The tree ring dataset for the critical Polar Urals site has gaps of up to 59 years, has cores with as many 7 breaks in them. The problems are so pervasive that COFECHA testing simply fails – an almost unprecented occurrence in a tree ring data set. Quality control procedures are not described in Briffa et al. , although such descriptions (e.g. use of COFECHA) are routine in site publications. Because of the importance of the Polar Urals site, it seemed prudent to me to carry out a COFECHA test on the dataset as archived at WDCP. COFECHA is a standard dendrochronological program for checking crossdating, which can be downloaded from the University of Arizona. The COFECHA test simply failed on this dataset, although I was able to obtain results in tests on nearly other tree ring site. The reasons for the total COFECHA failure turned out to be numerous NA values in the Polar Urals dataset. If a core had more than one break in it (represented by NA values), COFECHA failed. Also if a core had a lengthy break (e.g. over 10 years), COFECHA also failed. It took considerable trial and error to determine exactly why COFECHA was failing. In the course of this, I noticed that two series (862481 and 862482) had 0 values (rather than the usual NA identified of -999) for over 400 years each. Since the longest individual series was otherwise less than 350 years, this seemed implausible and suggested that two distinct series had been incorrectly merged. After several inquiries, Schweingruber’s group acknowledged the problem and agreed that identification numbers had been incorrectly duplicated. Briffa has not responded to inquiries about how they ensured that erroneous identification numbers did not compromise any of their calculations (or why the error was not corrected in the publicly distributed version. The error had obviously not been corrected in the 15 years since data collection in 1990 and remains uncorrected in the WDCP archive. In my calculations, I provided new identification numbers for portions of the two series. For subfossil trees, Briffa et al. usually only have one core. It is not known whether the two subfossil cores previously numbered 862481 and 862482 are anomalously two cores from one stump or from two different trees. EVen after fixing this, COFECHA still failed. After a while, I counted the total NA incidence in the Polar Urals dataset, determining the following: · Out of 93 cores, there were 81 intervals in 41 cores with NA values, amounting to a total of 275 missing values. · The longest missing interval was 59 years (862452) and there were 6 intervals of longer than 15 years and 14 intervals of 5 or more years. · Two cores had no fewer than 7 intervals with NA values, 18 cores had two or more distinct intervals. By comparison, the nearby Yamal dataset had no NA values. P. Jones (pers. comm., July 21, 2004) explained the numerous NA values as follows:
for density you have to track the measuring machine at right angles to the rings, so this means cutting the core into parts before putting into the X-ray machine. Often, some small pieces or individual rings are not amenable to measurement for density. Fritz set these to missing, as he couldn’t measure them – but for all his density measures (there is also early wood width, late wood width, minimum/maximum density and others) . This also included ring width. He knew exactly how many years were missed, but he didn’t go back and measure the ring width.
Jones explained further that they interpolated missing values for use in spline and RCS calculations, but kept track of the interpolations and did not use interpolated values in the chronology calculations. This is not discussed in the original publication. Despite this explanation, some segments had as few as 3 rings in them. I crosschecked this explanation with D. Larson, a dendrochronological specialist at the University of Guelph, who advised as follows regarding missing core:
When one breaks a core, it fractures easily along a spring wood boundary because that wood is weaker than summer wood with small cells. No wood actually falls away when a core is broken unless you use your teeth to break it. Or a hammer. If they have more than one missing ring at each end of a break, the series should not be used at all. If there are “Ålots’ of breaks to allow for the reorientation of the series in the radiograph, then that means that they were sloppy when they took the core and they were nowhere near the pith so the core is a tangent instead of a radius.
Under Larson’s criteria, a minimum of 26 cores in the dataset cannot be used. If the criteria are weakeened slightly to allow core use if there is only one break with less than 5 missing rings, then there are still 20 cores that cannot be used. In effect, COFECHA seems to implement a slightly weaker form of Larson’s criteria: if there is more than one break, COFECHA will not use the core; if there are long breaks, likewise COFECHA will not use the core. I haven’t determined the exact maximum break length – COFECHA will accept break lengths of a few years, but not lengthy breaks. There is circumstantial evidence that WDCP also encountered this problem. At WDCP, *.crn series have been calculated from nearly all of Schweingruber’s contributions. (Curiously, some of the chronologies extend into the 22nd century — this is due to a programming error at WDCP.) Anomalously, there is no chronology for any of the russ021 (Polar Urals) sites. Presumably this is due to COFECHA failure. Next post will be on crossdating problems at Polar Urals.