CCN activity and organic hygroscopicity of aerosols downwind of an urban region in central Amazonia : seasonal and diel variations and impact of anthropogenic emissions

During the Observations and Modeling of the Green Ocean Amazon (GoAmazon2014/5) campaign, sizeresolved cloud condensation nuclei (CCN) spectra were characterized at a research site (T3) 60 km downwind of the city of Manaus, Brazil, in central Amazonia for 1 year (12 March 2014 to 3 March 2015). Particle hygroscopicity (κCCN) and mixing state were derived from the size-resolved CCN spectra, and the hygroscopicity of the organic component of the aerosol (κorg) was then calculated from κCCN and concurrent chemical composition measurements. The annual average κCCN increased from 0.13 at 75 nm to 0.17 at 171 nm, and the increase was largely due to an increase in sulfate volume fraction. During both wet and dry seasons, κCCN, κorg, and particle composition under background conditions exhibited essentially no diel variations. The constant κorg of ∼ 0.15 is consistent with the largely uniform and high O : C value (∼ 0.8), indicating that the aerosols under background conditions are dominated by the aged regional aerosol particles consisting of highly oxygenated organic compounds. For air masses strongly influenced by urban pollution and/or local biomass burning, lower values of κorg and organic O : C atomic ratio were observed during night, due to accumulation of freshly emitted particles, dominated by primary organic aerosol (POA) with low hygroscopicity, within a shallow nocturnal boundary layer. The O : C, κorg, and κCCN increased from the early morning hours and peaked around noon, driven by the formation and aging of secondary organic aerosol (SOA) and dilution of POA emissions into a deeper Published by Copernicus Publications on behalf of the European Geosciences Union. 11780 R. Thalman et al.: CCN activity and organic hygroscopicity of aerosols downwind of an urban region boundary layer, while the development of the boundary layer, which leads to mixing with aged particles from the residual layer aloft, likely also contributed to the increases. The hygroscopicities associated with individual organic factors, derived from PMF (positive matrix factorization) analysis of AMS (aerosol mass spectrometry) spectra, were estimated through multivariable linear regression. For the SOA factors, the variation of the κ value with O : C agrees well with the linear relationship reported from earlier laboratory studies of SOA hygroscopicity. On the other hand, the variation in O : C of ambient aerosol organics is largely driven by the variation in the volume fractions of POA and SOA factors, which have very different O : C values. As POA factors have hygroscopicity values well below the linear relationship between SOA hygroscopicity and O : C, mixtures with different POA and SOA fractions exhibit a steeper slope for the increase in κorg with O : C, as observed during this and earlier field studies. This finding helps better understand and reconcile the differences in the relationships between κorg and O : C observed in laboratory and field studies, therefore providing a basis for improved parameterization in global models, especially in a tropical context.

Abstract. During the Observations and Modeling of the Green Ocean Amazon (GoAmazon2014/5) campaign, sizeresolved cloud condensation nuclei (CCN) spectra were characterized at a research site (T3) 60 km downwind of the city of Manaus, Brazil, in central Amazonia for 1 year (12 March 2014to 3 March 2015. Particle hygroscopicity (κ CCN ) and mixing state were derived from the size-resolved CCN spectra, and the hygroscopicity of the organic component of the aerosol (κ org ) was then calculated from κ CCN and concurrent chemical composition measurements. The annual average κ CCN increased from 0.13 at 75 nm to 0.17 at 171 nm, and the increase was largely due to an increase in sulfate volume fraction. During both wet and dry seasons, κ CCN , κ org , and particle composition under background con-ditions exhibited essentially no diel variations. The constant κ org of ∼ 0.15 is consistent with the largely uniform and high O : C value (∼ 0.8), indicating that the aerosols under background conditions are dominated by the aged regional aerosol particles consisting of highly oxygenated organic compounds. For air masses strongly influenced by urban pollution and/or local biomass burning, lower values of κ org and organic O : C atomic ratio were observed during night, due to accumulation of freshly emitted particles, dominated by primary organic aerosol (POA) with low hygroscopicity, within a shallow nocturnal boundary layer. The O : C, κ org , and κ CCN increased from the early morning hours and peaked around noon, driven by the formation and aging of secondary organic aerosol (SOA) and dilution of POA emissions into a deeper boundary layer, while the development of the boundary layer, which leads to mixing with aged particles from the residual layer aloft, likely also contributed to the increases. The hygroscopicities associated with individual organic factors, derived from PMF (positive matrix factorization) analysis of AMS (aerosol mass spectrometry) spectra, were estimated through multivariable linear regression. For the SOA factors, the variation of the κ value with O : C agrees well with the linear relationship reported from earlier laboratory studies of SOA hygroscopicity. On the other hand, the variation in O : C of ambient aerosol organics is largely driven by the variation in the volume fractions of POA and SOA factors, which have very different O : C values. As POA factors have hygroscopicity values well below the linear relationship between SOA hygroscopicity and O : C, mixtures with different POA and SOA fractions exhibit a steeper slope for the increase in κ org with O : C, as observed during this and earlier field studies. This finding helps better understand and reconcile the differences in the relationships between κ org and O : C observed in laboratory and field studies, therefore providing a basis for improved parameterization in global models, especially in a tropical context.

