Enhanced EOF method to better disentangle ecological processes
An extensive literature has focused on EOF and has outlined both its
usefulness and efficiency for dimensionality reduction but also its
limits. Several authors have flagged that extra care needs to be taken
when interpreting statistical structure as being physical or ecological
processes (Monahan et al., 2009). Typically, a single process can be
reflected through several modes of the EOF. In our application, we
decided to restrict the interpretation to the dimensions that best
highlight the signal of reproduction. Still, some reproduction signals
can be evidenced in the other dimensions of the EOF. Other methods (in
general linear transformations of EOF) have been developed to better
disentangle the underlying ecological process (Hannachi et al., 2007).
For instance, rotated EOF allows to obtain more distinct patterns by
relaxing orthogonality constraints. However, the choice of the rotation
criterion and the number of dimensions used for rotation have to be set
arbitrarily which implies to make additional choices. Developing
alternative methods could be of great interest to help disentangling
ecological processes.