Samples of atmospheric particles were collected aboard the U.S. Department of Energy Gulfstream aircraft (G-1). Flight patterns were flown between Terceira Island (38∘45′43′′ N, 27∘5′27′′ W) and Graciosa Island (39∘3′12′′ N, 28∘7′26′′ W), Portugal, and within 20–30 km of Graciosa Island (J. Wang et al., 2021). Flight plans were based on the projected meteorological conditions from various global forecast modes including Monitoring Atmospheric Composition and Climate, Global Forecast System, and European Centre for Medium-Range Weather Forecasts (ECMWF). A subset of collected samples was selected for analysis based on synoptic conditions (identifying DI vs. non-DI periods) and altitudes (clear MBL and FT layers) for each day. Samples analyzed were collected during the second intensive operation period of ACE-ENA, on the dates of 19 January 2018, 21 January 2018, 24 January 2018, 25 January 2018, 26 January 2018, 28 January 2018, 30 January 2018, 1 February 2018, 8 February 2018, 11 February 2018, 15 February 2018, 16 February 2018, and 19 February 2018​​​​​​​. These dates were selected due to unique transport episodes associated with the sampling periods. DI days we identified objectively using the Lagrangian analysis tool (LAGRANTO) version 2.0 (Sprenger and Wernli, 2015) and wind field data obtained from the ECMWF interim reanalysis (ERA-Interim) available 6-hourly, interpolated to 1∘ × 1∘ horizontal grid resolution, and at 60 vertical hybrid levels (Dee et al., 2011). DIs were identified by a systematic calculation of forward trajectories at altitudes higher than 600 hPa, while the DI trajectories were identified based on the vertical descent of the air masses. For a trajectory to be termed a DI, their pressure must increase (i.e., descend in altitude) by at least 400 hPa in 48 h (Raveh-Rubin, 2017). If such a DI trajectory is found within a 3∘ radius around Graciosa, the date is considered to be “DI”. In addition, backward trajectories for each sampling period were calculated for the end points at relevant flight altitudes (Figs. S1 and S2) using the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (Stein et al., 2015; Rolph et al., 2017). Atmospheric data from ERA-Interim are additionally analyzed for the atmospheric column at Graciosa, namely potential temperature, equivalent potential temperature, potential vorticity, and boundary layer height. The latter is diagnosed in ERA-Interim using the critical bulk Richardson number upon its first passing of the threshold 0.2 when scanning from the surface upwards (ECMWF, 2007).

The G-1 aircraft is equipped with sensor modules to deliver precise real-time inertial measurement, GPS, meteorological, and turbulence data such as position, altitude, temperature, pressure relative humidity, and three-dimensional winds. For particle collection, the G-1 was equipped with an isokinetic aerosol inlet, from which ambient aerosol was transported to individual instruments. Particle samples were collected using a custom-built time-resolved aerosol collector (TRAC) that autonomously collected particles on substrates at pre-set time intervals (Laskin et al., 2006). The TRAC is a single-stage impactor with an aerodynamic cutoff size (D50%) of 0.36 µm (Laskin et al., 2003) coupled to a rotating disk that can hold up to 160 samples. The disk was preloaded with microscopy substrates (carbon Type-B film on 400-mesh copper grids, Ted Pella, Inc.). The sampling was performed at a single spot on the center of each substrate for 7–10 min, depending upon the flight. After each flight, sample disks were taken off the TRAC, plated, and hermetically sealed prior to transport. Once samples were received in the lab, the sample grids used for the analysis were removed from the sealed plate and transferred into grid boxes stored at room temperature and dry conditions in a desiccator cabinet. Online measurements of aerosols aboard the G-1 include a passive cavity aerosol spectrometer 100X probe (PCASP, Dp=0.1–3.0 µm, 1 Hz resolution) and a fast integrated mobility spectrometer (FIMS, Dp=0.01–0.5 µm, 1 Hz resolution), which provided size distributions and concentrations of ambient particles (Kulkarni and Wang, 2006; Wang et al., 2018). During all research flights, a Nafion dryer reduced the relative humidity of the airstream in the sampling line. A CCN counter (Droplet Measurement Technologies) measured the concentration of particles that activate at a supersaturation of 0.14 %. A high-resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS) was deployed on board to characterize bulk non-refractory aerosol composition (i.e., organics, sulfate, ammonium, and chlorine) (DeCarlo et al., 2006; Zawadowicz et al., 2021). The particle size distributions and CCN concentrations were analyzed when the liquid water content was below 0.001 g m3 to avoid periods when cloud-shattering artifacts could influence the sampled particles (Korolev et al., 2011). The liquid water content was obtained by integrating the droplet size distributions measured by a fast cloud droplet probe (FCDP; Droplet Measurement Technologies).