Introduction
Atmospheric aerosols have a major impact on the radiative balance of the Earth's climate system by changing the microphysical structure, lifetime, and coverage of clouds. For the same liquid water content, high aerosol concentration leads to more, smaller cloud droplets and therefore higher cloud albedo (Twomey, 1977). The smaller droplet size also delays or inhibits warm precipitation, leading to increases in both cloud lifetime and coverage (Albrecht, 1989) and ultimately invigoration of convective clouds . Currently, the effects of aerosol on clouds remain one of the largest uncertainties in simulated climate change during the industrial era, and a large portion of this uncertainty is due to the natural aerosol properties and processes represented in models Ghan et al., 2013). The Amazon represents more than half of the planet's rainforest and is a rapidly changing region where deforestation, human activity, and natural resource needs are all at play in changing the ecosystem Batistella et al., 2009;Davidson et al., 2012). The Amazon basin also represents at times one of the cleanest continental regions on the planet where it is still possible to find extended periods with little or no impact of anthropogenic activity, although the longdistance transport of pollution is occasionally observed Hamilton et al., 2014;Martin et al., 2010b;Wang et al., 2016a, b;Williams et al., 2002). This makes the Amazon basin an ideal location to characterize aerosol under near-natural conditions and assess the impact due to urban emissions and biomass burning (Kuhn et al., 2010).
The biogenic activity of this region makes it a major source of organic carbon released into the atmosphere via isoprene and monoterpenes (Guenther et al., 2006(Guenther et al., , 2012Kesselmeier et al., 2002;Kuhn et al., 2007) which are mediated by biotic stress through heat, sunlight, and changes in CO 2 .
To understand the impact of aerosol on clouds and climate requires knowledge of the concentration of cloud condensation nuclei (CCN), which are particles that are able to form cloud droplets under relevant atmospheric conditions. The minimum supersaturation required to activate a particle into a cloud droplet can be predicted using the κ-Köhler theory based on particle size and the single hygroscopicity parameter κ, which combines a number of thermodynamic properties required for the description of water activity of the growing droplets . The value of κ is determined by the physicochemical properties of the solutes, including their molar volume, activity coefficient, and the effect on surface tension. For multicomponent particles, κ is the volume average of participating species. Hygroscopicity also describes particle growth under subsaturated conditions and can be derived from the particle growth factor (GF). However, particles sometime exhibit larger κ values for droplet activation (derived from CCN measurements under supersaturated conditions) than for particle growth (derived from particle GF under subsaturated conditions; e.g., Duplissy et al., 2008;Good et al., 2010;Mikhailov et al., 2013;Pajunoja et al., 2015;Wex et al., 2009). In this paper, "hygroscopicity" represents κ associated with droplet activation derived from CCN measurements unless noted otherwise.
A number of recent studies examined the sensitivity of predicted CCN concentration and cloud droplet number concentration to aerosol properties (e.g., Ervens et al., 2010;Kammermann et al., 2010;McFiggans et al., 2006;Mei et al., 2013b;Reutter et al., 2009;Rissman et al., 2004;Roberts et al., 2002;Wang, 2007;Wang et al., 2008). These studies show that the predicted CCN concentration is often sensitive to κ org , especially for aerosol under background conditions where organics tend to dominate submicron aerosol mass (Liu and Wang, 2010;Mei et al., 2013b). Using a constant κ org may lead to large biases in predicted CCN concentrations and aerosol indirect forcing (Liu and Wang, 2010). Therefore, it is imperative to understand organic hygroscopicity under background conditions, such as in the Amazon forest, as well as the variation of organic hygroscopicity due to anthropogenic emissions.
There have been several studies of aerosol hygroscopicity in the Amazon basin over the past 20 years Mikhailov et al., 2013;Pöhlker et al., 2016;Rissler et al., 2006;Roberts et al., 2001;Vestin et al., 2007;Whitehead et al., 2016;Zhou et al., 2002). Gunthe et al. (2009) performed size-resolved CCN measurements during the wet season in February and March 2008 as part of the AMAZE-08 campaign (Martin et al., 2010a). That study reported no diel cycle in the CCN concentration during periods with little or no influence of pollution. Pöhlker et al. (2016) measured size-resolved CCN spectra at a remote background site (Amazon Tall Tower Observatory, ATTO) over a 1-year period from March 2014 to February 2015 and observed no diel cycle and only weak seasonal trends in derived particle hygroscopicity (κ CCN ), while CCN concentrations had a pronounced seasonal cycle as the background aerosol concentration was strongly influenced by regional biomass burning during the dry season. During the SMOCC-2002 campaign (LBA-SMOCC; Large-Scale Biosphere-Atmosphere Experiment in Amazonia -SMOke, Aerosols, Clouds, Rainfall, and Climate), particle hygroscopicity was derived from the Humidified Tandem Differential Mobility Analyzer measurements in the state of Rondônia in the southwest of the Amazon region during the dry season from September to November of 2002 (Rissler et al., 2006;Vestin et al., 2007). The study concluded that the diel variation in the aerosol hygroscopicity could be linked to the structure and dynamics of the boundary layer. Local sources dominated nighttime aerosol properties with downward mixing from the residual layer aloft as the day progressed. All of these studies found that particle hygroscopicity increased with particle size (from the Aitken to accumulation modes), consistent with higher sulfate content at larger sizes . The same boundary layer evolution has been found to influence particle number and CCN evolution in a number of other related studies (Fisch et al., 2004;Martin et al., 2010a;Rissler et al., 2006;Vestin et al., 2007;Whitehead et al., 2010;Zhou et al., 2002).
In this study we present measurements of size-resolved CCN spectra at five particle diameters ranging from 75 to 171 nm downwind of Manaus, Brazil, in central Amazonia for a period of 1 year from 12 March 2014 to 3 March 2015. Particle hygroscopicity, mixing state, and organic hygroscop-icity are derived from the size-resolved-CCN activated fraction and concurrent aerosol composition measurements. The diel variations of these properties are examined for different seasons (i.e., wet season, dry season, and transition seasons) and for different types of representative air masses, including background conditions, as well as influences of urban pollution plumes and/or local biomass burning. During the wet season, the background air mass represents near-natural conditions, with occasional impact from anthropogenic emissions, while, in the dry season, the background is dominated by regional biomass burning aerosol particles. The relationship between organic hygroscopicity and particle oxidation level (i.e., O : C atomic ratio) is examined for both dry and wet seasons. Hygroscopicities associated with organic factors from aerosol mass spectrometry (AMS) positive matrix factorization (PMF) analysis are derived, and their relationship with the O : C ratio is compared with those from previous laboratory studies.

Activated fraction of size-selected particles
The CCN activation fraction of size-selected particles was measured using a differential mobility analyzer (DMA) coupled to a condensation particle counter (CPC; TSI Inc., 3010) and a cloud condensation nuclei counter (CCNC; Droplet Measurement Technologies, Boulder, CO;Frank et al., 2006;Mei et al., 2013a;Moore et al., 2010;. Aerosol particles were sampled with a total flow rate (Q a ) of 1.53 L min −1 from a height of 5 m a.g.l. and were dried to relative humidity (RH) below 20 % by a Nafion dryer immediately upon entering the instrument container. The dried aerosol particles then reached steady state charge distribution inside a Kr-85 aerosol charger (TSI, model 3077A) prior to being introduced into the DMA operated with a sheath flow rate (Q sh ) of 15.3 L min −1 to maintain a 10 : 1 sheath to aerosol flow ratio (Q sh /Q a ). The aerosol particles were size-selected by the DMA, and the size-selected particles were simultaneously characterized by a CPC (Q CPC = 0.53 L min −1 ) and a CCNC (see Fig. S1 in the Supplement for further details). This system had been operated in previous field campaigns by scanning the particle size while CCNC supersaturation was held constant (Mei et al., 2013a, b). During GoAmazon2014/5 the particle size classified by the DMA was stepped through seven particle diameters (51, 75, 94, 112, 142, 171, and 222 nm), while the CCNC supersaturation was also changed at each diameter by stepping the flow rates (Q CCN ranging from 0.2 to 1.0 L min −1 ) and temperature gradient ( T = 4.5, 5.5, 6.5, 8.0, and 10.0 • C). At a given supersaturation, data were acquired for a minimum of 30 s and until 1500 particles were counted by the CPC or up to a maximum time of 120 s. Depending on the aerosol number size distribution, the measurement cycled through the seven particle sizes in 1-2 h (see Figs. S1 and S2 for further details of the measurement setup and sampling protocol). The sampling sequence was designed so that the change of CCNC supersaturation was mostly accomplished by stepping flow rates, as the CCNC reaches steady state faster following flow changes than temperature changes. Change in the temperature gradient was kept at a minimum frequency but was necessary given the wide range of supersaturation explored. Given the low particle number concentration (e.g., ∼ 200 cm −3 under background conditions during the wet season), these approaches were important to achieve adequate counting statistics with good time resolution to capture changes of air mass within 10-20 min (Liu et al., 2016). The supersaturation of the CCNC was calibrated using ammonium sulfate aerosol, as described previously in the literature (e.g., Mei et al., 2013a), at each operational set point (Q CCN and T ), ranging from 0.075-1.1 %. Fluctuation of the temperature inside the instrument container, ranging from 20 to 30 • C over the course of a day, led to substantial variation in the absolute temperature inside the CCNC growth chamber. Calibrations were therefore repeated under a range of container and associated-growth-chamber temperatures. The dependence of the supersaturation on the temperature at the top of the CCNC column (instrument temperature T 1 ) was derived for each Q CCN and T pair and used to retrieve the supersaturation over the range of the instrument operating conditions (see Fig. S3 and further description in the Supplement).

