The key question, though, is how much variance remains at each frequency. Remember that the frequency response of a filter is a curve, not a step-function. If your filters retain half-amplitude at 7 and 9 years (as in the function arguments you state below), then you probably are losing a great deal of the variance in your band.
There are various ways to check this. One simple one would be to make a simple sinusoid with an 8-year period, and run your filters on it. How does the amplitude of the result compare to the (known) amplitude of the original sinusoid?
You should always experiment with the arguments to these canned filters. Try various choices and look at the results (for both the sinusoid and the actual data).
The frequency response of various filters (including the Lanczos filter) is well-described in the excellent book:
Bloomfield, P., 1976, Fourier Analysis of Time Series: An Introduction, John Wiley & Sons, New York, 258 pp.
Also see Duchon, C., 1979, "Lanczos filtering in one and two dimensions", J. Applied Meteorology, 18, 1016-1022.
Billy K On 25 Aug 2010, at 10:17 PM, nuncio murukesh wrote:
Hi users,
This is actually not a FERRET question but somebody who familiar
lsl_lowpass function may be able to help.
I would like to retain a particular frequency in a monthly time
series, say 7-9 year periodicity. For this purpose I constructed two
seperately filtered time series using lsl_lowpass function
filt1=lsl_lowpass(A, 108,108)
filt2=lsl_lowpass(A,84,84)
to retain 7-9 year periodicity filt1 is subtracted from filt2
Does this yield good results?
nuncio
--
Polar Remote Sensing Division
NCAOR
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ William S. Kessler NOAA / Pacific Marine Environmental Laboratory 7600 Sand Point Way NE Seattle WA 98115 USA william.s.kessler@xxxxxxxx Tel: 206-526-6221 Fax: 206-526-6744 Web: http://www.pmel.noaa.gov/people/kessler