A DMT passive cavity aerosol spectrometer probe (PCASP) X2 configured in an aviation pod with heated inlet was mounted at the Vasco top mast approximately 10 m above the water surface to provide optical measurements of dry-particle size distributions between approximately 125 nm and 3 µm. Additional aerosol instrumentation was located in a forward locker, below and slightly behind the aerosol inlet. Sampled air was fed to the locker via a 3 cm diameter, 4 m long inlet located next to the PCASP. Although the inlet was not aspirated, several high-flow-rate nephelometers sampling from the inlet ensured low residence time in the sample line. Further details and additional instrumentation are discussed in Reid et al. (2016). A size-resolved CCN system sampled air from the inlet just inside the instrument locker with an approximately 2 m, 0.635 cm diameter copper tube. A URG cyclone with an approximate 1 µm 50 % size cut was used to remove the largest particles from this CCN sampling line, including coarse-mode sea salt, to minimize wear and corrosion on downstream components. The sample was then dried using a Perma Pure poly-tube Nafion dryer with low-pressure sheath air to relative humidity (RH) values below 30 %, as verified by occasional in-line checks using a handheld Extech hygro-thermometer.

Approximately 1.1 L min-1 was drawn through the size-resolved CCN system, which was comprised of an X-ray neutralizer (TSI Model 3087) and a TSI 3080 electrostatic classifier with a long differential mobility analyzer (DMA) column (TSI Model 3081) for quasi-monodisperse particle sizing, preceded by a 0.071 cm orifice (0.69 µm 50 % cut point diameter) impactor. The DMA was operated with a sheath flow rate of 5 L min-1 and sample flow rate of 1.1 L min-1. The sample flow was then split between a TSI 3782 water-based condensation particle counter (CPC) with a flow rate of 0.6 L min-1 and a DMT cloud condensation nuclei counter (CCNc) with a flow rate of 0.5 L min-1. Flow rates used to calculate number concentrations were calibrated using a Gilibrator (Models 800285 & 800286) system, with in-line measurements conducted prior to each CCNc supersaturation calibration session as noted below.

The size-resolved CCN system measured CN and CCN (activated particles at a CCNc set point supersaturation) concentrations in each of 30 quasi-monodisperse size bins between 17 and 500 nm. The CCNc was operated at five temperature gradient settings and calibrated using ammonium sulfate (following the methods described by Petters et al., 2009) to measure the corresponding maximum environmental supersaturation within the CCNc column. The scan of all 30 size bins at each supersaturation took approximately 15 min, while a complete measurement over all five supersaturation settings took approximately 2 h due to pauses between settings while column temperatures stabilized. The measured CN and CCN particle counts were inverted using the methodology of Petters et al. (2009). The inversion yielded the dry ambient aerosol size distribution over the measured range (dN/ dlog10Dp for 17 ≤ Dp≤500 nm) and the equivalent distribution of CCN particles activated at each supersaturation (dCCN / dlog10Dp). The activated fraction spectrum was then calculated as the fraction of total particles that formed droplets (CCN/CN) at each diameter. Each activated fraction spectrum was then fit using a three-parameter fit similar to the approach of Rose et al. (2010). The diameter at which 50 % of particles in the fit had activated (D50) was used to calculate the associated hygroscopicity parameter, κ (Petters and Kreidenweis, 2007). Full calibration of the CCN system flow rates and supersaturations took 2 to 5 h to complete and was therefore conducted four times throughout the study on 15, 20, 27, and 29 September to limit measurement downtime. Each calibration session involved running a calibration scan at each CCNc temperature gradient setting (with two full scans conducted at each setting on 27 September), yielding a total of five calibrations per setting throughout the cruise. Calibrated supersaturation set points and their respective standard deviations were 0.14 % ± 0.01, 0.38 % ± 0.01, 0.52 % ± 0.01, 0.71 % ± 0.02, and 0.85 % ± 0.03, with no significant trend or calibration drift noted during the cruise. The measured range of 0.14 to 0.85 % supersaturation was selected based on values that could both be reliably measured by the CCNc instrument and represented supersaturations expected in the region where aerosol effects may be relevant, ranging from marine stratocumulus with peak supersaturations often below 0.2 % to highly convective clouds with supersaturations above 1 % (Reutter et al., 2009; Ward et al., 2010; Tao et al., 2012).

