Fog scavenging of organic and inorganic aerosol in the Po Valley

Introduction Conclusions References


Introduction
Fog scavenging is the process by which atmospheric particles transfer into the liquid phase of fog droplets.Fog scavenging affects particle microphysical and chemical properties, which are relevant to predict the effect of aerosol on climate and air quality.Indeed, fog scavenging reduces aerosol loading by promoting wet removal (Collett et al., 2001) and modifies aerosol hygroscopicity and particle size distribution by selectively removing water-soluble species (Collett et al., 2008).In addition, particle water uptake associated with fog promotes aerosol processing and secondary aerosol formation through heterogeneous-phase chemistry (Kaul et al., 2011;Ervens et al., 2011;Li et al., 2013).
Fog scavenging can take place through both impaction scavenging and nucleation scavenging.Impaction scavenging occurs when interstitial particles are incorporated into fog Published by Copernicus Publications on behalf of the European Geosciences Union.
droplets as a consequence of Brownian diffusion, inertial impaction, and phoretic effects.In a supersaturated atmosphere, aerosol particles can be activated to form fog droplets and be scavenged by nucleation.Generally, nucleation scavenging dominates in-cloud and in-fog aerosol scavenging (Seinfeld and Pandis, 1998;Elbert et al., 2000).
Several fog experiments have proven that nucleation scavenging is the dominant aerosol scavenging mechanism in the Po Valley.Fuzzi et al. (1988) observed that sulfate scavenging efficiency was lower compared to in-cloud measurements and attributed this difference to the lower supersaturation of fog compared to clouds.In the same area, Hallberg et al. (1992) observed that fog scavenging efficiency of different chemical compounds was related to their water solubility, with higher scavenging for water-soluble species like sulfate, and lower for carbonaceous hydrophobic compounds like elemental carbon.During the same experiment, the comparison of particle number size distribution in and out of fog showed that fog scavenging did not affect smaller particle concentration (mobility diameter D m below 300 nm) but did efficiently remove particles larger than 700 nm D m , leading to the conclusion that aerosol scavenging was mainly driven by nucleation (Noone et al., 1992).Elbert et al. (2000) investigated the relationship between atmospheric liquid water content and chemical composition of fog-water samples, and concluded that nucleation scavenging was the dominant particle removal mechanism.
The critical supersaturation required to nucleate fog droplets is usually around 0.01 and 0.03 % (Noone et al., 1992;Ming and Russell, 2004).However, in highly polluted environments like the Po Valley, soluble species in the gas phase and organic solutes in the liquid droplets might decrease the critical supersaturation required for particle activation, making cloud and fog formation possible even at ambient relative humidity below 100 % (Shulman et al., 1996;Laaksonen et al., 1998;Facchini et al., 1999b).
Although an increasing number of experiments investigated nucleation at supersaturation typical of clouds, little is known about the relative importance of chemistry and microphysics on cloud condensation nuclei (CCN) activity at supersaturation close to zero, which is typical of fog.Some studies show that particle size explained more than 84 % of the CCN variability at a supersaturation of 0.4 %. (Dusek et al., 2006).Other studies instead suggest that particle chemical composition can explain 40 % of the critical diameter variability at a supersaturation of 0.44 %; at lower supersaturation the influence of particle composition is expected to be even higher (Quinn et al., 2008).More recently, observations performed during a wide number of aircraft campaigns and ground-based experiments led to the conclusion that both size and compositional information are required to predict particle CCN activity (Hudson, 2007;Levin et al., 2014).
At the same time, the effect of fog scavenging on particle chemical composition, and especially on OA, is still poorly understood (Herckes et al., 2007).Facchini et al. (1999a) studied the distribution of different classes of carbonaceous species between interstitial aerosol and fog water, reporting a preferential partitioning of water-soluble organic carbon in fog droplets.Even when impaction scavenging is the dominant mechanism, a preferential removal of hydrophilic organic species was observed (Maria and Russell, 2005).Collett et al. (2008) investigated organic aerosol scavenged during four fog events in central California and observed a slight positive correlation with liquid water content (LWC).However, no information on particle microphysics and mixing state with other chemical constituents was available in these studies.
This paper analyzes the fog scavenging efficiency of major chemical components in fine particles through field observations in the Po Valley during fall of 2011 within the framework of the ARPA-Emilia Romagna supersite project.For the first time in this area, the effects of fog on particle chemistry were investigated at high time resolution, with particle microphysics taken into account.The objective of this study is to analyze the effect of scavenging on the chemical and the microphysical properties of aerosol after fog dissipation, and to investigate the role of chemical composition and mixing state of fine particles on nitrate scavenging and OA scavenging.In particular, scavenging of nitrate and OA is quantitatively discussed as a function of size distribution of the different chemical components and organic functional group composition.

