From Simulation to Observation: ML-Enabled Innovations in Cloud Microphysics

February 19, 2026

Emily de Jong

Hosted by Jim Hurrell

Abstract

Cloud microphysics—the processes governing droplets, ice, and aerosols at microscale—remains a leading source of uncertainty in atmospheric modeling and climate predictions. A fundamental challenge is the scale mismatch between sparse or coarse observations and the detailed microphysical data needed to advance our understanding of atmospheric processes and validate predictive models. This talk presents two complementary machine learning approaches that are transforming how we tackle this challenge: (1) data-driven reduced-order models that discover compact, physically-consistent equations for hydrometeor population evolution from high-fidelity particle-based simulations, and (2) cutting-edge generative AI methods that transform coarse satellite observations to infer detailed cloud properties like vertical structure. Together, these innovations bridge the observation-model gap and unlock new possibilities for understanding clouds and improving atmospheric predictions across scales.