As the SCS environment tended to have relatively few particles smaller than 50 nm, only the measurements at the 0.14 and 0.38 % supersaturation settings had complete activation curves that spanned the measured particle diameter range. For the higher supersaturation settings, the D50 diameter tended to occur at small diameters with low CN and CCN concentrations, thereby increasing uncertainty in the associated κ values. As a result, κ values were only reported for the 0.14 and 0.38 % supersaturation settings. In addition, rather than continuously running a full 2 h scan across all supersaturation settings, individual scans (approximately 15 min) were run more often for the 0.14 and 0.38 % settings to take advantage of this outcome. The range of κ values measured for the particles active at supersaturations of 0.14 and 0.38 % was typically between about 0.3 and 0.8, although the full range was between 0.2 and 1.1 (Fig. 2e, f). The D50 diameters for these hygroscopicity values spanned approximately 96 to 150 nm (κ range: 0.8–0.2) for the 0.14 % supersaturation setting, and 45 to 80 nm (κ range: 1.1–0.2) for the 0.38 % setting. The hygroscopicity measurements were therefore more indicative of accumulation mode hygroscopicity during the 0.14 % scans and Aitken mode hygroscopicity during those at 0.38 %. Additionally, particles outside these size ranges were not well quantified by these measurements.

Additional measurements of aerosol composition were used to validate identified source types impacting the measurements throughout the cruise. A series of PM2.5 filters were collected by 5 L min-1 MiniVol Tactical Air Samplers with sampling periods that varied between 1 and 2.5 days, and were analyzed for elemental concentrations by gravimetric, X-ray fluorescence (XRF), and ion chromatography methods, and organic and black carbon concentrations by the thermal-optical methods (Reid et al., 2016). Trace gas concentrations were measured intermittently throughout the cruise by whole-air gas samples for gas chromatography analysis, as discussed further in Reid et al. (2016).

A suite of weather-monitoring instruments was located on a 3 m bow mast to provide coincident meteorological measurements throughout the study. From this suite, wind speed and wind direction measurements from a Campbell sonic anemometer were used to identify gust front passage. An OTT Parsivel disdrometer was utilized to measure precipitation, from which only the rain rate measurements were used in this analysis.

Several remote-sensing and model products were used to characterize the wider SCS atmospheric environment and to identify potential aerosol sources. Moderate Resolution Imaging Spectroradiometer (MODIS) visible and infrared (IR) products were used to identify convection and squall line propagation across the SCS. The MODIS Collection 6 MOD08 Level 3 daily aerosol optical depth (AOD) products were utilized for AOD measurements in the region, though cloud cover obscured measurements throughout much of the study. MODIS active fire hot spot analysis and the Fire Locating and Modeling of Burning Emissions (FLAMBE) smoke flux product from Terra and Aqua were used to identify the locations and times during which fires were burning in the MC (Giglio et al., 2003; Reid et al., 2009; Hyer et al., 2013). Simulations from the Navy Operational Global Atmospheric Prediction System (NOGAPS) model were used to represent surface and 700 hPa winds, interpolated to 1∘ spatial resolution and averaged over the study period, to provide an estimate of typical aerosol transport pathways (Hogan and Rosmond, 1991; Xian et al., 2013). Finally, the Navy Aerosol Analysis and Prediction System (NAAPS) was used to predict smoke and sulfate aerosol mass concentrations at the surface along the Vasco ship track (Lynch et al., 2016).

