Spectral Estimation

The traditional periodogram method of spectral estimation using FFT is not appropriate for the present analysis because with a 110 year data set, we could only estimate the spectra at 0.009 cpy, 0.018 cpy, etc..., i.e. we would not have sufficient frequency resolution at low frequencies. Further, since windowing of the data would not be possible for these low frequencies, the standard deviation of the estimate of the spectra would be as large as the estimate itself [15]. At the higher frequencies, where windowing of the data is possible (due to which the confidence level in the estimates is higher), and where the frequency resolution is sufficent, we have verified that the ME spectral estimates agree with the traditional periodogram spectral estimates. A disadvantage of the ME method, however, is the problem of spurious peaks when a large number of poles is used. Consequently, we prefer to use a smaller number of poles and obtain slightly smoothed spectra in the present case. There is little drawback in doing this, since we estimate the spectra in all cases using the same number of poles. In the next section where we comment on possible important low-frequency processes, we filter the data before subjecting it to a ME spectral analysis, so that a spectral estimate with just a few poles will capture the underlying peaks.