Global temperature changes of the last millennium

Ronan Connolly1*, Michael Connolly1
1 Dublin, Ireland.
* Corresponding author. E-mail:

A review of the various global (or hemispheric) millennial temperature reconstructions was carried out. Unlike previous reviews, technical analyses presented via internet blogs were considered as well as the conventional peer-reviewed literature.

There was a remarkable consistency between all of the reconstructions in identifying three climatically distinct periods. These consisted of two relatively warm periods – the “Medieval Warm Period” (c. 800-1200 AD) and the “Current Warm Period” (c. 1900 AD on) – and a relatively cool period – the “Little Ice Age” (c. 1500-1850 AD). Disagreement seems to centre over how the two warm periods compare to each other, and exactly how cold, and continuous the cool period was.

However, many of the assumptions behind the reconstructions have still not been adequately justified. In addition, there are substantial inconsistencies in the data on which they are based, and between proxy-based and thermometer-based estimates.

R. Connolly, and M. Connolly (2014). Global temperature changes of the last millennium, Open Peer Rev. J., 16 (Clim. Sci.), ver. 1.0. URL:
First submitted on: January 8, 2014. This version submitted on: October 11, 2014

Creative Commons Licence
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.

Latest version

pdficon_large Download article – Version 1.0 (Submitted: October 11, 2014)

Supplementary Information is available through the FigShare website at

Older versions

pdficon_large Download article – Version 0.1 (Submitted: January 30, 2014)

Peer Reviews

  1. This is a thorough and comprehensive review and commentary. I would just add a few things. The Loehle work is one of the few which does not heavily weight certain series, and therefore avoids the data mining/spurious correlation problem. A little more emphasis could be given to the failure of the various reconstructions to statistically match each other (and therefore they do not corrorborate each other).

  2. Hi Craig Loehle,
    First off, I would like to apologise for the delay in replying.

    Secondly, thank you for your positive and supportive review, and also for being the first reviewer to submit a comment/review on the OPRJ forum. This is much appreciated!

    You make two main suggestions with regards to improvements of the article. These suggestions have prompted us to:

    (a) Consider including some discussion of the different reconstruction methods used by each of the estimates,
    (b) Provide some more discussion of the differences (i) between the various estimates and (ii) between the various proxy series.

    With this in mind, in version 0.2 of our article, we propose inserting a new section between Sections 2.1 and 2.2 briefly comparing and contrasting the various reconstruction methods used. We propose adding a few paragraphs to Sections 3.4 and 3.5 more explicitly emphasising the fact that the inconsistencies between different proxy series raises serious questions over the alleged strength of the “temperature signal” of the proxies. We also propose adding some discussion to Sections 5 and 6 of the differences between the estimates.

    Would these modifications satisfactorily resolve your concerns?

    For more details of what modifications we propose making, see the attached .pdf.

    Extended review/comment OPRJ-reply-to-Loehle.pdf

  3. First, my congratulations on the most comprehensive overview of the situation in paleoclimate that I have seen. A couple of comments.

    In your otherwise excellent section on the Graybill stripbarks, you’ve left out the story of the beautifully named “Ababneh Thesis”, discussed at

    Next, any weighting system for proxies based on any kind of correlation with the rising temperatures 1850-2000 will produce hockeysticks out of red noise. While you point this out in individual cases, it is a more general finding that is independent of the weighting system. It is particularly a problem in large multi-proxy studies.

    Because of this, any large multi-proxy analysis can only be successfully tested by some type of cluster analysis to identify the types of shapes represented by the individual proxies. See e.g. my study at

    Absent such a cluster analysis to separate out the types of proxy responses, it’s easy for the authors to pull out some but not all of the offenders. For example, in Mann 2008, various species of stripbark pines under individual species names are included along with the Graybill stripbark trees. Pulling out just the Graybill trees and re-running the analysis will only increase the weighting of the other stripbark species, so the results will change very little. This leads to a false conclusion that their results are “robust” … a much overused word.

    Again, great site, great overview paper.


  4. In light of the comments by Craig Loehle and Willis Eschenbach, we decided to update our original draft to include sections,
    • Discussing the different reconstruction methods used by the 19 proxy-based estimates, and their relative advantages/disadvantages
    • Providing a more detailed discussion of the lack of consistency between individual proxies, and the importance of carrying out rigorous “sensitivity studies”, including a discussion of Willis Eschenbach’s cluster analysis.
    • Contrasting (as opposed to mostly comparing) the inconsistencies between different estimates on the relative warmth of the 1400s, etc.

    We also included a more explicit reference to Ababneh’s update of the Sheep Mountain bristlecone pine chronology (as suggested by Willis Eschenbach). We also included some new references which we found since writing version 0.1, and added some more discussion of the debate over Regional Curve Standardization in the section on tree ring proxies.

    In addition, several people had suggested to us that our original draft placed too much emphasis on the problems with current paleoclimate research, without providing many suggestions for how these problems could be overcome. With that in mind, we decided to include a list of 10 recommendations in the conclusions that we believe could improve paleoclimate research. We added a section with suggestions on how the temperature signals in individual proxies could be tested.

    However, when we had added these sections, the original format of the paper became quite cumbersome, so we decided to restructure Sections 2, 3 and 5 into what we think is a more systematic format.
    In Section 2, we review the theoretical basis and different reconstruction methods used for the current global temperature proxy estimates. In Section 3, we will discuss some of the problems involved with the proxies used in the studies… In Section 5, the different proxy-based temperature estimates are compared and contrasted with each other.