The NOAA Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) Version 4.9 model (Draxler, 1999; Draxler and Hess, 1997, 1998) was used to generate daily 72 h back-trajectories (spawned at 0 Z, 8 AM local) from the Vasco location with arrival heights of 500 m to indicate likely marine boundary layer transport patterns. The Global Data Assimilation System (GDAS1) 1∘×1∘ HYSPLIT meteorological dataset was used to drive the model. Trajectory paths were found to be largely influenced by synoptic-scale changes in the regional meteorological state of the atmosphere, with no substantial differences due to arrival time of day. Arrival heights between 100 and 3000 m were examined. Trajectories with arrival heights below 1000 m were generally consistent and representative of boundary layer transport (Atwood et al., 2013; Xian et al., 2013), while higher heights tended to be increasingly influenced by free-tropospheric transport pathways with a more westerly component. As such, 500 m was selected to be representative of general shifts in synoptic-scale boundary layer transport pathways, though more complex vertical interactions and mixing from aloft are a potential influence in the region (Atwood et al., 2013). Trajectory lengths of 72 h were found to be sufficient to demonstrate general transport path differences between ocean-dominated regions of the central portion of the SCS and more terrestrially influenced regions that passed closer to Borneo and Sumatra.

The dN/ dlog10Dp dry-particle size distributions obtained every 15 min from the CCN system were first normalized by the particle number concentration summed over all bins. Each of these normalized size distributions was then parameterized by fitting a lognormal mixture distribution using an algorithm based on Hussein et al. (2005). This algorithm fit each distribution to one, two, or three lognormal modes, each defined by three lognormal distribution parameters (median diameter, geometric standard deviation, and fractional number concentration). Two modes were identified as the best fit for 690 of 731 data points in this dataset. In order to have a common point of comparison for classification purposes, the two-mode fit was used for all data points, yielding a parameterized Aitken and accumulation mode for each 15 min data point. Data points for which the fitting algorithm selected an additional third mode are noted in the clustering results in Sect. 3 and Supplement Fig. S1.

Parameters associated with each data point were then used to cluster the data points into groups based on similar observed aerosol properties. Cluster analyses have long been used to group observed aerosol size distributions into clusters of generally similar size distributions (Tunved et al., 2004), which can then be associated with various sources or atmospheric processes that shaped them (Charron et al., 2008; Beddows et al., 2009; Wegner et al., 2012). Similar cluster analyses have been utilized to classify aerosol types based on particle chemistry (Frossard et al., 2014), with Frossard et al. (2014) identifying clusters in marine aerosol observations associated with marine and non-marine aerosol types. As pointed out by those authors, the clustering approach can be superior to algorithms using simpler criteria to distinguish “clean” from “polluted” conditions, as more variables and a measure of similarity between data points are used to find the underlying population types. In this study normalized size distribution parameters were combined with total number concentration and particle composition information via hygroscopicity measurements to serve as input variables for the cluster analysis (parameters given in Table 1). As hygroscopicity data were available for at most one mode (during the 0.14 or 0.38 % supersaturation scans), Aitken and accumulation mode hygroscopicities were treated as missing for data points without this information. In order to account for missing data and adjust all clustering variables to the same scale, each variable was first standardized to a mean of 0 and standard deviation of 1, with missing data points imputed to a value of 0 (the mean value). As a result, the clustering distance function was insensitive to missing data but still included information on hygroscopicity when available.

A hierarchical cluster analysis was first conducted using the cluster.AgglomerativeClustering class of the Python scikit-learn package (Pedregosa et al., 2011) using the Ward linkage to help ascertain the number of clusters that can be found in the dataset. Each step in this process involved merging two data points or clusters into a new cluster based on those points with the shortest distance between normalized input measurements (Karl Pearson Euclidian distance function; Wilks, 2011). A dendrogram and associated measure of the distance between merged clusters for each subsequent clustering step was used to identify potential numbers of clusters appropriate for the dataset. The distance between merged clusters increases at steps that merge substantially different clusters (Wilks, 2011), in this case indicating 5, 8, 9, and 12 clusters as potentially appropriate for this data set.

