Re: statistics for Taylor Diagrams

This will be looked at.

[1] The NCL looks like it is the uncentered pattern correlation
[2] The ME, looks like it is the centered pattern correlation

Try

http://www.ncl.ucar.edu/Document/Functions/Contributed/pattern_cor.shtml

    wcc_gcm = pattern_cor(...)

On 5/16/13 8:50 AM, FERMEPIN Solange wrote:
> Hello everybody,
>
> I am trying to make some Taylor Diagrams. I was following the example
> taylor_3.ncl while using the statistics defined in taylor_7.ncl.
>
> I was getting ~0 correlation coeficient fot only two variables in my
> timeseries and reasonable values for all the others.
>
> I wrote again the pattern correlation by following "Taylor Diagram
> Primer" document posted in the site.
>
> I re run the script and now I get reasonable values (where before I got
> ~0) and exactly the same values for all the other variabales.
>
> Does anyone have any idea why? Is there any statistical reason that I am
> missing?
>
> I am using ncl version 6.1.2. Below the relevant parts of my code:
>
> Thanks a lot!
> Solange
>
>
> ;;; Statistics part of the script
>
> ratio = new((/ncase, nvar/),float)
> cc = new((/ncase, nvar/),float)
>
> gcm = new((/ntim,nlat,nlon/),float)
> obs = new((/ntim,nlat,nlon/),float)
> do nc=0,ncase-1
> do nv=0,nvar-1
>
> gcm = all_gcm(nc,nv,:,:,:)
> obs = all_obs(nc,nv,:,:,:)
>
> tot_area = sum(aire)
> tot_gcm = sum(aire*gcm)
> tot_obs = sum(aire*obs)
> mean_gcm = tot_gcm/tot_area
> mean_obs = tot_obs/tot_area
>
> wvar_gcm = sum(aire*(gcm-mean_gcm)^2)/tot_area
> wvar_obs = sum(aire*(obs-mean_obs)^2)/tot_area
>
> ;NCL
> wcc_gcm = (sum(aire*gcm*obs)-tot_gcm*tot_obs/tot_area)/ \
>
> ((sum(aire*gcm^2)-tot_gcm^2/tot_area)*(sum(aire*obs^2)-tot_obs^2/tot_area))^0.5
>
>
>
> ;ME
> ;wcc_gcm = sum(aire*(gcm-mean_gcm)*(obs-mean_obs))/tot_area/ \
> ; (wvar_gcm*wvar_obs)^0.5
>
>
> ratio(nc,nv) = (wvar_gcm/wvar_obs)^0.5
> cc(nc,nv) = wcc_gcm
>
> end do
> end do
>
_______________________________________________
ncl-talk mailing list
List instructions, subscriber options, unsubscribe:
http://mailman.ucar.edu/mailman/listinfo/ncl-talk
Received on Fri May 17 07:05:59 2013