Zencast, a new AI model from Google DeepMind, is accurate enough to compete with traditional weather forecasting. According to recently published research, it managed to outperform a leading forecasting model when tested on 2019 data.
AI isn't going to replace traditional forecasting any time soon, but it could join the arsenal of tools used to predict weather and warn the public about severe storms. Zencast is one of many AI weather forecasters models are being developed This can result in more accurate forecasts.
Gencast is one of many AI weather forecast models that can provide more accurate forecasts
“Weather affects basically every aspect of our lives… predicting weather is one of the great scientific challenges,” says Ilan Price, a senior research scientist at DeepMind. “Google DeepMind's mission is to advance AI for the benefit of humanity. And I think this is an important way, an important contribution on that front.”
Price and his colleagues tested Gencast against the ENS system, one of the world's top-tier models for forecasting, which is run by the European Center for Medium-Range Weather Forecasts (ECMWFAccording to research, Gencast outperformed ENS 97.2 percent of the time Published in this week's magazine Nature,
Gencast is a machine learning weather prediction model trained on weather data from 1979 to 2018. The model learns to recognize patterns in four decades of historical data and uses it to make predictions about what might happen in the future. This is very different from the way traditional models like ENS work, which still rely on supercomputers to solve complex equations to simulate the physics of the atmosphere. Gencast and ENS both produce collective forecastWhich presents a range of possible scenarios.
For example, when it came to predicting the path of a tropical cyclone, Gencast was able to provide an additional 12 hours of advance warning on average. Gencast was generally better at predicting cyclone tracks, extreme weather and wind power generation up to 15 days in advance.
One caveat is that Gencast tested itself against an older version of ENS, which now operates at a higher resolution. Peer-reviewed research compared GenCast predictions to ENS forecasts for 2019, looking at how close each model was to real-world conditions that year. According to ECMWF machine learning coordinator Matt Chantry, the ENS system has improved significantly since 2019. This makes it difficult to say how well Gencast might perform against ENS today.
Of course, resolution is not the only important factor when it comes to making strong predictions. ENS was already operating at slightly higher resolution than Gencast in 2019, and Gencast still managed to beat it. DeepMind says it conducted a similar study on data from 2020 to 2022 and found similar results, although it has not been peer-reviewed. But it did not have data to compare for 2023, when ENS began running at significantly higher resolution.
Dividing the world into a grid, Zencast works at 0.25 degrees resolution – meaning each square on the grid is a quarter degree of latitude and a quarter degree of longitude. In comparison, ENS used 0.2° resolution in 2019 and is now at 0.1° resolution.
Still, the development of Gencast “marks an important milestone in the evolution of weather forecasting,” Chantry said in an emailed statement. As well as ENS, ECMWF says it is also running its own version machine learning systemChantry says it “takes some inspiration from Gencast.”
Speed is an advantage for Gencast. It can generate a 15-day forecast in just eight minutes using a single Google Cloud TPU v5. Physics-based models like ENS can require several hours to do the same thing. Gencast bypasses all the equations that ENS has to solve, which is why it takes less time and computational power to prepare a forecast.
“Computationally, running traditional forecasts is much more expensive than models like Gencast,” says Price.
That efficiency could ease some concerns about environmental impact. Energy-hungry AI data centerswhich is already Contributed to the increase in Google's greenhouse gas emissions in recent yearsBut it's hard to say how Gencast compares to physics-based models when it comes to sustainability without knowing how much energy is used to train a machine learning model.
Gencast can still improve, including potentially scaling up to higher resolution. Additionally, Gencast makes predictions at 12-hour intervals compared to traditional models that typically do so in shorter intervals. This may make a difference to how these forecasts can be used in the real world (for example, to estimate how much wind energy will be available).
“We're kind of wrapping our heads around, is this good? And why?”
“You want to know what the wind will do throughout the day, not just at 6 a.m. and 6 p.m.,” says Professor Stephen Mullens, an adjunct instructor of meteorology at the University of Florida, who was not involved in the GenCast research.
Although there is growing interest in how AI can be used to improve forecasts, it still needs to prove itself. “People are watching it. I don't think the meteorology community as a whole is bought and sold on this,” Mullens says. “We are trained scientists who think in terms of physics… and because AI is not fundamentally like that, there is still an element that we are wrapping our heads around, is it any good? And why?”
Forecasters can check the gencast themselves; DeepMind released code For its open-source model. Price says he sees GenCast and more advanced AI models being used alongside traditional models in the real world. “Once these models are in the hands of practitioners, it increases trust and confidence,” Price says. “We really want it to have kind of a broader social impact.”