Sampling site
The measurements described in this work were performed at the meteorological station Giorgio Fea in San Pietro Capofiume, a rural background site located at about 30 km northeast of Bologna, in the eastern part of the Po Valley (northern Italy).Aerosol characterization was performed from 14 November to 1 December 2011.In the Po Valley, fall is usually characterized by stable meteorological conditions, with high relative humidity and low temperature, that lead to frequent fog events that often last for several hours (Fuzzi et al., 1992).
The meteorological conditions observed during the measurement period are reported in Fig. 1. High pressure and stable weather conditions characterized the entire campaign.Winds were generally below 2 m s −1 and temperature averaged 3 • C, with values around 10 • C during daytime and 0 • C at nighttime.Relative humidity (RH) was often close to 100 % throughout both daytime and nighttime, especially before 23 November.
Liquid water content (LWC) was measured continuously with a particulate volume monitor, PVM-100 (Gerber, 1991), at 1 min time resolution.LWC higher than 0.08 g m −3 is an indicator of fog presence.The LWC time trend is reported in Fig. 2a and indicates that fog formed almost every evening around 17:00 local time (LT) and persisted during the greater part of the night until the following morning, disappearing around 09:00 LT.During the entire campaign, a total of 14 distinct fog events were identified.

Aerosol sampling
Size-segregated aerosol particles were sampled by a Berner impactor (flow rate 80 L min −1 ) on aluminum and Tedlar foils.The Berner impactor collects particles on five stages, corresponding to the following particle aerodynamic diameter cutoffs: 0.14, 0.42, 1.2, 3.5, and 10 µm.Sampling was performed continuously during the entire period.Each day we collected two samples: a daytime sample from 09:00 to 17:00 LT, and a nighttime one from 17:00 to 09:00 LT.Particles collected on aluminum foils were analyzed for carbonaceous aerosol content by means of a thermal technique, while samples collected on Tedlar were analyzed by means of ion chromatography for quantification of water-soluble inorganic species.Submicron particles were collected on Teflon filters (37 mm diameter) downstream of a silica gel drier and a PM 1 cut cyclone (flow rate 16.7 L min −1 ).One 24 h sample was collected daily from 09:00 to 09:00 LT.Three additional samples were collected each day from 09:00 to 13:00, from 13:00 to 17:00, and from 17:00 to 09:00 LT of the following morning.
Submicron particles were also sampled on prewashed and prebaked quartz-fiber filters (PALL, 9 cm size) using a dichotomous sampler (Universal Air Sampler, model 310, MSP Corporation) at a constant nominal flow of 300 L min −1 located at ground level.A total of 30 samples were collected between 15 and 30 November.Typically, two filters were sampled every day, with a daytime PM 1 sample collected from 09:00 to 17:00 LT, and an evening/nighttime sample collected from 18:00 to 09:00 LT.Samples were stored in Petri dishes at 4 • C prior to analysis.

Offline aerosol chemical characterization
Size-segregated concentrations of inorganic ions were measured by means of ion chromatography analysis of ultrapure water extracts of Tedlar substrates.
Water-soluble organic carbon (WSOC) was detected by analyzing the Tedlar substrate water extracts using a multi N/C 2100S (Analytik Jena, Germany) elemental analyzer.WSOC was quantified as the difference between total watersoluble carbon and soluble inorganic carbon.Total carbon (TC) was determined by the analysis of aluminum foil with the same technique.The detection limit was 0.2 µg of carbon and the accuracy of the TC measurement was better than 5 % for 1 µg of carbon.
The dichotomous quartz-fiber filters were analyzed to identify organic molecular tracers using proton nuclear magnetic resonance (HNMR) spectroscopy according to Decesari et al. (2006).
Organic aerosol functional groups were analyzed by means of infrared spectrometry (Gilardoni et al., 2009;Russell et al., 2011).Teflon filters were analyzed by means of Fourier transform infrared (FTIR) spectrometry in transmission mode in the region 400-4000 cm −1 , using a Bruker TENSOR 27 FTIR spectrometer equipped with a deuterated triglycine sulfate (DTGS) detector.Spectrum resolution was 4 cm −1 .Filters were scanned prior and subsequent to aerosol collection in order to remove Teflon absorption signals from the aerosol absorption spectrum.Aerosol spectra were then processed with an automated algorithm to perform baseline, peak fitting, and peak integration according to the procedure described by Russell et al. (2009).The functional group identified during the experiment included alkyl (-CH 2 -), carboxyl (-COOH), hydroxyl (-OH), amine (-NH-), and organonitrate (C-NO 3 ) groups.Aromatic and unsaturated aliphatic (=CH-), carbonyl (C=O), and organosulfate (C-OSO 3 ) moieties were below detection limit at all times.Absorption intensity was converted into mass according to Lambert-Beer equation and based on calibration with standard representative of atmospheric aerosol organic molecules.Organic mass was defined as the sum of the different organic functional groups.Elemental composition of organic aerosol was instead determined based on the elemental composition of each group.