Additional information on the sampling conditions is presented in Table S1 of the Supplement and incudes sampling time and date, average sampling altitude, boundary layer height, particle concentration, and wind speed. The boundary layer height was calculated based on potential temperature measurements collected for each flight. The boundary layer is limited by a well-defined temperature inversion, resulting in a maximum value of the temperature gradient as a function of height (Stull, 1988). A summary of each flight (altitude and aerosol particle concentration vs. time) with the collection times highlighted is shown in Figs. S3 and S4. Guided by meteorological analysis and wind field data to identify DI periods, we performed offline microscopy analysis of collected particle samples across different atmospheric layers and transport episodes during the ACE-ENA campaign.

Morphology and elemental analysis of individual particles were performed using computer-controlled scanning electron microscopy coupled with energy dispersive X-ray spectroscopy operated at 20 kV (CCSEM/EDX; FEI Quanta 3D, EDAX Genesis). During CCSEM/EDX analysis particle samples were systematically imaged, and particles larger than 100 nm are recognized. Of note is that the particle size reported from CCSEM/EDX analysis is defined as the area equivalent diameter (AED, µm), which is based on fitting a circle with an area equivalent to the particle's 2D projected image. This is followed by an automated acquisition of individual EDX spectra for each particle (Laskin et al., 2005). EDX spectra with sufficient X-ray counting statistics (40–1500 photonss-1) were then processed to quantify relative atomic fractions of 15 elements: C, N, O, Na, Mg, Al, Si, P, S, Cl, K, Ca, Mn, Fe, and Cu. The EDX peak of Cu is heavily influenced by a background signal from the copper TEM grid and the sample holder made of beryllium–copper alloy. Therefore, quantified atomic fractions of Cu were excluded from particle-type classification of the analyzed particles. Two independent methods were employed for the particle-type grouping and classification: (1) k-means clustering and (2) rule-based particle classification. The k-means clustering is an unsupervised machine-learning algorithm designed to group similar datasets without user intervention (Rebotier and Prather, 2007; Moffet et al., 2012). The second approach for the categorization of particles utilizes a series of user-defined rules to separate analyzed particles into groups of typical elemental contribution (Laskin et al., 2012). For this work, the k-means clustering was used as a primary method for particle-type classification, while the rule-based approach was used as a complementary method to build confidence in the identification of different particle types. Details of the classification schemes are provided in the Supplement (Figs. S5 and S6) and in previous works (Moffet et al., 2012; Tomlin et al., 2020).

Scanning transmission X-ray microscopy with near-edge X-ray absorption fine-structure (STXM/NEXAFS) spectroscopy was used to elucidate the chemical mixing state of individual particles based on the NEXAFS spectral data acquired at the carbon K-edge (278–320 eV) (Hopkins et al., 2007; Moffet et al., 2010b, c). STXM/NEXAFS was performed at the synchrotron facilities on beamlines 11.0.2.2 and 5.3.2.2 in the Advanced Light Source, Lawrence Berkeley National Laboratory, and on beamline 10ID-1 at the University of Saskatchewan Canadian Light Source. STXM instrument operation is similar in both locations as described elsewhere (Kilcoyne et al., 2003). Briefly, a set of raster scan STXM images at each of the pre-set energy levels was acquired from a synchrotron monochromated incident light focused on the sample using a Fresnel zone plate. The transmitted light is detected at each of the energy settings, and spectra of individual particles could then be reconstructed based on the Beer–Lambert law from the intensity of transmitted light over the projection area of particles compared to the particle-free regions. The recorded intensity at each energy setting (E) across individual pixels was converted into optical density (ODE) as follows: 1ODE=-ln⁡I(E)I0(E)=μρt, where I(E) is the intensity of light transmitted through a particle, I0(E) is the intensity of incoming light (determined as intensity of light in the particle-free areas), μ is the mass absorption coefficient, ρ corresponds to the density, and t is the thickness of a particle. Sequences of STXM images are acquired at closely spaced energies of I0(E) to record a “stack” of images. Then, NEXAFS spectra from individual pixels of detected particles are extracted from the stack (∼96 energies over 278 to 320 eV range, 30–35 nm spatial resolution, 1 ms dwell time).

In addition, faster acquisition of STXM images at four key energies of 278 eV (pre-edge), 285.4 eV (C = C), 288.5 eV (-COOH), and 320 eV (post-edge) (15 × 15 µm, 30–35 nm spatial resolution, 1 ms dwell time) was employed to construct “maps” of individual particles using image processing methods reported in our earlier studies (Moffet et al., 2010a, 2013, 2016; Fraund et al., 2017). Briefly, a series of thresholds were used to identify the mapping components including inorganics (IN), organic carbon (OC), and soot–elemental carbon (EC). The total carbon (TC) was calculated as the difference between the carbon post-edge and pre-edge OD (TC = OD320eV-OD278eV). IN-rich regions were defined with pixels having an OD278eV/OD320eV ratio greater than 0.5. OC regions are those with abundant features corresponding to the carboxylic acid functional group (-COOH), defined by the difference between the intensity of the -COOH peak and carbon pre-edge peak greater than 0 (i.e., OD288.5eV–OD278eV>0). Finally, EC areas are identified by comparing the value of the sp2 / total carbon to that of highly oriented pyrolytic graphite (HOPG) according to (OD285.4eV/TC)×(ODHOPG, TC/ODHOPG,C=C)>0.35, which indicates extensive sp2 bonding of carbon corresponding to graphitic-like components (Hopkins et al., 2007).

