DeepMind's AI weather forecast model surpasses world-class competitor

By Alimat Aliyeva

Google DeepMind has introduced a new weather forecasting model using artificial intelligence, which outperforms traditional methods by offering higher accuracy in predicting weather up to 15 days in advance and better anticipating extreme weather events, Azernews reports.

The GenCast AI model evaluates the probability of different scenarios to assess trends across a wide range of variables, including wind energy and the movement of tropical cyclones.

"This marks a turning point in the development of AI for weather forecasting, as modern raw forecasts now come from machine learning models," said Ilan Price, a researcher at Google DeepMind. He added that GenCast can be integrated into operational weather forecasting systems, allowing meteorologists to gain better insights into trends and prepare for upcoming weather events.

What sets GenCast apart from previous machine learning models is its use of ensemble forecasts, which represent various possible outcomes—a method commonly used in traditional weather forecasting. The model was trained using a vast database from the European Center for Medium-Range Weather Forecasts (ECMWF), accumulated over four decades.

According to a publication in Nature, the GenCast model outperformed ECMWF’s 15-day forecast for 97.2% of 1,320 variables, including temperature, wind speed, and humidity. This marks a significant leap in accuracy and coverage compared to the GraphCast AI model introduced by Google DeepMind last year, which surpassed ECMWF forecasts for 3-10 days ahead by about 90% of the indicators.

AI-based weather forecasting models like GenCast are significantly faster than traditional methods, which rely on enormous computational power to process data. GenCast can generate a forecast in just eight minutes, while traditional methods take several hours to produce a similar forecast.

The researchers believe that GenCast can still be improved, particularly in its ability to predict the intensity of major storms. Additionally, its data resolution may be enhanced to match recent updates made by ECMWF.

ECMWF has praised the development of GenCast as "an important milestone in weather forecasting." The center also announced that it has incorporated key components of the GenCast approach into a version of its own AI-based forecasting system, which has been available since June and utilizes ensemble forecasts.

This innovation could revolutionize how meteorologists approach weather prediction, offering faster, more accurate, and more reliable data to help societies better prepare for extreme weather events.