Online aerosol chemical characterization
The size-resolved chemical composition of submicron aerosol particles was characterized online by a highresolution time-of-flight aerosol mass spectrometer (HR-TOF-AMS) (DeCarlo et al., 2006) and a soot particle aerosol mass spectrometer (SP-AMS) (Onasch et al., 2012).
The HR-TOF-AMS provided size-resolved chemical measurements of the nonrefractory sulfate, nitrate, ammonium, chloride, and organic mass in submicron particles (NR-PM 1 ).The HR-TOF-AMS was operating by alternating between "V" (higher sensitivity, lower mass resolution) and "W" (lower sensitivity, higher mass resolution) ion path modes every 5 min.The concentrations reported here correspond to the data collected in V mode.While operating in V mode, the HR-TOF-AMS acquires information about size distribution of particles, or particle time of flight (pToF) (Jimenez et al., 2003).The AMS has an effective 50 % cutoff for particle sizes below 80 and above 600 nm in vacuum aerodynamic diameter, D va , as determined by the transmission characteristics of the standard aerodynamic lens (Liu et al., 2007).All data were analyzed using standard AMS software SQUIRREL v1.51 and PIKA v1.10 (D.Sueper, University of Colorado-Boulder, Boulder, CO, USA) within Igor Pro 6.2.1 (WaveMetrics, Lake Oswego, OR).Positive matrix factorization (PMF) analyses on the HR-AMS data were performed using the PMF2.exealgorithm (v.4.2) in robust mode (Paatero and Tapper, 1994).The PMF inputs (mass spectral and error matrices) were prepared according to Zhang et al. (2011).The PMF solutions were then evaluated with an Igor Pro-based PMF evaluation tool (PET, v. 2.04) following the method described in Ulbrich et al. (2009) and Zhang et al. (2011).The HR-TOF-AMS collection efficiency (CE) was calculated according to Middlebrook et al. (2011) and averaged 0.48 ± 0.05.Concentration of AMS sulfate, nitrate, and ammonium corrected for CE and averaged over Berner impactor sampling periods were in agreement with offline measurements.
The SP-AMS was operated with both the laser and the tungsten vaporizers, alternating the laser on and off every 10 min.The refractory black carbon (rBC) data reported here are obtained in "laser on" mode, which provides measurements of the refractory component of the submicron aerosols, nominally rBC, and of the nonrefractory coating material associated with it.A relative ionization efficiency (RIE) of 0.2 (based on calibration with regal black particles) was applied to the rBC data.A CE of 0.6 was applied to all the SP-AMS data (refractory and nonrefractory components) based on comparison of SP-AMS rBC and BC measured with a multiangle absorption photometer (MAAP).Laser alignment issues did not allow for measurement of aerosol nonrefractory components (i.e.black carbon) before 17 November at 19:30 and between 21 November at 19:00 and 23 November at 16:00 LT.The size distribution data from SP-AMS were not available.

Aerosol optical properties
Optical measurements of light extinction (b ext , Mm −1 ) and light absorption (b abs , Mm −1 ) were respectively measured using a cavity-attenuated phase shift spectrometer particle extinction monitor (CAPS PMex) (Kebabian et al., 2007;Massoli et al., 2010) and a particle soot absorption photometer (PSAP) (Bond et al., 1999) valid for the singlewavelength model.The internal flow meter was calibrated with respect to a primary air flow calibrator (Gilian Gilibrator), the spot size was measured with a micrometer, and the aerosol scattering properties (b sc , Mm −1 ) were determined by an integrating nephelometer (Radiance Research single-wavelength M903) and considered in the absorption corrections according to Bond et al. (1999).PSAP data measured at filter transmissions < 30 % were rejected according to updated indications from PSAP manufacturer.The nominal CAPS data did not require any further corrections, as they were already corrected for gas-phase absorption during an automatic zeroing procedure and for the effects of dilution of the sample flow by the mirrors' purge flow (Massoli et al., 2010).The CAPS (630 nm) and M903 (539 nm) data were adjusted logarithmically to the PSAP wavelength (573 nm) according to (b x ) 573 = (b x ) λ (λ/573) α , where (b x ) λ is the extinction or the scattering coefficient measured at wavelength λ. α is the Ångström exponent calculated by best fitting the columnar aerosol optical depth within 440 nm and 870 nm measured with a POM-02L sun-sky multispectral radiometer (Prede Inc.).The particle single-scattering albedo at 573 nm (SSA) was calculated as The CAPS PMex monitors performed measurements with a precision (2σ ) of 1 Mm −1 at 1 s time resolution (decreasing to about 0.1 Mm −1 for 5 min integration times).The absolute error in the b ext coefficient was 5 % at most.
Both HR-TOF-AMSs, the SP-AMS, the CAPS, and the PSAP were connected to the same isokinetic inlet for particle sampling.

