Rob Wilson sent in a post on another thread arguing that bristlecones are not as bad a proxy as I would have everyone believe. Unlike realclimate, opposing views are not censored here. In fact, I’m happy to highlight them. I’ll read Rob’s note and reply on an another occasion. I’ll only note now that, in our discussions of bristlecones, especially in EE , we relied on specialist publications such as Graybill and Idso , Hughes and Funkhouser , and even (implicitly) IPCC 2AR, as questioning the validity of bristlecones as a temperature proxy, rather than arguing the point ourselves from first principles; otherwise I won’t editorialize further here, but will re-visit the topic on another occasion.
Rob: Certainly, if these data dominate a NH reconstruction through the use of PCA, this is probably not an ideal situation where presumably a NH reconstruction should be representative of the NH.
However, how valid are the BP data as a “Ålocal’ temperature proxy?
At the link below, I have uploaded some pictures comparing BP data with the North American (NA) mean series from our 2006 paper. We never included the BP data in our paper as we wanted to use as high latitude TR data as possible. However, the inclusion of the BP would have made little difference to the final NH reconstruction, although it would have depressed the MWP a little relative to the recent period.
Figure 1 – a comparison of our NA mean with a RCS chronology developed using BP RW data from three sites: Hermit Hill (N = 38; 1048-1983) and Windy Ridge (N = 29; 1050-1985) from Colorado and Sheep Mountain (N = 71; 0 — 1990) from California. The time-series have been normalized to the 1200-1750 period.
A couple of interesting points:
1. at least for the 1100-20th century period, there is a surprisingly strong common signal. From this comparison alone, one could conclude that, although the BP data represent a relatively small region (in a global sense), the data do seem to pick up the multi-decadal to centennial scale variability of the larger NA mean series. The deviation between the series prior to 1100 may simply represent the decreasing replication of NA sites – there are only two sites that go back prior to 1100 in the NA mean. As stated in our paper, I do not feel there is enough data prior to ~1400 anyway.
The BP RCS chronology correlates with “Ålocal’ July-September mean CRU gridded temperatures at 0.38 (Durbin-Watson = 1.70). This is admittedly not particularly strong (NB. no autocorrelation in the residuals though), but with the reasonable comparison with the NA mean, it does seem to suggest that temperature is likely the dominant controlling factor – especially at decadal to longer time-scales.
Now, I do not deny that all sorts of other factors may also influence growth and I am sure someone will say – “Åwell how about all the none explained variance?.’ Well – Dendro-reconstructions generally explain anywhere from 30-60% of the variance of the climate parameter that one is trying to reconstruct. This is a modeled mean response over a particular [calibration] time-period and does not look at the more complex situation for each year. Using regression, it is easy to test if multiple climate parameters (i.e. temperature and precipitation) effect the growth of trees. It is generally the case, that if the tree site is carefully selected (i.e. high elevation/latitude treeline for a temperature signal), then precipitation will have a minimal effect on growth and over the period of calibration, the correlation with precipitation will be close to zero.
2. I purposely normalized the data to the 1200-1750 period so that any possible inflation of BP index values due to CO2 fertilization would be accentuated relative to the NA mean. Interestingly, the NA mean index values are higher than the BP data. Assuming that no CO2 fertilization effect biases the NA mean (there is no evidence for this at all), then I see no evidence of this effect in the BP data either.
Figure 2 – comparison of the NA RCS mean with the recently published annual temperature reconstruction for the Colorado region from Salzer and Kipfmueller (2005). They also used BP data – independent to the data I used. There model explains 46% of the variance (DW = 1.64), so is appears to be a much stronger temperature proxy than the one I showed in Figure 1. Again we see reasonable coherence between 1100-1900 and again the BP data do not show higher values in the 20th century than the NA RCS mean. Salzer and Kipfmueller did not use RCS for detrending so it is possible that they have lost some long term information.
Take home message – I do not think the BP data are as bad as Steve would have us believe.
OK – a little more – a pounds worth.