This 12 months the Earth has been hit by a record variety of unpredictable extreme weather events made worse by climate change. Predicting them faster and with greater accuracy could enable us to organize higher for natural disasters and help save lives. A brand new AI model from Google DeepMind could make that easier. 

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

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

“Weather prediction is probably the most difficult problems that humanity has been working on for an extended, very long time. And in the event you take a look at what has happened in the previous couple of years with climate change, that is an incredibly essential problem,” says Pushmeet Kohli, the vp of research at Google DeepMind.  

Traditionally, meteorologists use massive computer simulations to make weather predictions. They’re very energy intensive and  time consuming to run, since the simulations have in mind many physics-based equations and different weather variables resembling temperature, precipitation, pressure, wind, humidity, and cloudiness, one after the other. 

GraphCast uses machine learning to do these calculations in under a minute. As a substitute of using the physics-based equations, it bases its predictions on 4 a long time of historical weather data. GraphCast uses graph neural networks, which map Earth’s surface into greater than 1,000,000 grid points. At each grid point, the model predicts the temperature, wind speed and direction, and mean sea-level pressure, in addition to other conditions like humidity. The neural network is then capable of find patterns and draw conclusions about what is going to occur next for every of those data points. 

For the past 12 months, weather forecasting has been going through a revolution as models resembling 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, resembling Pangu-Weather, and is capable of 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 within 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 since it shows that predictions will be made using historical data, says Aditya Grover, an assistant professor of computer science at UCLA, who developed ClimaX, a foundation model that enables researchers to do different tasks regarding modeling the Earth’s weather and climate. 

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

But GraphCast shouldn’t be perfect. It still lags behind conventional weather forecasting models in some areas, resembling precipitation, Dueben says. Meteorologists will still need to use conventional models alongside machine-learning models to supply higher predictions. 

Google DeepMind can be making GraphCast open source. That is a great development, says UCLA’s Grover. 

“With climate change on the rise, it’s extremely essential that big organizations, which have had the luxurious of a lot compute, also take into consideration giving back [to the scientific community],” he says.