AI can complement traditional weather models by offering guidance on tropical cyclones several days before they form

A new study reveals that artificial intelligence (AI) could supplement traditional weather models, particularly for predicting the formation of tropical cyclones several days before they develop. Led by Professor Sharan Majumdar from the University of Miami Rosenstiel School, researchers compared the European Center for Medium-Range Weather Forecasts' (ECMWF) conventional Integrated Forecasting System (IFS) with its newer AI-based system (AIFS).

The study, published in the journal Weather and Forecasting, examined African easterly waves and tropical cyclone development across the Atlantic Ocean from 2020 to 2024. While IFS improved over the period, the AI system showed greater potential in forecasting the development of stronger storms, especially when analyzed 84 to 120 hours before a storm's formation.

However, at shorter lead times of 36 to 48 hours, the AI system generally produced lower probabilities, particularly for weaker systems. The AI system also demonstrated accuracy in predicting the location of developing tropical systems, with smaller average position errors compared to conventional forecasts. The researchers emphasize that AI should complement, not replace, traditional numerical weather prediction models.

As ECMWF integrates AI into operational forecasting, the study suggests that combining AI-generated guidance with physics-based models could provide a more comprehensive picture for assessing the potential for tropical cyclone formation in the Atlantic Ocean.