Re: EOF explained variance

Dear Roger,
I have an NCEP derived product. I calculated the wind stress
curl of a region in South Atlantic and now I want to make an
EOF analysis of this field. I have a field with 9 latitude
points and 14 longitude points totalizing 126 points, and I
have the evolution of this field in 8036 times.

I think there are two ways to calculate EOF for a grid of
time series of observations. One method is obtain a
symmetric covariance matrix which is them decomposed into
eigenvalues and eigenvectors. And other way is using an
singular value decomposition (SVD) method which derives all
the components of the EOF analysis without calculate of the
covariance matrix.

First I try SVD but "out of memory" appears...

Because of this, I try to calculate the EOF "step by step"
with the covariance matrix, but isn't possible. The "out of
memory" appeared again.

Them I don't know what I can do to solve the problem??

Thanks

Leandro



"Roger Stafford" <ellieandrogerxyzzy@xxxxxxxxxxxxxxxxxxxxxx>
wrote in message <g5g92l$sp9$1@xxxxxxxxxxxxxxxxxx>...
"Leandro Calil" <leandro@xxxxxxxxxxxxxxxx> wrote in
message <g5g4i9$5rc
$1@xxxxxxxxxxxxxxxxxx>...
Dear Roger,
First, thanks for your help !!
I intend make a EOF analysis from a grid space that have 126
points and 8036 times.
I think the right matrix organization for calculate the
covariance matrix from my data is X(8036,126). Is that
right??
If it is right, I need to calculate cov(X). It is not this?
I need the total data variance. I hope reach this with the
sum of principal diagonal of my covariance matrix.
Thank you again...
Leandro

It isn't clear to me what you mean by "a grid space that
have 126 points and
8036 times". You can clear things up by telling us what
size you expect your
covariance matrix to have. Would it be 8036 by 8036 or
would it be 126 by
126? I am guessing it would be 126 by 126.

I don't believe you need to calculate the covariance
matrix to do your
empirical orthogonal function analysis (also known as
principal component
analysis.) Just use the 'svd' (singular value
decomposition) function directly
on your data matrix with each row being an observation,
(no doubt invoking
the "economy size" option to avoid a 8036 x 8036 size
matrix.) This will
automatically give you the orthogonal bases in descending
sizes of singular
values.

Roger Stafford


.

Relevant Pages

  • Re: EOF explained variance
    ... I have an NCEP derived product. ... EOF analysis of this field. ... symmetric covariance matrix which is them decomposed into ... The curl is a three-dimensional vector quantity. ...
    (comp.soft-sys.matlab)
  • Re: EOF explained variance
    ... I intend make a EOF analysis from a grid space that have 126 ... sum of principal diagonal of my covariance matrix. ... I am a intermediate user of matlab and I need your help... ... matlab shown "out of memory". ...
    (comp.soft-sys.matlab)