Commonly used models often excel on benchmark datasets but do not generalize across regions or varying wildfire conditions.