Climate Audit

Recently I noticed that there was an (otherwise digitally unavailable) version of the Briffa reconstruction in a second (undocumented) sheet attached (the worksheet tab denoted “Science3”) to the Jones et al 1998 archive at NCDC (see here). Using this version, I was able to replicate graphics that had hitherto been impenetrable. Although the spreadsheet itself is undocumented, it is possible to demonstrate that it is a version of the Briffa reconstruction using the age-banded method later described in Briffa et al 2001 (but without the questionable use of principal components in the latter publication).

There are four other versions of the Briffa reconstruction available digitally. The archive for Briffa et al (Nature 1998) – see here – contains two slightly different normalized versions (NHD1, NHD2) and a temperature reconstruction NHLMT which is a linear transformation of NHD1). These reconstructions were made using Hugershoff standardization. The archive for Briffa et al (JGR 2001) – see here – contains an age-banded version constructed with inverse regression on principal components after deletion of post-1960 values. A version of this data before the deletion is in the Climategate emails here (Oct 5, 1999).

Briffa and Osborn 1999
The version in the Jones et al 1998 spreadsheet was used in the Briffa and Osborn 1999 graphic using a gaussian smooth after deletion of pre-1550 and post-1960 values. The comparison is shown in a blink-graphic below; the match is so precise that it precludes any other version being the source of the Briffa and Osborn graphic.

Figure 1. Blink graph showing the provenance of the Briffa version in Briffa and Osborn 1999. Click on figure to see.

Matching the spreadsheet version (Science3) to Briffa and Osborn 1999 (Science) provides important information on the spreadsheet version. The caption to Briffa and Osborn 1999 stated that the Briffa version in the illustration was from Briffa et al 1998 (393 – volcanic), “processed to retain low-frequency signals”.

Comparison of NH temperature reconstructions, all recalibrated with linear regression against the 1881-1960 mean April-September instrumental temperatures averaged over land areas north of 20ºN. All series have been smoothed with a 50-year Gaussian-weighted filter and are anomalies from the 1961-90 mean…. northern NH tree-ring densities [1550-1960, from (3 – [Briffa et al 1998(393)]), processed to retain low-frequency signals] are in pale blue …

The Appendix to Briffa et al 2001 stated that the method used in Briffa and Osborn 1999 to “retain low-frequency signals” was age-banded standardization (described in the text of Briffa et al 2001):

One curve was produced by performing the age-banding procedure on all chronologies in the data set and by using an unweighted mean of all banded series from all locations. This is similar to the curve from 1650-1960 [sic – should be 1550-1960] presented by Briffa and Osborn 1999 (although we have since made very minor modifications to the age-banding procedure and the input data set). All other curves in Figure 4 were obtained by prior averaging of the age-banded density series into the nine subregions (as defined by Figure 1).

IPCC TAR First Order Draft
The Briffa version in the Science3 tab of the spreadsheet was also used in the spaghetti graph of the IPCC TAR First Order Draft, produced in the wake of the IPCC Lead Authors meeting in Arusha, Tanzania, only two months before the notorious trick email. This helps to further clarify both the provenance of the Briffa version in the Jones et al 1998 spreadsheet and the interesting Climategate emails between the Arusha meeting and the trick email.

The blink graph below demonstrates the use of the Briffa version from the Jones et al 1998 spreadsheet – this time truncated to 1625- 1950 or so. The emulation used Butterworth smoothing (in Mann style). (The caption is inaccurate on this and a couple of other points.)

Figure 2. Blink graph of IPCC TAR FOD Figure 2-25 excerpt illustrating provenance of Briffa reconstruction from the version in the Jones et al 1998 spreadsheet. Click on figure to see.

The original caption stated:

Fig.2.25: Comparison of warm-season (Jones et al, 1998) and annual mean (Mann et al, 1998;1999a) multiproxy-based and warm season tree-ring based (Briffa et al, 1998) millennial Northern Hemisphere temperature reconstructions. The recent instrumental annual mean Northern Hemiphere temperature record is shown for comparison. Also shown is an extratropical sampling of the Mann et al (1998) temperature pattern reconstructions more directly comparable in its latitudinal sampling emphasis to the Jones et al series. The self-consistently estimated two standard error limits (shaded region) for the smoothed Mann et al (1999a) series are shown. The horizontal dashed (zero) line denotes the 1961 1990 reference period mean temperature. All the series were smoothed with a 50 year Hamming-weights lowpass filter, with boundary constraints imposed by padding the series with its mean values during the first and last 25 years.

