GenCast, a new AI model from Google DeepMind, is accurate enough to rival traditional weather forecasts. It managed to outperform a leading forecasting model when tested on 2019 data, according to a recently published study.
AI won’t replace traditional forecasting anytime soon, but it could add to the arsenal of tools used to predict the weather and warn the public of severe storms. GenCast is one of many AI weather forecasts models under development this could lead to more accurate forecasts.
GenCast is one of several AI weather forecast models that could lead to more accurate forecasts.
“Weather affects basically every aspect of our lives…it’s also one of the great challenges in science, predicting the weather,” says Ilan Price, principal research scientist at DeepMind. “Google DeepMind’s mission is to advance AI for the benefit of humanity. And I think that’s an important path, an important contribution on this front.
Price and colleagues tested GenCast against the ENS system, one of the world’s most successful forecast models, run by the European Center for Medium-Range Weather Forecasts (ECMWF). GenCast outperformed ENS 97.2% of the time, study finds published this week in the journal Nature.
GenCast is a machine learning weather forecasting model trained on weather data from 1979 to 2018. The model learns to recognize patterns from four decades of historical data and uses them to make predictions about what might happen in the future . This is very different from the operation of traditional models like the ENS, which still rely on supercomputers to solve complex equations to simulate the physics of the atmosphere. Both GenCast and ENS produce overall forecastwhich offer a range of possible scenarios.
When it came to predicting the path of a tropical cyclone, for example, GenCast was able to give an average of 12 additional hours of warning. GenCast was generally better at predicting cyclone tracks, severe weather, and wind power generation up to 15 days in advance.
One caveat is that GenCast tested against an older version of ENS, which now runs at a higher resolution. The peer-reviewed research compares GenCast’s predictions to ENS’s 2019 predictions, seeing how close each model came to real-world conditions that year. The ENS system has improved significantly since 2019, according to Matt Chantry, ECMWF machine learning coordinator. So it’s hard to say how well GenCast might perform compared to ENS today.
Certainly, resolution isn’t the only important factor when it comes to making solid predictions. ENS was already working at a slightly higher resolution than GenCast in 2019, and GenCast still managed to beat it. DeepMind claims to have conducted similar studies on data from 2020 to 2022 and found similar results, although these have not been peer-reviewed. But it did not have data to make comparisons for 2023, when the ENS began operating at a significantly higher resolution.
By dividing the world into a grid, GenCast operates with a resolution of 0.25 degrees, meaning that each square on this grid corresponds to a quarter of a degree of latitude by a quarter of a degree of longitude. The ENS, in comparison, used a resolution of 0.2 degrees in 2019 and is now at a resolution of 0.1 degrees.
Still, the development of GenCast “marks an important milestone in the evolution of weather forecasting,” Chantry said in an emailed statement. Alongside ENS, ECMWF says it also runs its own version of a machine learning system. Chantry says he’s “inspired by GenCast.”
Speed is an advantage for GenCast. It can produce a 15-day forecast in just eight minutes using a single Google Cloud TPU v5. Physics-based models like ENS may need several hours to do the same thing. GenCast bypasses all the equations that the ENS must solve, which is why it takes less time and computing power to produce a forecast.
“Computationally, it’s much more expensive to run traditional forecasts compared to a model like Gencast,” Price says.
This efficiency could allay some of the concerns about the environmental impact of Energy-intensive AI data centerswho already have contributed to Google’s increasing greenhouse gas emissions in recent years. But it’s difficult to determine how GenCast compares to physics-based models when it comes to sustainability without knowing how much energy is used to train the machine learning model.
There are still improvements GenCast can make, including possible upscaling to higher resolution. Additionally, GenCast issues forecasts at 12-hour intervals, compared to traditional models which typically do so at shorter intervals. This can make a difference in how these forecasts can be used in the real world (to assess the amount of wind energy available, for example).
“We’re kind of thinking, is this right? And why?
“You would want to know what the wind will be like throughout the day, not just at 6 a.m. and 6 p.m.,” says Stephen Mullens, an assistant professor of meteorology at the University of Florida who was not involved in GenCast research.
Although there is growing interest in how AI can be used to improve forecasting, it has yet to be proven. “People look at him. I don’t think the weather community as a whole is being bought and sold,” Mullens says. “We’re trained scientists who think in terms of physics…and because AI is fundamentally not that, so there’s still one element that we’re kind of thinking about, is that right? And why?
Forecasters can view GenCast for themselves; DeepMind released the code for its open source model. Price says he sees GenCast and more improved AI models used in the real world alongside traditional models. “Once these models are in the hands of practitioners, it builds trust even more,” says Price. “We really want this to have some sort of widespread social impact.” »