Briffa’s Ph.D. student, Thomas Melvin, discusses the important impact of “modern sample bias” on RCS chronologies, discussing Tornetrask and Finnish sites in detail.
His PhD thesis, “Historical Growth Rates and Changing Climatic Sensitivity of Boreal Conifers” (May 2004) is online here and contains many interesting discussions of tree ring standardization. Here are some key comments, particularly from chapter 8:
Problems of the RCS technique are identified which are associated with tree age and diameter-related bias, arising from the use of ring-width to establish tree growth rates, regardless of tree diameter. These problems are manifest as “end effects” in chronology development and are most significant in the most recent century… Variation in growth rates between different trees growing at any specific point in time appears to be the normal situation and it is this variation which leads to modern sample bias….
The potential for modern sample bias occurs because the fastest growing trees of the earliest centuries and the young slower growing trees of the most recent centuries are generally missing from dendroclimatic samples taken from living trees at one point in time…
If samples are restricted to trees whose radius is 10cm at breast height, the samples will be from trees of varying ages (Enquist 1999)…
Samples will be from trees above some minimum radius, small trees are not generally sampled because they may suffer excessive damage from coring, and short ring sequences may be difficult to crossdate.
The impact of this on Briffa’s key sites of Polar Urals and Tornetrask is obvious as soon as the matter is stated. If, as Melvin says, there is a distribution of growth rates including both slow-growing and fast-growing trees, the minimum diameter for sampling living trees filters out slow-growing 20th century trees. The impact can be illustrated below. Figure 1 shows, for the Polar Urals site, a plot of cumulative ring widths by tree. The minimum cumulative ring width in 1990 is a little under 5 cm (scale at left is x100). The existence of a minimum sampling diameter is neatly illustrated by the lack of any samples in 1990 with cumulative ring width under the minimum. In subfossil collections, this is not the case: showing the important lack of homogeneity in sampling method. The bottom panel shows the simple mean restricted to trees with cumulative diameter above the 1990 minimum. Does 1032 look like the “coldest year of the millennium”, as claimed by Briffa et al [1995]?
FIGURE 1. POLAR URALS. Top: Cumulative ring width by tree, with red horizontal line showing minimum in sample year. Figure 2 shows the same thing for Tornestrask, with two different minimums – the higher being the minimum in the last year of sampling (1980) and the lower one being a lower value for sensitivity purposes.
FIGURE 2. TORNETRASK RING WIDTH. Top: Cumulative ring width; middle- chronology of mean ring widths for trees with diameter exceeding 1980 minimum.
Now consider the effect on Briffa’s arbitrary adjustment of the density chronology to the ring width chronology. Both chronologies are probably biased by this effect, but intuitively it seems to me that the bias is direct on the ring width chronology and probably much greater (since the age impact on MXD is less as well.) Thus an important contributor to the divergence of 20th century Tornetrask RW and MXD chronologies would be the impact of modern sampling bias – which would, in turn, not justify any coercion of the MXD chronology towards RW values.
Climate Audit 
7 Comments
I was reading the early part of the thesis and one thing I was struck by was the reliance on ‘curve fitting’ to decide what adjustments to the raw data were to be made as a first step. I’m not saying this is necessarily wrong, but I’d feel happier with a bit more justification. I believe that the theory is that any error induced in the record of an individual tree will be washed out since there are many overlapping trees. But he does have some discussion of problems there can be when filtering, in essence, throws out the signal with the noise, in a slightly different context, so I wonder if the same thing can happen with the curve fitting? Can we assume there will be no bias introduced by ‘curve fitting’?
Has anyone studied what CO2 fertilization would do to maximum density and ring width? Presumably, one could examine trees from a FACE experiment an measure this directly. There is a FACE experiment using loblolly pines at Duke University
Further to the last post – I apparently had an error in my html, since I also referrred to a reference I found at CO2science.org that claimed to find in experiments with loblolly pine that CO2 increases ring width more than wood density. Hence, the CO2 fertilizer effect might indeed help explain a bias like you referred to above.
Steve, how do I find out how cores taken from live trees are desiccated so that they are in a like state with cores taken from long dead, cured timber? This has to take place before RWs and MXDs are taken, right?. Does this desiccation shrink the outer RWs of a tree more than the inner RWs? How does it effect the relationship between RWs and MXDs if at all? I ask because I understand the outer rings of a tree are by far the most biologically active and carry the substantial portion of water and nutrients.
I also wonder if in identically cured cores, one taken from the same tree decades later, the later core will show any shrinkage in the RWs that correspond to the earlier cores outer rings. This due to some other effect. Lessened use atrophy? Hydrostatic pressure gradient effects?
Steve: My understanding is that they treat the wood, but I don’t know how. Probably the best text to check on would be Schweingruber’s.
BTW along these lines, I recall seeing an article mentioning very long-term changes in wood density in samples from the early Holocene. I don’t remember the reference and I’ve googled all conceivable combinaitons without success to try to retrace my steps. If anyone runs across any references to wood density in the early Holocene, I’d much appreciate the information.
Re Schweingruber’s book
I’ll go down to my library and try to get it via ILL. I can find a huge list of articles he published but not his book. Do you recall the title? But, more relevant is how the cores and sections that became the data in these articles were prepaired. I’ll just have to RTFA :).
When I google, poor signal to noise ratio!, I come up with all sorts of preparatory methods.
One had wet cores put in 95% alcohol for days at high temperature to extract the resin, then dried in telephone books before being glued into a shallow straight groove gouged in the upper surface of a plank. The next step is to sand down the mounted core to expose the rings. Obviously you can’t get MXDs from a core fixed by this method. Another site had the cores refrigerated. Perhaps they were confused with polar ice? They didn’t mention why this was done. It seems RWs from dead trees can come from disk-like sections. For some reason they don’t want to do this to live trees. Perhaps stumps are harder to hug. 🙂 It does seem that
RWs taken from dead trees vs live trees are prepaired quite differently. They are sure different before they are prepaired. So …
Do the different initial condition of these samples, different methods of acquisition, and different methods of preporation before data extraction introduce a bias in data from live trees relative to that taken from dead trees?
I would like to see a chart showing percent of data from live trees in Tornetrask vs year above a plot of the difference of the Tornetrask RW and MXD chronologies. A strong match in data rich years would be interesting.
Schweingruber, F.H. 1988. “Tree Rings: Basics and Applications of Dendrochronology”. Kluwer Academic Publishers, Dordrecht, Netherlands
How much extra work (or environmental damage or economic loss) would it be to actually cut the live trees down to get disc samples?