[ncl-talk] Questions about argument setting in eofunc

[Dennis Shea]

What do you mean by "normalizing" ?

[a] transforming to a normal (Gaussian) distribution.
[b] transforming to unit variance
[c] ???
===
Using the raw data:

[1] Using the covariance matrix emphasizes variance.
[2] Using the correlation matrix emphasizes patterns.

Correlations are 'normalized' by the product of the standard deviations.

   cor = SUM [(X(t)-Xave)*(Y(t)-Yave)}]/(Xstd*Ystd)

====

I'd suggest you experiment.

Incidentally, as noted in the 'eofunc' documentation,
the input data array need not be anomalies. NCL's function
does that 'under-the-hood'

Good luck



On Thu, Dec 8, 2016 at 9:28 PM, <Ziguang.Li at csiro.au> wrote:

> Hi there,
>
> I have some uncertainty about using eofunc to do empirical orthogonal
> function analysis (EOF). Following the online instruction, I should set
> attribution of optEOF, jopt, to assign correlation matrix or covariance
> matrix for calculating EOFs. I want to use normalized data as input, should
> I need to pre-process raw data as normalized values then set optEOF at jpopt=1
> before perform eofunc? Or just use anoamlies as input and set optEOF at jopt=1
> is enough and this function will automatically normalize input data?
>
> Looking forward to your reply.
>
> Ziguang
>
>
>
> _______________________________________________
> ncl-talk mailing list
> ncl-talk at ucar.edu
> List instructions, subscriber options, unsubscribe:
> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/ncl-talk/attachments/20161208/b946e73f/attachment.html