extremely interesting tornetrask and taymir chronologier on page 6!

Steve: Hi, Hans. These look the same as the corresponding Briffa 2000 chronologies – the measurement data for which became available in Sep 2009 as we all know

In the available data for annual anomalies, there is a clear “step change” that occurs between ’96 and ’98 of around 0.4 degrees in the NH and a similar, though smaller, step in the SH. These coincide very nicely with the unusually strong ’97 / ’98 El Nino event. Since then, there’s been a fairly flat ongoing trend at about the level of the “step”. That makes it completely unsurprising that “the last decade is the warmest on record”

However, in the headline graphs of trends, this step is smoothed which creates an impression of a rapid increase from before that time.

Now, my statistical education stopped around A level, about 25 yeras ago, but I seem to remember that it was generally bad practice to try and construct a trend across an event that causes a known and explained discontinuity in your data. Indeed, that’s one of the reasons they have to homogenise all the raw data before anyone sees it!

So, what am I missing and how do they justify treating temperatures before and after that step as a single series?

My post above is about the “highly artificial” “fudge factor” code. The way it was used and documented in the paper draft satisfies me personally that it’s not fraudulent.

What has “gone wrong” in the tree ring width and density calibrations, I believe, is the following two things. First, the rainfall matters a lot. Rainfall is not independent of the temperature. Researchers have tried find samples from areas where temperature is the constraint, not rainfall, to mitigate this. However, I don’t know how well the rainfall has been ruled out.

Second, and more importantly, most tree-ring studies don’t adjust for the age of the tree. The width and density of the tree ring are related to both how many years old that particular tree was when it grew the ring in question and what kind of year that year was.

Fortunately for us, at least one study adjusts for the age of the tree: http://people.su.se/~hgrud/documents/Grudd%202008.pdf

I just read Grudd’s study, and he demonstrates pretty convincingly that the reason why the tree ring density has stopped correlating with temperature in the post-1960 sample is that the trees in the previous studies were old. Grudd collected data on young trees and merged those to the data set. The presence of both old trees and young trees from all periods then allowed him to estimate the relation between the age of the tree when the ring grew and ring density and the year effect.

The paper suggests that Briffa / Briffa et al. (so many papers, can’t keep track) should have collected younger sample trees, should have removed the effect of the age of the tree, and then stuck with unadjusted density-based proxy. This would have eliminated the need for “Mike’s Nature trick” to “hide the decline.” There’s no decline in the density of recent rings of young trees.

Now, what’s the punch line from Grudd’s study? When you model the tree ring density right, you’ll get a relation between temperature and density that holds up well throughout the sample. Then, if you use that model to reconstruct temperature record (blue line in Fig 12), you’ll see that in Northern Europe was really warm for a couple of centuries about 1000 years ago, warmer than now.

Perhaps this has been done, but if so, than your discussion really shows that something went wrong?!

If I set up models, I try to find achieved real-life data and test if my model can at least replicate that, if not, than I have a problem?!

[Response: No. There was a draft, but it doesn't seem to ever have been published, and is very clear about why and how this was done. - gavin]

I read the paper at http://74.125.93.132/search?q=cache:www.cru.uea.ac.uk/~timo/papepages/pwosborn_summertemppatt_submit2gpc.pdf. Here’s what I believe is the relevant section, on page 21:

“Warm-season temperature reconstructions with extended spatial coverage have also been developed, making use of the spatial correlation evident in temperature variability to predict pasttemperatures even in grid boxes without any tree-ring density data. The calibration was undertaken on a box-by-box basis, and each grid-box temperature series was predicted using multiple linear regression against the leading principal components (PCs) of the calibrated, gridded reconstructions described in section 4.4. The PCs were computed from the correlation matrix of the reconstructions, so the calibration was in effect removed and similar results would have been obtained if the PCs of the raw, gridded density data had been used instead. The only difference is that the calibrated data with the artificial removal of the recent decline were used for the PCA. Using the adjusted data avoids the problems otherwise introduced by the existence of the decline (see section 4), though all reconstructions after 1930 will be artificially closer to the real temperatures because of the adjustment(the adjustment is quite small until about 1960 – Figure 5c). Tests with the unadjusted data show that none of the spatial patterns associated with the leading PCs are affected by the adjustment, and theonly PC time series that is affected is the leading PC and then only during the post-1930 period. Inother words, the adjustment pattern is very similar to the leading EOF pattern, and orthogonal to theothers, and thus only influences the first PC time series.”

