Looking at how the insurance market works for the science journal Nature, Myles Allen, an Oxford University physicist, thinks that problem is now largely solved. All you have to do, he says, is work out “a ‘mean likelihood-weighted liability’ by averaging over all possibilities consistent with currently available information”. Unpacked, it means that if past greenhouse gas emissions have increased flood risk tenfold, 90 per cent of the damage caused by a flood can be attributed to past emissions. Because carbon dioxide mixes itself in the global commons of the atmosphere, “an equitable settlement would apportion liability according to emissions”, argues Allen.

Link

1. In all stochastic time-series analysis we must remember that the sample autocorrelation function is not the same as the population (ensemble-generating) autocorrelation function. Even if the population acf is stationary, a sample acf is expected to dip up and down randomly as in this graph of ac(1). Of course, no one is saying the population acf is stationary either. It may well change as the climate system goes through its machinations. I believe I’ve made this point before about the danger of making inferences about acfs based on sample acfs from short* series. (*Where “shortness” is relative to the shape of the acf.)

2. I do not have the technical competence required to determine if Quenouille’s method as you propose is statistically valid. But inuitively it makes sense to me. I saw your question the first time around in another thread and thought: “I better not reply; I’m no expert”. But I agree; your suggestion may have merit. One must keep in mind that the sample ac coefficients used in Quenouille’s method are going to bounce around through time as in your graphic. It’s anybody’s guess as to whether this variation is meaningful. But it means that any analysis is context-dependent: you will not get the same correlations if you work from 18th, 19th, 20th c., etc. Climatologists must be wary of the contextual nature of all time-series analysis. Framing is everything.

Not quite sure what to make of this. Lag-1 autocorrelation in the calibration period 1850-1980 is 0.85. Maximum 131 year lag-1 autocorrelation is 0.87, minimum is 0.35, average is 0.66. Lag-1 autocorrelation for the entire 981 year dataset is 0.84.

I see no easy way to generate “red noise” with anything like that correlation structure. Again, I have to ask, why not use Quenouille’s formula for the significance of the correlation of two trends, as discussed here? I still haven’t gotten an answer as to why this would not be appropriate.

Best to you all,

w.

_________________________Series__Series________________________________
______Series_______Type__Start___End_____________Archived___________

______Guliya________IC_____OK______OK____Differs from Archived Version_
______Dunde________IC_____OK______OK____Differs from Archived Version_
______Dulan________TR_____OK______OK_____________Unarchived__________
____S Tibet 1_______TR____1100____1950____________Unarchived__________
____S Tibet 2_______TR____1100____1950____________Unarchived__________
____S Tibet 3_______TR____1100____1950____________Unarchived__________
____S Tibet 4_______TR____1100____1950____________Unarchived__________
____S Tibet 5_______TR____1100____1950____________Unarchived__________
____S Tibet 6_______TR____1100____1950____________Unarchived__________
____S Tibet 7_______TR____1100____1950____________Unarchived__________
____S Tibet 8_______TR____1100____1950____________Unarchived__________
____S Tibet 9_______TR____1100____1950____________Unarchived__________
____S Tibet 10______TR____1100____1950____________Unarchived__________
____S Tibet 11______TR____1100____1950____________Unarchived__________
____S Tibet 12______TR____1100____1950____________Unarchived__________
____East China_____DOC____OK______OK_____________Unarchived__________
_Great Ghost Lake__SED____OK______OK_____________Unarchived__________
_____Jiaming_______SED____OK_____1960____________Unarchived__________
_____Jinchuan______SED____OK_____1950____________Unarchived__________
______Japan________TR_____OK_____1950____________Unarchived__________
____Tornetrask______TR_____OK______OK____Differs from Archived Version_
______Yamal________TR_____OK______OK____Differs from Archived Version_
____Tornetrask______TR_____OK______OK____Differs from Archived Version_
______Taimyr________TR_____OK______OK____Differs from Archived Version_
_Methuselah Walk___TR_____OK_____1979________________OK______________

TR: tree ring DOC: documentary SED: sediment.

I invite anyone who knows of problems with any of the proxy series (timing, location, duplication, start/end dates, etc.) to add to this list. Also, if anything in this list is incorrect, please post a comment to that effect.