Aerosol chemical composition
Non-refractory submicron aerosol composition (organics, sulfate, nitrate, ammonium, and chloride) was measured by a high-resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS; Aerodyne Research Inc.; DeCarlo et al., 2006) during the two IOPs and by an Aerosol Chemical Speciation Monitor (ACSM; Aerodyne Research, Inc.; Ng et al., 2011) from July 2014 to March 2015. The AMS sampled from an inlet equipped with a PM 2.5 cyclone located at 5 m a.g.l. The ambient sample was first dried outside the container by a polytube Nafion dryer (Perma Pure, model PD-100T). Once inside the container the sample was further dried by a monotube Nafion dryer (Perma Pure, model MD-110) to achieve RH < 40 % and was split between the AMS and a Scanning Mobility Particle Sizer (SMPS, TSI, model 3081). Ambient measurements were obtained every 4 of 8 min. Further details of the AMS setup and operation are described in de Sá et al. (2017a).
The ACSM was a part of the Atmospheric Radiation Measurement (ARM) Mobile Facility-1 (AMF1) mobile aerosol observing system (MAOS). Aerosol was sampled through an inlet located 10 m above the ground. The aerosol sample was first dried through five large (40×1.75 cm I.D.) Nafion dryers before being distributed among various instruments including the ACSM. The ACSM sampling alternated between with and without an in-line filter using a three-way valve, such that aerosol-free background could be subtracted from the ambient measurement. A total of 28 ambient and background scans of the quadrupole mass spectrometer (unit mass resolution) were averaged to give one measurement every 30 min. The mass concentrations of organic species, sulfate, nitrate, ammonium, and chloride were derived from measurements using approaches described in Ng et al. (2011).
Refractory black carbon (rBC) was measured using both a Single Particle Soot Photometer (SP2; Droplet Measurement Technologies, Boulder, CO) and an aethalometer (Magee Scientific) co-located with the AMS and ACSM. The SP2 measures rBC using laser-induced incandescence, whereas the aethalometer measures equivalent black carbon (BC e ; Andreae and Gelencsér, 2006) using light absorption from particles collected onto a filter. While these are fundamentally different aerosol properties, both species (rBC and BC e ) were treated as equivalent in this study, and BC e concentration was adjusted to match that of rBC using the approach detailed in Sect. 2.2 of the Supplement.

Additional relevant measurements
Additional measurements of aerosol microphysics, trace gas concentrations, and atmospheric conditions used in this study are briefly described here. These measurements were part of the deployment of the ARM AMF1 facility during GoAma-zon2014/5 . Relevant aerosol measurements include dry-particle number size distributions from 10 to 480 nm by an SMPS and the number concentration of particles with diameters greater than 10 nm by a CPC (TSI Inc., model 3772). Mixing ratios of CO and O 3 were characterized by an Off-Axis Integrated Cavity Output Spectroscopy CO, N 2 O, and H 2 O analyzer (model number 908-0014, Los Gatos) and a UV Photometric O 3 analyzer (model 49i, Thermo Scientific Inc.), respectively. Oxides of nitrogen (NO, NO x , NO 2 , NO y ) were measured using a chemiluminescence technique (details given in Sect. S2.1 of the Supplement). Meteorological data included relative humidity, ambient temperature, wind speed and direction, and rain accumulation. The vertical profiles of atmospheric backscatter (clouds and aerosol) and boundary layer heights were estimated from ceilometer (model CL31, Vaisala) measurements.

Derivation of particle hygroscopicity and mixing state
The particle hygroscopicity parameter, κ , was derived from the activation spectrum (i.e., activated fraction as a function of supersaturation S) at the individual particle sizes using approaches detailed in the literature (Bougiatioti et al., 2011;Cerully et al., 2011;Lance et al., 2013;Mei et al., 2013a;Rose et al., 2008). The activation spectrum of size-selected particles was first corrected for the influence of multiply charged particles, which is estimated using the size distribution measured by the SMPS in MAOS and the activation spectrum measured at the sizes of the doubly and triply charged particles (see Sect. S3.1). The corrected activation spectrum of size-selected particles was then fit with a cumulative lognormal (Mei et al., 2013a;Rose et al., 2008) functional form (see Fig. S6 for examples): where R a is the activated fraction as a function of supersaturation S, E is maximum activated fraction, and (1−E) represents the number fraction of particles consisting of only nonhygroscopic species (e.g., uncoated rBC) that cannot serve as CCN under typical atmospheric supersaturations. S * is the supersaturation at which R a reaches 50 % of E and represents the median critical supersaturation of size-selected particles that serve as CCN. The value of σ s is related to the slope of the increasing R a with S near S * and reflects the heterogeneity of critical supersaturation, which to a large degree arises from the heterogeneity of the hygroscopicity among size-selected particles (Cerully et al., 2011;Mei et al., 2013a). The probability density function of hygroscopicity for size-selected particles is derived from the R a (S). The average hygroscopicity κ CCN and dispersion of the hygroscopicity σ (κ) /κ CCN for the size-selected CCN were then derived from the probability density function of hygroscopicity using the approach detailed in Sect. S3.4 in Supplement. The dispersion of the hygroscopicity reflects the composition heterogeneity (i.e., mixing state) among size-selected particles (Mei et al., 2013a). For simplicity, we use κ CCN to represent the average hygroscopicity of size-selected CCN in the following sections. As hygroscopicities reported in this study were derived from particle dry diameter and critical supersaturation, they represent "apparent hygroscopicity", which includes the potential impact due to the limited solubility of organics and the reduction of surface tension by surface active species (Sullivan et al., 2009).

