A Weird Jacoby-D'Arrigo Series

I mentioned a couple of days ago that Schweingruber et al [1993] seems to arrive at opposite conclusions to Jacoby and D’Arrigo [1989] as to whether anomalous post-1950 ring widths occur (let alone whether that is due to CO2 fertilization or temperature.) It was interesting to compare Schweingruber-Kienast results at Niwot Ridge. It turns out that Schweingruber has published results (cana060) for Churchill, Manitoba, an important site (cana158) in Jacoby and D’Arrigo [1989] (and several other Jacoby-D’Arrigo publications.) Here’s what it looks like – a typical Jacoby elevated 20th century, but inconsistent with Kienast and Luxmoore or Schweingruber et al 1993.

Figure 1. Jacoby-D’Arrigo chronology for Churchill MB (cana158)

Before showing the results, I’ll mention that I corresponded in early 2004 with D’Arrigo about this data set. MBH98 said that they had carried out certain quality control procedures for tree ring data sets to verify minimum correlations between the measurement data and WDCP crn series. I checked this for over 300 sites and one site that really stuck out as having a low correlation between the measurement and crn series was cana158 (it was about 0.05 by memory, as compared to a MBH minimum of 0.5, again by memory). In addition, the crn series went to 1982, while the measurement data only went to 1978, so there was some inconsistency between the data (as well as evidence that MBH had not carried out the quality control that they said that they had carried out.)

I wrote to D’Arrigo about this inconsistency and she said that the series should be withdrawn (although it’s used over and over). I re-checked a few days ago and the crn series is still at WDCP unamended. Anyway, after noticing the remarkable differences between Keinast and Luxmoore [1988] and Schweingruber et al [1993] on the one hand and Jacoby and D’Arrigo [1989] on the other, I’m now intriguing by any chances for detailed cross-checking. Figure 2 below shows two comparisons of the Jacoby-D’Arrigo chronology for Churchill MB – on the top against my emulation of the chronology calculation using STD methods and archived measurement data; on the bottom against the Schweingruber chronology for a sample in the same area (indeed Jacoby and D’Arrigo use Schweingruber’s MXD results from this sample). Both are materially different from the Jacoby-D’Arrigo chronology, especially in the 20th century.

Figure 1. Jacoby-D’Arrigo chronology for Churchill MB (cana158) Top: black – Jacoby chronology (cana158); blue – std emulation; Bottom – black – Jacoby chronology cana 158; red – Schweingruber chronology cana060.

I’m becoming pretty baffled by what’s going in tree ring world. I checked the identification numbers for the two different data sets. Schweingruber had identification numbers up to 19, but reported only 12 trees (missing 5,12-14, 16-18). He reported 24 cores – 2 consistently for each tree. Jacoby and D’Arrigo had identification numbers up 15, but reported only 9 trees (missing 2, 5-6, 8,11-12). They reported 4 cores for one tree (#7), 3 cores for one tree (#1); showed core id numbers reaching 3 for 2 trees where only 2 cores were reported (#9, #10); reported 2 cores for 3 trees with 2 being the highest core id# (#3, 14, 15) and only one core for 2 trees ( #4, 13). On a previous occasion, I mentioned a remarkable quote from Esper about dendro methodologies:

However as we mentioned earlier on the subject of biological growth populations, this does not mean that one could not improve a chronology by reducing the number of series used if the purpose of removing samples is to enhance a desired signal. The ability to pick and choose which samples to use is an advantage unique to dendroclimatology.

But there must be more going on here – the crn series as archived simply doesn’t match the measurement data. Maybe they’ve got some other measurement data squirreled away somewhere. It would be nice if something in dendro world tied together.


  1. Posted Apr 8, 2006 at 5:14 PM | Permalink

    The blue and black graph shows remarkably good correlation in the high frequencies (eyeballing it) over pretty much the whole chronology, but has two significant divergences from 1700-1750 and 1920-1970. I wonder what that means? Perhaps we’re seeing the “divergence” problem happening in earlier parts of the chronology as well as in the 20th century? Or perhaps some other biasing local factor.

