A recent article in the journal Nature described the new Google-backed artificial intelligence tool GenCast, which can significantly outperform conventional weather forecasting — with implications for saving money and lives by improving preparations for extreme weather.
Announced in early December, GenCast is a machine-learning model that boasts “better forecasts of both day-to-day weather and extreme events” than existing systems.
The technology, developed by the Google-owned company DeepMind, based in London, improves forecasting accuracy for up to 15 days into the future — a time frame that traditional models have found problematic.
“It’s a big deal,” Kerry Emanuel, a Massachusetts Institute of Technology professor emeritus, told The New York Times.
The newspaper explained that Emanuel and colleagues argued in 2019 that extending the time for reliable forecasts from 10 days to 15 days could help people avoid the worst of extreme weather and reap “enormous socioeconomic benefits.”
As the Times detailed, the “chaotic nature of Earth’s atmosphere” has historically put limits on how far into the future forecasters can peer accurately. GenCast achieved higher accuracy in tests using historical data against a current standard-bearer, the European Centre for Medium-Range Weather Forecasts.
According to the Times, the new model also “ran circles” around its DeepMind predecessor, released in late 2023. “It’s like we’ve made decades worth of improvements in one year,” Rémi Lam, who worked on both models, said.
Another improvement: GenCast is a “probabilistic” model rather than a “deterministic” one — meaning that it gives probabilities for outcomes rather than just yes/no predictions. This is the difference between a forecast with a percent chance of precipitation and a forecast that simply gives a best guess of whether it will rain at a particular time and place.
GenCast can also use smaller computers and produce results faster than conventional forecasting methods. Yet it still relies on traditional meteorology. The new tool uses AI’s ability to pick out patterns from data and create something new, but it’s anchored by 40 years of real-world data and current-day measurements, the Times reported.
Lam pointed out how this makes GenCast’s results less susceptible to some fallacies of other AI products that train on the internet’s chaos of facts and fictions. “We have a ground truth,” Lam told the Times.
Among other concerns, AI has been criticized for its massive energy needs. This is particularly a problem if the energy comes from sources that release heat-trapping gases that in turn cause planetary warming and amplify extreme weather.
However, DeepMind has argued that its application of the technology will help people deal with the consequences of changing weather patterns.
“Weather impacts all of us,” GenCast’s announcement stated. “As climate change drives more extreme weather events, accurate and trustworthy forecasts are more essential than ever.”
“… Consider tropical cyclones, also known as hurricanes and typhoons. Getting better and more advanced warnings of where they’ll strike land is invaluable.”
The DeepMind researchers highlighted how GenCast could also help the field of clean energy, such as by better predicting the reliability of wind power.
With DeepMind opening its models to the public, the world may not need to wait long for it to have influence.
Forecasters are interested in using the model. Matthew Chantry, a machine learning coordinator at ECMWF, told the Times that his center was already incorporating aspects of the tech.
“We are eager to engage with the wider weather community,” two of DeepMind’s lead researchers wrote in their blog. They acknowledged that such collaborations “are critical to our mission to apply our models to benefit humanity.”
Join our free newsletter for weekly updates on the latest innovations improving our lives and shaping our future, and don’t miss this cool list of easy ways to help yourself while helping the planet.
