Seminar Speaker: CliMAS Graduate Student, Wenhan Tang

Managing Microphysical Complexity: Aerosol Mixing State Impacts on Ice and Reduced-Order Modeling via Latent Space Dynamics

Atmospheric aerosols influence clouds and radiation, yet it remains challenging to understand the effects of particle diversity and include them in atmospheric models. One important aspect of this diversity is mixing state: how different chemical components are distributed among particles. My Ph.D. research examines when mixing state matters for cloud ice formation and how its effects can be incorporated into large-scale atmospheric models at lower computational cost.
 
First, we combine mathematical analysis and simulations that track individual particles to quantify how mixing state affects freezing within supercooled cloud droplets. We determine how much the frozen fraction can vary due to mixing state alone, using this range to quantify the uncertainty in freezing predictions when models do not explicitly represent mixing state. Mixing state becomes particularly important for freezing predictions when components differ strongly in freezing ability, effective ice-nucleating material is scarce, and particle sizes vary widely. We also show that, under strong competition for water vapor, mixing with soluble material can promote cloud droplet formation and thereby increase the number of ice crystals in clouds through subsequent freezing.
 
These findings highlight the need to include mixing-state effects in atmospheric models while keeping computational costs manageable. We introduce Aerosol Measure Embedding for Latent Dynamics (AeroMELD), a machine learning framework that summarizes complex aerosol populations using ten variables. This compact representation supports mixing and transport, while a learned neural network model predicts evolution through Brownian coagulation. In the evaluated coagulation benchmark, the framework captures mixing-state effects on cloud droplet formation, freezing, and optical properties while running thousands of times faster than particle-resolved simulations.
 
Overall, this work establishes a multiscale framework for understanding aerosol mixing-state effects on ice formation and explores reduced-order modeling of aerosol evolution in latent space. In the longer term, this framework could help bridge particle-resolved aerosol microphysics and computationally tractable representations for next-generation weather and climate models.

https://illinois.zoom.us/j/87247933909?pwd=Yg9a5afZrUDAodWJBhap3HvxzMVVeo.1

Meeting ID: 872 4793 3909     Pass Code: 202496