[PAOCQ] Sai Ravela (MIT)

“Physics-Coupled Machine Learning for Climate Resilience: Parsimony, Downscaling, and Serious Games”

While climate resilience must be risk-based, quantifying climate risk requires careful integration of projected hazards, exposure, and vulnerability under climate change. We showcase how generative AI can help map vulnerability and exposure, focusing on southwest Bangladesh, where communities engage in mixed-initiative serious games to make decisions using projected risks from sea-level rise, salinity intrusion, cyclones, and flooding.

We then turn to downscaling, a critical component of modern climate risk assessment, and demonstrate how simplified physics, statistics, and co-active learning can downscale climate models to decision-relevant scales, enabling new approaches to ensemble sampling and stochastic modeling. Yet many models, particularly those based on machine learning, are overparameterized, limiting trust and applicability. By reframing learning as a stochastic process, we show how entropy-based parsimony principles can provide new insights for building right-sized, physics-coupled generative models.

This reframing also leads us to Edward Lorenz, suggesting that learnability is a form of predictability and could serve as a guiding principle for the next generation of physics-coupled machine learning in climate risk science.

PAOC Colloquium —

Interdisciplinary seminar series that brings together the whole PAOC (Program in Atmospheres, Oceans, and Climate) community. Seminar topics include all research concerning the physics, chemistry, and biology of the atmospheres, oceans and climate, as well as talks about societal impacts of climatic processes.

Contact: paoc-colloquium-comm@mit.edu