Centering the input data is not equivalent to removing the linear trend beforehand about xeofs HOT 5 CLOSED

Ou, yes, ok, now it makes all sense 😅
I guess, the mix-up arose with the new version I installed of your package.
I used v0.7.0 until recently, in which the Principle Components were called components.
In the new version, you introduced a new nomenclature for the EOFs (eigenvectors; formerly eofs, now components) and the Principle Components (formerly components, now scores).

Apparently, this led to confusion on my side. Thanks for the exchange and the clarification! 👍🏽

from xeofs.

Yep, you're right ;) I forgot that the names were different before 1.0 - sorry for the confusion!

Closing but feel free to reopen if something's unclear.

from xeofs.

Hi Markus! Thanks for the kind words! The center parameter only removes the mean of the data but not the linear trend. Probably the term centering is used differently in some communities, sorry for the confusion.

There is currently no direct function in xeofs for removing global trends in the data.

I don't know if this is relevant for you, but you could try using Extended EOF /Multi-Channel Singular Spectrum Analysis to remove global non-linear trends in your data though :)

from xeofs.

Hi Niclas,

Thanks for the quick response!
Ok. After looking more closely at the code again, I guess (in case of S-mode EOF analysis, i.e. EOF or scores = spatial pattern, components = temporal evolution) the routine removes the field average for every single time step (sample_dims) and not the long-term mean, right? Maybe this can be formulated more clearly in the documentation.?
Because I assumed you subtract the long-term mean in every grid cell by setting center=True.

from xeofs.

Ahh, perhaps there is a small mix-up here?

Compagnucci & Richman (2007) define for m time steps at n stations (or grid points):

• S-mode treats the time series (m times) at each of the n stations (or grid points) as variables in the analysis; the domain is the geographical area.
• T-mode treats the spatial field, defined by all the n stations (or grid points), at each of the m times as variables; the domain is time.

And further,

S-mode analysis results in an nxn similarity matrix whereas the T-mode will result in an mxm similarity matrix.

So if the scores are the spatial patterns and the components are the temporal evolution, then it is T-mode.

Because I assumed you subtract the long-term mean in every grid cell by setting center=True.

Yes, this is true for S-mode, but not for T-mode

perhaps it's easiest to sum up like this:

from xeofs.