[GRASS-user] r.series and Fourier

[Glynn Clements]

Daniel Victoria wrote:

> A friend of mine asked me if it would be possible to do a Fourier
> analysis on a raster time series with Grass (or any other tool). Since
> I'm no expert in Fourier or time series the first thing that came to
> my mind was to send the grass rasters to R and do the processing
> there. But then I remember the r.series command. Is ti possible to add
> this type o analysis in the method list? Is it correct (mathematically
> I mean) to apply the fourier analysis  in a raster time series? Has
> anyone done this?

The main issue is that all of the existing r.series methods take a
series of input maps and produce a single output map.

An FFT would produce a series (or two series - real/imag or arg/abs)
of outputs. If you just want a single frequency component, you would
need to pass the frequency (or period) as an argument, and r.series
doesn't support that yet.

[Such a feature would also be useful for quantiles, rather than being
limited to median (50%), quart1 (25%), quart3 (75%) and perc90 (90%).]

A secondary issue is that GRASS' FFT function is optional, and is only
available if built with the --with-fftw option.

So, it's mostly a question of whether to add Fourier support to the
existing r.series module, or to add a separate r.series.fft module
(which could be based upon r.series).

For a separate module, most of the effort required is in
specification; if it's known exactly how the module is meant to
behave, implementation would be around an hour's work.

For inclusion into the existing r.series, the main issue is resisting
the temptation to add a bunch of other enhancements at the same time
(e.g. being able to generate multiple outputs in one run, arbitrary
quantiles, etc).

-- 
Glynn Clements <glynn at gclements.plus.com>