Re: EOF analysis on dataset w/ missing values

No details provided. I assume that you mean spatial
interpolation.

The EOF pattern is domain sensitive. If you do
extensive spatial interpolation that will affect
both the patterns and the PCs.

An extreme example: if you have SST would you interpolate over land
and then compute the EOFs. Would you expect the ocean patterns to be
tha same as using ocean only points? No.

==
Common sense must rule.

I'd compute the EOfs with missing data.

James Li wrote:
> Hi folks,
>
> I am using NCL to conduct an EOF analysis on a large dataset
> (time*lat*lon) w/ missing values. I tried interpolating missing values
> or not before the EOF analysis. Both methods worked well but produced
> patterns somewhat distinguishable enough to affect the follow-up
> interpretation. Wonder if anyone has similar experience or can give
> some advice on which method should I use? Thanks a lot.
>
> Best wishes,
> Jim Li
> _______________________________________________
> ncl-talk mailing list
> List instructions, subscriber options, unsubscribe:
> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
_______________________________________________
ncl-talk mailing list
List instructions, subscriber options, unsubscribe:
http://mailman.ucar.edu/mailman/listinfo/ncl-talk
Received on Wed Feb 3 21:11:41 2010