Richard Smith’s new paper doesn’t mention Graybill bristlecones, but once again, his paper does nothing more than discover what we already knew – that Graybill bristlecones have a HS shape. In the process, Smith amusingly discovers a “divergence” problem with lake sediments
Smith’s new paper describes the use of the methodology of his earlier paper to the “new” dataset used in McShane and Wyner 2010. Smith says that his earlier paper “used the NOAMER tree ring dataset, which consists of 70 temperature series constructed from tree rings for 581 years (1400-1980).” In the preceding post, I observed that the data set in question consisted of tree ring chronologies, which cannot be assumed to be “temperature series”.
The McShane-Wyner dataset considered by Smith consists of 93 Mann et al 2008 series that go back to 1000. Smith attributes the seeming instability of reconstructions to lake sediment records (observing that there are 12 within the dataset), pondering the possibility of “divergence” problems in lake sediments – a possibility that, according to Smith’s belief, had evaded the keen eye of paleos.
Smith defines divergence as follows:
Paleoclimatologists have coined the term “divergence” to describe cases in which the stationarity assumption appears to be breaking down within the timescale of observational data.
While this is a sensible definition, I’m not sure that this accurately characterizes its application by the Team, where the phenomenon is in practice limited to series that don’t go up. Smith’s definition would include series that go up too much. Rather than these examples being perceived as examples of “divergence”; they are welcomed by the Team.
The best known example of divergence concerns trees; see for example Briffa et al. 1998 or pages 48-52 of North et al. (2006). However, the problem does not (so far as is known) apply uniformly to all tree-ring proxies; the specific class of proxies for which it is known to be a problem are tree-ring latewood density records. However, most of the known records of this type go back no further than AD 1400; in particular, none of them are among the 93 proxies used in the present analysis (Dr. Michael Mann, personal communication). Therefore, it appears that the known divergence problem with tree rings is not responsible for the results in the present paper.
Smith attributes the instability to lake sediments, of which there are 12 in the McShane-Wyner dataset, two of which, as CA readers are well aware, are upside-down Tiljander series, the modern portion of which is hugely contaminated by bridgebuilding and agriculture. Smith suggests that non-stationarity in lake sediment series might be a problem – noting that, to his knowledge, this possibility had not been previously considered.
To the best of my knowledge, no previous study has explicitly identified lake sediment records as subject to this problem, though with the benefit of hindsight, it seems obvious that lake sediment deposits in the late 20th century would be affected by anthropogenic activity other than increasing CO2.
Thousands of blog readers around the world are familiar with the fact that Mann used the modern (contaminated) portion of the Tiljander series, ironically upside down. Ross and I even went to the trouble of reporting this in a short comment in PNAS (not cited by Smith on this point.)
In this case, Smith is not complaining about the Tiljander sediments going up too much. Actually his complaint is the opposite. Some of the sediment series in the Mann 2008 data set have a pronounced medieval warm period.
Smith therefore examines a reduced dataset of 81 proxies using inverse regression on principal components and once again gets a characteristic HS shape – one that looks for all the world like the original Mann reconstruction.
There’s a simple reason. Smith once again has created a bristlecone reconstruction. The 81 series in the new data set include 18 Graybill bristlecone chronologies, ALL of which were in the 70-series NOAMER dataset of his previous paper.
Last time, Smith had 20 Graybill bristlecones out of 70. This time, the Graybill bristlecones constitute 18 of 81 series in the data set. Surprise, surprise, he gets the same answer.