Re: Band-pass filter weights

Did you look at: http://www.ncl.ucar.edu/Applications/filter.shtml ?

The quotes below basically say the same thing. To adequately filter
a small band width (eg, 2-6 days) requires more, possibly many
more weights, than is required for a broader band-width (eg, 20-100
days). Please read the article:
   Reference: C. Duchon
     J. Applied Meteorology; August,1979; pp 1016-1022

This describes the Lanczos filter used in filwgts_lanczos:
 
http://www.ncl.ucar.edu/Document/Functions/Built-in/filwgts_lanczos.shtml

You can also use an fft to perform filtering though you must
be careful about Gibbs phenomenon, side lobes, etc
A good book on digital filtering is:
   The Scientist and Engineer's Guide to Digital Signal Processing
   Steven W. Smith, Ph.D.

---
I think you sent a question to the  Climate Data Operators (CDO)
group. I have no idea what filter they use but did the CDO not work?
---
I have taken Example 4 at the above filter page and altered it.
filters_1d.ncl uses one time series to generate a few plots.
Basically, the FFT and the Lanczos filters yield identical results
away from the beginning and ends of the overall series. Some think
the FFT is 'better' because it returns values at the beginning and end
whereas the Lanczos returns missing values. This is *NOT* true.
Those 'bonus' values are 'bogus' ... you can not get something
for nothing. The problem is that some people might actuall use
those bogus values.
filters_3d.ncl actually does some filtering on 3D OLR and plots
several different approaches to looking at the data.
It is *your job* to look at the printed output, the plots,
study the NCL language and learn about filters.
I am very busy and can not answer any more questions on this.
Good luck
On 8/6/12 12:39 AM, SB Ratna wrote:
> Dear User,
> I have daily OLR data for the period of 30 years. I would like to use
> Band-pass filter on various time scale using
> the function "*band_pass_latlon_time". *However, I am bit confuse to use
> the filter wights in this function,
>
> bpf   = (/20,100,201/), which is the filter on the scale 20-100 days and 201 is     filter weights.
>
> Could you please suggest me what weights should be used for the filtering the     daily data on (i) 2-6 days (ii) 10-90 days (iii) 1-3 years.
>
> I got confusion due to the below description provided in the NCL site.
>
> The description for the function"*band_pass_latlon_time"*is mentioned as below (http://www.ncl.ucar.edu/Document/Functions/Diagnostics/band_pass_latlon_time.shtml).
>
> "The user must be aware that the appropriate number of filter weights is a direct function of the band pass period. The shorter the period, the more weights that will be required to sufficiently resolve the desired period."
>
> However, the description of "*filwgts_lanczos"* is mentioned as below
> (http://www.ncl.ucar.edu/Document/Functions/Built-in/filwgts_lanczos.shtml).
>
> "A scalar indicating the total number of weights (must be an odd number;
> /nwt/ >= 3). The more weights, the better the filter, but there is a
> greater loss of data"
>
> Thanks.
>
> Regards
> SB ratna
>
> <http://sigads.rediff.com/RealMedia/ads/click_nx.ads/www.rediffmail.com/signatureline.htm@Middle?>
> Follow *_Rediff Deal ho jaye!
> <http://track.rediff.com/click?url=___http://dealhojaye.rediff.com?sc_cid=rediffmailsignature___&cmp=signature&lnk=rediffmailsignature&newservice=deals>_* to
> get exciting offers in your city everyday.
>
>
>
> _______________________________________________
> ncl-talk mailing list
> List instructions, subscriber options, unsubscribe:
> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
>

_______________________________________________
ncl-talk mailing list
List instructions, subscriber options, unsubscribe:
http://mailman.ucar.edu/mailman/listinfo/ncl-talk