Research flights were conducted under different synoptic conditions to allow for the characterization of common aerosols, trace gases, clouds, and precipitation. Figure 1a illustrates the typical flight pattern of the G-1 aircraft, which includes multiple legs at different altitudes while maneuvering perpendicular and along the wind direction. These patterns allowed for the full profile of aerosol and cloud layer along the MBL and lower FT altitudes. Figure 1b shows daily time series from 1 January to 28 February 2018 in relation to DI events identified from ERA-Interim reanalysis. The marked black dots indicate DI air masses within a 3∘ radius from 39∘ N, 28∘ W (i.e., the ENA site). The high frequency of the black dots (i.e., vertical distribution) indicates an increase in trajectories that satisfy the DI criterion at different pressure altitudes. For example, on 24 January 2018 in Fig. 1b, we see a series of DI air parcels (black dots) at different pressure altitudes ranging from 611 to 2360 m a.m.s.l. found to be below or above the boundary layer as indicated by the dashed red line. Guided by the frequency of the DI air masses, we selected a subset of the time-tagged particle samples for analysis by the complementary microscopy techniques as summarized in Table S1. To evaluate the consistency of the sources and long-range transport trajectories, we calculated back trajectories using the HYSPLIT model (Stein et al., 2015; Rolph et al., 2017). Figure 1c shows results of representative HYSPLIT 72 h back-trajectory calculations for the research flight on 24 January 2018, which identifies long-range transport of an air mass originating from North America. Trajectories were calculated every 6 h from 13:00 UTC on 24 January 2018 to 12:00 UTC on 22 January 2018 at three starting altitudes: 100, 2000, and 3000 m. This process was repeated with the same HYSPLIT input meteorological parameters for all research flights utilized in this work, as shown in Figs. S1 and S2.

A total of 38 particle sample grids from 13 (out of 19) research flights were analyzed. First, CCSEM/EDX analysis was carried out to characterize the particle-type composition typical of different synoptic scenarios. Figure 2 shows the results of the size-segregated particle-type population (right column) obtained from k-means clustering analysis of ∼36400 individual particles with the backscattering-mode scanning electron microscope (SEM) imaging of a representative subset of particles (left column), separated between MBL versus FT flight altitudes and between synoptic conditions of DI and non-DI sampling periods. The onboard FIMS instrument measurement provided particle size distribution data in a range of 0.01–0.5 µm. By superimposing the CCSEM/EDX particle analysis data with the FIMS size distribution data, we can approximate the representative composition and number concentration of potentially CCN active particles (>0.1 µm) in the MBL and FT during non-DI and DI periods, respectively. Note that the error bars in the particle number concentration indicate variation in the particle size distribution values averaged across different days and synoptic conditions. Also, comparison of AED and FIMS sizes needs to be considered with caution because particle flattering on the substrate results in overestimated AED sizes compared to more realistic FIMS values. Here, the AED-based particle distributions are scaled to match the Y axis of FIMS data and therefore to provide visual illustration of the chemical makeup of CCN particles.

The k-means algorithm identified four key clusters termed as carbonaceous, ammonium nitrates–sulfates, mixed sea salt, and aged sea salt based on the mean elemental contribution (Fig. S5). Note that the element fraction values obtained from individual EDX spectra were filtered to remove values less than 0.5 %. Carbonaceous is the dominant type and represents the majority of analyzed particles. It is defined based on the sole contributions of C and O elements in the particle EDX spectra. The second-most abundant cluster is the ammonium nitrates–sulfates, to which the contributions of N, O, and S are greater than 1 %. The aged sea salt and mixed sea salt clusters contain similar elemental signatures, with the latter containing significant amounts of refractory elements typical for sea salt and mineral dust including Mg, Cl, K, Ca, Mn, and Fe.

