The calculated eigenvalues are returned as an attribute named "eval"
The eigenvalue shift (es) is eqn 24 in North.
Let "ev" be the eigenvalues returned by eofunc and "neval"
be the number of computed eigenvalues
ev_shift = ev(0:neval-2)*sqrt(2.0/N)
where N is either the number of time steps or
the number of grid points ... depending upon
if the eofunc performs a transpose before computing
the EOFs or not. The eigenvalue difference is
ev_diff = ev(0:neval-2) - ev(1:)
Then, the "rule of thumb" sampling error is
ev_serr = (ev_shift/ev_diff)*ev
If successive eigenvalues are separated by anything close to
sampling_err then you may have sampling issues.
==============
Good luck
On 8/3/10 7:33 AM, 毛嘉富 wrote:
> Hi,
>
> As you know, we can derive the percentage variance explained by EOF modes from the EOF analysis with NCL.
> But how could we use NCL to get the unit standard deviation of the sampling errors associated with each percentage eigenvalue and tell the well distinguished modes from others (North et al., 1982).
>
> Thanks,
> Jiafu
> _______________________________________________
> 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 Aug 4 09:20:58 2010
This archive was generated by hypermail 2.1.8 : Tue Aug 17 2010 - 15:33:24 MDT