(512y) Modeling Summertime Evaporated Secondary Organic Aerosol Formed in the Aqueous Phase (aqSOA) in the Eastern United States

Aerosol liquid water (ALW) can affect the quantity and chemical composition of organic aerosols; however, the interaction between ALW and ambient organic aerosol compounds is highly uncertain. To characterize the behavior of organic aerosols under conditions of drying, water-soluble organic carbon in the particle phase (WSOC_p) in PM_2.5 was measured during two consecutive summers (2015 and 2016) in Baltimore in the eastern United States. The WSOC_p measurements were alternated through an unperturbed ambient channel (WSOC_p) and through a ‘dried’ channel (WSOC_p,dry) maintained at ~35% relative humidity (RH). Sample drying induced systematic evaporation of the WSOC_p during both summers. The quantity of evaporated WSOC_p was strongly related to RH levels, WSOC_p concentrations, isoprene emissions and NO_x/isoprene ratios. Significant differences in meteorological conditions, isoprene emissions, NO_x/isoprene ratios and overall WSOC_p concentrations were observed between the two summers. This led to major differences in the amount of evaporated WSOC_p during summer 2016 in comparison to that in summer 2015. The aim of this study was to characterize the interaction between ALW and ambient organic aerosol compounds and estimate the evaporated WSOC_P in the atmosphere. Hence, it was imperative to quantify the complex relationship between the dependent variable, i.e. evaporated organic aerosol, and the studied variables (i.e. isoprene emissions, NO_x/isoprene ratios, RH levels, and WSOC_p concentrations). Two estimation methods were conducted to estimate this relationship using the collected data in both summers. The first technique is a probabilistic method based on Gaussian machine learning which treats the dependent and the four independent variables as correlated Gaussian random processes to inversely estimate their correlation function. The second technique relies on advanced deep learning algorithms to propose a deterministic neural network relating evaporated WSOC_p mass with the independent variables. Data collected during summer 2015 is used to estimate the correlation function and the network parameters in first and second methods, respectively. The effectiveness of both techniques is assessed by applying both models to data collected during summer 2016 to forecast the evaporated WSOC_p mass.