All of which complexity just means that air travel carbon costs are a slightly more complex noneuclidean metric than just ‘spherical geometry’. But you should

All of which complexity just means that air travel carbon costs are a slightly more complex noneuclidean metric than just ‘spherical geometry’. But you should still be able to define a metric over the graph of airports which is based on costs. And what Fermat points tell us is that even in well behaved metric spaces like the plane there can be cases where minimizing distance means going to a node other than one of the starting point. So in more complex spaces it seems likely that that remains the case. Which was all the op was asking.