Derivation of organic aerosol hygroscopicity
The average particle hygroscopicity was then combined with the chemical composition data to derive the hygroscopicity of the organic component of the size-selected particles, κ org . Collectively, the AMS, ACSM, SP2, and aethalometer provided mass concentrations of organic species, sulfate, nitrate, ammonium, and rBC. The concentration of chloride was negligible ( 1 % of aerosol mass) and was not included in the analysis. Given the low concentrations during GoA-mazon2014/5, size-resolved mass concentrations at the time resolution of the CCN measurement were not directly derived from AMS particle-time-of-flight (PToF) mode data. For IOP1, measurements were classified into three groups based on bulk organic mass fraction and the characteristic mass size distribution of each species was averaged from measurements in each group. For the dry season, the measurements were classified into three groups each for day and night periods based on the bulk aerosol organic mass fraction, and the mass size distribution of each species was averaged from measurements in each of the six groups. The size-resolved mass concentrations of sulfate, nitrate, and organics at the time resolution of the CCN measurement were then derived by scaling the total mass concentration using the average mass size distributions for the corresponding group (based on the bulk organic mass fraction) of either wet or dry season (de Sá et al., 2017b). The shape of the NH + 4 mass size distribution was assumed to be the same as that of sulfate, as ammonium cations were primarily associated with sulfate. rBC was assumed to have the same size distribution shape as the total aerosol mass (i.e., mass fraction of rBC was independent of particle size); though this assumption may not always be appropriate, the effect is expected to be very small as the monthly average volume fraction of rBC was always less than 4 % (Fig. 1). A detailed description of the derivation of the size-resolved mass concentrations is given in Sect. S4.1.
In most cases, NH + 4 was insufficient to completely neutralize SO 2− 4 . The concentrations of both the organonitrate and inorganic nitrate during the two IOPs were retrieved from AMS data based on the ratio of ions NO + and NO + 2 (de Sá et al., 2017b;Fry et al., 2009). When the inorganic nitrate mass concentration was negligible (i.e., less than 30 ng m −3 ), as in most of the cases, the contributions of ammonium sulfate and ammonium bisulfate were calculated based on the mass concentrations of SO 2− 4 and NH + 4 (Nenes et al., 1998). In rare cases when the mass concentration of inorganic nitrate was greater than 30 ng m −3 , sulfate was as- sumed to be ammonium sulfate. During non-IOP periods, only the total nitrate mass concentration is available, all nitrate was assumed to be an organonitrate (Cerully et al., 2015;Lathem et al., 2013;Nenes et al., 1998;Zhang et al., 2005), and sulfate was fully neutralized by ammonium (see Sect. S4.2 in the Supplement for a sensitivity study of these assumptions for non-IOP periods). The -ONO 2 portion of the organonitrate was added back to the organic mass. We note that the amount of -ONO 2 added back was typically small given the low mass fraction of nitrate in the aerosol. Cloud condensation nuclei were assumed to be internal mixtures of (NH 4 ) 2 SO 4 , NH 4 HSO 4 , NH 4 NO 3 , organics, and rBC, and the volume fractions of the species were derived from the mass concentrations and densities. Densities of organics were estimated from the ratios of O : C and H : C measured by the AMS (Kuwata et al., 2012) and were on average 1450±100 and 1470±80 kg m −3 for IOP1 and IOP2, respectively. Densities of 1770, 1790, 1730, and 1800 kg m −3 were used for (NH 4 ) 2 SO 4 , NH 4 HSO 4 , NH 4 NO 3 , and rBC (Bond and Bergstrom, 2006;Park et al., 2004), respectively. In very rare cases, E was less than 100 %, suggesting some of the size-selected particles were nonhygroscopic. The nonhygroscopic particles were assumed to consist entirely of rBC (Mei et al., 2013a). The volume concentration of the nonhygro-scopic particles was derived as the product of (1−E) and the total volume concentration (i.e., the sum of volume concentrations of organics, (NH 4 ) 2 SO 4 , NH 4 HSO 4 , NH 4 NO 3 , and rBCs) at the size classified by the DMA. The volume concentration of rBC internally mixed within the CCN-active particles was then calculated as the difference between the total rBC volume concentration and the volume concentration of the nonhygroscopic particles. (Mei et al., 2013a, b). Assuming a κ value of zero for rBC, we can derive the hygroscopicity of the organic component of the CCN κ org as where x i is the volume fraction of the respective species. The κ values are 0.61, 0.7, and 0.67 for (NH 4 ) 2 SO 4 , NH 4 HSO 4 , and NH 4 NO 3 , respectively . The uncertainty in κ org using these calculations has been derived using the approach detailed in earlier studies (Mei et al., 2013a, b) and is on the order of 0.01-0.02 (which was generally between 10 and 20 %) for this dataset.

Derivation of κ for AMS PMF factors
PMF was applied to the AMS mass spectra (Lanz et al., 2008;, and six organic factors were identified for each of the two IOPs (de Sá et al., 2017b). For IOP2 the PMF analysis included data from 24 August to 15 October 2014, excluding a major regional biomass burning event from 16 to 23 August, which was treated separately in the PMF analysis. For IOP1 (wet season), the six factors were isoprene-epoxydiol-derived secondary organic aerosol (IEPOX-SOA), more-oxidized oxygenated organic aerosol (MO-OOA, i.e., highly oxidized organics), less-oxidized oxygenated organic aerosol ( Table 1. In this study, the O : C ratio was derived using the Improved-Ambient method (Canagaratna et al., 2015). For each IOP, hygroscopicities associated with the six factors were attributed based on multilinear regression of κ org with respect to the volume fractions of the factors (Levenberg-Marquardt algorithm, IGOR Pro, Wavemetrics): where κ i and x i are the hygroscopicity and volume fraction of the individual organic PMF factors. The volume fraction was derived from mass concentrations and the densities of the factors. κ org represents the average organic hygroscopicity at particle diameters (D p ) of 142 and 171 nm. As the PMF analysis is based on the mass spectra of the bulk submicrometer aerosol (i.e., MS mode measurements), an implicit assumption of Eq. (3) is that the bulk volume fractions of the factors represented those over the sizes at which κ org was derived (i.e., D p = 142 and 171 nm). The validity of this assumption is discussed in the results section. The robustness of the factor hygroscopicity derived through linear regression depends on the variation of the factor volume fraction during the measurement period. The HOA hygroscopicity was assumed as zero based on the results from previous studies (Cappa et al., 2009(Cappa et al., , 2011, and the hygroscopicity of the other five factors were derived by multilinear regression as described above.

Classification of seasons and air masses
The 1-year sampling period was divided into different seasons by grouping months according to the similarity of the aerosol properties and trace gas concentrations measured at the two background sites, T0a and T0t, as well as monthly accumulated rainfall. In this study, the seasons were defined as follows: For each season, the air masses arriving at the T3 site were classified into three different types: background, urban pollution, and local biomass burning based on trace gas and aerosol measurements at all sites. During the wet season, the background air mass represented near-natural conditions, with occasional impact from anthropogenic emissions, while, in the dry season, the background was dominated by regional biomass burning aerosol particles. Polluted air masses represent those with strong influence from urban emissions, which were mostly from Manaus. The localbiomass-burning type describes those air masses strongly influenced by local (i.e., fresh) biomass burning activities, which dominated over the impact from urban pollution, if any. For each season, background conditions were identified when CO and condensation nuclei (CN) concentrations were below the thresholds derived from measurements at the background T0a and T0t sites, and the NO y mixing ratio was below 1.5 ppb. Non-background conditions were identified by condensation nuclei and CO concentrations above the respective threshold levels. As biomass burning aerosol typically has a higher fraction of accumulation-mode particles, and the emissions from Manaus were more dominated by Aitken-mode particles, the fraction of particles with diameter less than 70 nm was used to differentiate air masses strongly influenced by local biomass burning from those with more impact from urban pollution (see Table S2 in Supplement for details). Contamination by the emissions from an on-site diesel generator, grass cutting activities, tractors, and other vehicles were evidenced by rBC concentrations above 1.0 µg m −3 or CN concentration above 10 000 cm −3 . Over the 1-year measurement period, background, urban pollution, and local biomass burning represented 12.4, 38.5, and 28.4 % of the CCN measurements, respectively (Table S3). We note that the air masses arriving at the T3 site often included contributions from different sources. The classification of the air masses using the above three types clearly represents a simplification but is very helpful for understanding the properties of aerosols influenced by the various major sources.

