- Google DeepMind AI can predict storms like Debi faster than normal methods
An AI developed by Google can accurately predict the weather up to ten days in advance, outperforming traditional forecasting methods and even top supercomputers.
The new Google DeepMind model, GraphCast, is trained using 40 years of meteorological data from weather stations, satellite images and radar data and can use machine learning to make predictions in less than a minute.
The tool was used to predict the path of Hurricane Lee in September, three days before traditional methods were able to do so, because it found the hurricane would make landfall in Nova Scotia nine days before it hit.
GraphCast’s forecasts could mean that storms, including Debi, which battered the south coast this week, could be spotted much earlier.
An AI developed by Google can accurately predict the weather up to ten days in advance, outperforming traditional forecasting methods and even the best supercomputers
Pushmeet Kohli, vice president of research at Google DeepMind, said: “Weather forecasting is one of the most challenging problems that humanity has been working on for a long time.
‘GraphCast is really very accurate compared to traditional systems and incredibly fast. It can make predictions in a minute. It’s very exciting.’
Developers said the technology “marks a turning point in weather forecasting.”
A scientific paper published Tuesday in the journal Science concluded that Google’s AI was more accurate than the most advanced forecasting system, the European Medium Range Weather Forecasting Model.
It was able to outperform the EMRWFM on 90 percent of the 1,380 metrics tested, including temperature, barometric pressure, wind speed, humidity and atmosphere levels.
GraphCast’s forecasts could mean that storms including Debi, which are hitting the south coast this week, could be spotted much earlier (waves in Folkestone, Kent).
Machine learning coordinator at ECMWF Matthew Chantry said weather forecasting systems were making “much faster and more impressive progress than we expected even two years ago.”
Remy Lam of Google DeepMind said that AI cannot yet replace the currently used supercomputers and that the new technology will complement the older system.
“We stand on the shoulders of giants to build those models,” he said, according to the BBC.
“AI models are trained based on data and that data is generated by traditional approaches, so we still need the traditional approach of collecting data to train the model,” Mr Lam said.