I'm trying to source localize a set of spatial filtering weights (which are obtained by methods similar to common spatial patterns). Therefore, the data are not time series. Is there any method to source localize such kind of data?

I found it was not allowed to "import MEG/EEG" with only a single time point. If I save a set of spatial filtering weights in the format of a time series, the results of "compute sources" seemed to depend on the temporal context, but in this case, the set of spatial filtering weights are actually independent from each other. So is it possible to do source localization for these spatial filtering weights?

However, you can only use it to localize sources for the modality the forward model was computed for.
EEG forward model requires EEG data in input (event one single time point).
MEG forward model requires MEG data in input (event one single time point).

I guess your maps of weights are not exactly looking like EEG or MEG topographies, and therefore I'm not sure what you could expect from this process.