A common meme in Team-world these days is that any issues or errors are minor and that none of them “matter”. As we peel back the layers of Kaufman et al, this is the first line of Team defence.
The rhetorical impact of Team reconstructions largely derives from the modern-medieval differential: is it in the red or is it in the black?
Thus when one sees study after study which has modern-medieval differentials that are always just slightly in the black, any prudent analyst would arch his/her eyebrow slightly and examine any accounting policies that may have contributed to getting the result in the black. (I use the term “accounting” intentionally, since the term implies that there be policies for the inclusion/exclusion of particular data sets and their truncation.) And let there be no doubt: when one is dealing with CPS reconstructions with very small data sets (20 or so), it is quite possible to affect the differential through a small subset of the data.
Sometimes the accounting exceptions look innocent enough e.g. MBH’s unique extension of Gaspe tree ring series back to 1400 from 1404. However, such variations in accounting policy invariably seem to enhance HS-ness in the composite and each such variation needs to be examined.
Obviously there are a few methods that I’ve learned to look for: does the study use Graybill bristlecones? Does it use Yamal? Does it use upside-down Tiljander? Are there any truncations or extensions of the series? Is the most modern version of the series used?
I noticed almost instantaneously that Kaufman used Yamal and upside-down Tiljander (however mitigating the impact of upside-down Tiljander by truncating it at 1800). He used two other Finnish series, both of which are, as far as I can tell right now, used in an orientation upside-down to that proposed by the original authors. Kaufman truncated the Blue Lake varve series because of supposed non-temperature inhomogeneity in the early portion of the series, but didn’t truncate the later portion of the Loso Iceberg Lake varve series where there was a definite inhomogeneity. Kaufman appears to have used an old version of the Hallet Lake series (which was replaced over a year ago in Nov 2008 at NCDC – otherwise, the inconsistencies between the Kaufman version and the NCDC version are inexplicable.) In addition to Yamal, Kaufman used two other Briffa versions, while not using seemingly plausible tree ring series at Tornetrask (Grudd) and Indigirka, Yakutia.
I’ve done a quick sensitivity analysis in which I’ve done a CPS average (980-1800 base) with the following variations:
1. Current version of 2- Hallet Lake and non-truncated version of 1-Blue Lake. (I haven’t checked whether this “matters”, but there didn’t seem to be any overwhelming reason to use an obsolete version of Hallet Lake or a truncated version of Blue Lake.)
2. The three Briffa series (Yamal, Tornetrask-Finland, Taymyr-Avam) are replaced by Polar Urals (Esper version), Tornetrask (Grudd version) and Indigirka (Moberg version). (I think that this is the sensitivity that carries the water here and my guess is that most of the difference arises from the Briffa data. I’ve provided materials that make this easy for anyone interested to check.)
3. The three Finnish proxies are used in the orientation of the original authors i.e. flipped from the Kaufman version.
Here’s the result. Obviously there is a lot in common in the general appearance of the two composites – the difference between the two is that there is nothing “unprecedented” about the 20th century in the latter case.
One can calculate the relative contribution of each accounting decision to the change in appearance. The largest contribution comes from Yamal versus Polar Urals. Each accounting decision has some impact on the modern-medieval differential. I’ve uploaded data sets and scripts so that interested readers can experiment for themselves and will attach an explanation script in the first comment.
I do not claim that the bottom graph is more reasonable or less unreasonable than the first graph. My point here – as on many other occasions – is that just calling something a “proxy” doesn’t mean that it is a “proxy”. The sensitivity of the modern-medieval differential to different roster selections means that the data is not consistent enough to yield a “robust” result.