Stay up up to now with free updates
Simply register for Climate change myFT Digest – delivered straight to your inbox.
Artificial intelligence has contributed to a breakthrough in accurate long-term weather and climate forecasting, in keeping with research that guarantees advances in each forecasting and the broader use of machine learning.
A team of researchers found that a Google-led model called NeuralGCM successfully combined artificial intelligence with traditional models of atmospheric physics to trace decades-long climate trends and extreme weather events equivalent to hurricanes, using a mix of machine learning and existing forecasting tools.
This combination of machine learning and established techniques could provide a template for refining using AI in other areas, from materials science to engineering design, the researchers say. NeuralGCM is far faster than traditional weather and climate forecasting and higher than pure AI models at longer-term predictions, they said.
“NeuralGCM shows that we will dramatically improve the accuracy and speed of atmospheric climate simulations by combining AI with physics-based models,” said Stephan Hoyer, senior engineer at Google Research and co-author of a paper on the work published in Nature.
The paper states that NeuralGCM has performed well in tests with a current forecast model based on atmospheric physics tools called X-SHIELDwhich is being developed by a division of the U.S. National Oceanic and Atmospheric Administration.
In one test, NeuralGCM identified almost as many tropical cyclones as traditional extreme weather trackers and twice as many as X-SHiELD. In one other test based on temperature and humidity readings in 2020, the error rate was between 15 and 50 percent lower.
NeuralGCM's calculations were in a position to generate 70,000 simulation days in 24 hours using considered one of Google's custom-built AI tensor processing units, the paper says. In contrast, X-SHiELD generated only 19 simulation days for comparable calculations, requiring 13,824 computing units.
Google collaborated with the intergovernmental European Centre for Medium-Range Weather Forecasts (ECMWF) to develop NeuralGCM.
The European company made its model publicly available in June, and Google has published the code for Open access to NeuralGCM. It uses 80 years of ECMWF observational data and reanalyses for machine learning.
Last 12 months, Google's DeepMind division introduced a pure AI-based weather forecasting model called GraphCast that outperformed traditional methods for periods as much as 10 days prematurely.
Even established weather forecasting agencies equivalent to the UK Met Office are conducting projects to integrate machine learning into their work.
Peter Dueben, head of Earth system modelling at ECMWF and co-author of the most recent paper, said that pure AI models are “often viewed with skepticism” by experts because they will not be based on mathematical equations from physics.
Combining the physics-based model with the deep learning model “seems to mix the perfect of each worlds,” he said, adding that the approach is a “big step toward climate modeling with machine learning.”
There continues to be “more work to do,” Dueben said, for instance to enable NeuralGCM to estimate the impact of rising COâ‚‚ on global surface temperatures. Other areas by which the model must be improved include the flexibility to simulate unprecedented climates, the paper said.
An expert not involved within the work, CĂ©dric M. John, head of knowledge science for environment and sustainability at Queen Mary University of London, said there was “compelling evidence” that NeuralGCM was more accurate than machine learning alone and faster than the “full physics” model. While there continues to be “room for improvement,” the probability of error ought to be measurable and upgrades possible, he said.
“Importantly, this hybrid model is in a position to capture a variety of forecasts well. The practical consequence of that is that it may possibly be used to derive an estimate of the uncertainty of the forecast,” said John.
Google is participating in a growing variety of environmental monitoring initiatives. The company is providing technological support for a satellite mission to watch emissions of methane and Partners NASAthe US space agency, is meant to assist local governments monitor air quality.
Climate Capital
Where climate change meets business, markets and politics. Read the FT's coverage here.
Are you interested by FT’s commitment to environmental sustainability? Learn more about our science-based goals here