Seasonal trend and size dependence of hygroscopicity and chemical composition
The monthly average κ CCN at the T3 site varied from 0.1 to 0.2 at five particle diameters ranging from 75 to 171 nm (Fig. 1a) and was substantially lower than the value of 0.3 ± 0.1 suggested for continental sites . This was due to the large organic volume fraction, up to 95 %, observed at the T3 site. In this study, measurements at 51 and 222 nm were not included, because the range of supersaturation sampled inside the CCN counter only adequately captured the activation spectrum for 51/222 nm particles with relatively high/low κ CCN values, leading to a positive/negative bias of the average κ CCN . The value of κ CCN exhibited similar seasonal variations at all five sizes. During the transition from wet to dry season, κ CCN decreased by 20-30 % with the absolute minimum of 0.116 occurring at 75 nm in September and October (Fig. 1a). The seasonal trend of κ CCN was mainly driven by the variation of κ org , which shows the lowest value in September during the dry season (Fig. 1b). Despite a strong increase in aerosol mass concentration from wet to dry season due to biomass burning emissions, the organic volume fraction exhibited little seasonal variation and was ∼ 90 % or higher at the four sizes from 94 to 171 nm (Fig. 1c). A minor increase in organic volume fraction in October might have also contributed to the lower κ CCN value observed. The species volume fractions at 75 nm are not shown due to the very low signal-to-noise ratio of the AMS PToF data in the small particle diameter range. No clear seasonal trend was observed for sulfate volume fraction, which ranged from 3 to 9 % at the four sizes. The lack of clear seasonal trends of sulfate and organic fractions is consistent with observations at the T0a site . Nitrate and rBC represented a small fraction of aerosol volume and were less than 1 and ∼ 4 %, respectively.
The average κ CCN increased with increasing particle size for all three air mass types and during all the seasons (Fig. 2), consistent with decreasing organic volume fraction with increasing particle size (Fig. 1c). The κ CCN at 75, 94, 112, 142, and 171 nm averaged for the 1-year measurement period were 0.130 ± 0.028, 0.144 ± 0.039, 0.148 ± 0.043, 0.164 ± 0.046, and 0.167 ± 0.042, respectively. The value of κ CCN and its size dependence under background conditions were largely consistent among different seasons and were in good agreement with those observed under near-natural conditions during the AMAZE-08 campaign at T0t in the wet season ) and during the 1-year period from March 2014 to February 2015 at the background T0a site (Pöhlker et al., 2016). For the air masses with strong influence from local biomass burning, the value of κ CCN and its size dependence are consistent with the κ value derived from particle growth factor measurements in July 2001, during a "recent biomass burning period" of the CLAIRE-2001 study (Rissler et al., 2004), which took place at Balbina, about 125 km northeast of Manaus. In contrast, κ values derived from particle growth factor measurements from 11 September to 8 October 2002, during the dry period of the LBA-SMOCC (Rissler et al., 2006), are substantially lower than κ CCN values observed at the T3 site for local biomass burning air masses at all sizes. As LBA-SMOCC took place in the state of Rondônia in southwestern Amazonia with extensive biomass burning activities during the dry season, the difference in κ could be due to the differences in fire condition and the type of vegetation burned. Previous studies show particles sometime exhibit larger κ values for droplet activation (derived from CCN measurements under supersaturated conditions) than for particle growth (derived from particle growth factor under subsaturated conditions); this could also contribute to the differences in κ values. Compared to κ CCN , κ org was largely independent of particle size for all three air mass types, indicating that the size dependence of κ CCN was mainly due to the size dependence of the organic volume fraction and particle composition (Fig. 1c-f). Dur-ing the dry season, aerosols classified as urban pollution and local biomass burning exhibited lower κ org values compared to background aerosols, contributing to the lower values of overall κ CCN .

Diel trends of particle and organic hygroscopicities
The diel variations of aerosol properties are presented in Figs. 3-7 for different air masses during the two IOPs. Aerosol properties derived from the size-resolved CCN measurements, including κ CCN , σ κ CCN /κ CCN , and κ org , and the volume fractions of different species were averaged at the three largest sizes (D p = 112, 142, and 171 nm). The fraction of organic mass at m/z = 44 (f 44 ) and O : C were derived from the AMS bulk measurements. Also shown are diel variations of planetary boundary layer (PBL) height, CN, and aerosol volume concentrations based on 5 min average data. Figures 3 and 4 show the diel variations of aerosol properties during the wet season of 2014 for background and urban pollution air masses, respectively. Only 0.7 % of the data were classified as local biomass burning (see Table S3), which is insufficient to evaluate the diel trends. During the wet season, the background air mass represents near-natural conditions, with minimum impact from anthropogenic emissions, although long-distance transport of African biomass burning may contribute to the aerosols observed Wang et al., 2016b). Background aerosol constantly exhibited relatively high hygroscopicity of ∼ 0.19 throughout the day. The lack of a diel trend in κ CCN is also in agreement with the results from the T0a site (ATTO), which is upwind of Manaus and served as a background site (Fig. 8). The particle composition averaged for the three particles diameters was dominated by organics, representing ∼ 90 % of the aerosol volume. The lack of a diel trend in κ CCN and κ org suggests little variation in particle composition throughout the day. The nearly constant κ org of ∼ 0.16 is also consistent with the lack of a diel trend in f 44 and O : C. The values of f 44 and O : C are ∼ 0.2 and ∼ 0.8, respectively, indicating that the aerosol under background condition during the wet season was dominated by the aged regional aerosol particles consisting of highly oxygenated organic compounds.

