Grass/Erdas PCA

[Anantha Prasad]

> From grass-lists-owner at max.cecer.army.mil Wed Apr 21 14:04:17 1993
> Date: Wed, 21 Apr 93 11:54:24 PDT
> From: lobo at Jasper.Stanford.EDU (Agustin Lobo)
> Sender: lists-owner at max.cecer.army.mil
> Reply-To: grassu-list at max.cecer.army.mil
> To: grassu-list at max.cecer.army.mil
> Subject: Re: Grass/Erdas PCA
> Content-Length: 1120
> 
> > From grass-lists-owner at max.cecer.army.mil Wed Apr 21 07:29:25 1993
> > Date: Wed, 21 Apr 1993 08:51:09 -0500
> > From: Anantha Prasad <prasad at ghat.inhs.uiuc.edu>
> > Sender: lists-owner at max.cecer.army.mil
> > Reply-To: grassu-list at max.cecer.army.mil
> > To: grassu-list at max.cecer.army.mil
> > Subject: Grass/Erdas PCA
> > Content-Length: 492
> > 
> > Hi Grass/Erdas Users:
> > Is anybody in this group aware of the fact that the covariance matrix calculated by r.covar (and hence the eigenvalues/vectors and Principal Components calculated by Grass) is quite different from the results obtained by using PRINCE in Erdas? (When it is run on identical sets of maps). What could be the reason for this? Is there some bug? I have tried with different options but still the problem remains. Thanks. Prasad (prasad at ghat.inhs.uiuc.edu  Ph: 217-244-2166)
> > 
> 
> I did communicate to the list and to bug.report a simmilar problem in March. I'm
> looking into this problem. My advice is that you should use m.eigensystem instead of r.pca. I'm not aware of any problem with r.covar, though. I will look into this asap, it's a critic question to me.
> Agus
> 
Hi Agus:
Since you had a similar problem before, I will explain further. The covariance matrix output by r.covar and the covariance matrix that PRINCE calculates are different too. So, the resulting eigenvalues/vectors and the PC's are different!!
(By the way, I use the output covariance matrix of r.covar as input to Splus's  (AT&T's Statistical Analysis Package) function 'eigen' (since I could not compile m.eigensystem due to f77 compilation problems) and then use the resulting eigenvectors and eigenvalues to create a map using r.mapcalc). 
So, you may want to look into r.covar output and PRINCE output too. Pl. let me know if you find out the reason (prasad at ghat.inhs.uiuc.edu  Ph: 217-244-2166)