First, we compared the change in particle-type population among samples in the MBL during non-DI and DI periods. The fraction of carbonaceous particles within the MBL contributed around 68 % in non-DI samples and decreased to 43 % in DI samples. Organic aerosol in the remote MBL has been suggested to originate from VOCs such as isoprenes, monoterpenes, formic acid, nitrogenized, and aliphatic organics released from biological activities near the sea surface, which undergo oxidation reactions leading to SOA formation (Facchini et al., 2008; Dall'Osto et al., 2012; Mungall et al., 2017). The lower fraction of carbonaceous particles during DI periods is counterbalanced by the increase in inorganics shifting from 32 % (non-DI periods) to 57 % (DI periods). Here, we operationally defined inorganics as the sum of mixed sea salt (4 %), aged sea salt (20 %), and ammonium nitrate–sulfate (33 %), which in fact may also contain organic carboxylic acids as components of aged sea salt. Regardless, shifting focus to the comparison of FT samples during non-DI and DI periods, we found that background carbonaceous particles contribute around 92 % (non-DI periods), decreasing to 46 % (DI periods). Similar to MBL observations, the shift in carbonaceous contribution can be attributed to an increase in inorganic influence during DI events changing from 8 % (non-DI periods) to 55 % (DI periods). We observed that most of the inorganic influence originates from ammonium nitrate and sulfate, contributing between 32 % and 33 % during DI periods regardless of sampling altitude (MBL vs. FT). Both carbonaceous particles and ammonium nitrate–sulfate can originate from ocean biological activity or anthropogenic sources. Typically, over marine areas, sulfate aerosol forms from oxidation of dimethyl sulfide (DMS), a common gas species emitted by biota. Sulfates are major components of accumulation-mode particles in the remote marine environment (Sanchez et al., 2018; Korhonen et al., 2008). Nitrate in marine particles can also come from vertical mixing in the ocean that surges nitrate-rich deep waters to the surface, followed by the aerosolization through wave motion (Zehr and Ward, 2002). However, the elevated contribution of ammonium nitrates and sulfates during the DI periods suggests likely influence from anthropogenic emissions originating from North America. Inorganic aerosols such as ammonium nitrates and sulfates are predominantly formed from the condensation of atmospheric precursors such as SO2, NH3, HOx, and NOx, which are common components of biomass burning emissions, urban areas, and agriculture activities among others (Reff et al., 2009). A study utilizing regional chemical models has found that the mass enhancements in inorganic aerosol can reach 23 % of carbonaceous enhancements as biomass burning processes accelerate secondary formation of inorganic aerosols (Souri et al., 2017). Uptake of S- and N-containing acidic species, as well as soluble organic acids, onto the preexisting sea salt particles modifies their composition through acid-displacement reactions that can be expressed in the general form of (Finlayson-Pitts, 2003; Laskin et al., 2012) 2*NaCl(aq)+HA(aq,g)→*NaA(aq)+HCl(aq,g), where *NaCl denotes sea salt, and HA represents atmospheric water-soluble acids (e.g., HNO3, H2SO4, CH3SO3H, and carboxylic acids). These reactions release the volatile HCl(g) product, leaving particles depleted in chloride and enriched in corresponding HA(aq) salts. Related to this acid-displacement chemistry, mixed sea salt and aged sea salt particle types were identified by the k-clustering analysis as illustrated in Fig. S5. The mixed sea salt particles contain key components of seawater (i.e., Na, Mg, and Cl; atomic fractions of Na and Cl >10 % with characteristic ratio of Cl/Na ∼0.6) and minor fractions (<2 %) of additional elements (e.g., Ca, Mn, Fe, Al, and Si), suggesting internal mixing of relatively fresh sea salt with other inorganic components without extensive chloride depletion. The other cluster of aged sea salt particles shows significant fractions of Na (∼10 %) but with substantially lower ratios of Cl/Na <0.1, which indicates chloride depletion (Fig. S5) due to atmospheric aging. Atomic fractions of C and N elements in this type of particle are much higher than those in the mixed sea salt cluster, while the fraction of S is much smaller. These observations suggest that in this geographical region acid-displacement reactions in the aged sea salt particles are mostly driven by water-soluble carboxylic acids (common components of SOA) (Laskin et al., 2012) and nitric acid (Finlayson-Pitts, 2003), while contributions by sulfonic or sulfuric acids are minor during the wintertime. Based on the k-means clustering, fractions of mixed sea salt range from 0.5 % to 3 %, while fractions of aged sea salt are overall more populous and range between 0.1 % and 20 % across all investigated samples. Additionally, both the aged sea salt and mixed sea salt cluster groups include minor contributions of Al and Si, indicative of possible mixing with mineral dust transported from the long-range continental sources.

To better discriminate particle-type groups according to their composition and the acid-displacement chemistry identified through the k-means clustering, a supplemental rule-based classification was performed using previously published definitions of particle-type classes common in marine environments (Laskin et al., 2012; Tomlin et al., 2020). Results of the particle-type characterization utilizing the rule-based assessment of their elemental composition (assigned into five major classes) are presented in Fig. S6. The applied rule-based classification scheme distinguishes among particle types common in the remote marine environment of sea salt, sea salt–sulfate, carbonaceous–sulfate, carbonaceous, and other (Fig. S6). For each sample, 600–3000 particles were analyzed, depending on particle loading on the substrates. The size-resolved particle-type classification identified using the rule-based schematic was overlaid on the acquired FIMS size distribution as shown in Fig. S7. Similar to the k-means clustering breakdown, we first compared the impact of DI events in MBL samples. Significant fractions of carbonaceous and carbonaceous–sulfate particles were identified in the background MBL samples, amounting to 86 % (non-DI periods) and decreasing to 49 % (DI period). Furthermore, the combined fraction of sea salt and mixed sea salt–sulfate is substantially smaller at around 10 % (non-DI periods) to 21 % (DI period). Fractions of uncategorized “other” particles contribute around 30 % (DI period), while only having a minimal contribution of 4 % during non-DI events. In contrast, background FT samples were dominated by carbonaceous and carbonaceous–sulfate, contributing as much as 95 % (non-DI periods) then decreasing to 55 % (DI period). Unlike the MBL samples, there was only minimal change in larger sea salt + mixed sea salt–sulfate from 2 % (non-DI period) to 4 % (DI period). However, the reduction in the carbonaceous and carbonaceous–sulfate contribution among FT samples during DI periods is associated with the large change in the “other” fraction shifting from 4 % (non-DI period) to 41 % (DI period). Based on the mean elemental composition of the “other” category, this group contains a combination of dust, sea salt, and carbonaceous components, suggesting extensive internal mixing of particles consistent with long-range transport (Froyd et al., 2019). This finding is also consistent with the k-means clustering results that indicated elevated contributions of particles with inorganic components during the DI periods. Overall, the particle-type fractions identified by both the k-means clustering and the rule-based classification schemes are consistent across all samples, suggesting that the mixing state population significantly changes from heavily organic-dominated to a mixture of inorganic–organic particle-type distribution, resulting in the observation of more complex particle compositions during DI periods.

