Hi Brian,
Good question! I had the same one a few years ago and my best answer is the attached paper. The short answer: when in doubt, use the mixed curve. OxCal allows you to mix the curves without specifying how much of each curve. Some papers do assume an even 50/50 mixture (it's an option in Rcarbon) but the truth is we don't know what the mixture was when the sample it no doubt varied over time. For that reason, I prefer this code (suggested by Christopher), which ranges from 100% IntCal to 100% SHCal. This increases probability ranges, but in practice the difference is quite small on most parts of the curve. The major advantage is you avoid making any assumption about which hemisphere the air is coming from or what the degree of mixing was at the time. Usually, we have no good information to even wager a guess.
Curve ("IntCal20","IntCal20.14c");
Curve ("SHCal20","SHCal20.14c");
Mix_Curve("Mixed","IntCal20","SHCal20",U(0,100));
This map recommends where to use each curve. It's based on wind patterns but also precipitation and the global 14C gradient – these are modern data that should roughly correspond to 14C content in the past atmosphere. As far south as Columbia, there is no major difference in atmospheric 14C from Seattle, consistent with the year-round winds that come in from the north Atlantic. As for ITCZ graphics, the SHCal20 paper has one that is much better than the one on Wikipedia.
The map of curve recommendations is a good starting point, but it's best to assess each location individually. For example, there are parts of coastal Peru that get all their rainfall during the austral summer (via the South American monsoon) and none during the winter. This means there is likely notable 14C input from the monsoon, which brings in air mixed from the Northern Hemisphere. So a mixed curve is best, even though it might be in the blue part of the map. When in doubt, go mixed.
Hope this helps
Erik