Google DeepMind’s weather AI can forecast extreme weather faster and more accurately

This year the Earth has been hit by a record number of unpredictable extreme weather events made worse by climate change. Predicting them faster and with greater accuracy could enable us to prepare better for natural disasters and help save lives. A new AI model from Google DeepMind could make that easier. 

In research published in Science today, Google DeepMind’s model, GraphCast, was able to predict weather conditions up to 10 days in advance, more accurately and much faster than the current gold standard. GraphCast outperformed the model from the European Centre for Medium-Range Weather Forecasts (ECMWF) in more than 90% of over 1,300 test areas. And on predictions for Earth’s troposphere—the lowest part of the atmosphere, where most weather happens—GraphCast outperformed the ECMWF’s model on more than 99% of weather variables, such as rain and air temperature 

Crucially, GraphCast can also offer meteorologists accurate warnings, much earlier than standard models, of conditions such as extreme temperatures and the paths of cyclones. In September, GraphCast accurately predicted that Hurricane Lee would make landfall in Nova Scotia nine days in advance, says Rémi Lam, a staff research scientist at Google DeepMind. Traditional weather forecasting models pinpointed the hurricane to Nova Scotia only six days in advance.

“Weather prediction is one of the most challenging problems that humanity has been working on for a long, long time. And if you look at what has happened in the last few years with climate change, this is an incredibly important problem,” says Pushmeet Kohli, the vice president of research at Google DeepMind.  

Traditionally, meteorologists use massive computer simulations to make weather predictions. They are very energy intensive and  time consuming to run, because the simulations take into account many physics-based equations and different weather variables such as temperature, precipitation, pressure, wind, humidity, and cloudiness, one by one. 

GraphCast uses machine learning to do these calculations in under a minute. Instead of using the physics-based equations, it bases its predictions on four decades of historical weather data. GraphCast uses graph neural networks, which map Earth’s surface into more than a million grid points. At each grid point, the model predicts the temperature, wind speed and direction, and mean sea-level pressure, as well as other conditions like humidity. The neural network is then able to find patterns and draw conclusions about what will happen next for each of these data points. 

For the past year, weather forecasting has been going through a revolution as models such as GraphCast, Huawei’s Pangu-Weather and Nvidia’s FourcastNet have made meteorologists rethink the role AI can play in weather forecasting. GraphCast improves on the performance of other competing models, such as Pangu-Weather, and is able to predict more weather variables, says Lam. The ECMWF is already using it.

When Google DeepMind first debuted GraphCast last December, it felt like Christmas, says Peter Dueben, head of Earth system modeling at ECMWF, who was not involved in the research. 

“It showed that these models are so good that we cannot avoid them anymore,” he says. 

GraphCast is a “reckoning moment” for weather prediction because it shows that predictions can be made using historical data, says Aditya Grover, an assistant professor of computer science at UCLA, who developed ClimaX, a foundation model that allows researchers to do different tasks relating to modeling the Earth’s weather and climate. 

DeepMind’s model is “great work and extremely exciting,” says Oliver Fuhrer, the head of the numerical prediction department at MeteoSwiss, the Swiss Federal Office of Meteorology and Climatology. Fuhrer says that other weather agencies, such as the ECMWF and the Swedish Meteorological and Hydrological Institute, have also used the graph neural network architecture proposed by Google DeepMind to build their own models. 

But GraphCast is not perfect. It still lags behind conventional weather forecasting models in some areas, such as precipitation, Dueben says. Meteorologists will still have to use conventional models alongside machine-learning models to offer better predictions. 

Google DeepMind is also making GraphCast open source. This is a good development, says UCLA’s Grover. 

“With climate change on the rise, it’s very important that big organizations, which have had the luxury of so much compute, also think about giving back [to the scientific community],” he says. 

Google DeepMind’s weather AI can forecast extreme weather faster and more accurately

This year the Earth has been hit by a record number of unpredictable extreme weather events made worse by climate change. Predicting them faster and with greater accuracy could enable us to prepare better for natural disasters and help save lives. A new AI model from Google DeepMind could make that easier. 

In research published in Science today, Google DeepMind’s model, GraphCast, was able to predict weather conditions up to 10 days in advance, more accurately and much faster than the current gold standard. GraphCast outperformed the model from the European Centre for Medium-Range Weather Forecasts (ECMWF) in more than 90% of over 1,300 test areas. And on predictions for Earth’s troposphere—the lowest part of the atmosphere, where most weather happens—GraphCast outperformed the ECMWF’s model on more than 99% of weather variables, such as rain and air temperature 

Crucially, GraphCast can also offer meteorologists accurate warnings, much earlier than standard models, of conditions such as extreme temperatures and the paths of cyclones. In September, GraphCast accurately predicted that Hurricane Lee would make landfall in Nova Scotia nine days in advance, says Rémi Lam, a staff research scientist at Google DeepMind. Traditional weather forecasting models pinpointed the hurricane to Nova Scotia only six days in advance.

“Weather prediction is one of the most challenging problems that humanity has been working on for a long, long time. And if you look at what has happened in the last few years with climate change, this is an incredibly important problem,” says Pushmeet Kohli, the vice president of research at Google DeepMind.  

Traditionally, meteorologists use massive computer simulations to make weather predictions. They are very energy intensive and  time consuming to run, because the simulations take into account many physics-based equations and different weather variables such as temperature, precipitation, pressure, wind, humidity, and cloudiness, one by one. 

