Hi Laura,
Sorry for my delay. All the strategies I've considered reduce to either the sum of the ICs to be merged scaled by some factor, or are not better than simply doing the sum. So, all and all, I'd say simply add the two spatial maps and carry on with the dual regression. Yes, it will add some noise from one IC on top of another, but when we look at the ranges of the values, these become diluted and shouldn't become an major issue.
Another option is the one I commented in the earlier email to Sheena in this thread: keep the original ICs, and then run a NPC (Non-Parametric Combination) over those that you'd like to combine. Given that these come from spatial ICA, we expect little overlap between the components, such that the Tippett combining function should be the most powerful. There are details on how to run
NPC here.
All the best,
Anderson