A nonhierarchical k-means cluster analysis was then conducted for each of these four potential cluster numbers using the scikit-learn cluster.KMeans class to refine the cluster members. The appropriate number of clusters was selected based on the k-means result with the least number of clusters that maintained physically distinct and temporally consistent aerosol populations for the associated clusters. In particular, as the time stamp of a data point was not included in the cluster analysis, clusters with smaller numbers of data points were considered distinct if they all occurred during a narrow time frame that could be associated with a transient atmospheric phenomenon. The time frames of all clusters were then compared against other aerosol and meteorological observations to ensure they were physically meaningful. The result was a set of eight identified aerosol population types with associated time periods corresponding to the 15 min CCN system data points. Finally, size distribution and hygroscopicity measurements were averaged for all time periods associated with each population type.

The daily positions of the Vasco during the 2 weeks in September 2012 that comprise this study are shown in Fig. 1a, together with HYSPLIT 72 h back-trajectories initiated within the MBL. Several extended periods at the same anchorage are noted by the range of dates spent at these locations. The daily fire hot spots are also indicated in this figure. Average NOGAPS surface winds in the boundary layer were from the southwest, often advecting air parcels from near Borneo, while lower free-tropospheric winds, such as those at 700 hPa, were more westerly due to the generally veering structure of the lower atmosphere in the SCS (Reid et al., 2016). Fires detected by MODIS occurred throughout much of Borneo and southern Sumatra during the study, with surface-level trajectories near the start and near the end of the study period passing close to active fires, whereas those during the middle period remained primarily over open ocean. Results from the NAAPS model, along with limited satellite AOD measurements not obscured by clouds during this period, confirmed this general smoke transport pathway. Accumulation mode aerosol mass concentration estimates (Fig. 1b) were initially generated from the PCASP measurements using a density of 1.4 µg m-3 (Levin et al., 2010), assumed to be representative of a combination of smoldering peat and agricultural fire emissions typical in the MC (Reid et al., 2012) that constituted the largest plumes observed during the study (Reid et al., 2016). Coincident model estimates generated by NAAPS along the ship track indicated generally similar results, with the highest mass concentrations occurring early and late in the measurement period, in general agreement with times during which back-trajectories passed over terrestrial sources and active fires. Air parcels advecting into the SCS and Sulu Sea during this period that originated from areas further to the north and west were cleaner than those from other sectors due to fewer emission sources and more precipitation along the trajectories. Changes in the observed particle concentrations also occurred on timescales shorter than these weekly large-scale variations, as shown in the aerosol observation time lines in Fig. 2a–c. Many of these higher-frequency fluctuations were associated with squall line passages and heavy local precipitation, as discussed further below. The time line of dN/dlogDp size distributions, as measured by the CCN system and shown in Fig. 2a, indicates that most of the particle number concentration fell within the 17–500 nm measurement range, except possibly during the highest-concentration periods. A more extensive comparison of model and satellite measurement in situ observations is discussed in Reid et al. (2016).

The cluster analysis was first conducted to investigate potential aerosol population types in the dataset, followed by physical interpretation of the results against cluster aerosol properties, coincident measurements, and meteorological conditions. The parameter values input to the cluster analysis are shown for each data point and variable in Fig. 3, and colored by cluster number for the results of the eight-cluster k-means analysis. The average value and intra-cluster standard deviation for each cluster parameter and cluster are given in Table 1. Normalized size distributions for each of these eight aerosol populations are shown in Fig. 4; the average CN and CCN number concentrations and hygroscopicities are given in Table 2. Equivalent normalized volume distributions are shown in Supplement Fig. S2. The cluster number associated with each measurement is similarly shown as the background color in Fig. 2 and marker color in Fig. 5. The aerosol properties, meteorological conditions, and likely transport pathways associated with data points in each cluster were then used to provide a physical interpretation of the results and identify each population type on the basis of its likely sources as discussed below. Clusters 1–4 were the most commonly found (representing 85 % of the total observations, Table 2), while clusters 5–8 represented special cases, generally of short duration, that could be identified by specific locations or sampling conditions.

