Re: Presently available EOF external functions

[Ansley Manke <ansley@xxxxxxxxxxxxx>]

Hi everyone,
Thanks to Billy for filling everyone in on the EOF functions that are
now available with Ferret.   Billy has provided us with code to do the
more standard method, which we am implementing and will be available
shortly - I hope with the next version of Ferret.   We will keep the
existing functionality available, for cases where data has gaps in the
time series.

To reiterate what Billy said, the functions EOF_SPACE, EOF_STAT,
and EOF_TFUNC run and give correct results, but are slow, as those
of you who have worked with them know.   We thought we'd comment
on this since there have been questions to the users group on EOF's and
this alerts others of you to the fact that they exist.   If EOF's are something
you'd like to try with Ferret, I'd suggest that you wait for the next version
if possible, and look for the announcement of the faster functions.

Ansley Manke

Billy Kessler wrote:

> Hello Ferreters -
>
> This seems to be an appropriate time to say a few things about the
> EOF external functions available in V5.n of Ferret. These are
> EOF_SPACE, EOF_STAT, and EOF_TFUNC. See the V5 documentation for
> descriptions of the use of these.
>
> These EOF routines were originally designed for a special purpose:
> to compute EOFs on gappy time series. Gaps in the time series
> introduce special complications which require a much more difficult
> (and CPU-consuming) code to deal with. Briefly, it is possible to
> make estimates of the EOFs and time functions (PCs) that minimize
> the expected square error in a least squares sense. (Ask me for
> details; the algebra is not easy to write down in an e-mail). I
> wrote these routines in fortran about 15 years ago, based on
> unpublished notes by Dudley Chelton.
>
> These EOF routines are now in ferret because someone asked the PMEL
> ferret team to make them into external functions. A more systematic
> approach would have made the more usual (and MUCH faster) type of
> EOF routines external functions first, but the squeaky wheel gets
> the grease, so this is the way it turned out. We are now working
> on making better EOF external functions available.
>
> So while the present routines definitely work, they are not ideal for
> probably the great majority of EOF work that most of us encounter.
> For complete time series, they are far slower than a usual EOF
> routine (roughly nx*nt times slower). Lots of needless computation
> is being done if there are no gaps. On the other hand, if you DO
> have a gappy time series, then these routines are very useful and
> hard to come by otherwise.
>
> Billy K
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> William S. Kessler
> NOAA / Pacific Marine Environmental Laboratory
> 7600 Sand Point Way NE
> Seattle Wa 98115 USA
>
> Tel: 206-526-6221                    o__    ____
> Fax: 206-526-6744                    _,>/'_  -----
> E-mail: kessler@pmel.noaa.gov       (_) \(_) ------
> Home page: http://www.pmel.noaa.gov/~kessler

--
Ansley Manke  Pacific Marine Environmental Laboratory  Seattle WA  (206)526-6246