Characterizing Human Explanation Strategies to Inform the Design of Explainable AI for Building Damage Assessment