After several years of trying, Briffa and Osborn have finally listed the sites used in Briffa et al 2001 and related publications. Briffa had reported results on large networks of over 300 sites collected by Schweingruber. Schweingruber had archived over 400 sites at WDCP.
However, it was impossible to tell exactly which sites had been used and up till now, Osborn and Briffa weren’t telling. While the webpage is obviously a huge improvement, there are some annoying problems. In some cases, the identification codes have typographical errors. I’ve noticed incorrect ids for adyboala, balyebda and lespiob in the data set for Briffa et al 2001.
Although Osborn says that all the sites are available at WDCP, most of the sites are there, but some are missing. I noticed 18 sites from the Tibetan Plateau. I’ve written Osborn to request that he remedy these things.
One of the first exercises that I carried out was to calculate a simple average of all the available ring width chronologies in the Briffa et al 2001 network (369 sites – 387 less the 18 unavailable sites.) In this case, I just used the Schweingruber chronologies without trying to re-do using RCS chronologies.
Here’s the result. As you see, there is a noticeable decline in the late 20th century – the “divergence problem.” If this is the average of a large sample of 369 sites, what are the odds that a random selection of say 8 sites has a sharp uptick in the late 20th century. Not very high.
Figure 1. Average of 369 available chronologies (as archived by Schweingruber) in the 387 site network of Briffa et al 2001.
Climate Audit 
19 Comments
I see the Dalton Minimum, the high temperatures from 1930 to 1950 and the decline to the mid-1970s. Perhaps these things are useful if interpreted properly.
Hi Steve,
still fuzzy after the holiday festivities, but I am sure from your plot above, that you are showing a mean of the MXD data rather than the ring-width data.
I might be wrong though
regards
Rob
Re #1
Unless you can account for the divergence and can properly model the relationship between tree ring width and temperature, then what you have is a recipe for delusion.
There may be no climate information in that simple composite – it maybe a statistical form of pareidolia that people can see a pattern and a relationship where none exists.
#2. Hi, Rob, no.THis is RW, I’ll do the MXD which is similar in general appearance as shown below. I’ve got a question about the underlying Schweingruber data here.
The data is collated from ftp://ftp.ncdc.noaa.gov/pub/data/paleo/treering/updates/wsl/indices/chronos actually the predecessor location at ngdc, using the format ftp://ftp.ncdc.noaa.gov/pub/data/paleo/treering/updates/wsl/hn-format-engl.txt In this case, the X (MXD) format is not in dimensionless 1000-centered units.
If one does running 100-year correlations between these two series, the correlations are pretty good, which encourages the view that there’s information in the data. However the current dendro policy (which you’re taking a brave stand against) of ignoring “adverse” results after 1960 is laughable to outside observers. I think that you’re going to have to throw out the linearity assumption and model these things with some sort of upside-down quadratic. This will generate two solutions in any reconstruction – a high temperature value and a low temperature value – and you’ll have to figure out some way of choosing between the two. Alternatively you’ll have to prove the linearity assumption and it’s becoming increasingly hard to maintain this in the face of actual 20th century results. In econometrics, you would not be able to assert model validity in the face of out-of-sample failure.
Running correlations between MXD and RW averages plotted above.
re: original post and #4
Steve,
I hate to ask this but what is being reported along the y axis in both charts? I presume years along the x axis, but I have no clue along the y. Thanks
The RW chronologies are in dimensionless units centered on 1 (multiplied to 1000). I’m not sure what units these particular Schweingruber MXD chronologies are in – hence the question to Rob. They are in the same units as Schweingruber’s Birminsdorff archive which appears to be different for MXD than the WDCP archive. The urls shown above provide whatever information is available.
This is big news.
Since tree rings are highly reliable indicators of temperature (only industry-funded denialist hacks believe otherwise) then we now know that the earth’s temperature has been DECREASING dramatically in the past half-century.
Therefore:
1) The thermometer record is wrong and should be calibrated against the highly reliable tree-rings immediately.
2) This trend should undoubtedly be blamed on humans. We will need an inventive young cliimatologist to whip up a model that shows this.
3) We should, of course, immediately do something extremely symbolic about this so that we can feel outraged at the world’s failure to immediately reorganize itself as we demand.
4) Since this trend is the opposite of the now-discredited warming trend we thought was occuring, we should immediately implement policies exactly the opposite of those advocated to combat global warming. To wit: we should immediately heavily subsidize fossil fuel use to increase the planet-saving blanket of carbon dioxide.
The time to act is NOW. Who’s with me?
What’s up with the, er, *volatility* prior to about 1700?
