Differences in Fine Particle Chemical Composition on Clear and Cloudy Days

Clouds are prevalent and alter fine particulate matter (PM2.5) mass and chemical composition. Cloud-affected satellite retrievals are subject to higher uncertainty and are often removed from data products, hindering quantitative estimates of tropospheric chemical composition during cloudy times. We examine surface PM2.5 chemical constituent 10 concentrations in the Interagency Monitoring of PROtected Visual Environments (IMPROVE) network in the United States during Cloudy and Clear Sky times defined using Moderate Resolution Imaging Spectroradiometer (MODIS) cloud flags from 2010-2014 with a focus on differences in particle hygroscopicity and aerosol liquid water (ALW). Cloudy and Clear Sky periods exhibit significant differences in PM2.5 mass and chemical composition that vary regionally and seasonally. In the eastern US, relative humidity alone cannot explain differences in ALW, suggesting emissions and in situ chemistry exert 15 determining impacts. An implicit clear sky bias may hinder efforts to quantitatively understand and improve representation of aerosol-cloud interactions, which remain dominant uncertainties in models.

Surface networks record PM 2.5 mass and chemical composition during clear sky and cloudy time periods alike. The difference between spatially and temporally aggregated PM 2.5 mass concentrations in the CONUS for cloudy and all sky 65 (cloudy + clear sky) conditions is estimated to be ±2.5 μg m -3 (Christopher and Gupta, 2010). Less attention has been given to clear sky and cloudy differences in PM 2.5 chemical composition, especially with regards to particle hygroscopicity and water uptake. Aerosol mass concentrations and chemical speciation including aerosol liquid water (ALW) influence AOT (Christiansen et al., 2019;Malm et al., 1994;Nguyen et al., 2016;Pitchford et al., 2007), cloud microphysics, and mesoscale convective systems (Kawecki and Steiner, 2018), including storm morphology and precipitation patterns (Kawecki et al., 70 2016). Particle chemical composition modulates particle size via water uptake. Particle size is a determining factor in light scattering by particles, which is important for aerosol radiative calculations. An implication of this work is that if particle hygroscopicity changes from clear sky to cloudy time periods, when aerosol-cloud interactions are most important, a quantitative understanding remains unclear.

75
In this work, we test the hypothesis that there are quantitative differences in PM 2.5 chemical composition between cloudy and clear sky time periods in ways important for water uptake. We employ a combination of satellite products, surface measurements, and thermodynamic modeling to analyze annual and seasonal trends in chemical climatology regions across the CONUS. We assess and quantify seasonal statistical significance (Kahn, 2005) for differences in distributions of RH, PM 2.5 , and chemical speciation during cloudy and clear sky times using surface measurements from the IMPROVE network 80 from 2010-2014 within the context of MODIS cloud flag values. Further, we examine one chemical climatology region in detail, the Mid South, as a case study. This region encompasses the location of the Atmospheric Radiation Measurement Southern Great Plains (SGP) site in an area of the CONUS that experiences varied weather patterns, a broad range of cloud conditions, and distinct seasonal variations in temperature and humidity (Sisterson et al., 2016).