Relative contributions of the particle-type fractions among separate DI events show substantial variability between different flights and MBL versus FT altitudes (Fig. S8). Furthermore, the dominant carbonaceous particle-type groups identified by CCSEM/EDX elemental analysis may exhibit significant differences in the spectral characteristics of carbon bonding, indicative of its long-range transport from North America during the DI periods. Furthermore, a previous study tracked the origin of air masses transported over long distances across the Atlantic Ocean to the Azores utilizing the Lagrangian Flexible Particle (FLEXPART) dispersion model to show detailed spatial resolution of air masses across different locations and altitudes (China et al., 2017). The influence of North American emissions on distant remote regions is well documented, with occurrences of continental pollutant transport events accompanied by strong influence from urban city emissions from Boston, Toronto, Detroit, and Chicago (Owen et al., 2006). On the other hand, extensive boreal wildfires in northern North America release large amounts of trace gases and aerosols into the atmosphere, which then can be transported to other remote regions including North America (Val Martín et al., 2006). In particular, boreal wildfires emit around 10 % of the annual anthropogenic aerosol black carbon in the Northern Hemisphere (Bond et al., 2004). The eastward transport of North American emissions begins as hot plumes of biomass burning emissions from wildfires rapidly rise to high altitudes (∼8 to 13 km a.g.l.) under favorable conditions (Zhu et al., 2018; Yu et al., 2019; Kloss et al., 2019). These plumes can be lofted into a warm conveyor belt preceding a cold front from an associated cyclone, which is followed by the entrainment of a cold descending airstream (from the same cyclone) that ultimately results in the air parcels containing continental emissions reaching the lower altitudes of the eastern North Atlantic (Owen et al., 2006; Zheng et al., 2020b). The transported aerosol undergoes substantial atmospheric aging through photochemical reactions (Hems et al., 2021), gas–particle partitioning (Vakkari et al., 2018), and coagulation (Ramnarine et al., 2019) processes as it travels across the Atlantic Ocean and descends into the MBL during the DI events.

