AI-supercharged Climate Extreme Event Risk Assessment

The social and economic costs of climate change are dominated by extreme weather events such as heat waves, droughts, and hurricanes. Mitigating and adapting to these events requires quantifying their frequencies, durations, and physical behavior in a changing climate. However, these statistics are impossible to estimate accurately with physics-based simulations due to computational limitations. Artificial intelligence (AI) weather models are much cheaper, but excel at short weather forecasts, which cannot alone provide climate statistics. In this project, the researchers will deploy newly developed mathematical tools to combine the strengths of AI and physics-based models and determine accurate climate statistics of rare events. The developed framework will be a scalable and generalizable methodology for assessing any extreme weather event risk under any climate change scenario. The project promises to deliver essential climate inputs, including uncertainty quantification, into models of the socio-economic impact of climate change.

“The most daunting hazards of a changing climate are the extreme events, like heat waves and rainstorms, that occur so seldom as to catch us unprepared, occurring with very small probabilities that are impossible to estimate from a limited historical record and prohibitively expensive to estimate from numerical simulation—that is, until cheap machine-learned weather models came online. For the first time we can simulate large samples of extreme events at low cost, but there is a crucial missing link that this project will address head-on: AI models can generate physically unrealistic but convincing scenarios, a dangerous combination. We will exploit AI weather models to explore many scenarios rapidly, but ultimately verify them with more trustworthy physics-based models, all within a rigorous mathematical framework tailored to describe, and help interpret, complex extreme event dynamics for urgently needed efficient, customizable, and physically grounded risk assessment for extreme weather.”

Justin Finkel, Postdoctoral Scholar, Department of Earth, Atmospheric, and Planetary Sciences, MIT