Stay tuned for free updates
Just register at Artificial intelligence myFT Digest – delivered directly to your inbox.
Google DeepMind has unveiled an artificial intelligence weather forecasting model that outperforms traditional methods in forecasting up to 15 days ahead and is better at predicting extreme events.
The tool, known as GenCast, measures the likelihood of multiple scenarios to accurately assess trends from wind energy production to tropical cyclone movement.
GenCast’s probabilistic technique is a new milestone in the rapid progress in using artificial intelligence to provide better and faster daily weather forecasts, an approach increasingly used by traditional forecasters.
“(This) represents something of an inflection point in the development of artificial intelligence for weather forecasting, with today’s raw forecasts now coming from machine learning models,” said Ilan Price, a Google DeepMind scientist.
“GenCast can be incorporated into operational weather forecasting systems, offering valuable information to help decision makers better understand and prepare for future weather events.”
What’s new about GenCast compared to previous machine learning models is the use of so-called “ensemble” forecasts that represent different outcomes, a technique used in traditional forecasting today. GenCast is trained on four decades of data from the European Center for Medium-Range Weather Forecasts (ECMWF).
The model outperformed ECMWF’s 15-day forecast by 97.2% of 1,320 variables, such as temperature, wind speed and humidity, the paper said. published in Nature on Wednesday.
The result is a further increase in the accuracy and scale of Google’s DeepMind breakthrough GraphCast model introduced last year. GraphCast outperformed ECMWF forecasts in about 90 percent of the metrics for forecasts three to ten days ahead.
Artificial intelligence forecasting models are typically faster and potentially more efficient than standard forecasting methods that rely on heavy computing power to handle equations derived from atmospheric physics. GenCast can create its forecast in just eight minutes compared to hours for a traditional forecast – and with a fraction of the electronic processing required.
The GenCast model could be further improved in areas such as its ability to predict the intensity of large storms, the researchers said. The resolution of its data may be increased according to the updates made this year by ECMWF.
ECMWF said the development of GenCast was a “significant milestone in the development of weather forecasting”. It said it had integrated “key components” of the GenCast approach into a version of its own artificial intelligence forecasting system, with live ensemble forecasts available from June.
The innovative machine learning science behind GenCast still needs to be tested in extreme weather events, ECMWF added.
The development of GenCast will further heat up the debate over how widely AI should be deployed in forecasting, with many scientists favoring a hybrid technique for some purposes.
In July, Google presented The NeuralGCM modelwhich combines machine learning and traditional physics to achieve better results than AI alone for long-term forecasting and climate trends.
“There are open questions and debate about the optimal balance between physics and machine learning predictive systems. The wider scientific community, including (us), is actively investigating this,” ECMWF said.
The Met Office, the UK’s national weather service, is exploring how to use the “exciting” developments in its own AI-powered forecasting models, said Stephen Ramsdale, chief forecaster in charge of AI.
“We support the greatest value from a hybrid approach that combines human assessment, traditional physics-based models and AI-based weather forecasting,” he added.
Climate capital
Where climate change meets business, markets and politics. Check out the FT’s coverage here.
Curious about FT’s commitment to environmental sustainability? Learn more about our science-based goals here
