AI helps drive breakthroughs in weather and climate forecasting

Stay up to date with free updates

Simply register for the Climate change myFT Digest — straight to your inbox.

Artificial intelligence has helped make breakthroughs in accurate long-term weather and climate predictions, according to research that promises advances in both forecasting and the broader use of machine learning.

A team of scientists found that a Google-led model, NeuralGCM, using a combination of machine learning and existing forecasting tools, successfully harnessed AI from conventional atmospheric physics models to track decades-long climate trends and extreme weather events such as cyclones.

This combination of machine learning with established techniques could provide a template for refining the use of AI in other fields, from materials discovery to engineering design, the researchers suggest. NeuralGCM was much faster than traditional weather and climate forecasting and better than AI-only models at long-term predictions, they said.

“NeuralGCM shows that when we combine AI with physics-based models, we can dramatically improve the accuracy and speed of atmospheric climate simulations,” said Stephan Hoyer, a senior fellow at Google Research and co-author of a paper describing the work published in Nature.

The paper says NeuralGCM was faster, more accurate, and used less computing power when tested against a current forecasting model based on atmospheric physics tools, called X-SHiELD, which is being developed by a branch of the U.S. National Oceanic and Atmospheric Administration.

In one test, NeuralGCM identified nearly the same number of tropical cyclones as conventional extreme weather trackers, and twice as many as X-SHiELD. In another test based on temperature and humidity levels in 2020, the margin of error was between 15 and 50 percent lower.

NeuralGCM’s calculations were able to generate 70,000 simulation days in 24 hours using one of Google’s custom AI tensor processing units, the paper said. In comparison, for similar calculations, X-SHiELD generated only 19 simulation days and required 13,824 computing units to do so.

Google collaborated with the intergovernmental European Centre for Medium-Range Weather Forecasts (ECMWF) to develop NeuralGCM.

The European group made its model public in June, and Google has made the code for NeuralGCM open access. It uses 80 years of ECMWF observational data and reanalysis for machine learning.

Last year, Google’s DeepMind division introduced a purely AI-based weather forecasting model called GraphCast, which outperformed conventional methods for periods up to 10 days ahead.

Established weather forecasting agencies such as the UK’s Met Office also have projects underway to integrate machine learning into their work.

Peter Dueben, head of ECMWF’s Earth system modelling division and co-author of the latest paper, said models based solely on AI were “often viewed with scepticism” by experts because they were not based on mathematical equations devised from physics.

The combination of the physics-based model with the deep learning model “seems to provide the best of both worlds,” he said, adding that the approach is a “big step toward climate modeling with machine learning.”

There was still “more work to be done,” such as making NeuralGCM able to estimate the impact of CO₂ increases on global surface temperatures, Dueben said. Other areas where the model needed to be improved included its ability to simulate unprecedented climates, the paper said.

An expert not involved in the work, Cédric M. John, head of data 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 was still “room for improvement,” the potential for error should be measurable and upgrades possible, he suggested.

“Importantly, this hybrid model is good at capturing a set of predictions, and the practical implication of this is that an estimate of the uncertainty of the prediction can be derived,” says John.

Google has become involved in a growing number of environmental monitoring initiatives. It is providing technological support for a satellite mission to track global warming methane emissions and is working with NASA, the U.S. space agency, to help local governments monitor air quality.