    We also corrected some minor typos and rephrased some sentences which weren’t originally very clear.

    The new version (1.0) of our paper is available here, but we have kept a link to the original version (0.1) in the “Older Versions” archive above, for posterity.

  5. Lately I’ve gotten into the bad habit of not commenting until a thorough examination is complete. But that results in frequently never making any comment at all (because I get distracted by other things). With that background, here are some first thoughts, which hopefully will not turn out to be final thoughts, time permitting. But who knows.

    There are a couple of interesting facets to this review. One is that the review includes discussions from web sites, which is an entirely new direction for sure. Doing so cannot be easy, since internet discussions typically vary wildly in terms of quality and coherence, and ad hominem attacks are quite high in web-based paleoclimate discussions, making it hard to know how much personal acrimony tints the arguments. This factor is huge, making for example, great difficulty in trusting the full objectivity of statements by Michael Mann and Steven McIntyre for example, who very clearly and obviously detest each other with a serious passion. Not only that, but each also has a set of followers, includig other blog writers, that more or less detests the opposing “camp”. It’s a very sorry state of affairs and the two sides have collectively done nothing so much as create an enormous public confusion and distrust. So to wade into these discussions in an attempt to illuminate the strictly technical issues, is a definite service to the public, many of whom must have had their understanding of the science informed almost entirely by these various web sites. This is very commendable.

    Also, there can be informative discussions at such sites that may never have appeared in the scientific literature for one reason or another. I’ve never been one to argue for the inherent superiority of peer-reviewed literature over discussions elsewhere; the only thing that matters in the end is the validity of the arguments made, and truly bad work appears in the literature on a daily basis. The peer reviwed literature system is seriously flawed in a number of ways, and the defense of a thesis because it’s “passed peer review” is an often meaningless and invalid argument.

    As for the content, the review focuses almost entirely on the issues involving how to aggregate spatially dispersed (i.e. site specific) paleo-temperature estimates into a hemispheric or global mean. I’ve only glanced cursorily at this (see below), but it appears to be quite thorough, referencing many sources. It also looks to be written in as non-technical of a language as possible, again very helpful. I hope to be able to comment on some of the actual issues raised at some point.

    Given that the review’s goal is a comprehensive assessment of the issues involved, my main comment at this point is that it concentrates on issues which are definitely important (potentially, if not actually) but are nevertheless one step removed from more primary issues which must first be addressed and resolved. Simply stated, the various issues involving aggregating site-specific paleo-estimates into larger scale “reconstructions” addressed by C&C only become relevant when it is first known that the site-specific estimates themselves, are in fact, reliable. There is a definite “troubleshooting hierarchy” in the field, stemming directly from the sequence of steps involved in creating any reconstruction.

    With respect to tree ring analysis, which is the basis for the vast majority of terrestrial reconstructions, the two most primary issues are (1) whether linear relationships between driver (climate) and response (ring characteristic) can in fact be assumed monotonic (usually, linearly so). This is an enormous, common and highly suspect assumption, and represents the key issue addressed by Loehle (2009), although not only by him.

    Secondly, the mathematical methods by which the non-environmental signal is removed from the ring response series, must be valid. Further, this issue must be evaluated with respect to the characteristics of the actual, existing tree ring data sets to which they have been applied, which very strongly involves the age structure of the sampled trees. This issue has been identified If these methods are faulty, one WILL necessarily mis-estimate the long term trend in the ring response, a result which will be highly critical for all subsequent estimates of relative climatic states over time whenever an actual trend exists, i.e. will defeat the entire goal of the process. This is the essential point in my series of blog posts (and rejected PNAS manuscript two years ago) on analytical problems in dendroclimatology (starting here:

    C&C include some discussion of the use of pseudo-proxies in evaluating statistical analysis methods. My final point here is that I am 100% in favor of this approach (not just in this field, but in observational science generally), and am in fact convinced that only via systematically conducteed simulations of realistic tree ring data sets can methodological problems be identified (and potentially corrected). The problem is that this approach has only been applied to the types of issues that C&C address here, and not to the types of issues that Loehle and I address. They need instead to be addressed to ALL of the actual or potential statistical and mathematical analysis steps involved in creating a large scale reconstruction.

    Hopefully more to follow, but we shall see on that score.

    Extended review/comment CC-comment.pdf

  6. As was “discussed” on Twitter, I do not at all agree with section 2.5’s thesis, one which has been used as one of the principle arguments against dendroclimatology practice generally, at various blogs, especially McIntyre’s, but several others also. It’s been way over-simplified to the point of damaging and it’s time to put a stop to it.

    This is a very serious misunderstanding of the likely importance of spurious correlation in dendroclimatology, one that has been repeated over and over at various blogs. I think I’m about the only one that has objected to it. But whenever I object to it, I am immediately met by hostile and irrational objections by some of those who favor that argument, so I just drop it, because I’m not wasting my time. However, since these people are so persistent, I’ll probably have to spell it out on a blog post at some point.

    The bottom line is we are all very much aware of spurious correlation and have been for 100 years or more. The question is how likely the issue is to be a real problem, biasing temp reconstructions, and the answer is “not very”. And it’s really not that hard to understand either.

Do you want to submit a review or comment?