The forecast is looking good for this new weather program.
Google DeepMind, the search engine’s AI-focused lab, has introduced a game-changing weather forecasting model that is said to deliver 10-day predictions in less than a minute — “at unprecedented accuracy.”
GraphCast outperforms traditional and high-tech weather models by at least 90% as it emphasizes historic weather data, according to a research paper on the AI model released Tuesday.
Google touts more than a million global grid points for hyperlocal data. GraphCast requires two pieces of information — the state of the weather now and six hours earlier as it predicts the weather six hours in the future and up to 10 days ahead.
“We believe this marks a turning point in weather forecasting,” the Science paper reads, “which helps open new avenues to strengthen the breadth of weather-dependent decision-making by individuals and industries, by making cheap prediction more accurate, more accessible, and suitable for specific applications.”
The model was tested against the European Centre for Medium-Range Weather Forecasts’ gold-standard system, called High RESolution forecast (HRES).
GraphCast was found to be 99.7% more spot-on in some cases.
The ECMRWF is already experimenting with the tool, thanks to open-source coding that anyone can use.
The system is also trumpeted by Google to be groundbreaking in detecting extreme weather events ahead of their arrival.
During training, it was able to more accurately predict cyclone movement than HRES. In September, it also predicted Hurricane Lee would reach Nova Scotia nine days before landfall.
Traditional models took three extra days, Google said.
Additionally, GraphCast excels at predicting flooding through atmospheric data on rain patterns.
What makes this interface different from traditional approaches is that it doesn’t run algorithms and mega equations like that of supercomputers. Instead, deep learning is utilized.
“Deep learning offers a different approach: using data instead of physical equations to create a weather forecast system,” wrote GraphCast team member Remi Lam. “GraphCast is trained on decades of historical weather data to learn a model of the cause and effect relationships that govern how Earth’s weather evolves, from the present into the future.”
Meanwhile, Google recently released a new, separate 24-hour forecast model called MetNet-3 that delivers incredible accuracy as well.
“Pioneering the use of AI in weather forecasting will benefit billions of people in their everyday lives,” Lam added. “But our wider research is not just about anticipating weather — it’s about understanding the broader patterns of our climate.”
AI use in weather forecasting is something experts have referred to as a “quiet revolution” that is substantially changing the industry, Science.org reported.
“It’s very, very exciting to know we can generate global predictions that are skillful, really cheaply,” Maria Molina, an AI-centered research meteorologist at the University of Maryland, told the outlet Tuesday.
And, this is only the beginning, says Christopher Bretherton, an atmospheric scientist for the Allen Institute for AI.
The institute’s 40-year training data could become instrumental in catalyzing new models that future AI would be trained on, Bretherton says, “and then run them 100 times faster.”