Background conditions
Aerosol number and volume concentrations exhibited a minimum at ∼ 310 cm −3 and ∼ 0.3 µm 3 cm −3 , respectively, just before sunrise. The number and volume concentrations started increasing after sunrise and peaking at 400 cm −3 and 0.8 µm 3 cm −3 in the afternoon. These diel variations are partially attributed to the wet scavenging of accumulation-mode particles, which dominate the submicron particle concentrations under background conditions, and the mixing of the particles from the residual layer aloft down to the surface as the boundary layer develops in the morning. During the night, the radiative cooling at the surface leads to a shallow nocturnal boundary layer with low and variable winds. RH near surface was near 100 %, and fog or mist was identified by the weather station (Present Weather Detector, Visalia) 62 % of the time during the 1-year measurement period. The gradual decreases of particle number and volume concentration during these fog events were due to the wet deposition of the accumulation-mode particles activated into droplets. Similar decreases of particle number concentration were previously reported during night fog events in the tropical rainforest in Borneo (Whitehead et al., 2010). After sunrise, the boundary layer deepened on average from less than 200 up to 800 m as a result of solar heating (Fig. 3g). Consequently, particles in the residual layer aloft (Fisch et al., 2004;Rissler et al., 2006), which were not impacted by the fog, were mixed down to the surface, leading in part to the observed increases in both number and volume concentrations. Such mixing of particles from the residual layer in the morning had been observed previously in the Amazon basin during the dry season (Rissler et al., 2006). The formation of secondary organic aerosol (SOA) as a result of photochemical oxidation (in both gas and particle phases) likely contributed to the increase in volume concentration Martin et al., 2010a;Pöschl et al., 2010).

Air masses impacted by urban pollution
Air masses arriving at the T3 site frequently had passed over urban and industrial areas upwind. When the air mass was influenced by the urban pollution, κ CCN and its dispersion exhibited clear diel variations (Fig. 4). The value of κ CCN was lower during the night at 0.15, and it increased from the early morning hours, peaking at a value of 0.19 around noon (LT, UTC −4 h). The dispersion was anticorrelated with κ CCN , exhibiting higher values (i.e., increased heterogeneity in particle chemical composition) during night and a minimum value around noon. To a large degree, the diel trend of κ CCN was due to the variation of κ org . The value of κ org was lower during night at 0.10 and increased to 0.16 at noon. The increase in κ org is consistent with the variation of O : C, which increased during the early morning and reached the highest value of 0.8 around noon time. The pollution strongly affected the particle number and volume concentrations, both exhibiting similar diel trends. Under polluted conditions, particle number concentration ranged on average from 1500 to 2300 cm −3 , which is an increase by a factor of ∼ 5 from that under background conditions. In comparison, the increase in the volume concentration was only about a factor of 2 (i.e., from a range of 0.3-0.75 to 1.0 to 1.3 µm 3 cm −3 ), as the urban pollution is dominated by Aitken-mode particles that make a relatively small contribution to aerosol mass and volume concentration.
The diel variations of κ CCN , its dispersion, κ org , and O : C are explained as follows. During night, particles in freshly  (Pöhlker et al., 2016). The top-right panel includes κ derived from particle growth factor measurements in July 2001, during a "recent biomass burning period" of the CLAIRE-2001 study (Rissler et al., 2004), and from 11 September to 8 October 2002, during the dry period of the LBA-SMOCC (Rissler et al., 2006). emitted pollution, which are dominated by primary organic aerosol (POA) and have low hygroscopicity, are mixed with more aged particles within a shallow nocturnal boundary layer (Bateman et al., 2017). In the absence of photochemical oxidation and aging, this external mixture leads to higher dispersion of particle hygroscopicity as well as overall lower O : C and κ org . As the pollution aerosols are mainly from isolated point sources, they are confined in the shallow nocturnal boundary layer during night, and the residual layer above the T3 site is expected to consist of aged background aerosols. Therefore, unlike under background conditions, the mixing of aerosol aloft in the residual layer down to the surface cannot by itself explain the increase in particle number and volume concentration during the day, both of which were substantially above the background values. These increases under polluted conditions might be due to the stronger urban influence at T3 during the day. The strong increase in CN concentration at 16:00 UTC (12:00 LT) could be caused by the arrival at the T3 site of the Manaus plume emitted during early morning traffic hours. To a large degree, the increases in O : C and κ org are due to the formation and aging of SOA in the pollution, while the development of the daytime boundary layer, which leads to dilution of pollution and mixing with aged particles from the residual layer, can also contribute to the increases. The condensation of secondary species and photochemical aging also leads to more homogenous composition among particles (Mei et al., 2013a), and therefore lower dispersion of κ CCN , as was observed. The O : C reached a maximum average value of 0.8, similar to that under the background conditions. This suggests that the formation and photochemical aging of SOA quickly led to highly oxygenated organic compounds (i.e., within several hours; de Sá et al., 2017b).

Dry season aerosol
Background conditions Figures 5-7 show the diel variations of the aerosol properties observed during the dry season for background, urban pollution, and local biomass burning air masses, respectively. During the dry season, the background aerosol (Fig. 5) is strongly influenced by regional biomass burning, and air masses arriving at the T3 site often pass through urban and industrial areas along the Amazon River and in northeast Brazil , indicating that the background aerosol is also impacted by more aged urban and industrial emissions  . Despite different aerosol sources and processes, the particle hygroscopicity, dispersion, and κ org exhibited similar values as those of background aerosol during the wet seasons and a lack of obvious diel variations. This is also consistent with the absence of a significant diel trend of κ CCN observed at T0a (ATTO) during the dry season (Fig. 8). The O : C value increased slightly from 0.8 during night to 0.9 in the afternoon, possibly due to further oxidation and aging of background aerosols during the daytime. The high value of O : C is consistent with the relatively high value of κ org (0.15) and is close to that observed under background condition during the wet season, indicating highly oxygenated organic aerosol. The number and volume concentrations were lower just before dawn and increased during the early morning, again a result of wet scavenging of particles by fog followed by the mixing of aerosol aloft in the residual layer down to the surface as the boundary layer develops in the morning.

Air masses impacted by urban pollution and local biomass burning
For urban pollution (Fig. 6) and local biomass burning (Fig. 7) air masses during the dry season, κ CCN (urban pollution: 0.12-0.20; local biomass burning: 0.10-0.17), its dispersion (urban pollution: 0.4-1.0; local biomass burning: 0.4-0.9), κ org (urban pollution: 0.10-0.15; local biomass burning: 0.08-0.14), and O : C (polluted: 0.7-0.85; biomass burning: 0.7-0.86) showed similar values and diel variations as those under polluted conditions during the wet season. This is consistent with the picture that freshly emitted particles (in either the case of urban pollution or local biomass burning) lead to overall lower O : C, κ org , and higher κ CCN dispersion during night, followed by increases in O : C, κ org , and a decrease in the dispersion during daytime, which are mainly driven by the formation and photochemical aging of SOA, with contributions from the mixing of background aerosol aloft in the residual layer down to the surface and dilution of fresh emission as the boundary layer develops. Compared to urban pollution, local biomass burning air masses exhibited lower κ CCN and κ org values during night and stronger diel variations. In the afternoon, κ org and O : C reached high values of 0.14 and 0.86, respectively, as observed for background aerosols. For urban pollution air masses, aerosol number and volume concentrations showed similar trends for both seasons. The increases in number and volume concentration from early morning to noon were similar, about 1000 cm −3 and 0.5 µm 3 cm −3 , respectively, for both wet and dry seasons. The percentage increases were less pronounced in the dry season due to the higher background values. In contrast to urban pollution, local biomass burning showed higher aerosol number and volume concentrations at night and decreased during the morning. Local biomass burning activities typically peaked during evening hours, consistent with frequent classification of the nighttime aerosol as local biomass burning ( Fig. 7i; Vestin et al., 2007). Despite the wet removal of particles by fog, the strong emission from local biomass burning, largely confined within the shallow nocturnal boundary layer, led to higher surface aerosol concentrations than those in the residue layer aloft, which likely represented the regional background. As the boundary layer deepened in the morning, the mixing with aerosol from the residual layer led to decreases in both aerosol number and volume concentration observed at the surface ( Fig. 7e and f).

