Google DeepMind has announced a breakthrough in tropical cyclone forecasting with its WeatherNext artificial intelligence model, which can provide forecasters with up to an extra day of useful warning time compared with previous forecasting systems.
The AI model has demonstrated state-of-the-art performance in predicting cyclone tracks, intensity and wind structure. According to Google DeepMind, WeatherNext’s three-day forecasts can achieve accuracy comparable to that of earlier systems at the two-day range, representing a significant improvement in the ability to anticipate dangerous storms.
The technology uses artificial intelligence (AI) trained on decades of global weather patterns along with specialised data on extreme tropical cyclones. Instead of producing only one forecast, the system can generate multiple possible scenarios, allowing meteorologists to assess different ways a storm could develop.
The model’s potential was demonstrated during the 2025 Atlantic hurricane season, when WeatherNext helped the US National Hurricane Center forecast Hurricane Melissa’s rapid intensification and landfall in Jamaica. The system predicted that the storm could reach Category 5 strength and make landfall in Jamaica five days in advance, giving authorities additional time to prepare communities and organise emergency responses.
Cyclone forecasting has traditionally faced a major challenge: models that accurately predict a storm’s path have often struggled to forecast its intensity. WeatherNext aims to overcome this limitation by providing accurate predictions of both track and intensity.
Google DeepMind has also made its WeatherNext technology available to the wider research community, with the goal of encouraging further scientific research and improving early-warning systems.