Finally, I am looking to collect concise statements of any other problems with the Juckes et al. “Millennial temperature reconstruction intercomparison and evaluation” study. These include theoretical, practical, and ethical problems. When the collection is complete, I will post them on the “Climate of the Past” online review web site. The discussion there is open until 21 December 2006, so I propose 7 December 2006 as a closing date for these comments on this thread.

Likely this should be a new thread.

My best to everyone,

w.

PS – Is there a way to post a table, such as the one clumsily implemented above, to this site?

I’m compiling a list of the issues raised here, for eventual posting on the CoP discussion site. I’ll post it here first and ask for comments …

w.

Averaging 12 temperature-sensitive tree-ring
series from various parts of Tibet (r = 0.52 to 0.79, p < 0.01), Wu
and Lin [1981] established a reconstruction of yearly average
temperatures anomalies. However, there is a data gap from the
7th to 11th century in this series (Figure 2).

Thus, rather than nine individual series, Yang contains 8 individual series, plus an average of 12 other individual series. Of the total of 20 individual proxies, 12 of them (the group from Tibet) don’t have data from 600 AD to 1100AD.

w.

Willis, I don’t understand your post here. The Yang composite doesn’t have 20 proxies. It has 9 proxies: 1) Guliya; 2) Dunde – in both cases, Yang uses a very smooth version inconsistent with the most recent versions archived in 2004 by Thompson; 3. an unarchived Dulan tree ring series; 4. an unarchived S Tibet tree ring series 5. an unarchived E CHIN documentary 6. an unarchived Great Ghost Lake,Taiwan sediment series 7. an unarchived Jiaming sediment series 8. an unarchived Jinchuan sediment series 9. an unarchived Japan tree ring series (dC13 as I recall).

However, according to Yang’s paper, the “unarchived S Tibet tree ring series” is not actually a single series, but an average of 12 individual tree ring proxies in Southern Tibet. According to Yang,

Averaging 12 temperature-sensitive tree-ring
series from various parts of Tibet (r = 0.52 to 0.79, p

Thus, rather than nine individual series, Yang contains 8 individual series, plus an average of 12 other individual series. Of the total of 20 individual proxies, 12 of them (the group from Tibet) don’t have data from 600 AD to 1100AD.

w.

They make the assumption that proxies that correlate well with recent instrumental data are good proxies for temperature. Your graph clearly shows that is not the case, as the proxies differ widely in the non instrumental period, so as you say, they can’t all be correct. I suppose the deviations of the proxies from each other in the non instrumental period must have been measured somewhere along the line? In fact in Juckes’ Table 1 only 5 of the 34 proxy series shown have an R higher than 0.5 when compared to the Northern Hemisphere temperature record, so even the claim that the Proxies correlate well with recent temperature is suspect.

From Juckes Paper

MM2005c also suggest that the MBH1998 “North American proxy PC1″ is a statistical outlier as far as its correlation to Northern Hemispheric, temperature Tnh, is concerned. Table 1 shows, however, that other tree ring series and other proxies have higher anomaly correlations with Tnh. Pages 1019 (foot),1020 (top)

What does this mean? Does it mean that Manns PC1 R=0.49 is ok because other tree ring proxies have a higher R?

Where is the logic in this?

Lastly, a more general question. If a sample is taken, it can be measured and used to evaluate the characteristics of the population as a whole. The larger the sample, the more confidence can be placed in its characteristics being similar to the global population.

I would have thought that the sample data collected to form the basis of these proxy data sets is so small compared with the overall population of each group, that the amount of confidence that can be placed on them accurately representing the population as a whole is very low. In addition, aside from the issue of cherry picking data, as Bender has pointed out (survivor bias), it is virtually impossible for any sampling of trees surviving today to be in any meaningful way reprentative of the tree population in the time period being measured.

As to archiving, Briffa has calculated a completely different chronology from Yamal than the one archived by Hantemirov and Shiyatov. Juckes has cited Hantemirov and Shiyatov as an authority, but didn’t use the version that they archived. Instead he used Briffa’s completely different version without any disclosure. Briffa has refused to archive measurement data for Yamal so that observers can check for themselves. Likewise Briffa has refused to archive measurement data for the Tornetrask update and Taimyr. Given these flat-out refusals, I’m not sure what Juckes’ criteria really are.