Background marine: data points associated with this cluster occurred throughout the study, typically following rain in the vicinity of the Vasco or transport from areas further removed from terrestrial regions. In addition, this type was observed following shortly after periods associated with each of the other identified clusters, often appearing as a transition between other types (Fig. 2). The measured properties of this population type were similar to the background marine aerosol reported in many prior studies (Hoppel et al., 1994; Jensen et al., 1996; Brechtel et al., 1998; O'Dowd et al., 1997; Heintzenberg et al., 2004; Allan et al., 2009; Good et al., 2010). The population featured a bimodal size distribution with a Hoppel minimum near 90 nm. The inner quartile range (IQR: middle 50 % of observations between the 25–75 % percentiles) of number concentrations ranged from 382 to 623 cm-3, with on average 42 % of the total number concentration residing in the accumulation mode as specified by the bimodal fit. Modal hygroscopicities were found to average 0.65 for the accumulation mode and 0.46 for the smaller Aitken mode, while activated fractions were generally moderate across the range of measured supersaturations as compared to other identified population types (Table 1). Each of these findings further reinforced the classification of this cluster as a typical background marine aerosol.

Precipitation: this distribution was found during periods immediately following extensive precipitation at or near the Vasco (Fig. 2d). Air masses had been substantially scrubbed of particles, and accumulation mode particles had been preferentially removed. While the number concentrations of large-mode particles were lower than those in the background marine periods, the number concentrations of smaller particles, particularly those below 40–50 nm, were comparable to the background marine type. The longest contiguous period of this type occurred on 14 September immediately following the passage of a squall line observed in the satellite visible and IR products (not shown) that left a clean air mass with fewer than 200 cm-3 measured in the 17–500 nm range in its wake. Number concentrations tended to be lower than the background marine type with an IQR: from 227 to 441 cm-3. Hygroscopicities were similarly lower than the background marine population with κ values of 0.54 and 0.34 for the accumulation and Aitken modes, respectively. Total CCN concentrations across all supersaturations were found to be lower than in background marine air masses due to the combination of fewer total particles, generally smaller particle sizes, and lower hygroscopicities.

Smoke: data points associated with this aerosol type occurred primarily in two events on 14 and 25–26 September, during which back-trajectories were at their furthest south, near burning regions in Borneo (Fig. 1a). Normalized size distributions indicated that particles were largely concentrated in a single accumulation mode with a tail of smaller particles. This type was associated with the highest total particle number and estimated submicron mass concentrations observed during the cruise, with the exception of measurements taken in the urban plume of Puerto Princesa. The standard deviations in the normalized size distribution parameters for the dominant accumulation mode in this population (Fig. 4, Table 1) were small, even while number concentration varied widely (IQR: 1802 to 2780 cm-3; 81 % accumulation modal fraction). Accumulation mode hygroscopicities were lower than either the background marine or precipitation types, with average κ values of 0.40. Aitken mode hygroscopicities showed the opposite behavior from the first two population types with higher κ values of 0.54, though the measured uncertainties (Fig. 2e and f) and standard deviations were considerably higher than for the accumulation mode (0.25 and 0.03, respectively). Interestingly, activated fractions were highest among all population types across the full range of measured supersaturations, owing to the large number fraction of particles in the accumulation mode, while CCN concentrations were the highest of all types (except those measured in port) due primarily to the larger total particle number concentration in these smoke plumes.

Accumulation mode lognormal median diameters around 200 nm with a tail of smaller particles, elevated concentrations of carbon monoxide and benzene, and potassium in filter samples during this period (Reid et al., 2016, and Fig. 2c) were all consistent with expectations for aged biomass burning smoke (Yokelson et al., 2008; Akagi et al., 2011; Reid et al., 2015; Sakamoto et al., 2015). Additional examination and attribution of this event to biomass burning in Sumatra and Borneo are discussed further in Reid et al. (2016). Finally, while smoke is considered the dominant aerosol source during these periods, anthropogenic pollution may still have been co-emitted along the transport path and contributed to measured results.