    Either way it doesn’t look good for those trying to extract a meaningful low-frequency signal from them. Which one is correct? The high frequency signals seem much more reliable.

  2. TCO
    Posted Apr 8, 2006 at 8:23 PM | Permalink

    I think wiggle matching theory would imply that the high frequency agreement would be reasonable for trees from the same population. But low may not be (not sure why…competition?) Anyhow, if there is some effect of people creating series that bias result by leaving out the “apples”, that might be the effect one would see.

  3. Steve McIntyre
    Posted Apr 8, 2006 at 11:14 PM | Permalink

    #2 – there’s something else going on besides cherry picking. In the top panel, I’ve benchmarked my emulation of Jacoby’s chronology calculations on another data set, where I can replicate very closely. Here I’ve worked with the cherry picked measurements (if indeed they have been picked) and still get a different answer. I think that there might be something wrong with the calculation. We already know that D’Arrigo asked the chronology be withdrawn from WDCP.

    I’m not sure that I’ll be able to locate other overlaps, but I’m definitely going to look.

  4. CasualBrowser
    Posted Apr 9, 2006 at 1:08 AM | Permalink

    Although they report different total core numbers, is there any way to compare Jacoby and Schweingruber cores to see which cores Jacoby left out?

    Were the cores Jacoby left out, and Schweingruber used, reported by Jacoby (but not used)?

  5. Steve McIntyre
    Posted Apr 9, 2006 at 8:39 AM | Permalink

    #3 – they both started numbering from 1, but the samples are not the same where the numbers coincide. They seem to be from different programs.

  6. JerryB
    Posted Apr 9, 2006 at 9:52 AM | Permalink

    Re: “The ability to pick and choose which samples to use is an advantage unique to dendroclimatology.”

    I sometimes wonder whether some practices, that might be valid in dendrochronology, got carried over to dendroclimatology in which they would be invalid.

  7. mark
    Posted Apr 9, 2006 at 11:06 AM | Permalink

    Lots do. PCA is used primarily for image processing where the cause and effect is known. It also is used in situations where the weights are applied to data with stationary statistics for noise. I.e. just because the weighting is valid in the 20th century does not mean it is valid for the 18th because the noise in the samples is not known to be the same then as now.


  8. jae
    Posted Apr 10, 2006 at 10:34 AM | Permalink

    Getting much clearer: you can pick a set of tree rings to construct ANY type of curve you wish.

  9. TP
    Posted Apr 15, 2006 at 3:12 PM | Permalink

    Dear Steve,

    I am curious as to why your blue curve in the upper half of the last figure [the second Figure 1], while standardized, still has a long, downward trend in ring index from the late-1700s until 1950?

    You are aware of the well-documented phenomenon of narrower ring-widths as trees get larger, right? It is simple geometry: if a tree has a constant rate of growth, raw ring widths must decline as the stem expands. Therefore, a constant raw ring width or increase in raw ring width as the stem increases in diameter indicates increased tree growth rates. Allometric equations, used in forestry to predict the amount of aboveground biomass [wood], support this line of thinking. This is a primary reason as to why ring widths are ‘standardized’.

    So, the long downward trend in ring index since the 1700, when the trees were smaller, until the 1950s, when the trees were undoubtedly much larger, can be explained if the geometric growth trend described above is not removed. I am curious at to your standardization method. You say it is standard, but that is rather vague. Can you elaborate?


  10. Steve McIntyre
    Posted Apr 15, 2006 at 8:48 PM | Permalink

    #9. Each core was standardized by fitting a “generalized negative exponential” i.e. a negative exponential with an asympotote to each core following usual procedures. I did this in R, but I’ve verified the algorithm against ARSTAN.

    The bottom graphic where the red chronology is somewhat similar to my emulation of Jacoby is from Schweingruber and is archived rather than calculated. The inability to replicate Jacoby-D’Arrigo is odd, but D’Arrigo said that the chronology needs to be withdrawn (but has not done so.) The chronology is used in a number of studies.

%d bloggers like this: