If you found two disjoint sections that seemed positive on their own, did you try looping both separately in the same model? Wondering how localized the structures are.