The graphic itself shows that it was prepared by Ian Macadam on Sept 27, 1999 – only a few days after the fevered exchange among Mann, Briffa, Jones and Coordinating Lead Authors Chris Folland of the UK Met Office and Tom Karl of NOAA. Ian Macadam was a UK Met Office employee, presumably reporting to Folland, who turns up in a few Climategate emails. Mann mentioned him on Sep 22, 1999, a few days before the FOD graphic as follows:

I am perfectly amenable to keeping Keith’s series in the plot, and can ask Ian Macadam (Chris [Folland]?) to add it to the plot he has been preparing (nobody liked my own color/plotting conventions so I’ve given up doing this myself). The key thing is making sure the series are vertically aligned in a reasonable way. I had been using the entire 20th century, but in the case of Keith’s, we need to align the first half of the 20th century w/ the corresponding mean values of the other series, due to the late 20th century decline.

It’s interesting that the very first incident of hide-the-decline in IPCC literature was in a graphic prepared by the UK Met Office.

The following day (Sep 23 – 139. 0938108054.txt), in an email addressed to Jones (also sent to Folland, Briffa and Karl), Mann notes that Phil Jones had sent him the Briffa version used in the FOD spaghetti graph – one that differed from the version in the published article – something Mann described as “tenuous”:

Thanks for your comments Phil, ….

I am definitely using the version of the Briffa et al series you sent in which Keith had restandardized to retain *more* low-frequency variability relative to the one shown by Briffa et al (1998). So already, the reconstruction I’m using is one-step removed from the published series (as far as I know!) and that makes our use of even this series a bit tenuous in my mind, but I’m happy to do it and let the reviewers tell us if they see any problem.

Mann also refers here (for the first time) to yet another forthcoming version. This version – age-banding with principal components – was sent to him by Osborn on Oct 5, 1999, too late for inclusion in the FOD, but used in the final graphic. It was eventually published in Briffa et al 2001. Mann:

If I understand you correctly, there is yet a new version of this series that is two steps removed from Briffa et al (1998)? Frankly, at this stage I think we have to go w/ what we have (please see Ian Macadam’s plot when it is available–I think the story it tells isn’t all that bad, actually) for the time being. Things as you say will change following review anyways.

Please check out the data here ASAP:
ftp://eclogite.geo.umass.edu/pub/mann/IPCC/MILLENNIUM/
This directory has all the series, aligned as I described to have
a 1961-90 base climatology (or in the case of your series, a pseudo 1961-90 base climatology achieved by actually matching the mean of your series and the instrumental record over the interval 1931-60 interval). These are the data that Ian Macadam is hopefully presently plotting up, and I don’t think the discrepancies between the different series are as bad as we percieved earlier (other than the late 19th century where
you are somewhat on the warm side relative to the rest). Please confirm ASAP that we have the right version of the series (note, these have all been 40 year lowpassed)…

Perhaps one of the reasons why the “discrepancies between the different series” were not “as bad as we percieved earlier” was that the FOD graphic deleted Briffa values prior to 1625. If the full Briffa version were used, it would have, of course, yielded the same discrepancy as the deleted portions of the Briffa and Osborn 1999 spaghetti graph.

This subsequent curve used almost exactly the same data, but the result was much closer to the Mann/Jones reconstructions, as noted in Phil Jones’ reply (141. 0938121656.txt):

One important aspect Keith will address is whether you’re using the latest Briffa et al curve. We know you’re not but the one with the greater low frequency and therefore much better chance of looking much better with the other two series, isn’t yet published. We know it looks better in plots we have here.

Unfortunately, Mann’s archive at the University of Massachusetts has been deleted, including the data in ftp://eclogite.geo.umass.edu/pub/mann/IPCC/MILLENNIUM/.

While the information on the provenance of the Briffa version in the Jones et al 1998 archive has to be pieced together, the net result is, in my opinion, a convincing demonstration that the Briffa version in the Science3 tab of the NOAA spreadsheet is an age-banded version constructed from a network virtually identical to the Briffa 2001 network, but without using principal components, and that, after truncation, this version was used in Briffa and Osborn 1999 and (later that year) in the IPCC TAR FOD spaghetti graph.