Data and Methods 85
Cloudy and clear sky classifications are determined using publicly available data (National Aeronautics and Space Administration, 2018) from MODIS on the Aqua and Terra satellites. Pairing of satellite and surface PM 2.5 mass measurements typically works best in rural and vegetated locations, where the spectral properties of the background tend to be dark and vary little over the space of a satellite grid cell (Hauser, 2005;Jones and Christopher, 2010). For this reason, we use rural IMPROVE network sites that are located primarily in national parks, although improvements have been made for 90 retrievals over bright surfaces (Hauser, 2005;Hsu et al., 2004Hsu et al., , 2006Hsu et al., , 2013Zhang et al., 2016). We use 500 m resolution pixels that contain the IMPROVE sites. Retrievals are flagged as cloudy if QA flags specifically identified clouds as preventing retrieval, or if 2.1-micrometer reflectance was too high (r>0.35) and the fraction of 500 m sub pixels that were cloudy was greater than 44.4%. We choose 44.4% because it is a fundamental limit of the algorithm (Remer et al., 2013). IMPROVE monitors are frequently under a MODIS swath with valid retrievals even if the pixel containing the IMPROVE 95 station is not successfully retrieved. As an alternative to the IMPROVE pixel, we employ a method for quality assurance, a 17x17 grid. This allows for any retrieval within a 50 km x 50 km area to represent the IMPROVE station. If all 17x17 pixels are not retrieved, then the state over the monitor is determined to be cloudy. The 17x17 grid approach is much more likely to attribute non-retrieved data to clouds (98.5%) than the containing pixel approach, which attributes 89.8% of non-retrieved data to clouds. Misidentifying non-retrievals as cloudy is unlikely to substantially affect interpretation, as the sample size is 100 large (N>70,000 total observations, and N>1500 for an individual region).
IMPROVE network data were downloaded on 13 July 2015 and 26 May 2016 from public archives (http://vista.cira.colostate.edu/Improve/) (IMPROVE Network, 2019) for 132 unique sites across the CONUS with complete data records for the years 2010-2014 (Fig. S1a). IMPROVE data is collected every 3 days. We investigate 24-hour average 105 PM 2.5 mass, ALW, RH, sulfate (SO 4 2-), nitrate (NO 3 -), and total organic carbon (TOC) mass concentrations. Other species affect particle hygroscopic properties but are not widely measured in routine networks. For example, we investigate TOC as a whole even though primary and secondary species affect water uptake differently. There is no direct measurement of either in routine monitoring network operations, although fractionation can sometimes be used to infer information about sources and formation processes (Aswini et al., 2019;Cao et al., 2005;Chow et al., 2004). We group IMPROVE sites across the 110 CONUS into 22 chemical climatology regions defined by the IMPROVE network ( Fig. S1b) (Hand et al., 2011;Malm et al., 2017). PM 2.5 mass and composition is provided directly from the IMPROVE database, while ALW is estimated.
ALW is a function of RH, particle concentration, and chemical composition. We estimate ALW using a metastable assumption in the inorganic (K + -Ca 2+ -Mg 2+ -NH 4 + -Na + -SO 4 2--NO 3 --Cl --H 2 O) aerosol thermodynamic equilibrium model 115 ISORROPIAv2.1 (Fountoukis and Nenes, 2007). We use the reverse, open-system problem because only aerosol measurements are available. Particle mass concentration inputs of SO 4 2and NO 3 are taken from IMPROVE measurements.
Ammonium ion is not considered due to limited measurement availability. Dust and organic species are also not considered because water uptake properties are not well constrained (Jathar et al., 2016;Metzger et al., 2018), and there is large spatial heterogeneity in dust. Our approach to employing ISORROPIA introduces uncertainties (e.g., pH estimates would be 120 unreliable (Guo et al., 2015)), but neglect of dust does not affect overall interpretation of ALW mass (Fig. S2), consistent with an earlier sensitivity using this technique that included organic species (Nguyen et al., 2015). The temperature and RH were extracted from the North American Regional Reanalysis (NARR) model (Kalnay et al., 1996) similar to Nguyen et al. 2016. 125 Cloudy and Clear Sky differences in ALW are investigated in two ways. First, we compare ALW estimated using 24-hour average chemical composition and meteorology and group results into Clear Sky and Cloudy bins using the MODIS cloud flag. We use these daily values when comparing ALW within chemical climatology regions. Second, we investigate trends across the eastern US to isolate the effect of chemical composition. We select the eastern US since ALW concentrations are largest in this region (Fig. S3), and it is in cloud often and consistently (cloud fraction 30-50% year-round) (Fig. S4). This 130 makes statistical comparisons between Cloudy and Clear Sky times more robust than in the dry western US, where ALW concentrations and cloud fraction are low in most seasons. We group 24-hour average chemical composition and meteorology into Clear Sky and Cloudy bins and take monthly medians. We perform ALW estimations using the medians via three ISORROPIA calculation scenarios: 1) Clear Sky chemical composition and Clear Sky meteorology ("Clear Sky" scenario), 2) Cloudy chemical composition and Cloudy meteorology ("Cloudy"), and 3) Clear Sky chemical composition 135 and Cloudy meteorology ("Mixed") (Table S1, Fig. S5). We use monthly medians to avoid complications that arise from differing numbers of Cloudy and Clear Sky days in the Mixed scenario. To investigate meteorology and chemical composition impacts separately, we perform the Mixed scenario in order to reproduce studies in which cloud free growth factors (Brock et al., 2016) are eventually applied in models that contain cloudy meteorological conditions (Bar-Or et al., 2012). When the Mixed scenario is significantly different than Cloudy, we can reject the hypothesis that RH and temperature 140 alone explain the difference. Wet deposition is unconstrained in this analysis, but cloud droplets typically evaporate (Pruppacher and Klett, 2010).
Growth factors used in the Mid South region are estimated from a modified Kohler equation (Brock et al., 2016;Jefferson et al., 2017) (Eq. 1). We use RH from the NARR and estimate , the particle hygroscopicity, from IMPROVE-measured 145 chemical composition mass concentrations and individual species κ values ( =0.5, =0.7) (Petters and Kreidenweis, 2007). Here, is the hygroscopic diameter growth. (1)

