Communities facing severe tropical storms could soon gain precious time to prepare and evacuate. A new artificial intelligence model developed by DeepMind, named WeatherNext, can predict the track and intensity of deadly cyclones three days in advance—matching the accuracy that standard forecasting systems only manage two days out.
Traditional weather forecasts rely on physics-based simulations run on supercomputers, which can take days to calculate complex atmospheric changes. In contrast, WeatherNext generates a 15-day forecast in under a minute using specialized AI hardware. To achieve this speed, the model works on a lower-resolution atmospheric grid and relies on patterns learned from 20 terabytes of climate data and 5,000 historical storms.
Meteorologists describe the rapid arrival of AI in weather forecasting as an extraordinary shift, highlighting how fast and cost-effective these tools have become. However, atmospheric scientists stress that human meteorological expertise remains essential to evaluate whether AI predictions are realistic before issuing public safety warnings.
Some researchers also advise caution before abandoning traditional forecasting methods entirely. Because machine learning models are trained on past data, they might struggle to anticipate unprecedented, extreme weather events driven by climate change. Additionally, weather and climate systems share much of their computer code, leading experts to warn that discarding physics-based weather tools could harm long-term climate research.