The Reference also has a Table 1 of other proxies and paleo-temperature records that show periodicities of between, “55-76 years.”

So, the pregnant question, and admittedly my hobby-horse, can/does the BEST methodology preserve a hypothetical 60 year frequency in data that is chopped up by the scalpel? I do not see how it can on a theoretical basis, and from P.Solars’ pics above, there seems little evidence that it does.

Lui-2011 Fig.2 Power Spectrum shows large power density at frequencies below 0.05/yr (i.e. cycle times >= 20 years). They note significant power spikes at 1324, 800, 199, 110 Years (and by visual estimate, ~70 and 27 yrs, too).

Their Fig. 4 effectively shows how sequentially adding the low frequency components approaches the long term character of the 2489 year sequence.

I bring these graphs to your attention only as a good illustration of how LOW FREQUENCY content in the data is essential to the study of multi-decade, nay multi-century, temperature trends.

To recap, BEST’s methodology of the suture is to trust the short term trends, and pay little attention to absolute temperature values. If their mean segment lenght is 12 years (as per P.Solar above, Nov.4), then it is as if BEST took Lui’s Fig.2 and cut away EVERYTHING LEFT of 0.08 leaving little but white noise scraps to suture back together for a century+ “signal”. BEST has some low frequency content in its final sutured product, but since all the original low frequency was destroyed by the scalpel, what low frequecy reappears is highly suspect.

I compliment Lui et al, on a study that highlights Fourier analysis of the data. I cannot speak to the quality of Lui-2011 paper because I have not and cannot validate HOW THEY retained their low frequency data from “tree ring to computer”. I do not know how they retained DC shift and the DC trend prior to the Fourier analysis. I’m also a little puzzled how they got a 1324 year freq component from a 2485 year time series, since integer cycles of the processed signals are the product. Perhaps a variable window size was applied…

My point is that the process as I see it effectively destroys the data we seek. Proof that it is authentically recreated in the suturing process are scant. There is a low frequency returned in the final produce, but it is not original. It is counterfeit.

Based upon what must be happening to the data in the Fourier Domain, I am having the same reaction to the BEST process as most of you would have to a process that seems to violate the 2nd Law of Thermodynamics.

I share your concerns about this data processing. If you try to find the OLS trend over a 12y period … ANY 12y PERIOD in the last 200y of B-est you’ll find it is ZERO!

can you show the calculation? I don’t follow it.

http://img38.imageshack.us/img38/1905/atsas.png

Another factor that could cause spurious anomalies of the type Steve discusses in his 10/31 324PM comment above is repainting of the old Stephenson Screens/Cotton Region Shelters. As Anthony noted when Surface Stations was just getting started, these screens were supposed to be repainted periodically to maintain constant long-run whiteness. However, if they darken for a few years and then suddenly become white again, the BEST algorithm might detect an anomaly and thereby eliminate the correction!

As Steve points out, it would be very useful to see an analysis of the BEST slice points, both by size and by time.

An abrupt downwards change would likely be interpreted as an anomaly (in the true sense of the work) and “corrected”. The gradual recovery in the following decade would not cause a correction in the other sense.

Unless this is studied and accounted for volcanoes are likely to result in spurious warming “corrections”.

The effects of all these methods need fully auditing. The authors should be doing this as a matter of course before even writing their draft papers. This is not a option to looked into 5 years later as “further research”.

Any net positive or negative correct needs to be identified. If it can be quantifiably attributed to a known cause, fine.

All corrections should be logged with time, magnitude and location and the resulting data studied. If there are clusters of corrections in 1963 ,for example, you have issues that need correction before publication.

The current position of B-est that their adjustments “should be trend neutral” is absolutely unsatisfactory. This needs to be studied, documented and the results published.

Without such verification the method has no scientific value.

http://tinypic.com/r/24qu049/5

Accepting that Hadcrut is land and sea, it still seems like the longer frequencies have been decimated as Stephen suggested.

For apples to apples this should be done with CRUtemp land only but it’s getting late.