I guess my concern is that spatial data and its interactions are quite distinct from traditional tabular data, or imagery data. I'm less interested in learning

I guess my concern is that spatial data and its interactions are quite distinct from traditional tabular data, or imagery data. I'm less interested in learning how to classify trees and roofs from a satellite image, and more in the interaction of various layers of differing types. That said, I could probably adjust inputs so that there is a tabular training set. Gut feel is that there will still be some big holes in my knowledge by travelling a traditional ML route.