Linda and I indeed diverge on this particular question.
A lot will depend, I think, on variables that we have too limited information on. The problem we are facing here is akin to a two-dimensional stepping stone model, where you get isolation-by-distance in two dimensions, and we can consider genetic neighbourhoods (as defined by Wright 1943) as discrete demes of size N, exchanging a fraction m of migrants with adjacent neighbourhoods. This was discussed in great theoretical length by
Maruyama 1971 and frankly it's too complex for me to understand entirely.
As I understand it, it goes more or less like this
Depending on N and m, you can get very low to very high FST across the matrix. If the population-specific FST (HSi/HT for the ith population) is low (<0.2), there is little inbreeding in the local population relative to the total population, and I wouldn't worry about local Ne. What Linda refers to, I think, is local Ne, which reflects the number of breeders of a subpopulation (or neighbourhood or deme). but this local Ne is not the same as the inbreeding Ne.
My assumption may be wrong, but I was under the impression that (at migration-drift equilibrium) the inbreeding Ne of a subpopulation relates to the Ne of the metapopulation like HS to HT
I cannot recollect any study where FST values across continuously distributed populations are larger than 0.2, which means that each time, the local deme has (on average) more than 80% of the gene diversity of the total population.
I even have a hard time finding studies showing FST>0.1 in continuously distributed populations.
The weakness of thinking in terms of FST in relation to gene flow is that it assumes migration-drift equilibrium. But at least it's a framework I understand.
If we are to give guidelines on how to consider substructure in continuously distributed populations, it needs to be firmly embedded in population genetic theory. As far as I know, we have no clear cut framework to say that we should evaluate continuous populations in chunks (how large?), where each chunk should have Ne>500.
If Linda or anyone else has a better explanation, I'd be happy to try to get to the bottom of it.
cheers
Joachim