🚀 go-pugleaf

In article <1992Oct8.152654.17216@news.miami.edu> Arthur Mariano, 
mariano@sgrmag.rsmas.miami.edu writes:
>An AR(1) filter or Markov model can either
>act as a high or low pass filter, while an AR(2m) model is used for
>detecting m sinusoids. Many oceanographic data sets are red and have
>significant energy in a broad mesoscale range. AR methods which use

I don't know why there is so much confusion about this.  The same is
true of the FFT.  The FFT of the covariances is identical to the squared
FFT of the time series, in the case of regularly spaced data.  This is
a mathematical identity.  However, when there are many gaps in the data,
it is still usually possible to form a "complete" set of covariances to
transform into a power spectrum.  (The number of degrees of freedom in
each estimate of the covariance varies according to the gaps, etc., so
there may be some subtleties in assigning confidence limits.)

For either approach, squaring the FFT of the timeseries or FFTing the 
covariances, it is probably wise to separate the mean and trend from
the data first.  Cheers,

	-jerry