Re: a problem with FFTA

[Ansley Manke <ansley@xxxxxxxxxxxxx>]

Hi everyone,
Just a PS to my note yesterday.  Using monthly data requires
some caution, as the units of "month" on different monthly 
axes do not necessarily mean the same thing.  So one might 
get a "drift" when regridding, especially noticeable with 
long time series, and which might have an impact when doing
time series analysis like an FFT.

An approach that sidesteps this risk is

  DEFINE AXIS/units=days/T=<lo>:<hi>/npoints=`T[gt=sst,t=<lo>:<hi>@ngd]` regt

In the example from the post, this would be: 
  LET ntim = T[gt=sst,T=01-JAN-1960:01-DEC-1997@ngd]
  DEFINE AXIS/units=days/T=01-JAN-1960:01-DEC-1997/NPOINTS=`ntim` regt

Note the units are days, but the length of the time step
is still a "month".  We nowuse npoints equal time steps.

Be careful that the endpoints match points on the original axis.  

Or you could use use index limits and do it this way:
 
  SHOW GRID sst       ! see that the time axis has 456 points
  LET tt = T[gt=sst]
  DEFINE SYMBOL trng "`tt[L=1]`:`tt[L=456]`"
  DEFINE AXIS/UNITS=`tt,RETURN=tunit`/T=($trng)/NPOINTS=`tt[t=($trng)@ngd]` regt


Ansley Manke