Campaign overview
Figure 2a shows online nonrefractory PM 1 (NR-PM 1 ) together with PM 1 time trend and offline chemically reconstructed PM 1.2 .PM 1 was estimated as the sum of rBC and nonrefractory submicron aerosol components, quantified by SP-AMS and HR-TOF-AMS, respectively.PM 1.2 represents the sum of sulfate, nitrate, ammonium, water-soluble organic mass (WSOM), and water-insoluble carbon mass (WICM) from the analysis of the first three stages of size-segregated Berner impactor.WSOM was calculated from the watersoluble organic carbon, using an OM to OC ratio of 1.6, in agreement with elemental analysis of high-resolution AMS data.WICM was derived from water-insoluble carbon concentration using a conversion factor of 1.1.The concen-trations of AMS sulfate, nitrate, and ammonium corrected for CE and averaged over Berner impactor sampling periods were in agreement with offline measurements.The good agreement between online and offline measurements indicates that sampling artifacts of submicron particles were negligible and confirmed that the CE correction of the online AMS data was accurate.
Total PM 1 concentration, whose average was 22 ± 15 µg m −3 over the entire campaign, varied with fog, with higher PM 1 values when fog was not present.PM 1 concentration averaged 32 ± 14 µg m −3 under clear conditions (LWC < 0.08 g m −3 ) and 10 ± 6 µg m −3 in fog (LWC > 0.08 g m −3 ), corresponding to an average decrease of 60 % in mass during fog events.Concentration reduction occurred rapidly in time.Events characterized by a rapid increase in LWC were associated with a decrease in PM 1 in less than 15 min (Fig. 2a).
PM 1 concentration in the range 10-30 µg m −3 was in agreement with previous measurements performed at the same site during winter 2008 (Carbone et al., 2010).These values, especially for out-of-fog conditions, are higher than the average PM 2.5 reported for several rural background and urban background sites (Putaud et al., 2010), and higher than the EU target limit of 25 µg m −3 set by the EU air quality directive for PM 2.5 (EU/50/2008).
Figure 2b shows the time trend of the submicron chemical components as measured by the HR-TOF-AMS.Submicron mass was dominated by organics and ammonium nitrate, as also observed in previous campaigns (Carbone et al., 2010).Average mass fraction was 54±12 % for organic matter, 27± 10 % for nitrate, 10 ± 3 % for ammonium, and 6 ± 3 % for sulfate.
Previous work has shown that the OA loadings in the rural Po Valley during fall are usually dominated by traffic emissions and wood burning for residential heating purposes (Gilardoni et al., 2011).To identify OA main sources during fall 2011 in San Pietro Capofiume, PMF analyses were performed on the organic mass spectral matrices.For this database, we chose a four-factor solution with rotational forcing parameter f Peak = 0 (Q/Q exp = 4), yielding a hydrocarbon-like OA (HOA) component, a low-volatility oxygenated OA (LV-OOA) component, and two biomass burning OA (BBOA) components, which were recombined into one BBOA factor.A detailed summary of key diagnostic plots of the PMF results and a discussion of the factor solution choices are reported in Supplement (Figs. 1S, 2S, 3S, and 4S and related text).Figure 3 shows HOA, LV-OOA and BBOA from top to bottom.During the campaign the HOA had an average mass of 2 µg m −3 , whereas the BBOA was low (0.5 µg m −3 ) until 22 November and then increased significantly during the rest of the campaign due to a decrease of ambient temperature and probably increase in emissions from wood burning for residential heating.After 22 November, BBOA averaged 2.5 µg m −3 , except for a 8 h event occurring between 28 and 29 November, when 16 µg m −3 of S.  BBOA was measured.The LV-OOA time series show a more regular pattern through the 2-week time period, with loadings ranging from 1 µg m −3 at night to 5 µg m −3 during the day, thereby following a time trend opposite to that of LWC.
In fact the highest concentrations of LV-OOA were observed when LWC was lower, i.e. out of fog.Interpretation of PMF factors is clearer when looking at the diurnal profiles of the three factors (Fig. 3), which show a strong diurnal cycle in LV-OOA (peaking between 12:00 and 17:00 LT) and a mild opposite trend in the HOA factor (slightly lower during daytime than at nighttime).The BBOA diurnal trend is instead mostly flat due to efficient fog removal during nighttime, when higher emissions are expected.Table 1 summarizes the Pearson correlation coefficients (r) between the three PMF factors and several gas-phase and particle tracers.HOA correlates best with NO (r = 0.62) whereas BBOA correlates with rBC (r = 0.75) and biomass burning tracers, i.e. levoglucosan (r = 0.71) and potassium (r = 0.80).The LV-OOA factor correlates strongly with secondary inorganic ions, i.e. sulfate and nitrate (r = 0.85 and r = 0.90, respectively), and has a stronger correlation than BBOA with more oxidized species as oxalic acid (r = 0.96).The LV-OOA correlates also with amines, calculated as the sum of monomethyl, dimethyl, and trimethyl amines from offline HNMR analysis (r = 0.71).
The elemental composition (H / C, O / C, N / C) and the OM/OC calculated using the standard AMS data analysis code (Aiken et al., 2007) are also reported in Fig. 3.The av-  (Saarikoski et al., 2012).The relatively large oxygenation of this HOA is likely due to the fact that the SPC station is a rural area away from direct urban and fresher (less oxidized) emissions.The average H / C and O / C values for the BBOA component are 1.59 and 0.24, respectively, consistent with literature BBOA mass spectra (Mohr et al., 2012;Saarikoski et al., 2012).The average H / C

Fog scavenging
Figure 4 compares the size-resolved reconstructed mass and particle composition in the range 50 nm-10 µm during night and day.Data are obtained from offline analysis of Berner impactor samples, collected during day-and nighttime.Most of the nighttime collection periods overlapped with fog events, so nighttime data are representative of in-fog conditions, while daytime average composition represents out-offog conditions.

