Google’s AI Is So Smart It Can Even Predict the Weather

Predicting the weather is notoriously difficult, but now a team at DeepMind, Google's AI division, believe they might have cracked it with a new weather forecasting AI that they say outperforms the best available systems.

The AI is described in a study published on Tuesday in the journal Science.

"We believe this marks a turning point in weather forecasting," wrote the authors, led by Remi Lam. They said the tool, called GraphCast, could help make "cheap prediction more accurate, more accessible, and suitable for specific applications."

Of particular note is that the AI-powered forecaster did a good job of predicting extreme weather events, including hurricane storm tracks, atmospheric rivers, and extreme heat and cold events — none of which it was specifically trained to do.

GraphCast uses machine learning, and was trained using a total of 36.7 million "parameters" based on historical weather data. It can produce a 10-day weather forecast in under a minute on a single Google machine-learning optimized device — a sharp contrast to traditional weather models that use vast amounts of computing power to produce their predictions.

The DeepMind researchers compared GraphCast with the European Centre for Medium-Range Weather Forecasts' tool known as HRES, which is considered among the best in the world today. They found that GraphCast outperformed HRES on 90% of its forecasts, with measures of accuracy and lead time included.

"Our approach should not be regarded as a replacement for traditional weather forecasting methods, which have been developed for decades, rigorously tested in many real-world contexts, and offer many features we have not yet explored," the DeepMind researchers said.

Instead, the tool shows that the machine learning-powered approach "is able to meet the challenges of real-world forecasting problems and has potential to complement and improve the current best methods."