comparing model performance for a location seems such an obvious thing but it's basically non existent, I wonder why