ISORROPIA-lite is a lean and accelerated version of the widely used ISORROPIA-II (Fountoukis and Nenes, 2007) aerosol thermodynamics model and it focuses on the simulation of the composition of the inorganic fraction of the atmospheric aerosol that is in equilibrium with the gas phase. It assumes that the aerosol exists only in the metastable state at low RH, and the activity coefficients of ionic pairs are always obtained from precalculated look-up tables. It estimates aerosol water associated with each one of the aerosol components. Furthermore, ISORROPIA-lite has an important additional feature compared to ISORROPIA-II, as it considers the effects of SOAW on inorganic aerosol thermodynamics. The resulting increase in the total water mass drives more of the water-soluble gaseous species to the particle phase to satisfy equilibrium. SOAW, WSOA in ISORROPIA-lite, is calculated using the well-established κ-Köhler theory of Petters and Kreidenweis (2007): 1WSOA=ρwρSOACSOAκ1RH-1, where ρw is the density of water, ρSOA the SOA density, CSOA the SOA concentration, κ the SOA hygroscopicity parameter, and RH the relative humidity on a 0–1 scale. More details about the ISORROPIA-lite can be found in Kakavas et al. (2022).

PMCAMx (Karydis et al., 2010; Tsimpidi et al., 2010) is a three-dimensional chemical transport model based on CAMx (Environ, 2006), which simulates horizontal and vertical advection and dispersion, dry and wet deposition, and aqueous, gas, and aerosol chemistry. The mechanism used in this work for gas-phase chemistry simulations is the Carbon Bond 5 chemistry scheme (CB05; Yarwood et al., 2005), which includes 190 reactions of 79 gas species. To describe the aerosol size and composition distribution, 10 size sections (from 40 nm to 40 µm) are used, assuming that all particles in each size bin have the same composition. Therefore, PMCAMx predicts PMx concentrations where x can be, among other choices, 1, 2.5, and 10 µm. Equilibrium is always assumed between the bulk aerosol and gas phases. The partitioning of semi-volatile inorganic species between the gas and particulate phases is simulated by ISORROPIA-lite. Weighting factors based on the effective surface area of each size bin are used to distribute to the various size bins the mass transferred between the two phases in each time step (Pandis et al., 1993). For the simulation of organic aerosols, the volatility basis set (VBS) approach (Donahue et al., 2006) is used. POA is simulated using eight volatility bins (from 10-1 to 106 µg m-3) at 298 K, while for SOA four volatility bins (1, 10, 102, and 103 µg m-3) at 298 K are used (Murphy and Pandis, 2009). An organic aerosol phase is assumed with its components forming a pseudo-ideal solution and being in equilibrium with the gas phase (Strader et al., 1998). The influence of water on the partitioning of the organic components of the particulate matter between the gas and aerosol phases is assumed to be negligible in this version of PMCAMx (Koo et al., 2003). Low-volatility organic compounds (LVOCs) and extremely LVOCs are implicitly included in the lowest volatility bin of this version of the VBS used in PMCAMx. These compounds are always in the particulate phase in these simulations, and therefore the addition of lower bins increases the computational cost without changing the predicted organic aerosol concentration. For the major point sources, the NOx plumes are simulated using the plume-in-grid (PiG) approach (Karamchandani et al., 2011; Zakoura and Pandis, 2019). The variable size resolution model (VSRM) of Fahey and Pandis (2001) is used for the simulation of the aqueous phase. The model is based on the chemical mechanism of Pandis and Seinfeld (1989) with the addition of Ca2+ to the list of particle components as well as H2SO4 in the gas phase (Fahey and Pandis, 2001).

We applied PMCAMx over the continental United States during 2010. The modelling domain includes northern Mexico and southern Canada and covers a 4752×2952 km region (Fig. S1 in the Supplement). The model grid consists of 10 824 cells with horizontal dimensions of 36×36 km. The meteorological inputs were provided by the Weather Research Forecasting model (WRF v3.6.1) using a horizontal resolution of 12×12 km. RH levels above 95 % were rare; therefore, there was no need to screen the few high-RH values outside clouds. The gaseous and primary particle emissions were developed by Xing et al. (2013). More details about the meteorological inputs and the emissions can be found in Skyllakou et al. (2021).

To quantify the SOAW effects on inorganic aerosol thermodynamics, three simulations were performed. The first was a simulation neglecting SOAW and including only inorganic aerosol water. Two additional simulations were performed: one where κ of SOA was assumed to be equal to 0.1 and one with κ=0.2 (Kuang et al., 2020) to examine how SOA hygroscopicity affects total fine aerosol water content and PM levels and composition. Even if greater values of the hygroscopicity parameter (e.g. κ=0.3) are possible (Kuang et al., 2020), these represent rather extreme cases for the simulation of average effects over the US. Therefore, the two simulations used in this study provide a good estimate of the corresponding uncertainty. Previous studies have estimated secondary organic aerosol density values of 1–1.4 g cm-3 (Turpin and Lim, 2001; Kostenidou et al., 2007). An SOA density of 1 g cm-3 was assumed in the simulations. SOA exists mostly in submicrometer particles, so our subsequent study focuses on PM1.

