I was planning a new post on the truncation of the inconvenient 20th century downturn in the Briffa 2001 graphic in the IPCC graphic, that I observed some time ago here. Since data truncation is in the news, I was going to update the graphic in this post to better show just how cynical the IPCC truncation was (and will still do so.)
In the course of doing so, I thought that I’d check Briffa’s correlation claims and, surprisingly or unsurprisingly, found that my calculations were consistently lower than his. Also that the correlations over the full period of overlap were lower than the reported correlations over 1880-1960.
Briffa et al 2001 made 9 regional reconstructions shown in his Plate 2. See here with digital versions of 9 regional reconstructions here. Here is an example of one of the regional reconstructions together with Briffa’s description:
Plate 2 shows all of the new-low-frequency April-September regional reconstructions, plotted here as decadally smoothed anomalues in black. They are bracketed by their 1 and 2 standard error confidence estimates. These are time and time-scale-dependent and take into account the residual temperature variance in calibration and uncertainty (Standard errors) in the regression coefficients, as well as acconting for autocorreltation in the residuals (using a method suggested by P. Bloomfield, pers comm 1990) as described by Briffa et al (submitted 2000 – Holocene 2002).
Briffa checked the correlations of these 9 reconstructions with regional temperature collations, describing the procedure as follows:
Each regional ABD MXD time series (See Figure 1) has been calibrated against an equivalent regional April-to-September mean temperature series, produced as an aggregate of the co-located instrumental surface anomalies (with respect to 1961-90 from land stations only [Jones et al 1999]). We used simple linear regression fitting the regression equations over the period 1881-1960 or over the total available period prior to 1960 when the instrumental record was shorter (see Table 1). The period after 1960 was not used to avoid bias in the regression coefficients that could be generated by an anomalous decline in tree density measurements over recent decades that is not forced by temperature Briffa et al Nature 391, 1998
Here is the table showing the claimed correlations between the regional ABD density reconstructions and regional Apr-Sept temperature.
I then collated CRUTEM3 (and CRUTEM2) gridcell temperatures for the gridcells in the Briffa regions and calculated the Apr-Sept average over all the gridcells in each region and then calculated correlations both for the entire record and the 1880-1960 period (as in Briffa). The barplot below compares the reported Briffa values (on the right) for each of the 9 regions with my replications – far left – for full period; 2nd – 1880-1960 CRUTEM3; third – 1880-1960 – CRUTEM2. In all cases, the selected 1880-1960 yields a higher correlation than the full period – the 1880-1960 selection minimizes the effect of the Divergence Problem. In a couple of cases, I got much lower correlations than reported: CAS and TIBP. In the CAS region, the series were virtually uncorrelated over their full length, while Briffa reported a correlation of over 0.5.
The opportunism of the 1880-1960 period for “low-frequency” comparison is shown in the following figure. In 4 of 9 regions, low-frequency correlations are negative over the total period (due to Divergence). Briffa et al did not report these correlations, but only the correlations over the opportunistically chosen 1880-1960 period.