Guest commentary by Darrell Kaufman (N. Arizona U.)
In a major step forward in proxy data synthesis, the PAst Global Changes (PAGES) 2k Consortium has just published a suite of continental scale reconstructions of temperature for the past two millennia in Nature Geoscience. More information about the study and its implications are available at the FAQ on the PAGES website and the datasets themselves are available at NOAA Paleoclimate.
The main conclusion of the study is that the most coherent feature in nearly all of the regional temperature reconstructions is a long-term cooling trend, which ended late in the 19th century, and which was followed by a warming trend in the 20th C. The 20th century in the reconstructions ranks as the warmest or nearly the warmest century in all regions except Antarctica. During the last 30-year period in the reconstructions (1971-2000 CE), the average reconstructed temperature among all of the regions was likely higher than anytime in at least ~1400 years. Interestingly, temperatures did not fluctuate uniformly among all regions at multi-decadal to centennial scales. For example, there were no globally synchronous multi-decadal warm or cold intervals that define a worldwide Medieval Warm Period or Little Ice Age. Cool 30-year periods between the years 830 and 1910 CE were particularly pronounced during times of weak solar activity and strong tropical volcanic eruptions and especially if both phenomena often occurred simultaneously.
Figure: Thirty-year mean relative temperatures for the seven PAGES 2k continental-scale regions arranged vertically from north to south.
The origin of the ‘PAGES 2k Network‘ and its activities can be found here and consists of nearly 80 individual collaborators. The Consortium’s collection of local expertise and proxy records was transformed into a synthesis by a smaller team of lead authors, but the large author list recognizes that the expertise of the wider team was essential in increasing the range of data used and interpreting it.
In addition to the background available at the FAQ, I think it is important to also highlight some aspects of the analytical procedures behind the study and the vital contributions of three young co-authors.
The benefit of the ‘regions-up’ approach embodied in the PAGES-2k consortium is that it made it easy to take advantage of local expertise and include a large amount of new data that would have been more difficult to assemble for a centralized global reconstruction. However, being decentralized, the groups in different regions opted for different methodologies for building their default reconstructions. While justifiable, this does raise a question about the impact different methodologies would have. To address this, the synthesis team (ably led by Nicholas McKay) applied three particular reconstruction methods to all of the regions, as well as looking at the basic area-averaged and weighted composites. He further analyzed the site-level records individually and without many of the assumptions that underlie the regional temperature reconstructions. These results show that the long-term cooling trend and recent warming are dominant features of the dataset however you analyze it. There is a sizable fraction of the records that do not conform to the continental averages, highlighting the spatial variability and/or the noise level in specific proxies.
One of the new procedures used to reconstruct temperature is an approach developed by Sami Hanhijärvi (U. Helsinki), which was also recently applied to the North Atlantic region. The method (PaiCo) relies on pairwise comparisons to arrive at a time series that integrates records with differing temporal resolutions and relaxes assumptions about the relation between the proxy series and temperature. Hanhijärvi applied this procedure to the proxy data from each of the continental-scale regions and found that reconstructions using different approaches are similar and generally support the primary conclusions of the study.
Regions where this study helps clarify the temperature history are mainly in the Southern Hemisphere. We include new and updated temperature reconstructions from Antarctica, Australasia and South America. The proxy records from these three regions come from many sources, ranging from glacier ice to trees and from lake sediment to corals. Raphael Neukom (Swiss Federal Research Institute WSL and University of Bern) played a key role in the analyses across the Southern Hemisphere. He used principal components regression (Australasia), a scaled composite (Antarctica), and an integration of these two approaches (South America) to create the time series of annual temperature change.
Inevitably, assembling such a large and diverse dataset involves many judgement calls. The PAGES-2k consortium has tried to assess the impact of these structural decisions by using multiple methods, but we hope that this synthesis is really just the start of a more detailed analysis of regional temperature trends and we welcome constructive suggestions for improvements.
References
- . , "Continental-scale temperature variability during the past two millennia", Nature Geoscience, vol. 6, pp. 339-346, 2013. http://dx.doi.org/10.1038/ngeo1797
- S. Hanhijärvi, M.P. Tingley, and A. Korhola, "Pairwise comparisons to reconstruct mean temperature in the Arctic Atlantic Region over the last 2,000 years", Climate Dynamics, vol. 41, pp. 2039-2060, 2013. http://dx.doi.org/10.1007/s00382-013-1701-4
Re 10:
In response to my original suggestion of calculating the correlation between nearby proxies as a cross-check on their quality Dr. Steig responded with,
“in my experience, the correlation among weather station data is no higher than among similarly spaced paleoclimate data”.
As I mentioned earlier (#35) I had taken this to mean that neither weather stations nor proxies were very well cross-correlated, but in light of the BEST team’s data I guess I need to revise that. Taken as written it implies that proxies should have a higher degree of cross-correlation than weather stations, but I doubt that was the intent and will assume it to mean they should be the same.
