In my original script, the array "pc" (principal components) is of shape (/8,22000/). I preformed the spectral analysis for all the PCs. However, to make it easier to find out what my problem was, I modified the quoted parts in the previous mail but didn't manage to delete all the "ipc" subscripts. :-)[3] What is "ipc" subscript?
maxlag = dimsizes(pc)/5[2] what is maxlag? According to Chatfield [see esacr documentation for reference] maxlag should not be larger than dimsizes(pc)/4. Using this length means that the autocorrelation approach will not yield the same information for higher freq.
It's true that the first order Markov process is not consistent with my spectrum. The AR(3) process fits much better. (I haven't yet tried any orders higher than 3). Even if I take the spectrum of the fitted AR(3) process as reference, which seems to fit quite well to my spectrum obtained by applying "specx_anal" directly to the original time series, the situation seems different if I use the auto-correlation coefficients (Figure 2 of the attachment). I thought the two figures should be similar and wondered if I have make some mistakes in getting the reference curves in Figure 2. That was why I sent the previous e-mail.Well, when I look at the black dots ... I'd say your spectrum is not consistent with a first order Markov process.
_______________________________________________
ncl-talk mailing list
ncl-talk_at_ucar.edu
http://mailman.ucar.edu/mailman/listinfo/ncl-talk
This archive was generated by hypermail 2.2.0 : Mon Nov 21 2005 - 17:26:33 MST