Long Instrumental Series

I’ve been looking recently at some issues in the post-1820 proxy data and re-noticed some things that I’d seen before but lacked my present context.

Most of you are familiar with the CRU temperature series since 1850. Of course, there are instrumental series that go before that. MBH98 used 12 long instrumental series. There are 14 such series archived at the UVA FTP website (in standard deviation units). Here is a graph showing the average of these series up to 1980 (as archived.
Figure 1. Average of MBH98 long instrumental series up to 1980.

Below I show the individual times series.

Obviously, the long instrumental temperature series do not show the iconic hockey stick shape. This is such an obvious graph to plot that one wonders why it wasn’t shown in MBH98 itself. Here is a plot of the underlying series, which are almost entirely European in origin. The two series shown in red were excluded in MBH98 for unstated reasons – one of them is the well-known De Bilt series (beloved of Hans Erren). Mann et al. have never reported the source of the instrumental data. It was incorrectly cited in MBH98 as Jones and Bradley [1992]. The Corrigendum only said "NOAA". Nature refused to require more specificity.

I’ve tracked most of the series as being exact linear transformations of data versions in ndp-041, but some are still puzzles.

Figure 2: Long instrumental temperature series at MBH98 FTP site (presumably standard deviation units).

Proposed identifications columnwise top to bottom: Minneapolis, Paris, Geneva, Kremsmuenster, Wien, De Bilt, Wroclaw, Warszawa, Stockholm, St. Peterburg, Bergen, Trondheim, Arkhangel, Central England. Red – archived but not used. Minneapolis, Paris, Stockholm, St. Peterburg, Bergen, Trondheim, Arkhangel have correlation of 1 to ndp-041 version; C England version is identical to JJA version at C1500 archive [check]. Wien and Wroclaw have correlations >0.975; Geneva and Kremsmuenster >0.8. I was looking at these series in connection with the attenuation of variance in earlier portions of the MBH98 reconstruction. This does not occur in the long instrumental series – I’ll show which ones are affected on another day.

Another curiosity: remember how Briffa complained that the proxies were not keeping up with temperature post-1960? I showed some graphs showing that this was a widespread problem in both MXD and ring width series. Here we have the opposite situation: in this dataset, the "proxies" are outstripping the instrumental series. So we have (at least) two interesting problems here:

1) why is the trend in the long instrumental series so different from the temperature PC1 and the CRU average?

2) why are MBH98 tree ring data outstripping instrumental temperatures (e.g. bristlecones growth), when tree rings are not doing so on an overall basis?

To answer the first question, one needs to do a detailed site-by-site analysis of the CRU dataset. This is prevented by Phil Jones’ refusal to let potential critics look at the underlying station data. (Look at the Top 15 Reasons for withholding data – he’s the guy with "25 years invested in this work".) Given that his work has been funded by the U.S. DOE, I doubt that he will be able to stonewall forever – there’s more than one way to skin a cat. The second question leads to interesting issues of cherrypicking proxy series. I’m trying to work up a statistical method of demonstrating cherrypicking and have some thoughts on how to do this, but they are still preliminary.


  1. Paul
    Posted May 26, 2005 at 12:59 PM | Permalink

    It’s obvious why the red series that is second from the bottom on the left is excluded (De Bilt?), it’s warmer from 1850 to 1950 than at any other time.

  2. Dave Dardinger
    Posted May 26, 2005 at 1:46 PM | Permalink

    Those are a bit strange. It’s easy to see the warm spot in the 1930s & 1940s but where is the recovery from the little ice age? Are those standard deviations from the mean of the individual station or from some global average curve?

  3. John A
    Posted May 26, 2005 at 1:55 PM | Permalink

    Yep, the CET record is recognisable by its length and also by the spike at 1739/1740 which had a shockingly cold winter recorded both in the UK and in the colonies of New England.

