Exploring the intersection of AI and climate physics: The role of machine learning in advancing climate science

Big data and the rapid increase in computing power are transforming climate science, where machine learning is playing a key role in mapping the physics of our changing climate.

“What’s happening within the field is revolutionary,” says Annalisa Bracco, associate chair and professor in the School of Earth and Atmospheric Sciences. He said that’s because many climate-related processes – from ocean currents to the melting of glaciers and weather patterns – can be described with physical equations. , these advances have the potential to help us understand and predict climate in critically important ways.

Bracco is the lead author of a new review paper that provides a comprehensive perspective on the intersection of AI and climate physics.

An international collaboration between Georgia Tech’s Bracco, Julien Brazard (Nansen Environment and Remote Sensing Center), Henk A. Dijkstra (Utrecht University), Pedram Hassanzadeh (University of Chicago), Christian Lessig (European Center for Medium-Range Weather Forecasts) Results ), and Claire Monteleone (University of Colorado Boulder), the paper, “Machine Learning for the Physics of Climate,” was recently published In nature review physics,

“One of the goals of our team was to help people think deeper about how climate science and AI intersect,” says Bracco. “Machine learning is allowing us to study the physics of climate in ways that were previously impossible. With increasing amounts of data and observations, we can now investigate climate at scales and resolutions we have never been able to before Were able to do it.”

connecting hidden dots

The team showed that ML is making a difference in three key areas: accounting for missing observational data, building more robust climate models, and enhancing predictions, especially in weather forecasting. However, the research also highlights the limitations of AI – and how researchers can work to fill those gaps.

“Machine learning has been fantastic in allowing us to expand the time and spatial scales for which we have measurements,” says Bracco, explaining that ML can help fill in missing data points – researchers’ sources of reference. To create an even stronger record. However, like cutting holes in a shirt, it works best when the rest of the material remains intact.

“Machine learning can draw conclusions from past situations when observations are abundant, but it can’t yet predict future trends or collect the data we need,” Bracco says. “To keep moving forward, we need scientists who can determine what data we need, collect that data, and solve problems.”

climate modeling, weather forecasting

Machine learning is often used when improving climate models that can simulate changing systems like our atmosphere, oceans, land, biochemistry, and ice. “These models are limited by our computing power, and run on a three-dimensional grid,” Bracco explains. Below grid resolution, researchers need to approximate complex physics with simple equations that computers can quickly solve, a process called “parameterization.”

Machine learning is changing this, she says, offering new ways to improve parameterization. “We can run a model at extremely high resolution for a short period of time, so that we don’t need to parameterize many physical processes – using machine learning to get equations that best fit what’s happening at smaller scales. Guess what,” she explains. “We can then use those equations in a coarse model that we can run for hundreds of years.”

While a complete climate model based solely on machine learning may remain out of reach, the team found that ML is advancing our ability to accurately predict weather systems and certain climate events, such as El Niño.

Previously, weather forecasting was based on knowing initial conditions – such as temperature, humidity and barometric pressure – and running a model based on physics equations to predict what might happen next. Now, machine learning is giving researchers the opportunity to learn from the past.

“We can use information about what happened when similar initial conditions occurred in the past to predict the future without solving the underlying governing equations,” Bracco says. “And all this while using orders-of-magnitude fewer computing resources.”

human relations

Bracco emphasizes that while AI and ML play a vital role in accelerating research, humans are at the core of progress. “I think the personal collaboration that gave rise to this paper is in itself a testament to the importance of human interaction,” she says, recalling that this research was the result of a workshop held at the Kavli Institute for Theoretical Physics. The result was one of the team’s first in-person discussions since the COVID-19 pandemic.

She adds, “Machine learning is a fantastic tool – but it’s not the solution to everything.” “There is also a real need for human researchers collecting high-quality data, and for interdisciplinary collaboration across different fields. I see this as a major challenge, but there is a great opportunity for computer scientists and physicists, mathematicians, biologists and chemists to work together. There’s a big opportunity.” ,