Given the discussions above of geographic factors in modifying the correlation of weather stations, it also occurred to me that there must be decades of temperature data from commercial weather reporting for many of the areas where proxies originate. Accordingly, I feel justified in asking again: could these data be used to come up with a “target” cross-correlation fro the proxies? And could these cross-correlations form a basis for assessing the signal to noise ratios in the proxies?
I’m really having trouble understanding why this idea seems to be dismissed out of hand.
[Response:I didn’t mean to appear to have dismissed anything out of hand. I was merely responding to a specific assumption you were making, which may not always be correct. I think your suggestion about cross-correlations for for assessing the signal to noise ratios in the proxies is good. –eric]
Re 51:
Eric, thanks for your response.
Now if we could only do something about the SNR of the blog comments … ;-)
[Response: Indeed. –eric]
> watcher
> my hypothesis that closely spaced weather stations
> should be better correlated than more distant ones
But they say they make the adjustments they describe.
They consider
“the temperature measurement for a given place and time as the sum of four terms: an average global temperature Tavg; the positional variation caused by latitude or elevation; the measurement bias, or offset variable; and the temperature associated with local weather”
They adjust away the latter three terms, leaving the first one, right?
(Someone will surely correct this oversimpification)
Seems like you assume the caption for a figure is the complete statement of the adjustment, then your hypothesis — derived from reading that one caption — is that the paper looks wrong.
Could be your reading of that caption is incomplete.
Just sayin’, doubt yourself first.
They’re nice folks over at BEST, they do have a FAQ and publish their work. You might do better asking them there.
Watcher:
The thin skin appears to be mostly yours. It seems relatively straightforward that paleo data are difficult, and if you assume good faith on the part of the majority of the scientific community, along with relative intelligence, you might not be so ready to take offense when you suggest that they do their jobs, and that they have not already done so. Because the temperature record is important and has been under constant attack, well supported by industry which has vast funds to put into an alternative universe of think tanks and detractors, scientists have strained every muscle and every brain cell to come up with ways to measure, be honest about uncertainty, qualify it, and extend the record.
This effort is worthwhile, but defending it is wearying, as the same old suspicions pop up hundreds and thousands of times, unchanging and relentless and unaccepting of on-topic responses. You may not be aware that you are treading a well-trodden path, but your first statements looked familiar.
We are never going to get perfect data for thousands of years ago, but this is a worthy effort. If we start with the premise “what can we learn from this” instead of “what’s wrong with it” we could get on.
Perhaps you are genuinely interested in learning more, but your approach was guaranteed to raise the hackles of a community under constant attack from forces that are consistently hostile and prone to use what they know about science to destroy rather than build knowledge (and these attacks are not limited to science, it gets personal when your family gets physical threats).
It seems obvious that when data sets are different by necessity, but measure the same thing (temperature over time), it will be easy to attack the correlations by saying they’re not perfect, but life isn’t perfect. The suggestion that different measurements of temperature should not be compared is silly if you think about it. The best we have is to find ways to use them together, with care and honesty to qualify the ways in which they diverge from what we’ve been able to learn. We’re stuck with it, on our only planet.
One can hope you came here because you realized that some of the world’s top climate scientists are freely sharing with the rest of us in an effort to improve communication. If opening questions sound like a suspicious policeman with an agenda, it’s not surprising that others more familiar with the subject matter than you are call out your assumptions.
OK, one more time. Just because I don’t know when to stop.
Hank Roberts (53 etc) still seems to think I’m criticising the BEST reconstruction, when all I’ve done is take one of their background analyses and used it to make AN ENTIRELY DIFFERENT POINT which Dr. Steig has been gracious enough to concede may have some merit.
Susan Anderson (54) makes a number of points which I would like to address.
Trust me, my skin is pretty thick and my somewhat repetitive postings have been born out of a frustration at not being understood. I suppose I need to accept some responsibility for not expressing myself clearly, but in at least some cases (Hank shall remain nameless here) folks take their preconceptions about what they are expecting and react to those.
As for suggesting folks aren’t doing their jobs: I mentioned that I was trained as a scientist, and in fact I continue to work as such — though more in the engineering line these days. I’m constantly involved in technical discussions which I treat very differently from social discussions. In a technical discussion, if you think you have a valid point, you worry that bone until you get your point across. It has nothing to do with motivations or other peoples’ feelings, because in the end Nature is the arbiter and it doesn’t matter who THINKS they’re right.
If you review my allegedly hackle-raising approach, you’ll see that at no time did I impute any motives or threaten any families. I gave a technical assessment (qualitative, admittedly) of some data. I gave a technical suggestion on how it could be weighted during analysis. Since this is a blog about scientific discussion I felt it an appropriate way to proceed, and still do.
At a dinner party I wouldn’t tell the host their cake would be better with butter instead of shortening. I’d just offer to bring dessert next time.
