New AI Weather Model Outperforms Top Global Forecasting Systems

Google DeepMind has unveiled GenCast, an artificial intelligence (AI) model that surpasses the world’s leading weather forecasting system in speed and accuracy, according to a paper  published in Nature on Wednesday.

Unlike conventional models, GenCast can deliver forecasts up to 15 days in advance within minutes, a dramatic improvement over the hours required by existing systems, notes Nature in a press release.

The model’s capabilities show GenCast outpacing the European Centre for Medium-Range Weather Forecasts’ ensemble model (ENS) in predicting extreme events like hurricanes and heatwaves, as noted in Nature.

Researchers believe this breakthrough signals a new era of quicker, more reliable AI-driven weather predictions. Nature explains that GenCast relies solely on historical weather data, allowing it to uncover complex patterns between variables such as temperature, humidity, and wind.

Nature reports that this data-driven approach contrasts with traditional systems, which simulate atmospheric physics using supercomputers. The model generates ensemble forecasts—a set of predictions from slightly varied starting conditions—which provide not only expected outcomes but also the likelihood of their accuracy.

This probabilistic method proved superior in nearly every metric when GenCast was tested against ENS using weather data from 2019. It also excelled in forecasting extreme conditions, such as tropical cyclone tracks and severe temperature swings, as reported by Nature.

GenCast’s speed is another advantage, producing a complete 15-day forecast in just eight minutes using an AI processing chip. Traditional models take hours. This efficiency could prove vital for early warnings in emergencies, such as hurricanes, noted Nature.

DeepMind has also released the code and non-commercial parameters of GenCast, aiming to democratize weather forecasting research.

“This is a really great contribution to open science,” says Matthew Chantry, a machine-learning coordinator at the European Centre For Medium-Range Weather Forecasts in Reading, UK, as reported by Nature.

“We need to understand how these models perform in the most extreme weather events”, and publishing the model and data publicly will enable the research community to evaluate them, he added.

Dr. Kerry Emanuel, a professor emeritus of atmospheric science at MIT who was not involved in the DeepMind research, noted that GenCast is likely to complement existing methods rather than replace them, as reported by The New York Times.

Each type, he said, has its own strengths and weaknesses in predicting the riot of variable phenomena that constitute the weather. “The status quo isn’t going to disappear,” Dr. Emanuel said. “Perhaps the two of them working together will prove to be the best way forward,” reported The Times

As AI models like GenCast advance, they promise faster, more accurate weather forecasts, empowering communities to make better-informed decisions in the face of extreme weather.