Results of the elemental microanalysis of particles presented above provide statistics on broad particle classes identified and show the significant contribution of organic-dominated particles in the region well. However, CCSEM/EDX analysis is limited in providing detailed information on the carbon speciation within individual particles and other metrics of particle internal composition (chemical mixing state). To investigate chemical differences in the carbon components of particles we employed STXM/NEXAFS spectromicroscopy methods, which provide spatially resolved carbon bonding speciation and differentiate between EC and OC regions within individual particles (Moffet et al., 2010a, c). It is also worth mentioning that the definitions of carbonaceous particles identified by CCSEM/EDX and described in the previous section is somewhat different from OC particles defined by STXM/NEXAFS. The former corresponds to the distribution of organics across a population of particles (i.e., external mixing), while the latter is related to the multi-component internal heterogeneity of individual particles (i.e., internal mixing). Figure 3a shows an illustrative carbon K-edge map of individual particles from one of the DI period samples (the cumulative map of all ∼4300 particles from all samples analyzed in this study is included in the Supplement, Fig. S9). The carbon K-edge composition map distinguishes three main components based on the spectral information (Moffet et al., 2010a) as described earlier: IN (blue), OC (green), and EC (red). Each pixel within an individual particle may contain either single or multiple components (i.e., components can overlap) that are grouped to yield five typical classes based on the internal mixing between OC, EC, and IN components: (1) IN, (2) OC-EC-IN, (3) OC-EC, (4) OC-IN, and (5) OC. The size-resolved histograms of these five classes superimposed with the onboard particle size distribution data measured by FIMS are shown in Fig. 3b to highlight the organic and inorganic contributions within individual particles as a function of particle size. A mixture of organic and inorganic particles (OC-IN) appears to be the dominant class across all samples, contributing 40 %–76 % to the total particle population. Furthermore, the consideration of multiple sources of EC from wildfires (Park et al., 2007), residential wood smoke (Allen and Rector, 2020), agricultural burning (Liu et al., 2016; Holder et al., 2017), and urban emissions (Paredes-Miranda et al., 2013) in North America led us to expect a large contribution of EC within our sample. However, OC-EC and OC-EC-IN particles contributed only 0.4 %–1.3 % to the total particle population. EC–soot lifetime is primarily governed by its wet deposition rate, which is dependent on the particle's affinity to absorb water (Barrett et al., 2019). Freshly emitted soot particles are hydrophobic; however, atmospheric processes can increase the hydroscopicity properties of soot particles through the accumulation of OH-initiated oxidation of organics during long-range transport and atmospheric aging (Dzepina et al., 2015), leading to decreased atmospheric lifetime of EC regardless of initial composition (Khalizov et al., 2013; Browne et al., 2015; China et al., 2015). IN particles (i.e., inorganics such as sea salt and sulfates) appear to be consistent with the particle-type observations inferred from CCSEM/EDX data. Singe-component IN particles contribute up to 15 % in the MBL at the time of non-DI periods, while their contribution during DI decreases to ∼0.8 %. Subject to long-range transport, IN-dominant particles also accumulate substantial OC components when encountering DI, and as they entrain into the MBL and create ensembles of ambient particles with complex multi-component internal mixing states through different atmospheric processes such as condensation (Mozurkewich, 1986) and coagulation (Holmes, 2007). Consistently, fractions of single-component OC particles within the MBL during DI periods increased (from 7 % to 22 %) and slightly decreased in the FT layer (from 26 % to 20 %). These observations suggest that entrainment of aerosols with higher extents of internal mixing (from long-range transport) are present in the MBL and can contribute to the regional aerosol composition, which in turn may modify aerosol–cloud interactions typical for the area.

NEXAFS spectra (285–294 eV) of individual particles were used to assess the carbon chemical bonding environment allowing, us to identify representative types of OC-containing particles (Moffet et al., 2010a). Figure 4 shows the representative NEXAFS spectra acquired for 103 individual carbon-containing particles. This resulted in the identification of six carbon “types”, as shown along with their illustrative secondary electron-mode SEM imaging. Each carbon type is classified based on characteristic spectral features such as peak positions and relative intensities. For all spectra shown in Fig. 4a, the individual contribution of carbon energy transitions was quantified via spectral deconvolution. Details on the deconvolution process are described in previous works (Moffet et al., 2010b, 2013; Tomlin et al., 2020). Figure 5a shows the deconvolution fit of the averaged NEXAFS spectra for each carbon type identified across different sampling conditions, with Fig. 5b illustrating the contribution of each functional group based on the individual peak area. It is worth noting that the difference in absorption between the post-edge (OD320eV) and pre-edge (OD278eV) energies is a measure of the amount of total carbonaceous material in the particles.

The type 1 (biological) class has some contribution from alkene groups (C* = C at 285.4 eV) with significant enhancement of aliphatic hydrocarbons (C*–H at 287.7 eV) and alcohol groups (C*–OH at 289.5 eV). These spectra appear to be similar to the reported NEXAFS spectrum for phospholipids, a constituent of cell walls (Lawrence et al., 2003; Nováková et al., 2008). Lipid material is concentrated in the sea surface microlayer through the rupturing of phytoplankton cell membranes (i.e., cell lysis) (X. Wang et al., 2015​​​​​​​). A majority of lipid compounds produced by phytoplankton in seawater include glyceroglycolipids, phospholipids, and triacylglycerols containing significant amounts of aliphatic and alcohol groups (Harwood and Guschina, 2009). The transition of aliphatic-rich organic species into the aerosol phase is governed by the bursting of bubble films (Blanchard, 1989) enriched in lipid organic species found on the surface of seawater (X. Wang et al., 2015). The type 2 (homogeneous organic particles) class has almost equivalent peak contributions from each reported functional group as shown in Fig. 5b. The NEXAFS spectrum for type 2 is quantitatively similar to those reported for organic particles from anthropogenic emissions in urban areas of Mexico City (Moffet et al., 2010b) and central California (Moffet et al., 2013). As the aerosol plume is transported away from the source of emission, organic mass increases, while the fraction of C = C decreases (Doran et al., 2007; Kleinman et al., 2008; Moffet et al., 2010b). As a result, organic functional groups build up with particle age such as carboxylic acids, carbonyl, alcohol, and other carbon–oxygen functional groups. It has been suggested that formation of these homogeneous organic particles likely results from the accumulation growth of primary emitted particles as they travel further away from their emission source (Moffet et al., 2010b). Type 3 (soot) had the largest contribution of C* = C at 285.4 eV spectral feature (42 % of peak area contribution). Based on reported literature, this spectrum is comparable with atmospheric particles collected during various field studies of biomass burning emissions (Hopkins et al., 2007). Interestingly, particles collected during aircraft measurements during the Aerosol Characterization Experiment in Asia (ACE-Asia) campaign (Maria et al., 2004) from emissions over mixed combustion sources had nearly identical percentage of sp2 values of around 41 % (Hopkins et al., 2007).