S. Gilardoni et al.: Fog scavenging of nitrate and organic aerosol
Aerosol mass concentration of particles smaller than 140 nm and larger than 3.5 µm accounted for less than 10 % of aerosol loading, and was not affected by fog scavenging.Particles with aerodynamic diameter between 140 nm and 1.2 µm were those mostly affected by the presence of fog.Fog scavenging removed 50 to 70 % of their mass.Occasionally mass concentration of particles larger than 3.5 µm slightly increased during fog, likely due to the collection of fog droplets on the upper impactor stages, as also observed during previous campaigns (Fuzzi et al., 1992).
Comparison of size-resolved composition shows that fog scavenging removes water-soluble inorganic species more efficiently than organic matter, increasing the contribution of carbonaceous material up to 80-90 % in interstitial particles.In addition, scavenging changes the properties of this carbonaceous fraction increasing the WICM to WSOM ratio in submicron particles from 0.2-0.3before fog to 0.6-0.8during fog.The enrichment of OA in insoluble species is consistent with the results reported by Facchini et al. (1999a) based on fog and interstitial aerosol chemical characterization.
The effect of scavenging on the size and chemical composition of the submicron aerosol was further analyzed at higher resolution using the online pToF measurements from the HR-TOF-AMS (Fig. 5).The figure shows that the mean mode diameter shifted from 300-400 nm to 200-250 nm D va when going from out-of-fog to in-fog conditions.The largest change in mass was observed for nitrate, followed by organics.It is also interesting to see that organic and inorganic components were partially internally mixed (organic modes peaking at slightly smaller D va ) both in fog and out of fog.
Table 2 reports mass scavenging efficiency of major chemical components of submicron particles.Based on the HR-TOF-AMS measurements, mass scavenging efficiency (η) was calculated by comparing the concentration of each chemical species before fog formation and right after fog formation according to Noone et al. (1992).
The variability of scavenging efficiency among the different chemical species can be explained by their hygroscopicity (κ).Nitrate and ammonium showed the highest mass scavenging efficiencies, on average 71 and 68 %, respectively.Black carbon, the most hydrophobic component, was the species least efficiently scavenged (39 % on average).OA showed the largest variability, with η ranging between 20 and 60 %, in agreement with previous observations (Collett et al., 2008).Collett et al. (2008) observed a slight correlation between OA scavenging and LWC.During the Po Valley experiment, the variability of η observed among the different fog events could not be explained by the variability of LWC.
In fact, the correlation coefficient (r 2 ) of organic η and LWC was 0.02, indicating that LWC explained only 2 % of the organic scavenging variability.
The scavenging efficiency of sulfate was slightly lower than the one of nitrate (61 %).The difference between sulfate and nitrate scavenging observed during events 5, 7, 9, 10, 11, 12, and 13 could be explained by fog processing, i.e., in situ formation of sulfate though oxidation of SO 2 in the aqueous phase.Fog processing would contribute to sulfate formation, compensating in part for the removal associated with scavenging.Fog processing plays a major role in secondary organic and inorganic aerosol formation (Kaul et al., 2011;Sun et al., 2013), and its effects on aerosol composition and properties need further investigation.
Fog scavenging of different aerosol chemical components was previously investigated in the Po Valley fog system by Hallberg et al. (1992) and by Facchini et al. (1999a).The average scavenging efficiencies reported by Hallberg et al. (1992) for sulfate (18 %) and EC (6 %) were significantly lower than those measured during the present study.On the other hand, sulfate scavenging observed at the same site during a different experiment was 60 % (Facchini et al., 1999a) and agrees well with the values here reported.The scavenging efficiencies of ammonium and nitrate measured during the fall 2011 experiment are comparable to the upper bound of the variability range reported by Facchini et al. (1999a) (0.3-0.7).In addition, Facchini et al. (1999a) observed that scavenging of OA was lower compared to that of inorganic species; nevertheless, scavenging of WSOM was similar to that of nitrate and ammonium.The results of the present study confirm those observations.
Previous scavenging studies lack information about the size distribution of different chemical components, which, together with hygroscopicity, might have a significant role in determining the overall scavenging variability.Scavenging follows particle activation by water uptake, and based on Köhler theory, activation of particles larger than 300-400 nm is more likely to occur than for smaller particles, especially at low supersaturation typical of fog events.Ammonium and nitrate mass size distribution peaked around 400-500 nm D va , organic distribution had a maximum above 250 nm D va , and rBC, whose size distribution was not measured, is expected to peak around 100 nm D va ("primary soot" mode); a much less intense rBC mode can be sometimes present in the accumulation region at 400-450 nm D va ("aged soot" mode) for aged aerosol (Onasch et al., 2012;Massoli et al., 2012).
Figure 6 shows the trends in SSA obtained from combining CAPS extinction with PSAP absorption coefficients, and CAPS extinction with nephelometer scattering coefficients, according to Eq. ( 1).Good agreement between the two SSA trends was observed, with an average error below 2 %.The discrepancy observed between 21 and 23 November is likely due to the uncertainty of b abs , which was often below 0.05 Mm −1 during these days.The SSA shows a diurnal trend with high SSA values during the day without fog (reaching 0.9 during midday) and low SSA values at night in fog, with SSA values as low as 0.7.This is consistent with the time trends shown in Figs. 2 and  that fog scavenging removed water-soluble species (hygroscopic and light-scattering) and left behind particles enriched in carbonaceous material (less hygroscopic and more lightabsorbing).The smoother diurnal trend in SSA observed between 21 and 25 November corresponds to times when the fog events were less pronounced, LWC was low, and temperature was varying less (cf.Figs. 1 and 2).

