How Do We Learn to Live with Extreme Events?

“There’s no question the planet is warming. But the most uncertainty exists about the events that affect us the most—occurring one city, one forest, one continent at a time,” said climate scientist Daniel Swain.

Cyclones, floods, heat waves, tornados, and other kinds of extreme weather emerge from the same processes that make up the climate, explains Swain, who works at the University of California Los Angeles’ Institute of the Environment and Sustainability.

But for any given disaster, how do we pinpoint whether climate change was a culprit? Swain recently co-published a primer on the young field of “extreme-event attribution,” arguing that the media and even other scientists often misunderstand the data.

At the meeting, Swain will give an overview of recent advances in real-world observations and modeling, and explain how society will need to adapt to climate-driven extreme weather.

One of the biggest dilemmas with extreme weather is predicting it: How do we know what to expect, and when?

Scientists at the University of Paris-Saclay seek answers by building a weather dictionary that relies on word search technology.

A machine learning technique called Latent Dirichlet Allocation picks out topics from text. The group applied the strategy to produce totally reimagined weather maps.

“We have exported the popular linguistic technique to the study of the climate to understand the ‘language’ of weather extreme events. What are the recurring topics when the atmosphere speaks, through the wind, to us?” said complex systems researcher Davide Faranda.

The team compared grid points of sea-level pressure to words and successfully identified cyclones and anticyclones known to meteorologists, like the Genoa Low, the Scandinavian High, and the Azores anticyclone.

Breaking down weather into simple motifs makes it clearer to study the effects of climate change. “It provides an easy way to study extreme events such as heat waves and cold spells and identify their precursors,” said Faranda.

Artificial intelligence has revolutionized climate predictability, but many roadblocks remain. University of California San Diego computer scientist Rose Yu has discovered a way to significantly improve AI’s ability to forecast the climate.

The main problem is that while deep learning makes powerful and accurate predictions, these don’t always adhere to the actual laws of physics. Yu and her colleagues have developed workarounds that build physics into an algorithm to model turbulent flows.

“We have solved highly-challenging problems in physical science around climate models and COVID-19 simulations. I demonstrate how to principally integrate physics in AI models and algorithms to achieve both prediction accuracy and physical consistency,” said Yu.

Computational approaches like Yu’s could improve how we predict everything from extreme weather events and climate change to the next pandemic, and even traffic patterns within a city.