Field and laboratory studies showed that sea salt particles can react with atmospheric water-soluble organic acids, leading to chloride depletion within particles (Laskin et al., 2012; B. Wang et al., 2015). Consistent with these previous studies, fresh sea salt typically has an intact rectangular inorganic core with a carbon outer shell arising from a thin layer of carboxylic acid coating as indicated by the peak for R(C* = O)OH at 288.5 eV. Accordingly, type 4 is referred to as “fresh sea salt” in this work. In addition, the minor quantity of carbonaceous material in type 4, as inferred from the small difference between the post- and pre-edge energies (OD320eV–OD278eV) apparent from Fig. 5a, further supports the observation of freshly emitted sea salt particles. In contrast, type 5 (aged sea salt–organics) is made up of sea salt particles that have reacted with carboxylic acid components of organic aerosol condensate, which results in a substantial contribution of the R(C* = O)OH at 288.5 eV peak while retaining the carbonate peak C*O3 at 290.4 eV. Of note is that type 5 (aged sea salt–organics) contains significantly more carbon mass than type 4 (fresh sea salt–organics), as indicated by the NEXAFS spectrum. Finally, the type 6 (K-dominated) class is identified based on the appearance of characteristic potassium peaks at 297.1 eV (K*L2) and 299.7 eV (K*L3), with a percent contribution of ∼51 % relative to the total peak area. Potassium-salt particles are common markers of biomass burning smoke (Andreae, 1983; Li et al., 2003). Large fractions of KCl particles are commonly emitted from both flaming and smoldering fires, while atmospheric aging can transform them into K2SO4 and KNO3 through multi-phase acid-displacement reactions similar to those of NaCl (Li et al., 2003). However, these K-dominated particles can also be released as mixed secondary particles containing fractions of organic species, methylsulfonic acid, trimethylamine, SO42-, NH4+, and K from potential biogenic sources in oceans (Willis et al., 2017).

Organic volume fraction (OVF) is a practical parameter to assess reactivity (Worsnop et al., 2002; Folkers et al., 2003) and hygroscopicity (Wang et al., 2008; Schill et al., 2015; Ruehl et al., 2016) of mixed inorganic–organic particles. Based on the STXM/NEXAFS measurements of individual particles, OVF is defined as a ratio of the optical thickness of the organic components (torg) divided by the total optical thickness of the particle (torg+tinorg) (Moffet et al., 2010a; Pham et al., 2017; Fraund et al., 2019). STXM images collected at the carbon K-edge were used to calculate the OVF. The values of absorbance at the pre-edge (278 eV) and the post-edge (320 eV) energies are related to the inorganic mass and the sum of inorganic + organic mass, respectively. Assuming specific values for densities (ρ) and mass absorption coefficients (μ) for the organic and inorganic components, values of torg and tinorg can be determined, allowing OVF calculation (Fraund et al., 2019). For this study, we assumed that the inorganic component of particles corresponds to (NH4)2SO4 based on the particle elemental composition identified by CCSEM/EDX analysis, while oxalic acid (C2H2O4) is used as a proxy for the organic component. Oxalic acid was chosen to represent biomass burning (Yamasoe et al., 2000) and vehicular exhaust (Kawamura and Kaplan, 1987). Of note, based on previous reported studies, is that assumptions of chemically different organic components have a minor effect on the resulting OVF values, while choice of the inorganic components resulted in a larger variation in the OVF calculations (Pham et al., 2017; Fraund et al., 2019). Here, we estimate the systematic error in OVF when assuming different inorganic–organic components, as shown in Table S2. Assuming NaCl to be the inorganic component instead of (NH4)2SO4 yields a difference of ∼35 %. On the other hand, assuming the organic component to be oxalic acid yields a ∼5 %–30 % difference in OVF when compared to other organics such as sucrose, adipic acid, and glucose.