Effect of OA local sources on scavenging uncertainty
Local emissions that add to the preexisting aerosol in concomitance with the beginning of a fog formation event could lead to the underestimation of fog scavenging.For the OA, sources include a regional component (LV-OOA), a local component (BBOA), and a third component (HOA) that has both local and regional sources.The concentration of HOA and BBOA factors is expected to increase during the evening hours, as confirmed by diurnal time trend observed when fog was not present (22 and 23 November) and consistent with results reported by Saarikoski et al. (2012) for the same season.These local sources might lead to an underestimation of organic scavenging efficiency during evening fogs.
Assuming that the 22 and 23 November could be used as a reference period to identify the diurnal behavior of HOA and BBOA, we can use these days to obtain a rough estimate of the scavenging underestimation.The diurnal trend of HOA and BBOA factors for the reference period is reported in Fig 5S .The trend of HOA between 16:00 and 19:00 LT during days with no fog shows an increase of about 0.4 µg m −3 h −1 , while during days with fog the increase is about 0.1 µg m −3 h −1 .The difference corresponds to a removal rate of 0.3 µg m −3 h −1 .Similarly, the average diurnal trend suggests that the removal rate of BBOA factor is about 1 µg m −3 h −1 .Considering that the average organic concentration after fog formation in the evening events is about 7 µg m −3 and that the integration time used to calculate scavenging efficiency is 2 h, the uncertainty on organic concentration associated with neglecting HOA and BBOA local sources would range between 8 % (during the first period, when BBOA contribution was negligible) and 36 % (during the second period, when BBOA mass fraction was more significant).Taking into account this uncertainty, the average scavenging efficiency of organics would increase from 50 to 58 %, which is within the variability of the data.

Discussion
The objective of this section is to investigate the effect of particle composition and microphysical properties on nucleation scavenging.Equation ( 2) assumes that the same air parcel is sampled before and right after fog formation; the intrusion of fresh air masses and the contribution of aerosol local sources would change the composition of the aerosol, making nucleation scavenging calculation inaccurate.For this reason, (i) we calculated scavenging efficiency based on observations over a short time interval (30 min before and after fog formation) and (ii) we excluded fog events associated with fog transport or intrusion.The time interval of 30 min was chosen as the shortest interval that allowed an accurate pToF measurement.Intrusion events were identified based on the analysis of meteorological parameters (Table 2).Fog events 2, 3, 4, 5, 6, 8, 9, 10, 12, and 13 were characterized by temperature decrease, stagnant conditions (i.e.wind speed below 2 m s −1 ), and almost complete scavenging of particles above 700 nm.In agreement with previous observations in the Po Valley, we classified these events as radiation fog (Noone et al., 1992;Wobrock et al., 1992;Whiteaker et al., 2002).Thus, changes in aerosol concentration should be attributed mainly to nucleation scavenging.However, we identified those events characterized by no or very small temperature variation and higher wind speed as fog transport/intrusion events (1, 7, 11, and 14).The scavenging of particles larger than 700 nm was consistently below 70 % (Table 2).
Fog events characterized by intrusion were also characterized by lower scavenging efficiencies.During these events, the prevailing wind direction was from north, northwest, and west, where major traffic roads and urban areas are located.It is likely that transport of pollutants to the measurement site was responsible for the apparently lower scavenging efficiency.Removing these intrusion events from the list of investigated fog events reduces the variability of observed scavenging.The standard deviation of the average scavenging efficiency of nitrate, for example, goes from 18 % when all events are taken into account to 6 % when only radiation fog events are considered.The standard deviation of organic average scavenging efficiency is halved (from 22 to 11 %).
To study the effect of chemical composition and particle size on scavenging, we investigated scavenging efficiency size distribution for the main submicron chemical components (nitrate, as representative of inorganic aerosol, and organics) only during radiation fog events.

Nitrate scavenging
Figure 7a reports the size distribution of η for nitrate mass concentration corresponding to the different radiation fog events.η is calculated comparing nitrate size distribution from pToF measurements averaged over 30 min before fog formation and over 30 min after fog formation.Except for events 4 and 9, all fog events showed similar variation of η, with values of about 0.5 at 200 nm D va , corresponding to a geometric diameter of 133 nm (assuming spherical particles with density 1.5 g cm −3 ).
The largest variation of η was observed below 200 nm (from 20 to 80 %), while above 400 nm, η ranged between 90 and 100 % for all the events.We explained η variability based on particle chemical composition, particularly by taking into account the contribution of organic matter to particle mass.While ammonium nitrate particles are hydrophilic, organic particulate matter is expected to be more hydrophobic (Petters and Kreidenweis, 2008).Figure 8a shows that η of nitrate was higher for larger particles and smaller for smaller particles.Within the same size range, η decreased with increasing organic mass fraction, i.e. increasing the contribution of a more hydrophobic component.Similarly, Fig. 8b reports η variability as a function of κ (Petters and Kreidenweis, 2008).κ is a measure of particle hygroscopicity, and was calculated assuming that aerosol was internally mixed according to κ = i i κ i ; (3) i and κ i are the volume fraction and the hygroscopicity of the single component i, respectively.To calculate the volume fractions, the following densities were used: 1.72 g cm −3 for nitrate, 1.78 g cm −3 for sulfate, 1.75 g cm −3 for ammonium, 1.4 g cm −3 for chloride, and 1.27 g cm −3 for organics (Duplissy et al., 2011).κ i was assumed equal to 0.7 for nitrate sulfate, and ammonium; 1.3 for chloride; and 0.15 for organics (Petters and Kreidenweis, 2007;Jimenez et al., 2009;Gunthe et al., 2009;Chang et al., 2010).
Within each size range, more hygroscopic particles were more efficiently scavenged.The hygroscopicity had a larger effect on η of smaller particles, while for particles larger than 300 nm, hygroscopicity, and thus chemical composition, affected scavenging weakly.The correlation coefficients r 2 corresponding to particles below 200 nm D va (0.51) indicates that hygroscopicity explained 50 % of nitrate scavenging variability.

