- 1Department of Earth and Environmental Sciences, The University of Manchester, Manchester M13 9PL, UK
- 2Department of Climate, Meteorology and Atmospheric Sciences, University of Illinois Urbana-Champaign, Urbana IL, 61801, USA
- 3Department of Mechanical Science and Engineering, University of Illinois Urbana-Champaign, Urbana IL 61801, USA
The aerosol chemical mixing state refers to the distribution of chemical components within individual aerosol particles, which affects their physical properties and interactions with the environment, such as light absorption, cloud formation, and potential health impacts. Accurate estimations of aerosol chemical mixing states are crucial for assessing both climate and health impacts. While particle-resolved models can track changes in aerosol compositions, they often struggle to capture real-world mixing states due to limitations in input data quality, such as emission inventories used in simulations.
In this study, we developed a deep learning foundation model based on particle-resolved simulations and fine-tuned it with limited observational data. The process-guided fine-tuned model improved R² by 300% compared to a fully data-driven baseline, effectively mitigating the challenges posed by sparse observational data and uncertainties model simulations.
Our approach enables dynamic estimations of aerosol mixing states in real-world environments, offering scalability and continuous learning.
How to cite: Jiang, F., Zheng, Z., Coe, H., Topping, D., Riemer, N., and West, M.: Integrating Simulations and Observations: A Foundation Model for Estimating Aerosol Mixing State Index, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6321, https://doi.org/10.5194/egusphere-egu25-6321, 2025.
