How to systematically develop an effective AI-based bias correction model?