CAT12 cannot be used reliably for the cerebellum surface computation for the moment.
What Brainstorm does to create this surface is simply compute an isosurface of the cerebellum label of the ASEG atlas from FreeSurfer. This provides a correct shape, but not a proper reconstruction of the surface of the cerebellum, and with no registration to any atlas.
Is there a way to register an atlas like AAL in the 3D cortex view of the cerebellum so I can finally use it for the source localization?
AAL is not available on the surface, it is a volume parcellation in MNI-space that is projected into the subject space using non-linear MNI normalization of the subject T1 volume with SPM or CAT.
There is no solution to obtain an AAL parcellation of the surface.