COMPUTATIONAL SOLUTIONS FOR OPTIMIZING SPATIOTEMPORAL MODEL PERFORMANCE AND EFFICIENCY OF ORGANIC PARTICLES IN THE ATMOSPHERE

Air pollution and global radiative forcing are influenced by natural (biogenic) and human-generated (anthropogenic) emissions into the atmosphere. Consequently, there is strong motivation to model the complex transport and chemistry of atmospheric emissions. Chemical transport, air quality and climate models are used by stakeholders to assess future changes in the environment and atmosphere. In the atmosphere, particles suspended in air (also known as aerosols) exist in a variety of mixing states and morphologies. However, aerosols are typically assumed to be externally mixed or homogeneously mixed in air quality and climate models due to the computational burden of complex morphology parameterizations. The overall goal of this dissertation was to improve physical accuracy of organic aerosols in air quality models without sacrificing computational efficiency. In Chapter 2, chamber experiments measuring the formation of isoprene epoxydiol (IEPOX) secondary organic aerosols (SOA) were simulated in a box model with aerosol parameterizations from the Community Multiscale Air Quality (CMAQ) model. This work identified the critical parameters in the formation of secondary organic aerosols (SOA) from isoprene (IEPOX-SOA) with core-shell considerations to be organic diffusivity, phase separation, reactive uptake equation, and hygroscopicity. An optimized model (normalized mean bias = 0.077) that explicitly accounts for core-shell morphology, aerosol water growth, and surface area was developed for future regional-scale modeling implementations. In Chapter 3, dark brown carbon (D-BrC) from wildfires was modeled with the Weather Research and Forecasting modeling system coupled with Chemistry (WRF-Chem) and an offline radiation model was used to assess aerosol mixing influences on AAOD. Core-shell D-BrC increased absorbing aerosol optical depth by up to 5.2% and core-shell mixing enhanced absorption effects from D-BrC by up to 11.9%, highlighting the importance of its inclusion in models. The first two projects highlighted the importance of having accurate organic aerosol physiochemical and microphysical properties. However, explicit modeling is computationally intense. In Chapter 4, a physics-informed atmospheric chemistry modular neural network (PACMANN) modeling framework was developed and implemented on a case study for IEPOX-SOA with core-shell considerations. The physics-informed neural network was capable of emulating CMAQ core-shell parameterization for IEPOX-SOA with an average normalized mean bias ≤ 20% and significantly improved computational efficiency.