Comments on: The Three AVHRR PCs http://climateaudit.org/2009/02/18/the-three-avhrr-pcs/ by Steve McIntyre Sat, 25 May 2013 07:32:56 +0000 hourly 1 http://wordpress.com/ By: Ryan O http://climateaudit.org/2009/02/18/the-three-avhrr-pcs/#comment-177351 Tue, 24 Feb 2009 22:21:38 +0000 http://www.climateaudit.org/?p=5265#comment-177351 Re: Jeff C. (#96), Here’s the PCs:
.

]]> By: Jean S http://climateaudit.org/2009/02/18/the-three-avhrr-pcs/#comment-177350 Sat, 21 Feb 2009 08:13:27 +0000 http://www.climateaudit.org/?p=5265#comment-177350 re: neigs
In the RegEm code neigs controls the number of eigenvalues actually calculated in the TTLS regression (peigs.m). Of course, it should be larger than “the number of eigenvalues actually used” (regpar). By default, neigs is set to the maximum (i.e., all eigenvalues are calculated), so you don’t have to worry about that. The purpose of neigs is purely computational: there is no point of calculating all eigenvalues of a large matrix if you only plan to keep a few of those.

]]> By: Jeff C. http://climateaudit.org/2009/02/18/the-three-avhrr-pcs/#comment-177349 Sat, 21 Feb 2009 04:36:37 +0000 http://www.climateaudit.org/?p=5265#comment-177349 Re: Jeff C. (#95), After thing for a few minutes (should do that before I type)the fact that it flipped probably isn’t that significant. Ryan – can you put up the plots of the 3 PCs for this recon?

]]> By: Jeff C. http://climateaudit.org/2009/02/18/the-three-avhrr-pcs/#comment-177348 Sat, 21 Feb 2009 04:25:41 +0000 http://www.climateaudit.org/?p=5265#comment-177348 Re: Ryan O (#94), Don’t forget your day job! Seriously, this is good stuff. The PCA is supposed to use similar methodolgy to the RegEM except they replace missing values with the monthly mean instead of letting RegEM do it. Why would the eigenvector flip? Doesn’t make any sense, particularly when they say “The first three principal components are statistically separable and can be meaningfully related to important dynamical features of high-latitude Southern Hemisphere atmospheric circulation” as they do in the paper.

There is also a 15 station recon using RegEM that may have some interesting features.

]]> By: Ryan O http://climateaudit.org/2009/02/18/the-three-avhrr-pcs/#comment-177347 Sat, 21 Feb 2009 03:49:06 +0000 http://www.climateaudit.org/?p=5265#comment-177347 Hopefully not too OT . . . I started looking at the PCA reconstruction and eigenvector 2 looks to be the inverse of the AVHRR eigenvector 2. I didn’t know where else to put this . . .
.

]]> By: Ryan O http://climateaudit.org/2009/02/18/the-three-avhrr-pcs/#comment-177346 Fri, 20 Feb 2009 23:59:16 +0000 http://www.climateaudit.org/?p=5265#comment-177346 Re: Ryan O (#89),
The R script download link wasn’t working. Here it is again.
http://www.mediafire.com/?imkymyqz4fw

]]> By: Mark T http://climateaudit.org/2009/02/18/the-three-avhrr-pcs/#comment-177345 Fri, 20 Feb 2009 22:25:10 +0000 http://www.climateaudit.org/?p=5265#comment-177345 Uh, I was speaking in general terms, but that’s cool, too, since I didn’t notice the distinction anyway – good to know when I actually delve into it. It’s getting confusing jumping between threads, I think.

Mark

]]> By: Ryan O http://climateaudit.org/2009/02/18/the-three-avhrr-pcs/#comment-177344 Fri, 20 Feb 2009 22:18:11 +0000 http://www.climateaudit.org/?p=5265#comment-177344 Re: Mark T (#90),

Maybe I’ll look at this stuff more this weekend. I’d be willing to bet, btw, that if you look at the impacts the various PCs have, PCs 3-6 will be unnoticeable pre-1980, and much larger post-1980, i.e., the percentage of variance they describe pre-1980 is near zero, but much larger post-1980, which averages out to the values used in the recon.

.
This one is a different situation than the AWS recon. In this case, there are only the 3 PCs because the entire recon is generated exclusively from them. Without the processed satellite data, we can’t see what 4-6 would be.
.
And I would be willing to bet that PCs 4-6 are significant, because the trends of the grid points corresponding to the manned station locations are very different than the trends from the actual ground observations.

]]> By: Mark T http://climateaudit.org/2009/02/18/the-three-avhrr-pcs/#comment-177343 Fri, 20 Feb 2009 21:11:49 +0000 http://www.climateaudit.org/?p=5265#comment-177343 Hehe, sounds like I have a “learn R” notation on my agenda now. :) Thanks.

Maybe I’ll look at this stuff more this weekend. I’d be willing to bet, btw, that if you look at the impacts the various PCs have, PCs 3-6 will be unnoticeable pre-1980, and much larger post-1980, i.e., the percentage of variance they describe pre-1980 is near zero, but much larger post-1980, which averages out to the values used in the recon.

Mark

]]> By: Ryan O http://climateaudit.org/2009/02/18/the-three-avhrr-pcs/#comment-177342 Fri, 20 Feb 2009 20:54:57 +0000 http://www.climateaudit.org/?p=5265#comment-177342 Thanks, Mark. :)
.
If you want to see the eignenvectors in action based on how they predict temperature, here’s a little movie in R. The two text files are 1) with the AWS recon and manned stations centered and spliced; and, 2) with just the manned stations centered and spliced. They’re easy to pick out . . . as are the eigenvectors.
.
To use, set sat_recon to the matrix you want (either spliced.manned or spliced.all). The rest of the instructions are commented in the last 4 lines of the script.
.
http://www.mediafire.com/file/imkymyqz4fw/MeansPlot.R
.
http://www.mediafire.com/file/mnyw15yy51m/spliceall.txt
.
http://www.mediafire.com/file/5qdzcuxynut/splicemanned.txt

]]>