Google’s DeepMind division has introduced a new artificial intelligence (AI) agent, GenCast, that can accurately forecast the weather for up to 15 days. The model outperformed the world’s best predictions of deadly storms and everyday weather, making it a significant breakthrough in the field of weather forecasting.
GenCast was trained on a massive archive of weather data from 1979 to 2018 and uses a radically different approach than traditional methods, which rely on room-size supercomputers. The AI agent runs on smaller machines and studies atmospheric patterns to learn subtle dynamics that result in the planet’s weather.
The model’s forecasts are probabilistic, providing a range of percentages for the likelihood of rain in a specific region on a given day. This approach is considered more nuanced and sophisticated than deterministic forecasts, which give no indication of uncertainty.
Comparative testing showed that GenCast can predict hurricane track predictions consistently outdid those of the European center. However, the team acknowledged its heavy reliance on conventional weather readings and noted that it provides little information about hurricane intensity.
GenCast will likely complement current methods rather than replace them, as each type has its own strengths and weaknesses in predicting variable phenomena. The DeepMind team hopes to encourage other weather experts to test their new technology by sharing online their AI agent and underlying computer code.
The breakthrough is part of a broader trend towards public openness in the field of AI research, which is helping “lots and lots of people engage in this revolution.” Google’s willingness to share its AI advance publicly has paved the way for further innovation and collaboration.
Source: https://www.nytimes.com/2024/12/04/science/google-ai-weather-forecast.html