150
Statistical significance for differences in measurement distributions of PM 2.5 chemical composition and properties between Cloudy and Clear Sky time periods from 2010-2014 is determined using the Mann-Whitney U Test in R statistical software (R Core Team, 2013). The Mann-Whitney U Test is a non-parametric test that compares two samples to assess whether

Hygroscopicity and Chemical Composition
Distributions in monthly particle chemical composition across the eastern US in 2010-2014 are sufficiently changed between MODIS-defined Cloudy and Clear Sky times to affect hygroscopicity and alter predicted ALW mass concentrations beyond 165 differences that would arise from changes in meteorology alone (Fig. 1). These findings are consistent with an analysis in the desert southwest US that shows that chemical composition is an essential factor for improving cloud condensation nuclei predictions (Crosbie et al., 2015). The only difference between the Mixed and Cloudy ALW calculations is that the Mixed scenario employs Clear Sky chemical composition (rather than Cloudy chemical composition) extrapolated to Cloudy meteorology. This type of scenario can occur in model development or satellite validation applications when PM 2.5 -AOD 170 relationships or growth factors remain unmeasured for Cloudy periods (Brock et al., 2016;van Donkelaar et al., 2010;de Hoogh et al., 2016;Tian and Chen, 2010). Previous work using climate models shows that application of ALW uptake that is influenced by incorrect chemical composition significantly affects top of atmosphere radiative forcing estimates and attribution of anthropogenic climate impacts (Rastak et al., 2017). When Clear Sky chemical composition is extrapolated to Cloudy period meteorology ("Mixed"), monthly median ALW concentrations in the eastern US, in all seasons except winter, 175 are significantly different from our best estimate, which employs the actual chemical composition during cloudy periods ("Cloudy"). Interestingly, monthly median Clear Sky and Cloudy scenario ALW concentrations do not differ significantly except during winter despite higher Cloudy RH (Fig. 2). This is consistent with chemical composition as a determining factor in ALW (Carlton and Turpin, 2013;Liao and Seinfeld, 2005), CCN (Crosbie et al., 2015), and extinction (Pitchford et al., 2007) on cloudy days because the pattern in ALW is opposite the pattern in RH. 180 Clear Sky/Cloudy patterns in SO 4 2and NO 3 mass concentrations, which affect particle hygroscopicity, vary regionally and seasonally. When aggregated over the eastern US, ALW estimates for the Mixed case are largest during summer and spring and can be explained by elevated Clear Sky SO 4 2and NO 3 concentrations and high Cloudy RH (Fig. 2). Generally, Mixed ALW concentrations in the eastern US are higher than for the Cloudy scenario because Clear Sky chemical composition 185 facilitates greater hygroscopicity and Cloudy RH is elevated (Table S2). A notable exception is the Ohio River Valley during winter, where Cloudy SO 4 2-, NO 3 -, and RH are higher than Clear Sky. In this case, Cloudy period ALW concentrations are higher than for the Mixed scenario. These findings highlight that a changing PM 2.5 chemical composition has a determining effect on ALW mass concentrations (Nguyen et al., 2016), a critical element in the estimation of aerosol-cloud interactions and particle radiative impacts. During cloudy periods, when the accurate prediction of ALW and aerosol-cloud interactions is 190 most critical, in situ knowledge of PM 2.5 chemical composition is required.