Hygroscopicity of PMF factors and the variation of organic hygroscopicity with oxidation level
The hygroscopicities associated with the AMS PMF factors were estimated through multivariable linear regression using different subsets of the data as well as the entire dataset for each of the two IOPs (IOP1 in Fig. 9 and IOP2 in Fig. 10). The different subsets included measurements during day, night, certain sampling periods, and ranges in particle hygroscopicity dispersion. Comparison of the hygroscopicities derived from the different subsets of data allowed us to examine the robustness of this approach. Uncertainty in the derived κ for individual factors was determined by the number of points available to fit in the time series, with greater data coverage and therefore lower uncertainty during the dry season. For the wet season (IOP1), the hygroscopicities associated with PMF factors derived using different subsets of the data are largely in agreement with those derived from the entire dataset. There are notable differences between the hygroscopicities of MO-OOA and Fac91 factors derived using data under background conditions only and those derived using the entire dataset. Such difference could be partially due to the limited data under the background conditions during IOP1 (Fig. 3). For the dry season (IOP2), the hygroscopicities of PMF factors derived using measurements under background conditions or data with hygroscopicity dispersion less than 0.4 are quite different from those derived using other data subsets and the entire dry season dataset. The agreement among the PMF factor hygroscopicities derived using different sub-datasets during the wet season and the disagreements for the dry season are attributed to the applicability of the underlying assumption that the bulk volume fractions of PMF factors (i.e., derived from MS mode data) represented those at the sizes of CCN measurements. For the wet season, the average f 44 was largely independent of particle size from 130 to 400 nm (Fig. S10), which is the size range that dominated the bulk aerosol mass concentration measured by AMS. This is consistent with the assumption that the bulk volume fractions of the PMF factors represent those at the two CCN sizes (142 and 171 nm). For the dry season, the f 44 averaged over local biomass burning air masses and the entire IOP2 exhibited an increase with particle diameter from 100 to 300 nm (Fig. S10). For periods with hygroscopicity dispersion less than 0.4 or under background conditions, the average size distribution of f 44 was noisier due to fewer data points. Nevertheless, the size distribution shows f 44 was largely independent of the particle size under these conditions, consistent with the assumption described above. In the following discussion, we focus on the PMF factor hygroscopicities de-rived using all data during the wet season and under background conditions in the dry season.
The MO-OOA factors for the two IOPs exhibit very similar O : C and κ values. The O : C values were 1.19 and 1.24, and κ values were 0.20 and 0.21, for IOP1 and 2, respectively ( Table 1). The O : C and κ values are consistent with those of some typical SOA compounds, such as malonic acid, which has an O : C value of 1.33 and a κ value of 0.23 (Kumar et al., 2003), and succinic acid, which has an O : C value of 1 and a κ value of 0.23 (Hori et al., 2003). For the LO-OOA and IEPOX-SOA factors, the hygroscopicities vary between the two IOPs. The κ values of IEPOX-SOA were 0.18 and 0.08 during IOP1 and IOP2, respectively, and the κ values of the LO-OOA factor varied from 0.12 to 0.20 between IOP1 and IOP2. The difference in κ may be partially due to the change of O : C values of the factors derived for the two IOPs. The difference in SOA precursors and therefore composition in LO-OOA (Ng et al., 2010) may also contribute to the difference in its κ values between the two IOPs. The variation of IEPOX-SOA κ between the two IOPs could be a result of the different RH conditions, which may strongly influence the composition of IEPOX-SOA (Riva et al., 2016).
During the wet season, a factor with high contribution from m/z = 91 was identified. The Fac91 factor correlates with several tracers for anthropogenic emissions, including NO x , benzene, toluene, trimethylbenzene (TMB), and xylenes, but not high NO x isoprene products (e.g., methylglyceric acid; de Sá et al., 2017b). This factor likely represents SOA formed from aromatics emitted from urban areas, possibly combined with a mixture of freshly oxidized biogenic compounds within the urban-influenced air (de Sá et al., 2017b). The Fac91 factor has a hygroscopicity value of 0.10 and a much lower O : C ratio of 0.328 compared to those of MO-OOA and LO-OOA.
The less-oxidized organic factors identified by the PMF analysis were HOA for both IOPs, BBOA for IOP1, and fresh and aged BBOA for IOP2. These factors represent primary OA, except that the aged BBOA of IOP2 likely included contributions from oxidized POA or SOA. The hygroscopicity of the HOA factors was fixed as zero in the multivariate regressions. All BBOA factors have a distinctive m/z = 60 peak and correlate with biomass burning traces including levoglucosan and vanillin (de Sá et al., 2017b). The retrieved hygroscopicity values for the BBOA factors are substantially lower than those of SOA factors, especially for the fresh-BBOA factor during IOP2. The extremely low hygroscopicity suggests that the fresh BBOA, likely produced by local fires, behaves very similar to HOA in terms of CCN activation despite a substantially higher O : C. Figures 11 and 12 show that for SOA factors, including IEPOX-SOA, LO-OOA, MO-OOA for both IOPs, and Fac91 for IOP1, the κ value increases with increasing O : C, and the variation of κ with O : C agrees with the linear relationship derived from laboratory studies of SOA CCN activities (Lambe et al., 2011). The low hygroscopicities of the HOA and the BBOA factors, which are below the linear relationship for SOAs, are also consistent with laboratory results of POA and oxidized POA (Lambe et al., 2011). Cerully et al. (2015 derived κ of LO-OOA, MO-OOA, and IEPOX-SOA factors from data collected in the southeastern US during the Southern Oxidant and Aerosol Study (SOAS). A different name, isoprene-OA, was used for the IEPOX-SOA factor in Cerully et al. (2015), as, while this factor is mainly attributed to SOA formed from IEPOX uptake, it might not be entirely due to IEPOX (Schwantes et al., 2015;Xu et al., 2015a, b). The O : C values calculated using the Improved-Ambient method (Canagaratna et al., 2015) (Pöhlker et al., 2016) and at the T3 site under background conditions (this study) during the (a) wet season (April and May 2014) and (b) dry season. The size-resolved CCN data at T0a was collected by stepping the particle size at given CCNC supersaturations (Rose et al., 2008). Data displayed for T0a are averaged over critical particle diameters ranging from 44 to 175 nm, while the T3 data are averaged from measurements at 112, 142, and 171 nm.
zon2014/5. While the IEPOX-SOA factors identified using different datasets share many similar features, they are not identical and can consist of different groups of compounds. Such differences may be due to varying degrees of oxidation in different environments between the two field campaigns. For comparison with earlier field studies, the values of κ org and O : C were averaged according to the hours of the day over particle diameters of 142 and 171 nm for data under polluted conditions during IOP1 and all data during IOP2 (Figs. 11 and 12). For the 1 h diel averages, the slope of κ org vs. O : C, derived through a bivariate least squares fit (i.e., orthogonal distance regression), is steeper than that derived from laboratory studies of SOA hygroscopicity, especially during IOP2. This steep slope during IOP2 is consistent with the results from earlier field studies (Mei et al., 2013b), although there is a clear offset between the . Hygroscopicity of AMS PMF factors for IOP1 (i.e., wet season) retrieved by multilinear regressions using all data (red square), data from UTC 12:00-24:00 (cyan circle), data from UTC 00:00-12:00 (blue triangle), data under background conditions (red triangle), data under polluted conditions (red diamond), and data with hygroscopicity dispersion σ κ CCN /κ CCN less than 0.6 (black diamond).  Figure 10. Hygroscopicity of PMF factors for IOP2 (i.e., dry season) retrieved by multilinear regressions using all data (red square), data from UTC 12:00-24:00 (cyan circle), data from UTC 00:00-12:00 (blue triangle), data with strong influence from local biomass burning (red right triangle), data under background conditions (red upside-down triangle), data from 21 August to 14 September 2014 only (red left-pointing triangle), data from 15 September to 15 October only (brown diamond), and data with a dispersion (σ κ CCN /κ CCN ) < 0.4 (black diamond). two relationships. The O : C ratios from Mei et al. (2013b) and Lambe et al. (2011) were scaled by a factor of 1.27 to account for changes in the method of calculating the O : C ratio (Improved-Ambient, Canagaratna et al., 2015), while all O : C values from this work were calculated using the Improved-Ambient method. This offset between the field studies may be partially due to the different precursors of the SOA for the campaigns, with a higher anthropogenic volatile organic compound fraction expected for CalNex and CARES, which took place near Los Angeles and Sacramento, respectively. In addition, biomass burning represented a much smaller fraction of the organics during CalNex and CARES (Mei et al., 2013a, b Figure 11. The variation of PMF factor hygroscopicity, 1 h diel average of organic hygroscopicity, and O : C ratio at 142 and 171 nm for urban pollution air masses. Also shown are the relationships between κ org and O : C reported by earlier field and laboratory studies. which the increase in O : C was mainly driven by oxidation. As POA exhibits hygroscopicity values well below the linear fit between SOA hygroscopicity and O : C, mixtures with different POA and SOA fractions lead to a steeper slope for the increase in κ org with O : C, as shown by the results from this and previous field studies (Mei et al., 2013b).