  4. Posted May 26, 2005 at 4:11 PM | Permalink

    re #1 That’s a data artifact in the GHCN version: an updated version was already published in 1996
    A van Engelen and Nellestijn, JW, 1996: Monthly, seasonal and annual means of air temperature in tenths of centigrades in De Bilt, Netherlands, 1706-1995. KNMI report from the Climatological Services Branch.
    Data here:http://hanserren.cwhoutwijk.nl/co2/LABRIJN.xls

  5. Steve McIntyre
    Posted May 26, 2005 at 6:42 PM | Permalink

    Dave, In the CEng series, the biggest downspike was in the late 17th century. The CEng series goes back to 1659, but the MBH98 version trunctes prior to 1730. We pointed this out in MM03. They subsequently attributed this version to Jones and Bradley [1992], the well-known article on the LIA. The cold late 17th century portion is also deleted from this version – without notice. It’s a little strange when they then argue that there was no LIA. A couple of years ago on sci.environment, they argued that this truncation was the “right” thing to do because of changing homogeneity, but then what to do about tree ring records, which are much worse for homogeneity. Also the version starting in 1659 is used in Jones et al 1998 and Jones and Mann 2004. So it’s hard for the Hockey Team to figure out a consistent position. Regards, Steve

  6. Paul
    Posted May 26, 2005 at 7:49 PM | Permalink

    I remember two experiments with amazing results from my career in experimental high energy physics. The first was the “split A2″. Many subatomic particles are evanescent, they live 10^-21 seconds or so, and their signature is a resonance bump that is a Lorzentian, something like a Gaussian but with longer tails. The A2 had the remarkable feature of a deep hole right at the peak, and it was a 5 sigma effect. No one could understand it because it violated quantum mechanics and probability. The second experiment was the apparent discovery of quarks by an experiment at Stanford. They were clearly seeing electric charges that were 1/3 and 2/3 the charge of an electron. Both experimental results turned out to be wrong and the result of cherrypicking data (and ruined a few careers). The moral is that you have to include all data unless you can show that your equipment blew up in the process and the data you took was bad. If the experiment was properly calibrated and running you have to include the results, even if that Nobel Prize disappears in the statistical averaging.

    Unless there is an experimental reason to exclude the red series or a statistic that clearly isolates the red series from the others there is no reason to exclude them from the analysis. Just by eye, the autocorrelation function is probably going to be very similar for all the series, a sharp peak with a tail.

    The best method to avoid fooling yourself in a case like this is to develop the algorithms using dummy data generated by, for example, a Monte Carlo. Once you are sure that the algorithms are correct, then use them on all the real data, once. That way you can’t do the natural thing and fool yourself with exciting results from an unrepresentative small subsample of your data.

  7. Michael Mayson
    Posted May 26, 2005 at 8:52 PM | Permalink

    Steve, I wonder if the big downspike in the late 17th century was as a result of major volcanic activity in Iceland. See this article http://www.newscientist.com/channel/earth/mg18625002.400
    “In Europe they called 1783 the Year of Awe”

  8. John A
    Posted May 27, 2005 at 5:09 AM | Permalink

    Re: #7

    A boring pedant writes: 1783 was in the 18th Century.

  9. Michael Mayson
    Posted May 27, 2005 at 5:17 AM | Permalink

    Re #8 Pedant you may be – but you are also right! Damn!

  10. JBK
    Posted May 27, 2005 at 5:56 PM | Permalink

    Interesting that the 1816 Tambora “year without summer” downspike is visible on the average graph and on most of the the individual plots.

  11. Posted Aug 7, 2005 at 5:03 AM | Permalink

    re #4:
    After a webserver crash the new url for the Labrijn ‘De Bilt’ series is:


  12. Posted Aug 14, 2005 at 2:23 AM | Permalink

    I am still puzzled what the source of the “De Bilt”data in GHCN is.
    I know of two sources in the literature:
    1) Labrijn, A., 1945, Het klimaat van Nederland gedurende de laatste twee en een halve eeuw., KNMI mededeelingen en verhandelingen 49. (L1945)

    2) A van Engelen and Nellestijn, JW, 1996: Monthly, seasonal and annual means of air temperature in tenths of centigrades in De Bilt, Netherlands, 1706-1995. KNMI report from the Climatological Services Branch (vEN1996)

    Both datasets don’t show the 1850 jump, vEN1996 found some clerical errors in L1945. However, only in the pre 1734 data acquired by Cruquius a substantial difference is visible between the two datasets. Also the 1950 inhomogeneity was corrected by vEN1996.

  13. Posted Feb 12, 2008 at 5:40 PM | Permalink

    as requested in post 85 unthreaded #31

    Armagh, CET and De Bilt have inhomogeneities in the early part.




Get every new post delivered to your Inbox.

Join 3,203 other followers

%d bloggers like this: