Los Angeles, California – Google DeepMind has introduced a new AI model called GenCast, which has shown remarkable accuracy in weather forecasting, rivaling traditional methods. Recent research suggests that GenCast outperformed a leading forecast model when tested on 2019 data. While AI is not expected to replace traditional forecasting soon, it could become a valuable tool in predicting weather patterns and alerting the public about severe storms.
GenCast is among several AI weather forecasting models under development that aim to enhance forecast accuracy. Ilan Price, a senior research scientist at DeepMind, emphasized the significance of advancing AI to benefit humanity through more accurate weather predictions. In a study, Price and his colleagues tested GenCast against the high-tier European Centre for Medium-Range Weather Forecasts (ECMWF) system, with GenCast outperforming ENS 97.2% of the time.
GenCast, a machine learning model, uses weather data from 1979 to 2018 to identify patterns and make future predictions. Unlike traditional models like ENS, which rely on supercomputers to solve complex equations simulating atmospheric physics, GenCast operates by recognizing patterns in historical data. This difference allows GenCast to provide additional advance warning for events like tropical cyclones and predicts tracks and extreme weather up to 15 days in advance.
One significant advantage of GenCast is its speed in producing forecasts. GenCast can generate a 15-day forecast in just eight minutes using a single Google Cloud TPU v5, whereas physics-based models like ENS may require several hours for the same task. Despite this efficiency, there are still areas where GenCast can improve, such as potentially scaling up to higher resolutions and providing predictions at more frequent intervals.
The development of GenCast marks a milestone in weather forecasting evolution, according to ECMWF. The organization is also exploring its version of a machine learning system inspired by GenCast. While there is growing interest in using AI to enhance forecasts, the meteorological community remains cautious. Stephen Mullens, an assistant professor at the University of Florida, notes that scientists are still evaluating the effectiveness of AI in weather forecasting and its compatibility with physics-based models.
Forecasters can access the code for GenCast, an open-source model released by DeepMind. Price envisions a future where improved AI models work in tandem with traditional methods to enhance trust and confidence in weather predictions. As AI technology continues to advance, its integration into forecasting practices aims to have a widespread social impact, benefiting communities worldwide.
