They are different species too:
1)”Dendroclimatology of high-elevation Nothofagus pumilio forests at their northern distribution limit in the central Andes of Chile.”
2)”Araucaria araucana tree-ring chronologies in Argentina”
The negative correlation of the second paper was noted by the original authors:
“Regional tree growth is strongly negatively related to temperatures during summer and fall in the previous-growing season and spring in the current-growing season, respectively.”
They interpret this a due to lack of water making warmth a negative.
First reference has a similar conclusion about negative correlation of growth to temperature: “Results indicate that, at the northern sites (Vilches and Laguna del Laja), the tree-ring growth of N. pumilio is positively correlated with late-spring and early summer precipitation and that higher temperatures reduce radial growth, probably because of an increase in evapotranspiration and decrease in water availability.” As does this related source:”Tree growth at the northernmost regions shows a positive correlation with annual precipitation (PC1-prec) and negative correlation with mean annual temperature (PC2-temp), under a Mediterranean-type climate where water availability is a major limiting factor.”
Thanks!
Revkin covers Pages2K at DotEarth and includes a long guest post by Kaufman:
http://dotearth.blogs.nytimes.com/2013/04/22/study-charts-2000-years-of-continental-climate-changes/#more-49117
I posted a link to the three relevant CA threads with a brief intro about half an hour ago… not sure if the comment will be approved or not (it hasn’t yet, but I’m not sure how long DotEarth typically takes to moderate comments).
Jason: it could take as long as overnight. I posted something similar, although more general. I’ve never had a problem with being censored there. Revkin wants both sides to be heard.
Couple of questions from an ignoramus:
1) Is this The Big Meta-Study that combines all knowledge on the last two millenia? If it is, what are the conclusions? Was there a Medieval Warm Period, or not?
2) What are the final conclusions of the statistics-critics? Are there reliable conclusions that can be drawn from all this, or not, and what are they?
3) Are the data and methods (finally) completely out? If so, could those who disagree with how various statistics were done – redo them? Are there competing studies done by skeptics?
I’m trying to see if there are Big Picture conclusions that believers and skeptics should agree on.
Steve: this study has been out for only a few days. It takes more than a minute or two to figure out what they’ve done. They’ve made a much better than usual attempt to document what they’ve done, but it still takes time. At first blush, pretty much every criticism of (say) Kaufman et al 2009 on varvology or of Gergis/Neukom applies to the new variations.
Were any of the hockey team involved in this ? Although one can never see behind the scenes, I suspect that they are fairly radioactive to many (though certainly not all) members of the field.
6 Trackbacks
[…] Over at Climate Audit, Steve has been looking at the resurfacing of some Hockey Stick science papers, and the fun continues here and here. […]
[…] The Climate Auditor is onto it, and dissembles the new paper here: PAGES2K Reconstructions […]
[…] been reporting on it for a number of days. See Steve McIntyre’s posts here, here and especially here. WattsUpWithThat has discussed the paper here and here. SkepticalScience responded to the paper as […]
[…] been reporting on it for a number of days. See Steve McIntyre’s posts here, here and especially here. WattsUpWithThat has discussed the paper here and here. SkepticalScience responded to the paper as […]
[…] McIntyre su Climate Audit (qui, qui e […]
[…] De klimaatveranderingen van de laatste twee millenia vertoonden een sterke regionale expressie. Om het ruimtelijk-tijdelijk patroon op te helderen heeft het PAGES 2K project de temperaturen van de afgelopen millenia gereconstrueerd voor zeven regio’s op alle continenten. Wat de reconstructies gemeen hebben is de bijna overal aanwezige opvallende afkoeling die pas tegen het einde van de negentiende eeuw eindigt en overgaat in een periode van sterke opwarming, vrijwel overal op aarde. Op een schaal van decennia en eeuwen vertoont de temperatuurvariabiliteit duidelijke regionale patronen, met meer overeenstemming binnen de beide halfronden dan tussen hen onderling. Er blijkt géén sprake van synchrone en wereldwijde warmme en koude intervallen zoals een middeleeuwse warme periode (Medieval Warm Period of MWP) of Kleine IJstijd (Little Ice Age of LIA), maar alle reconstructies laten tussen 1580 en 1880 koude condities zien, met in sommige gebieden gedurende de achttiende eeuw kortdurende warmere intermezzo’s. De transisitie naar deze koudere concities gebeurde in het Arctisch gebied, Europa en Azië eerder dan in Noord-Amerika of in de regio’s van het Zuidelijk Halfrond. De recent opwarming keerde de langdurige trend tot afkoeling. Gedurende de periode 1971–2000, was de aan het oppervlak gerelateerde gewogen gemiddelde gereconstrueerde temperatuur hoger dan in in welke periode dan ooit in de afgelopen 1.400 jaar. Opvallend ook nu weer is dat de opwarming in het Arctisch gebied sneller gaat dan in de rest van de wereld. Deze conclusie sluit heel goed aan bij de waarnemingen dat het Noordpoolijs als sneeuw voor de zon verdwijnt en dat Groenland, gletsjers en andere ijskappen sneller smelten dan ooit tevoren. > climateaudit.org: Overzicht van PAGES2K reconstructies […]