# Re: How to remove interdecadal variation by using NCL

From: Dennis Shea <shea_at_nyahnyahspammersnyahnyah>
Date: Wed, 25 Oct 2006 06:56:06 -0600 (MDT)

I forgot

[5] If appropriate, I would suggest removing
any linear trend in the time series prior
to applying the filter.

On Wed, 25 Oct 2006, Dennis Shea wrote:

> >(periods over 8 years) on the data through harmonic analysis.
> > Is it right to use fourier_info function of NCL?
> > Please let me know how to remove that variation
>
> [1] No. It is not appropriate to use harmonic analysis.
> This assume the data are cyclic. Obviously, any
> time series is not cyclic.
>
> [2] The classic approach is to use a filter. In the
> case you describe, it would be a low pass filter.
> The filter consists of weights which would be
> passed over the data via "wgt_runave"
>
> http://www.ncl.ucar.edu/Document/Functions/Built-in/wgt_runave.shtml
>
> [3] You can design your own filter using "filwgts_lancos"
>
> http://www.ncl.ucar.edu/Document/Functions/Built-in/filwgts_lancos.shtml
>
> The filter design is a trade-off between "cut off frequency" and
> the number of weights. You did not state the sampling interval.
> Then use the weights as input to wgt_runave. If you have (say)
> 51 weights, then you will "lose" 25 points at each end of the series.
>
> [4] There is another approach: (a) taper the data; (b) forward fft;
> (c) set coefficiebts to 0.0, or weight the coefficients;
> (d) backward fft. This 'sounds' better BUT you will get
> unexpected behavior at the beginning and end of the
> reconstructed [filtered] series.
>
> Good luck
>
> _______________________________________________
> ncl-talk mailing list
> ncl-talk_at_ucar.edu
> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
>
_______________________________________________
ncl-talk mailing list
ncl-talk_at_ucar.edu
http://mailman.ucar.edu/mailman/listinfo/ncl-talk
Received on Wed Oct 25 2006 - 06:56:06 MDT

This archive was generated by hypermail 2.2.0 : Wed Oct 25 2006 - 16:39:38 MDT