From scores to causality and confidence: operational acceptance of AI weather forecasts

Abstract This paper argues that the operational value of machine learning weather prediction (MLWP) cannot be assessed through verification skill alone. Instead, acceptance depends on support for the core functions of operational meteorology: understanding, translating and communicating. Current MLWP systems show strong performance but lack the interpretability, diagnostic depth and stability required for trust in safety‐critical contexts. The paper highlights the need for closer collaboration between developers and operational users and identifies a key future role for artificial intelligence (AI) in supporting data triage and human decision‐making.