Re: [ferret_users] low pass filter

["William S. Kessler" <william.s.kessler@xxxxxxxx>]

Do not use any filter blindly!  Always make some test plots of the  
results of any filter!

There is no such thing as a filter that removes all variance in a  
frequency band, while preserving all variance in other bands. Any  
usable filter has a curved frequency response: it will remove some of  
the variance you are interested in, while retaining some that you  
would like to remove. For this reason, you must experiment and  
consider the properties of any filter before using it for more than a  
quick look at the data.

Try several different filters, starting with the simple ones (@SBX,  
@SPZ). Plot the filter results overlaid on the unfiltered time series  
to learn about what it is doing, and how choices of filter type and  
parameters affect the results. After doing several such tests, you can  
generally see by eye how the different filter options interact. (If  
your data set is very large, perhaps make a smaller version so the  
computation doesn't take too long and the plot is comprehensible).

In general, geophysical data is "dirty": there are so many processes  
going on, and so many different things correlated with each other and  
interacting with each other, that fine tuning a filter is often a  
waste of time. Unlike, say, astronomy, where the precise  
characteristics of a filter can be crucial to extracting a signal, our  
science is just very messy. That's the way climate is. Geophysical  
reality is generally not composed of pure harmonic tones. For example,  
I doubt if it makes very much difference to use Ferret's triangle or  
Hanning filters (as long as the filter lengths are the same).

The Lanczos filter used in LSL_LOWPASS is relatively sophisticated; is  
there a reason why it is necessary for your purpose? Definitely plot  
its results on top of @SPZ to see if it is worth the extra computation.

Before using LSL_LOWPASS, I recommend several good sources of  
information:

Duchon, C. E., 1979: Lanczos filtering in one and two dimensions. J.  
Appl. Meteor., 18, 1016–1022.

Bloomfield, P., 1976, Fourier Analysis of Time Series: An  
Introduction, John Wiley & Sons, New York, 258 pp.
(This book is where the algorithm for LSL_LOWPASS comes from)

And see the Appendix of Jones et al for a plot of the frequency  
response of the Lanczos filter:
Jones, C., D. E. Waliser, and C. Gautier, 1998: The influence of the  
Madden–Julian oscillation on ocean surface heat fluxes and sea  
surface temperature. J. Climate, 11, 1057–1072.

Billy K

On 20 Feb 2011, at 5:59 AM, golla nageswararao wrote:

> Hi all,
>   I am having a basic doubt in low pass filter.
>   I am having 10 min interval timeseries data of winds.
>   I want to do remove period less than 10 days from my data.
>   Can anybody say that whether my following syntax is right or not?
>  let u_lp=lsl_lowpass(u,730,730)
>
> Thanks in advance.
>
> --
> With Best regards,
> G.NageswaraRao,

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
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