and various individual pages such as

http://www.cgd.ucar.edu/ccr/ammann/millennium/CODES_MBH.html

I was unable to access any of them although I was able to access other files at UCAR. On earlier occasions, I have blocked from Mann’s FTP site and from Rutherford’s FTP site. Could someone check as to whether they can access Ammann’s webpage. If you can, then the Team has made another block. It seems surprising that employees of federally funded agencies can engage in such petty personal forms of discrimination, if indeed, that’s what Ammann has done.

]]>Hate to disagree with ya Greg, but I disagree with ya ;)

And you are indeed correct Spence.

]]>Hate to disagree with ya Greg, but I disagree with ya ;)

Removing the mean from a sample period will zero the DC bin whether the data has a trend or not.

Remember the definition of F(n) for a DFT of time series f(k)

F(n) = 1/N SUM[from k=0 to N-1] f(k).e^(-j.k.2.pi.n/N)

The DC bin occurs at n=0, i.e. the exponential term k.2.pi.n/N becomes zero

e^(-j*0) is e^(0), which is unity

Therefore the DC bin is

F(0) = 1/N SUM[from k=0 to N-1] f(k)

… which looks pretty much like the mean of f(k) to me. If the mean of the series is zero for a real transform, then the DC bin is zero, whether the data has a trend or not.

]]>POI. Removing the mean will not zero the DC bin. To zero the DC bin you also have de-trend the sample period.

]]>I found the result still not convincing.

eg:

http://home.casema.nl/errenwijlens/co2/pfisterbgaden.gif

http://home.casema.nl/errenwijlens/co2/errenvsluterbacher.htm

I think we are talking slightly at cross-purposes here. I’m not sure you quite follow my line of reasoning, where I’m coming from and where I’m trying to get to. This may be partly my fault for doing a brain dump to the blog rather than explaining in a structured way.

Just to clarify, I didn’t mean to come across as saying you assumed there are no other frequencies in the reconstruction, I simply stated that it would be wrong to make that assumption. I’m trying to clarify my point by including all details.

Clearly frequencies below 1/period can, and do, exist. That doesn’t preclude the fact that certain conditions have to be met to capture those frequencies in the samples.

Okay, your getting close to where I’m coming from here. Let me try and explain it in a different way.

We start off with a 580-year reconstruction. Clearly that contains resolvable frequencies using a harmonic model down to 1/580 year. We are analysing that by looking at a 45-year portion of that reconstruction. As you point out, these low frequencies don’t map cleanly to the shorter distribution. When we chop off the DC bin, analogous to removing the mean in the R2 statistic (a step that does not take place in the RE statistic), we modify how the DFT responds to these lower frequencies.

What I’m actually interested in is one step further back from this – what is the consequence of doing this on an unconstrained system, and a system constrained by matching a 79-year portion for mean and variance, then assessing over a 45-year sampled portion of that curve.

My assertion is that the low frequency components coupled in via the assumptions and constraints associated with the harmonic model are the same as the assumptions and constraints of zeroing the DC bin of the DFT. Therefore any consequences in terms of correlation distance could be assessed by looking at the band pass associated with the DC bin of a 45-point DFT, on an unconstrained system. And that is equivalent to applying a sinc function to an unconstrained system.

So, have I helped or confused the issue further?

]]>