Organic scavenging
Figure 7b reports the size distribution of η for organic mass concentration.η curves for organics are slightly lower than nitrate curves (average nitrate is reported in black) and showed a larger variability.η for organics reached 0.5 around 250 nm D va , and increased with increasing diameter, reaching values above 0.9 only for particles larger than 600 nm D va .Event 2 is not reported because the organic loading was too low to allow for an accurate analysis of scavenging size trend.
The similarity between organic and nitrate η curves supports the hypothesis that most of the organic mass was internally mixed with nitrate, as also suggested by the similar size distribution of pToF curves (Fig. 5).In fact, if organics were not internally mixed with water-soluble species, their scavenging below 1 µm at very low supersaturation observed in fog would be close to zero (Petters and Kreidenweis, 2008).It follows that the largest part of organic mass scavenged in submicron particles was internally mixed with nitrate.Based on this conclusion, and assuming that all the scavenged organic aerosol was internally mixed with the scavenged nitrate, we can estimate the organic mass fraction internally mixed with nitrate to be equal to the organic scavenging efficiency normalized over nitrate scavenging efficiency in each size bin.According to this, 50 to 90 % of organic mass resulted internally mixed with nitrate in the range 150-700 nm (D va ).The lower mixing ratios were observed for particles smaller than 200 nm D va , on average 35 %, in agreement with values expected for primary and less processed particles.
The correlation of size-segregated scavenging efficiency of nitrate and organics suggests that OA scavenging is controlled mainly by mixing with more hydrophilic species, at least for larger particles (about 350 nm D va ).This result does not exclude that OA scavenging variability could be affected by OA properties.With this in mind, the OA composition was further investigated by FTIR spectroscopy.
To verify the consistency between AMS and FTIR results, we compared the oxygen to carbon ratio (O / C), estimated by means of elemental analysis of AMS data (Aiken et al., 2008) to the one calculated via the contribution of oxygenated and nonoxygenated functional groups quantified by FTIR spectroscopy.Among oxygenated groups, FTIR identified organonitrates, which are not included in the organic oxygen budget in the AMS analysis.Thus, we compared the O / C ratio from AMS to the FTIR ratio calculated without the contribution of organonitrates.Figure 9 shows that the agreement between the two techniques was satisfactory (generally within 10 %) for samples characterized by a shorter collection period (below 15 h) (Maria et al., 2003).In the following paragraphs we limit our considerations to those samples characterized by a collection period shorter than 15 h.Functional group analysis by FTIR showed that organic mass was mainly composed of alkylic, carboxylic, and hydroxyl groups, representing on average 44, 28, and 22 % of organic mass, respectively.Organic nitrogen species, i.e. amines and organonitrates, accounted for a small fraction of the OA loading.
It is worth noting that the organonitrate functional group represents a small fraction of organic mass; nevertheless its presence could affect the O / C ratio.For example, when organonitrates comprised 6 % of organic mass, the O / C ratio was 14 % higher than the O / C measured by AMS or calculated by FTIR functional group excluding nitrate contribution.
Table 3 reports the scavenging efficiency of organic aerosol for four fog events when organic functional group composition from FTIR was available right before fog formation.The scavenging efficiency of organic aerosol was slightly correlated with the average O / C ratio obtained from the FTIR data.Samples characterized by the highest O / C ratio were also associated with the highest OA scavenging (18 and 28 November).The higher scavenging of more oxygenated aerosol is also confirmed by higher scavenging efficiency of the organic oxygen compared to the total OA (Table 2).This result can be attributed to the highest polarity, and thus hygroscopicity, of oxidized organic aerosols and to a more efficient mixing of oxidized/aged organics with secondary inorganic aerosol, which is more hydrophilic.We speculate that the effect of mixing might be more relevant than the effect of polarity and hygroscopicity by itself.If oxidized organics were removed by fog more efficiently than less oxidized species only because of being more hygroscopic, then the scavenging of organic oxygen should have been constant or only slightly variable, as observed, for example, for nitrate.The fact that organic oxygen scavenging efficiency varied by 18 percentage points during radiation fog events indicates that a different effect is controlling its scavenging, more likely efficiency of mixing with highly hygroscopic species, such as ammonium nitrate.
This conclusion is further supported by the analysis of the main oxygenated organic functional groups.Table 3 reports the molar ratio of the carboxylic (COOH) to the hydroxyl (OH) group.For a molecule with a defined number of carbon atoms, the COOH and OH groups are expected to affect its solubility in different ways.In fact, the COOH group has a dipole moment larger than the one of an OH group, and thus the COOH group would increase its solubility more than a OH group.The fog events of 28, 29, and 30 November showed a very similar COOH to OH ratio, indicating that the relative contribution of oxygenated functional groups was similar.Nevertheless, organic oxygen scavenging efficiency varied from 57 to 75 %.The limited number of available measurements and the lack of information on the average OA molecular weight and carbon chain length do not allow us to understand the effect of the chemical nature of organic oxygen on the variability of organic scavenging during the Po Valley experiment.