Conclusions
Size-resolved CCN spectra at five particle diameters ranging from 75 to 171 nm were characterized downwind of Manaus, Brazil, in central Amazonia for a period of 1 year from 12 March 2014 to 3 March 2015 during GoAmazon2014/5. For each season, the air masses arriving at the site were classified into different types, including background, urban pollution, and local biomass burning. During the wet season, the background air mass represented near-natural conditions, at times with impact from anthropogenic emissions, while, in the dry season, the background was dominated by regional and long-distance biomass burning aerosol particles. Polluted air masses represented those with strong influence from urban emissions, which were mostly from Manaus. The local-biomass-burning type describes those strongly influenced by local (i.e., fresh) biomass mass burning activities that dominate the impact from urban pollution, if any. Particle hygroscopicity, mixing state, and organic hygroscopicity were derived from size-resolved CCN activation fraction and concurrent aerosol composition measurements. The monthly mean κ CCN exhibits the lowest values during the dry season, largely due to lower κ org when aerosol was often strongly influenced by local biomass burning. The κ CCN increased with particle size during all seasons, consistent with decreasing organic volume fraction with increasing particle size. Under background conditions, the value of κ CCN and its size dependence were largely consistent among different seasons, despite the very different aerosol sources. During the dry season, aerosols classified as urban pollution and local biomass burning exhibited lower κ org values compared to background aerosols, contributing to the lower values of overall κ CCN . Under background conditions during both wet and dry seasons, the largely constant diel trends of κ CCN and κ org suggest little variation in particle composition throughout the day. The constant κ org of ∼ 0.15 is consistent with the lack of a diel trend in f 44 and O : C. The high values of f 44 and O : C indicate that the aerosols under background conditions are dominated by the aged regional aerosol particles consisting of highly oxygenated organic compounds. When the air mass is influenced by urban pollution or local biomass burning, κ CCN , κ org , f 44 , and O : C exhibit clear diel variations. The value of κ CCN (0.1-0.2) is lower during the night and increases from the early morning hours, peaking around noon (LT, UTC −4 h). This diel trend of κ CCN is largely driven by the variation in κ org (0.08-0.15), consistent with the variation of O : C. The dispersion of κ CCN is anticorrelated with κ CCN , exhibiting higher values during night and a minimum value around noon, indicating an increased heterogeneity in particle chemical composition during nighttime. These diel variations for air masses strongly influenced by urban pollution and local biomass burning indicate that, during the night, freshly emitted particles, dominated by POA and with low hygroscopicity, are mixed with more aged particles within a shallow nocturnal boundary layer. In the absence of photochemical oxidation and aging, this external mixture leads to higher dispersion of particle hygroscopicity as well as overall lower O : C and κ org . The increases in O : C and κ org during daytime are driven by the formation and aging of SOA and dilution of POA emissions into a deeper boundary layer, while the development of the boundary layer, which leads to mixing with aged particles from the residual layer, likely also contributes to the increases.
The hygroscopicities associated with individual PMF organic factors were derived through multivariable linear regression. For the SOA factors, κ increases within increasing O : C, and the variation of κ with O : C agrees well with the linear relationship derived from laboratory studies of SOA hygroscopicity (Lambe et al., 2011). The low hygroscopicity of HOA and the BBOA factors, which are below the linear relationship, are also consistent with laboratory results of POA and oxidized POA (Lambe et al., 2011). In contrast, the slope of κ org (i.e., overall organic hygroscopicity) vs. O : C is much steeper when compared to that derived from laboratory studies of SOA hygroscopicity, especially for IOP2. Such difference is due to the increase in O : C being driven primarily by oxidation in laboratory SOA studies, while the variation in O : C of ambient organics is to a large degree due to the variations in volume fractions of POA and SOA factors, which have very different O : C values. As POA factors show hygroscopicity values well below the linear fit between SOA hygroscopicity and O : C, mixtures with different POA and SOA fractions lead to a steeper slope for the increase in κ org with O : C, as shown by the results from this and earlier field studies (Mei et al., 2013b).
Data availability. The datasets used in this publication are available at the ARM Climate Research Facility database for the GoAmazon2014/5 experiment (https://www.arm.gov/research/ campaigns/amf2014goamazon).