GraphCast uses machine learning to do these calculations in under a minute. Instead of using the physics-based equations, it bases its predictions on four decades of historical weather data. GraphCast uses graph neural networks, which map Earth’s surface into more than a million grid points. At each grid point, the model predicts the temperature, wind speed and direction, and mean sea-level pressure, as well as other conditions like humidity. The neural network is then able to find patterns and draw conclusions about what will happen next for each of these data points. 

For the past year, weather forecasting has been going through a revolution as models such as GraphCast, Huawei’s Pangu-Weather and Nvidia’s FourcastNet have made meteorologists rethink the role AI can play in weather forecasting. GraphCast improves on the performance of other competing models, such as Pangu-Weather, and is able to predict more weather variables, says Lam. The ECMWF is already using it.

When Google DeepMind first debuted GraphCast last December, it felt like Christmas, says Peter Dueben, head of Earth system modeling at ECMWF, who was not involved in the research. 

“It showed that these models are so good that we cannot avoid them anymore,” he says. 

GraphCast is a “reckoning moment” for weather prediction because it shows that predictions can be made using historical data, says Aditya Grover, an assistant professor of computer science at UCLA, who developed ClimaX, a foundation model that allows researchers to do different tasks relating to modeling the Earth’s weather and climate. 

DeepMind’s model is “great work and extremely exciting,” says Oliver Fuhrer, the head of the numerical prediction department at MeteoSwiss, the Swiss Federal Office of Meteorology and Climatology. Fuhrer says that other weather agencies, such as the ECMWF and the Swedish Meteorological and Hydrological Institute, have also used the graph neural network architecture proposed by Google DeepMind to build their own models. 

But GraphCast is not perfect. It still lags behind conventional weather forecasting models in some areas, such as precipitation, Dueben says. Meteorologists will still have to use conventional models alongside machine-learning models to offer better predictions. 

Google DeepMind is also making GraphCast open source. This is a good development, says UCLA’s Grover. 

“With climate change on the rise, it’s very important that big organizations, which have had the luxury of so much compute, also think about giving back [to the scientific community],” he says. 

Noise-canceling headphones could let you pick and choose the sounds you want to hear

This is a subscriber-only story.

Future noise-canceling headphones could let users opt back in to certain sounds they’d like to hear, such as babies crying, birds tweeting, or alarms ringing.

The technology that makes it possible, called semantic hearing, could pave the way for smarter hearing aids and earphones, allowing the wearer to filter out some sounds while boosting others. 

The system, which is still in prototype, works by connecting off-the-shelf noise-canceling headphones to a smartphone app. The microphones embedded in these headphones, which are used to cancel out noise, are repurposed to also detect the sounds in the world around the wearer. These sounds are then played back to a neural network, which is running on the smartphone; then certain sounds are boosted or suppressed in real time, depending on the user’s preferences. It was developed by researchers from the University of Washington, who presented the research at the ACM Symposium on User Interface Software and Technology (UIST) last week.

The team trained the network on thousands of audio samples from online data sets and sounds collected from various noisy environments. Then they taught it to recognize 20 everyday sounds, such as a thunderstorm, a toilet flushing, or glass breaking.

It was tested on nine participants, who wandered around offices, parks, and streets. The researchers found that their system performed well at muffling and boosting sounds, even in situations it hadn’t been trained for. However, it struggled slightly at separating human speech from background music, especially rap music.

Mimicking human ability

Researchers have long tried to solve the “cocktail party problem”—that is, to get a computer to focus on a single voice in a crowded room, as humans are able to do. This new method represents a significant step forward and demonstrates the technology’s potential, says Marc Delcroix, a senior research scientist at NTT Communication Science Laboratories, Kyoto, who studies speech enhancement and recognition and was not involved in the project. 

“This kind of achievement is very helpful for the field,” he says. “Similar ideas have been around, especially in the field of speech separation, but they are the first to propose a complete real-time binaural target sound extraction system.”

“Noise-canceling headsets today have this capability where you can still play music even when the noise canceling is turned on,” says Shyam Gollakota, an assistant professor at the University of Washington, who worked on the project. “Instead of playing music, we are playing back the actual sounds of interest from the environment, which we extracted from our machine-learning algorithms.”

Gollakota is excited by the technology’s potential for helping people with hearing loss, as hearing aids can be of limited use in noisy environments. “It’s a unique opportunity to create the future of intelligent hearables through enhanced hearing,” he says.

The ability to be more selective about what we can and can’t hear could also benefit people who require focused listening for their job, such as health-care, military, and engineering professionals, or for factory or construction workers who want to protect their hearing while still being able to communicate.

Filtering out the world

This type of system could for the first time give us a degree of control over the sounds that surround us—for better or worse, says Mack Hagood, an associate professor of media and communication at Miami University in Ohio, and author of Hush: Media and Sonic Self-Control, who did not work on the project.

“This is the dream—I’ve seen people fantasizing about this for a long time,” he says. “We’re basically getting to tick a box whether we want to hear those sounds or not, and there could be times where this narrowing of experience is really beneficial—something we really should do that might actually help promote better communication.”

But whenever we opt for control and choice, we’re pushing aside serendipity and happy accidents, he says. “We’re deciding in advance what we do and don’t want to hear,” he adds. “And that doesn’t give us the opportunity to know whether we really would have enjoyed hearing something.”