> worry that bone until you get your point across
Please continue. I’m not a scientist, just another reader here. I’ve been trying to figure out what you’re trying to ask, without success; I haven’t figured out what the assumption is that you’re starting from. That’s just me, most likely.
Someone else will be able to understand you, if you persist.
Best bet — ask your question in a way that will attract the interest of one of the scientists here; they usually notice and reply rather quickly once you ask a question that’s clear enough to answer. (And when they do, folks like me learn from both question and answer. Seriously, do persist.)
Watcher,
I’d bet that the analysis you propose has been done many times by many people, but it has an issue with the “hide the incline”* that contrarians point out – that the instrument record has a larger increase than some proxies would suggest. Thus, I’m not sure success is even possible until we figure out why there has been a divergence with some proxies during the modern era.
*of course, contrarians use the word “decline”, which is ludicrous. The thermometer record has INCREASED, not decreased as compared to some proxies.
Re 57:
Jim,
I’m afraid I have to disagree with much of what you said.
If it has been done many times then it hasn’t been published, at least not within the last few years since I’ve been paying attention to this stuff. As was pointed out by another commenter there are professional climate guys here and I’m confident that if I was missing some existing body of literature on the subject it would have been pointed out.
It has nothing directly to do with “hide the decline” since the question being asked is very simply “are proxies correlated with each other to the same degree that thermometers are correlated with each other”. Either or both can be moving up, down or sideways without prejudicing the result of the correlation. It doesn’t require that the correlations be calculated over the same times frames. It doesn’t even require that the sign or magnitude of the temp/proxy connection remain constant*. If two weather stations read similar temperatures over the course of last year the odds are good they did so last century as well. Or last millennium. When the correlations between a corresponding pair of proxies is calculated no assumptions about calibration to temperature are required. It thus avoids the criticism of “pre-selection” of proxies based on the short overlap period with the instrumental record.
* I’m on thinner ice here, but I think that’s true!
[Response: Of course, it’s not simply “how well are they correlated”, because that depends on the timescale. One could have proxies that are poorly correlated with each other on annual timescales, but well correlated on longer timescales. The physics determining the long term variability could be related to the temperature, but the physics pertaining to the short term variability might not. Actually, a reverse sort of example of this would be raw tree ring data, where the very lowest frequencies are determined by the age of the tree (and hence the need for multiple trees of different age, corrections for the growth curve, etc. etc.).–eric]
Re Eric’s response to 58
If I might paraphrase your response: “The devil is in the details”. Quite right.
I can imagine a pair of thermometers, one at each edge of Oklahoma, say. Having driven across Oklahoma one can’t help but be struck by the OMG sameness of the place, so overall one would expect a pretty tight correlation between the two stations. Further imagine a cold front passing West to East. The temperature variation would have a time lag so the correlation would be there, but “out of phase” if you will, using instantaneous temperature. If you used a daily max/min you’d probably reduce that effect. With weekly or annual averages maybe get rid of it altogether.
So yes, I agree you’d need to examine whether it’s better to use monthly, annual, etc. data to correlate with the (at best) annual data from proxies. Lots of tedious work, but then that’s what grad students are for, isn’t it?
[Response: ;) re grad students. An example of what you’re talking about is my (at the time) grad student’s work on Mt. Logan. The “proxy” is annual snow accumulation rate, the climate target atmospheric circulation. Winter is the best time period. The proxy works just as well (or better) than the instrumental precipitation data. http://faculty.washington.edu/steig/papers/recent/RupperSteigRoe.pdf,
Watcher asks, “are proxies correlated with each other to the same degree that thermometers are correlated with each other”
A good question, though it makes me wonder how many Oklahoma twins are in the proxy record.
Re 59:
Thanks for the link. The paper is not something I’m going to pretend to have digested in a quick scan, but what’s clear right away is that even ignoring the two high altitude locations (Mt. Logan and Wolverine Glacier) correlation patterns for instrumental precipitation are pretty complex. One hopes that overall temps would be simpler or the method wouldn’t be of much use.
Re 60:
Probably none. It’s just a “thought experiment”. It’s often useful to consider the most idealized case you can come up with before layering on the complications that arise in the real world. And before anyone piles in: yes, Oklahoma exists in the real world.
At least parts of it do ;->
Re: 59, 62
In fact, Oklahoma (or any other of the Plains states) is a bad example of a place where you’d expect tight correlation. The east-west gradients are strong and are related mostly to soil moisture. Annual rainfall ranges from >50 in. in the east to <20 in the Panhandle. Differences in the annual temperature are almost 8 F (warmer on average in the south and east), but day to day differences in the spring are typically reversed from the annual averages. Today, for instance, highs in the west were low-to-mid 90s, with the east in the low-to-mid 80s. Vegetation in the east is typical of eastern deciduous forests and, in the west, the scrub vegetation associated with the high desert.