The number of sites in the composite declines. Many dendro composites adjust the variability of these series to reflect changes in number of series contributing to a composite using a method of Briffa and Osborn – a method which hasn’t been reflected on in general statistical literature to my knowledge.
Yes, thermometers don’t have teleconnections, so they are prone to sampling error. In addition, Mann’s AD1000 network is more accurate than thermometer-based land temperature in 1850.
But I’m not prepared to pursue my line of inquiry any longer as I think this is getting too silly. (Time to have a glass of wine, Happy New Year once more! )
I would like to see 95% CI plotted on that first graph to give me a sense of how low the variability is, especially during the recent period of divergence. With 369 samples the SE should be quite small, pulling the confidence intervals in tight to the mean curve. I am frankly surprised by this high degree of divergence in such a large sample. Impressive.
I wonder if Dr Wilson knows “why bristlecone pine growth has really increased in the 20th century”. Since he lost his wager in #2 (“I’m sure … that you are showing a mean of the MXD data”) maybe he could undertake to correspond with Dr Hughes on this issue?
🙂
I am still of the opinion that changing pH of rainfall and the amount of atmospheric CO2 have more long term impact on tree growth in the Western US than temperature does. In the Western US, soils are mainly high pH with some native uncultivated desert soils over pH 8.0 due to natural lime content. Increasing acidity of rainfall would allow greater availability of nutrients in these high pH soils.
What I would expect to see if this was a main factor in tree growth would be an overall gradual increase in growth until recently when steps were taken to reduce the emissions from coal fired power plants. This might be less pronounced as one gets closer to the West coast where impact from Asia might be greater than from North American power plants.
A plot of tree ring width for a given site against a plot of rainfall pH and atmospheric CO2 would interest me but I doubt such data are widely recorded. In other words, my opinion is that nutrient availability is not being weighted as highly as it should be in explaining long term variations of tree growth. In the Western US, decreasing pH of rainfall (acidic rain) would be a fertilizer in most locations.
While I certainly will acknowledge that the climate is indeed warming, the proxies being used are also subject to influences beyond temperature that are not being fully taken into account.
Happy New Year Steve (and others)
personally, I try never to use archived chronologies from the ITRDB as it is not always clear what detrending methods were used to derive the final chronology. Rather, I think it is better to download the raw non-detrended data (ring-width in millimetres and MXD in g/cm3) and work with those. However, from my experience, the archived chronologies in the ITRDB will come in four flavours: (1) RAW – simple mean in original measurements; (2) STD – the so called standard chronology after detrending the raw data. Data are dimensionless indices with a mean of 1 (as are those of RES and ARS); (3) RES – the residual chronology – all data were AR modelled to remove autocorrelation; (4) ARS – This chronology type follows the method of Cook (1985) where the common persistence is added back into the RES chronology. More information can be found at:
http://www.ncdc.noaa.gov/paleo/treeinfo.html
If the Schweingruber chronologies are in a different format, I am not sure what that would be.
As for the quadratic relationship – as we discussed at the AGU, this is only one possible answer to a complex problem. I can honestly say that for all the TR data that I have personally developed, the use of a quadratic relationship would not have improved the ‘fit’ during the calibration period.
However, as you now know, many groups are trying to address the divergence issue and we should have a much clearer picture in a few years.
bestens
Rob
#14. Rob – This was just a quick study and it doesn’t do any harm to examine Schweingruber’s actual chronologies. As to the need to consult measurement data, I obviously support this – this is one reason why I’ve been requesting that Briffa archive measurement data for Taymir, the Tornetrask update and Yamal – all key sites used repetitively in Team multiproxy reconstructions, including, date I say, the long chronologies in D’Arrigo et al 2006. (I know that you’d archive the data if it were up to you, but it’s still unarchived 7 years later.) The password protection on the Euro tree ring data is also pretty silly.
Rob – about 20% of the Schweingruber ring width series do not converge when one tries to fit a one-site RCS generalized neg exponential to them.
I would guess that as most of the Schwreingruber data utilise living data of a similar age class, then RCS detrending would not be appropriate anyway.
#17. I don’t follow. I can understand that it might not be “necessary”, but that’s a different issue. The convergence problem mostly comes from the growth curve not being convex – I’ve plotted a number of the failure cases and a Hugereshoff type fit would probably resolve the issue. But then the fitting seems a bit opportunistic. HEre are age-RW plots for some of the convergence failures – you can see why a neg-exp fit doesn’t work right away.
yes – in these cases a more flexible curve would be better
I generally use a flexible spline – i.e. 10% of the mean RC length