Differences in daily mass concentrations of fine particle chemical constituents between Cloudy and Clear Sky periods across the CONUS are spatially and temporally different among PM 2.5 mass and its chemical constituents except in the Northwest region (Figs. 3-7, Tables S3-S7, Fig. S6). These patterns cannot be adequately described as a function of MODIS cloud 195 fraction (Figs. S6-S7). If meteorological processes and physical transport are the only controlling factors, then patterns in mass concentrations among PM 2.5 and constituents should not vary. However, they do, suggesting differences in emissions and/or in situ chemical production of PM 2.5 during Cloudy and Clear Sky time periods. Where differences are significant for ALW, Cloudy ALW is higher than Clear Sky in all seasons, with few exceptions (Fig. 3, Table S3). Water uptake contributes to particle growth with a determining impact on particle size and radiative properties. PM 2.5 mass, greater during 200 Clear Sky times in most regions and seasons, has nearly an opposite pattern to ALW spatial and seasonal trends (Fig. 4). The largest ALW differences are observed in the central and eastern US during winter. Wintertime Cloudy SO 4 2mass concentrations are greater than Clear Sky (Fig. 5, Table S5), and the highest NO 3 mass concentration differences are observed during Cloudy times in winter when temperatures are coldest (Fig. 6, Table S8). This promotes thermodynamic stability of nitrate in the condensed phase, increasing particle hygroscopicity and facilitating ALW. 205 Outside of winter, significant SO 4 2mass concentrations are typically higher on Clear Sky days in the eastern US (Fig. 5, Table S5). Higher Clear Sky SO 4 2concentrations during summertime are associated with heat waves and stagnation events, which are characterized by a lack of ventilation in high pressure systems (Jacob and Winner, 2009;Wang and Angell, 1999) and higher electricity demand (Farkas et al., 2016) associated with emissions that form sulfate. 210 TOC mass concentrations are nearly always higher during Clear Sky times than Cloudy (Fig. 7, Table S7) in all chemical climatology regions across the CONUS, with the largest differences during summer and fall. Precursor VOC emissions (e.g., biogenic) and subsequent derived PM that contributes to OC differ by season and region (Donahue et al., 2009;Gentner et al., 2017;Youn et al., 2013). Increased sunlight under clear sky conditions leads to higher biogenic VOC emissions 215 (Sakulyanontvittaya et al., 2008) and enhanced photolysis rates that facilitate hydroxyl radical production important to secondary organic aerosol formation (Tang et al., 2003). Organic aerosol hygroscopicity and water uptake is highly uncertain (Christiansen et al., 2019;Nguyen et al., 2015), and yet has profound impacts on top-of-atmosphere radiative forcing calculations (Rastak et al., 2017). We note that TOC is also influenced by primary sources of OC including wildland fires in the west and prescribed burning in the east which are not influenced by cloud presence (Spracklen et al., 2007;Tian et al., 220 2009;Zeng et al., 2008).

PM 2.5 Mass Concentrations
Significant differences in PM 2.5 mass concentrations measured at IMPROVE monitoring locations are observed between Cloudy and Clear Sky conditions in the majority (>60%) of regions in any given season during 2010-2014 ( Fig. 4 and Table   S4) and do not correlate with MODIS cloud fraction during any season in any region (Fig. S8). In all regions, Clear Sky 225 PM 2.5 concentrations are generally higher than Cloudy. Satellite AOT products used to derive PM 2.5 may overestimate the atmospheric burden across the CONUS, particularly during summertime. Median All Sky PM 2.5 concentrations are also https://doi.org/10.5194/acp-2020-184 Preprint. Discussion started: 23 March 2020 c Author(s) 2020. CC BY 4.0 License. significantly different and typically lower than Clear Sky in multiple chemical climatology regions (Table S9). This suggests the clear sky bias in satellite data may impart a positive bias when assessing surface PM 2.5 trends in model applications for air quality, weather, and climate. 230

