Machine learning for error quantification in simulating the climate impacts of atmospheric aerosols

Nicole Riemer, University of Illinois at Urbana-Champaign

Usage Details

Atmospheric aerosols vary in their chemical composition, resulting in different “aerosol mixing states”, which we define as the distribution of aerosol chemical species among particles in a population. Oversimplified assumptions of aerosol mixing state in atmospheric models can introduce errors in estimations of weather and climate-relevant aerosol properties. A realistic representation of the aerosol mixing state can be achieved in principle with a Particle-resolved Monte Carlo (PartMC) numerical model but at a computational cost that is prohibitive for direct implementation in Earth System Models (ESMs).

This project introduced a method for estimating the spatial and temporal distribution of aerosol mixing state in ESMs using the “aerosol mixing state index” as a mixing state metric, which is a proxy for error introduced by simplified aerosol representations. We developed a data-driven workflow, leveraging machine learning algorithms to emulate detailed PartMC simulations in terms of the aerosol mixing state quantification. The coarse-grained modeling was applied to large-scale EMSs such as Community Earth System Model (CESM) to demonstrate the global distribution of aerosol mixing state indexes.