Scavenging closure
We verified the closure of nitrate and organic scavenging efficiency during radiation fog using a simple model based on size-segregated chemical composition.For each fog event the efficiency of nitrate scavenging was modeled according to the following equation: where [NO − 3 ] i is the concentration of nitrate before fog formation in size bin i (as dM / dlogD va ) and lens i is the transmission efficiency of the aerodynamic lens of the HR-TOF-AMS for size bin i .Values of nitrate concentration below detection limit were replaced by half of the detection limit.η i is the scavenging efficiency of nitrate in size bin i.The dependency of η i on κ varies with particle size (D va ), as shown in Sect.4.1.Thus, we estimated η i from κ and D va based on the equations corresponding to the linear fits reported in Fig. 8b: (5) κ was estimated from particle chemical composition according to Eq. ( 3).On average the model overestimates the observations by 3 %; only in one case the model underpredicts the measured scavenging by 5 % (absolute value).Thereafter, we modeled organic scavenging based on the simulated nitrate size-segregated scavenging efficiency and the average mixing with nitrate: [Org] i is the concentration of organic in size bin i before fog begins (as dM / dlogD va ), η i is the scavenging efficiency of nitrate calculated with Eq. ( 5), lens i is the transmission efficiency of the aerodynamic lens of the HR-TOF-AMS, and Mix is the fraction of organic internally mixed with nitrate.
For each fog event, Mix was estimated as the average ratio between organic and nitrate size-segregated scavenging efficiency.The difference between model and observation varies between −11 and +8 % (absolute values).The slightly larger discrepancy of OA scavenging is likely due to the simplified description of mixing with nitrate and the fact that the model does not include the effect of OA properties on its scavenging, such as hygroscopicity.

Conclusions
The employment of high time resolution techniques allowed for the investigation of chemical and microphysical properties of fine particles immediately before and after fog formation during 14 distinct fog events.
The highest scavenging efficiencies were observed for nitrate, ammonium, and sulfate, at 70, 68, and 61 %, respectively.Scavenging of organic aerosol averaged 50 %, while the lowest values characterized black carbon (39 % on average).
Fog preferentially removed water-soluble components and oxidized organic aerosol in particles with a vacuum aerodynamic dry diameter larger than 400 nm, leaving an interstitial aerosol enriched in water-insoluble and less oxidized carbonaceous species.
Although scavenging of water-soluble species like nitrate was close to completeness above 400 nm D va , it was strongly affected by chemical composition in smaller size ranges, especially when less water-soluble species like OA, were present.These results indicate that at very high relative humidity, OA affects significantly particle ability to take up water and form droplets.This implies that models require an accurate description of particle microphysics in order to describe both particle wet removal and atmospheric processing through heterogenous-phase chemistry.
Nucleation scavenging of OA varied between 45 and 63 %.A slight correlation between O / C ratio and scavenging efficiency was observed.Nevertheless, the correlation of sizesegregated scavenging efficiency with that of nitrate, and the similarity of nitrate and OA size distributions before and after fog formation, suggests that organic scavenging was controlled by mixing with water-soluble inorganic species.The small variability of the degree of oxidation of OA and functional group composition during the experiment made it impossible to discriminate the role of OA composition on its scavenging.
To summarize, fog represents a significant sink of pollutants on a regional scale, especially in highly polluted areas characterized by cold winter and stagnant atmospheric conditions like the Po Valley.Here we have shown that scavenging is very efficient for nitrate and inefficient for BC.Scavenging of organic component is highly variable, and further studies are required to distinguish particle-mixing effects from organic composition effects.
The Supplement related to this article is available online at doi:10.5194/acp-14-6967-2014-supplement.

Figure 1 .
Figure 1.Time trends of temperature (red), relative humidity (blue), and wind speed (black) at San Pietro Capofiume during the research campaign.

Figure 5 .
Figure 5. Mass size distribution of major constituents of submicron particle out of fog (straight line) and in fog (dotted line) averaged over the entire campaign.

Figure 7 .
Figure 7. Mass scavenging efficiency (η) size distribution of nitrate (a) and organics (b) for each fog event; the black line in (b) corresponds to average nitrate scavenging efficiency.D va is vacuum aerodynamic diameter.

Figure
Figure Size-segregated scavenging efficiency of nitrate as a function of organic mass fraction (a) and κ (b); markers are color-coded as a function of particle diameter (D va ).

Figure 9 .
Figure 9.Comparison of oxygen to carbon ratio from analysis of organic aerosol by FTIR and AMS.For the comparison, FTIR ratio is calculated ignoring the contribution of organonitrate.Black triangles are samples with collection time below 15 h.

Table 1 .
Correlation coefficients r of the PMF factors with the main external gas-phase and aerosol tracers measured at SPC during the campaign.The values in bold denote the highest correlations for each factor.The asterisks indicate data from filter analysis.

Table 2 .
Mass scavenging efficiency (η) of the different chemical species, scavenging of particle larger than 700 nm (N > 700 nm), and meteorological parameters characteristic of each fog event.Org SO 4 Org oxygen N > 700 nm LWC (g m −3 ) T drop ( • C) WS (m s −1 ) 3 and with the fact

Table 3 .
Comparison of scavenging efficiency of organic aerosol and organic oxygen with elemental analysis and organic functional group data.