Re: regCoef using the modes of some variables

From: Adam Phillips <asphilli_at_nyahnyahspammersnyahnyah>
Date: Wed Jul 13 2011 - 14:29:50 MDT

Hi Maria,
I have little experience w/SVD analysis, but for EOF's I frequently do
something like this:
; arr_UW = anomalous array dimensioned lat x lon x time
; arr = arr_UW but sqrt cosine weighted

neof = 3
evecv = eofunc(arr,neof,False)
pcts = eofunc_ts(arr,evecv,False)
pctsS = dim_standardize(pcts,0)
finarr = arr(:neof-1,:,:) ; preallocate array w/metadata
do gg = 0,neof-1
    finarr(gg,:,:) = (/ regCoef(pctsS(gg,:),arr_UW) /)
end do

For SVD's I would think it would be similar, but instead of regressing
the PCs on the original anomalous field regress the left (right)
singular vector(s) against the original anomalous array similar to what
I did above.

I'm not sure if that is what you are after... Perhaps others more
experienced in SVD analysis can chime in here..

On 07/13/2011 11:47 AM, Maria Chirinos wrote:
> Hi NCL-Talk,
> Does anyone know how I can calculate a linear regression using the modes
> of certain variables that have previously calculated with SVD? I need
> something like eof2data but in my case would be mode2data . I think I
> should combine the modes obtained with their respective time series, but
> I'm not sure how.
> Any help is welcome.
> _______________________________________________
> ncl-talk mailing list
> List instructions, subscriber options, unsubscribe:

Adam Phillips                      
NCAR/Climate and Global Dynamics Division       (303) 497-1726
P.O. Box 3000				
Boulder, CO 80307-3000
ncl-talk mailing list
List instructions, subscriber options, unsubscribe:
Received on Wed Jul 13 14:29:55 2011

This archive was generated by hypermail 2.1.8 : Mon Jul 18 2011 - 15:57:58 MDT