Case Study: The Mid South
ALW concentrations are significantly higher during Cloudy times than Clear Sky in the Mid South during all seasons (Table   1, Fig. 8). RH in the region is high year-round during Cloudy and Clear Sky periods alike, with the median greater than 60%.
Gas-phase water vapor mixing ratios are sufficiently high that water availability is not limiting for ALW in the region for any season. Aerosol mass concentrations and chemical composition vary, however, and the effects on particle hygroscopicity 235 can be seen in contrasting Cloudy and Clear Sky ALW concentrations among the seasons. For example, during Clear Sky conditions, the highest ALW mass concentrations occur during summer and spring, which correspond to the highest SO 4 2concentrations in the Mid South, and not when Clear Sky RH is highest (i.e., during winter). The largest absolute ALW concentrations and estimated growth factors occur during Cloudy times in the winter and spring, when NO 3 mass fraction and RH are highest. This is consistent with independent humidified nephelometer measurements by Jefferson et al., who find 240 that aerosol growth rates are highest in the winter and spring at the SGP site within the Mid South chemical climatology region, and identify nitrate and RH as determining factors (Jefferson et al., 2017).
NO 3 concentrations are generally lower than SO 4 2in the Mid South, but NO 3 is more hygroscopic and provides influence over ALW patterns. Sulfate is traditionally considered dominant in determining absolute ALW mass concentrations in this 245 region, and sulfate mass fraction is highest in summer (Carlton and Turpin, 2013;Gasparini et al., 2006). Similar to other regions of the CONUS, SO 4 2mass concentrations are greatest during summertime Clear Sky conditions due to transport (Parworth et al., 2015), increased rates of photochemistry (Stone et al., 2012), and increased electricity sector emissions during heat waves and stagnation events (Appel et al., 2011;Farkas et al., 2016), which generally occur on sunny days.
Sulfate mass fraction is lowest in winter, when NO 3 concentrations are high due to cooler temperatures and transport of 250 precursor species from nearby agricultural and surrounding urban areas (Parworth et al., 2015). Year-round NO 3 concentrations are higher during Cloudy conditions than Clear Sky, which are associated with lower temperatures. Under Cloudy conditions, the highest ALW concentrations and estimated growth factors occur during winter and spring, when NO 3 mass fraction and RH are highest. In another continental location, the Po Valley in Italy, NO 3 was found to control ALW concentrations with implications for secondary organic aerosol (Hodas et al., 2014). The Mid South is also a continental, 255 agricultural area and aerosol growth may be subject to similar mechanisms.

Conclusions
Across the CONUS, statistically discernible differences among PM 2.5 and chemical constituent concentrations under Cloudy and Clear Sky conditions cannot be explained solely by physical mechanisms. The chemical properties of aerosol are important to explain differences in water uptake and particle composition under different meteorological conditions. While 260 meteorological phenomena such as pressure systems, winds, and air mixing affect PM 2.5 and chemical component concentrations, they are insufficient to explain chemical constituent differences between Cloudy and Clear Sky times. In situ chemical formation processes are necessary to fully explain temporal and spatial patterns. Spatially and seasonally, PM 2.5 and particle speciation information that lends insight into water uptake, particle properties, and particle growth is incomplete when information is gathered only during Clear Sky times. The work presented here indicates aerosol growth due to water 265 uptake is greatest during satellite periods identified as Cloudy in many regions, when satellites are unable to remotely sense particle properties and impacts. This limits understanding of atmospheric particle burden and its climate-relevant physicochemical properties, which have implications for the prediction of weather (Kawecki and Steiner, 2018), air quality, and climate. This indicates that the clear sky bias affects accurate representation of ALW on cloudy days and suggests that without in situ chemical information, aerosol-cloud interactions and subsequent estimates of radiative forcings in models 270 (Lin et al., 2016;Vogelmann et al., 2012) will remain a large uncertainty.