Figure 6 shows representative chemical mixing state maps and OVF values of particles sampled during different atmospheric transport episodes during this study. Particles appear to have a varying amount of organic coating for different sampling episodes as shown on the OVF maps. The comparison of the OVF map and the carbon speciation map illustrates overlap between the two mapping schemes. Finally, histograms show particle fractions at varying OVF values during different atmospheric transport episodes. Layers of organics are seen encapsulating inorganic cores. As expected, background particles collected in the MBL show inorganic NaCl cores (as indicated by a rectangular core morphology) with modest organic coating (OVF <30 %), consistent with a previous report (Chi et al., 2015). However, during the DI periods, the majority of particles have equal or greater fractions of organic-to-inorganic components (40 %–60 % OVF), while only a few particles exhibit core–shell morphology typical for background particles (i.e., non-DI periods). Furthermore, FT particles during non-DI periods have OVF <10 % when compared to FT samples during DI periods (10 %–20 % OVF). In general, samples collected at the FT altitudes show reduced OVF values compared to the MBL samples regardless of the occurrence of DIs. Core–shell particle morphologies were also observed in FT samples, although not frequently (see Fig. S9). FT samples were dominated by inorganic–organic particles in the size range of 0.20–0.25 µm, which are likely mixed sulfate–organic particles based on the size-resolved particle-type datasets obtained from CCSEM/EDX analysis. A recent study conducted in central Oregon found that the organic mass fraction from FT samples was between 27 % and 84 %, while sulfate mass fractions ranged from 39 %–50 % (Zhou et al., 2019). Based on these reported studies the elevated contributions of organics and sulfate in the FT may be attributed to the enrichment of organonitrates and organosulfate compounds originating from biogenic sources in the absence of wildfire influence. However, FT organic and sulfate aerosol mass is also known to be associated with urban and biomass burning emissions (Bahreini, 2003; Dunlea et al., 2009; Roberts et al., 2010; Y. Wang et al., 2021). Studies in the northeastern Pacific found that submicron aerosol mass was dominated by sulfate and organic components originating from aged Asian pollution plumes (Dunlea et al., 2009). FT organic and sulfate particles can then experience long-range transport and aging as the air parcels are carried across the Atlantic and descend into the MBL of the ENA site (China et al., 2017). To summarize, we observe enhancements in the OVF values of individual particles during the DI periods, quantified as 2.06 ± 0.16- and 1.11 ± 0.04-fold increases in OVF for the MBL and FT samples, respectively, assuming (NH4)2SO4–oxalic acid components. The larger total OVF in the MBL (relative to FT samples) regardless of DI events is most likely due to the additional contribution of marine organic sources within the boundary layer. The background organic concentration in the MBL is different than FT due to other sources of organics such as dissolved organic matter on the seawater surface (Doval et al., 2001; Miyazaki et al., 2018). The transport of organics from the ocean surface directly into the atmosphere is primarily driven by turbulent winds (O'Dowd et al., 2004; Prather et al., 2013), resulting in the enhancement of the background organic concentration in the MBL. Furthermore, the observed enhancements in OVF in the MBL during DI periods could be the result of organic-rich air parcels (originating from North America) descending from the FT into the MBL, leading to changes in the total organic concentration (Zheng et al., 2020b; Y. Wang et al., 2021).

CCN activity of individual particles is governed by both their size and chemical composition. In particular, condensation of organic carbon onto atmospheric inorganic particles can impact the efficiency at which particles of mixed organic–inorganic composition can act as CCN and INPs due to changes in particles' hygroscopicity and viscosity (Beydoun et al., 2017; Ovadnevaite et al., 2017; Altaf et al., 2018). To account for the effects of organics on aerosol hygroscopicity, we use the κ-Köhler equation (Petters and Kreidenweis, 2007) to estimate the hygroscopicity parameter κ corresponding to mixed inorganic–organic particles​​​​​​​: 3κ=(1-forg)κinorg+forgκorg, where forg is the OVF value derived from the STXM data, κorg=0.1 is the hygroscopicity of the organic component, and κinorg=0.6 is that of (NH4)2SO4 (Petters and Kreidenweis, 2007). We derived κ values for different synoptic and atmospheric layer conditions using the size-resolved OVF ratio shown in Fig. S10, and we found that κMBL, DI=0.41 and κFT, DI=0.33 for DI periods and that κMBL, non-DI=0.48 and κFT, non-DI=0.36 for non-DI periods. The lower κ values under DI periods are consistent with enhancements in the organic contribution. Of note is that the values of κ obtained here using Eq. (2) need to be considered the low-limit values, which might be somewhat higher considering possible contributions from more hygroscopic components of particles related to original and aged sea salt (κNaCl=1.3 and κNa2SO4=0.8).

Using κ, we can calculate the critical size of a dry particle (Fig. S11) that can be activated under the supersaturation of 0.14 % (setting of the CCN counter deployed on G-1) (Petters and Kreidenweis, 2007). The theoretical CCN number concentrations are then estimated by integrating the FIMS-measured aerosol size distributions above the critical dry particle diameter. Figure 7 shows the results for the OVF-based calculations of theoretical CCN concentrations compared to the onboard CCN measurements at 0.14 % supersaturation. There is general agreement between calculated and measured CCN concentration, but FT cases appear to have better agreement. The uncertainty in the calculated CCN is due to the supersaturation fluctuation of the CCN counter (0.13 %–0.15 %), as shown in Fig. S11. The large error bars in the measured CCN are a result of the variability of the measured CCN during different sampling periods. We also note that the exact value of κorg may play a role in affecting the CCN calculation. So, theoretical CCN concentrations were also calculated using κorg=0, and the results are compared against the measured CCN concentrations in Fig. 7b. However, the impact of this change in κorg does not significantly change the agreement between the calculated and measured CCN concentrations. This result shows that calculating the CCN concentration using OVF values derived from the STXM data and the κ-Köhler theory can be a good estimate of the actual CCN concentrations.