The annual average PM1 water ground-level concentrations neglecting SOAW are shown in Fig. 1. Higher PM1 water concentrations from 8 to 18 µg m-3 are predicted in the north-eastern part of the US due to the higher inorganic PM1 concentrations (Fig. 2) and RH levels in that area. When SOAW is present in the simulations, total PM1 water levels increase everywhere, with greater fractional increases in the south-eastern US (by up to 50 % when κ=0.1 and by up to 100 % when κ=0.2 in Alabama and north-western Mexico) due to higher SOA levels (Fig. S2). In the north-eastern US, lower fractional increases are predicted (10 %–15 % when κ=0.1 and 20 %–30 % when κ=0.2). In general, assuming a κ of SOA equal to 0.2 instead of 0.1 increases the corresponding amount of SOAW by about a factor of 2. Figure 1 shows the distributions of fractional increase change in the annual PM1 water levels at ground level from SOAW. Total PM1 water average concentrations increase from 20 % to 30 % in about 60 % of the modelling domain when κ=0.1. For κ=0.2, the corresponding increase is from 40 % to 60 %.

Predicted SOA levels are higher during summertime (Fig. S2) since the emissions and oxidation rates of volatile organic compounds (VOCs) are higher (Zhang et al., 2013; Freney et al., 2014; Skyllakou et al., 2014; Fountoukis et al., 2016). However, even during wintertime, fresh biomass burning emissions exposed to NO2 and O3 can form significant amounts of SOA in periods with low OH levels (Kodros et al., 2020). Higher total PM1 water concentrations are predicted during winter (Fig. 3) since the RH levels and inorganic fine aerosol concentrations are higher, especially nitrate and chloride, which increasingly partition to the aerosol phase as temperature decreases (Guo et al., 2017). However, PM1 chloride concentrations are low (less than 0.1 µg m-3) with higher concentrations in Kansas because of biomass burning episodes. Higher fractional increases in fine aerosol water levels (up to 5 times) due to SOAW are predicted during summer in the south-eastern part of US, where SOA concentrations are higher. This corresponds to increases in average fine aerosol water concentrations of up to 8 µg m-3.

Ammonium nitrate and ammonium sulfate are the inorganic salts that contribute the most to the total PM1 water levels (Fig. S3). SOAW also contributes significantly to total PM1 water levels, especially in the south-eastern US (about 30 % and 50 % of total PM1 water when κ=0.1 and κ=0.2 respectively), when the mass fraction of SOA in dry PM1 exceeds 30 %.

Higher dry PM1 concentrations are predicted for the eastern part of the US (up to 15 µg m-3) in the base case (Fig. 4). These dry PM1 levels increase slightly by up to 0.6 % and 1.2 % due to SOAW when κ=0.1 and κ=0.2 for SOA is assumed respectively. The greatest annual average fractional increase in total dry PM1 levels is predicted in California (1 % when κ=0.1 and 2 % when κ=0.2). The probability density (Fig. 4) indicates that in about 60 % of the modelling domain, total dry fine aerosol concentrations increase of up to 0.3 % when κ=0.1. For κ=0.2, the corresponding increase is from 0.4 % to 2 %. The areas of the greatest PM1 increase correspond to regions where aerosol pH tends to be relatively high (Pye et al., 2020). In these areas, nitric acid and ammonia can condense and increase aerosol mass because of the increase in water from the SOA. Because of this partitioning change, the predicted gas-phase concentrations of semi-volatile inorganic components decreased on average when SOAW was considered (Fig. S4). SOAW had a negligible absolute impact on the low fine chloride concentrations in this period (Fig. 2). However, in periods during which chloride salts and SOA contribute significantly (e.g. during intense biomass burning periods), fine chloride concentrations could also change (Metzger et al., 2006; Fountoukis et al., 2009; Gunthe et al., 2021).

Skyllakou et al. (2021) found that PMCAMx had a small fractional bias (5 %) and a fractional error (25 %) for the annual average PM2.5 concentrations of 1067 measurements stations in the US. The performance of PMCAMx regarding annual average OA is considered good in these simulations, with a fractional bias of 5 % and a fractional error of 26 % in the 306 stations in the US. For daily average concentrations, the performance is also quite encouraging, with a fractional bias of 15 % and a fractional error of 56 %. Given that the effect of the extension of the model on the total fine PM mass is small (of the order of 1 %), this does not result in any noticeable change in its already very good performance for dry fine PM. Therefore, the major change in the model predictions is in the aerosol water concentrations.

The annual average results indicate that SOAW mainly affects fine aerosol water levels. To better analyze the effects of SOAW, we focus on the temporal evolution of the predicted levels of PM1 components at four sites (Fig. S1) with different characteristics (Table S1). We have chosen one city from the west, one from the south, one from south-east, and one from the north-east. They are all in different environments with different major sources and climatological conditions. The presence of SOAW increased PM1 water concentrations at all sites from 1 % by almost 1 order of magnitude (Fig. 5). However, these fractional increases correspond most of the time to PM1 water concentration increases of a few µg m-3 (Fig. S5) because they occur under low RH levels. During periods of higher RH (80 % to 100 %), the PM1 water levels are predicted to increase by up to 100 µg m-3 (e.g. in Toronto).

Total dry PM1 concentrations during most of the simulated period increase on average by less than 1 % in all sites (Fig. 5) due to SOAW. There are periods, however, with greater fractional increases (by up to 10 %) and even small decreases (by up to 5 %) in total dry fine aerosol levels in the examined sites. The decreases can be explained by SOAW increasing the size of particles and therefore their dry deposition rate (Nenes et al., 2020). Depending on SOA hygroscopicity, increases by up to 1.5 µg m-3 for nitrate and 0.5 µg m-3 for ammonium are predicted (Fig. S5). Fine nitrate increases of 10 % were more frequent in the examined sites; however, greater increases, by up to 200 %, are predicted during the simulated period (Fig. 6). As expected, greater increases can occur more often with greater assumed SOA hygroscopicity.