I don’t find the argument particularly convincing as far as the substance of this hypothetical experiment goes. But that’s not the point. The point is that I think this is conclusive evidence that the “very artificial” “fudge factor” code is _not_ fraudulent.

Tuomo says:
6 December 2009 at 8:28 PM
Given that this is a comments thread to a CRU-email context post: Here’s a question about what exactly “Mike’s Nature trick” is. By Mike’s Nature trick I mean what exactly was done by Jones that he referred by that term.

If you read the web, you see a lot of claims. Some obviously wrong, some more intriguing. My question to you: Are the following claims true?

- Briffa’s proxy data is valid as a temperature proxy only if it correlates with the instrumental temperatures.

[Response: Basically true. You could also use other independent temperature proxies in earlier times. - gavin]

- In Briffa’s original proxy data series, the proxy series correlates positively with the instrumental series before 1940, but there’s no positive reliable positive correlation in the post 1940 data.

[Response: Not quite. The divergence happens post 1960. - gavin]

- (Based on the email correspondence and code samples, someone could guess that) Jones deleted the post-1960 values of the Briffa series, replaced them with some materially different values.
- The Jones’s materially different values are most likely instrumental temperature measurements or other values selected to be close to the instrumental measurements
- Jones then smoothed the spliced series.

[Response: This is a misreading. The only goal was a smoothed blend of the proxy and instrumental data to indicate the long term and recent changes without it being too cluttered. The post-1960 data in the Briffa reconstruction isn't relevant to that. But smoothing requires some decision about what to do with the end point problem (in this case starting in 1935 since it was a 50 year smooth). Jones used the instrumental data so that values from 1935 to 1985 are a blend of the proxy and instrumental data. I'm not quite sure what the criterion was at the 1999 end point of the instrumental period. - gavin]

- The end result was a “proxy” series that looked like an accurate reconstruction of late 20th century temperatures when compared to the instrumental measurements.

[Response: No. The end result was described as the proxies and instrumental record.- gavin]

- The results were published in World Meteorological Organization WMO-No. 913 (http://www.wmo.ch/pages/prog/wcp/wcdmp/statemnt/wmo913.pdf)

[Response: The smooth was used in that brochure. - gavin]

My question to you are which ones of these claims are true and which are not? I guess with more resolution I’d like to know which claims are demonstrably wrong (proven innocent), which claims could be true but that there’s not credible evidence for (not proven guilty), and which claims are demonstrably true (guilty).

What is of particular interest to me is whether the output of the “highly artificial” “fudge factor” code was ever used in any published papers. I have a co-author that uses “if 6==9″ as debugging toggle, and code sections in those parts do not (at least intentionally) make it to our papers. He argues that this might be a similar case.

[Response: No. There was a draft, but it doesn't seem to ever have been published, and is very clear about why and how this was done. - gavin]

- Briffa’s proxy data is valid as a temperature proxy only if it correlates with the instrumental temperatures. In Briffa’s original proxy data series, the proxy series correlates positively with the instrumental series before 1940, but there’s no positive reliable positive correlation in the post 1940 data.

- (Based on the email correspondence and code samples, someone could guess that) Jones deleted the post-1960 values of the Briffa series, replaced them with some materially different values.

- The Jones’s materially different values are most likely instrumental temperature measurements or other values selected to be close to the instrumental measurements. The end result was a “proxy” series that looked like an accurate reconstruction of late 20th century temperatures when compared to the instrumental measurements.

- Jones’s graphs used the code fraglemtn with “highly artificial” “fudge factor.”

- The results were published in World Meteorological Organization WMO-No. 913 (http://www.wmo.ch/pages/prog/wcp/wcdmp/statemnt/wmo913.pdf)

I’d like to know which claims are demonstrably wrong (proven innocent), which claims could be true but that there’s not credible evidence for (not proven guilty), and which claims are demonstrably true (guilty).

What is of particular interest to me is whether the output of the “highly artificial” “fudge factor” code was ever used in any published papers.