Mixed marine: this population was characterized by periods during which the background marine type mixed with other sources of aerosol. Most of the data points associated with this type had transport pathways and biomass burning sources similar to those for the smoke population type, but with number concentrations and size distribution parameters between those of the background marine and smoke types (IQR: 782 to 1160 cm-3; 68 % accumulation modal fraction), indicating there was insufficient smoke for it to dominate the properties of the marine background. Accumulation and Aitken mode hygroscopicities of 0.48 and 0.54, along with activated fractions and CCN concentrations, were similarly indicative of mixing between smoke and background marine sources.

While periods of smoke mixing with a background marine air mass appeared to constitute the majority of data points in this cluster, several other periods point to other phenomena of interest being included in this type, perhaps indicating this cluster was relatively more complex than other population types. Short-lived intrusions (2 to 5 h) of accumulation mode particles were regularly observed in both the CCN system and PCASP datasets (e.g., 18–23 Z on 22, 23, and 24 September), after which the size distributions quickly returned to background marine conditions. These excursions were largely constrained to the pre-dawn hours (sunrise occurs around 22 Z), when the boundary layer was thinnest and when precipitation was occurring in the vicinity of the Vasco. Several prior studies have shown that smoke and anthropogenic pollution aerosol within the wider MC region can be lofted into and transported in the lower free troposphere (Tosca et al., 2011; Robinson et al., 2012; Zender et al., 2012; Campbell et al., 2013; Atwood et al., 2013). The influence of a free-tropospheric aerosol layer as a source of MBL aerosol and CCN has been identified in other remote oceanic regions as well (Clarke et al., 2013). One possible explanation for these events (and possibly for the observed organic and ultrafine events that were characterized by increases in gas phase volatile organic compounds (VOCs) as noted in the next clusters) is therefore that aerosol may have been mixed down into the MBL from a layer aloft, perhaps on the edge of rain shafts. Alternatively, they may also be due to intermittent plumes of aerosol that survived stochastic precipitation removal events along a boundary layer transport pathway or human terrestrial activities in the pre-dawn hours. In addition, air masses influenced by anthropogenic pollution may have been included in this cluster as well, but without sufficiently different impacts on aerosol parameters to result in a distinct cluster.

Organic event: an approximately 4 h period starting at 1 Z on 23 September had measured particle concentrations between 200 and 325 cm-3, but with significantly (p<0.001) larger median diameters than either the precipitation or background marine types (Fig. 3). Both Aitken and accumulation mode particles had among the lowest hygroscopicities measured during the cruise, with κ values around 0.2. During this event measured concentrations of numerous VOCs were much higher than in gas canisters collected approximately 6 h before and after it, with no associated increase in carbon monoxide (Reid et al., 2016; Fig. 2c). The particles had lower hygroscopicities and larger sizes than the background marine particles observed just before this event. While the source of this event is uncertain, Robinson et al. (2012) found occasional organic aerosol above the boundary layer they attributed to biogenic secondary organic aerosol (SOA) formation during an airborne campaign in the outflow regions of Borneo, while Irwin et al. (2011) reported κ values between 0.05 and 0.37 in a terrestrial, biogenically dominated MC environment. Such a source would be consistent with the observed population, perhaps due to growth of a background marine population by condensation of organics, although we lack the ancillary data needed to establish this.

Ultrafine event: this cluster was associated with an approximately 20 h period on 17–18 September, which included the highest concentration of particles below about 30 nm observed throughout the study (Fig. 2a) and coincided with a period of elevated VOC measurements at the start of this event (Fig. 2c). A filter during this period showed very low potassium concentrations, while benzene was among the lowest values measured during the study, indicating that biomass burning was not the likely source for this event. Anthropogenic, shipping, and marine and terrestrial biogenic emissions are known sources of such compounds; isoprene, a common biogenic VOC, was not observed during this event, and a brief period of elevated dimethyl sulfide, associated with marine emissions from phytoplankton, was observed shortly before – but not during – this event (Reid et al., 2016).

A tri-modal best fit was indicated by the Hussein et al. (2005) algorithm for a number of these data points (Fig. 2a and Supplement Fig. S1). The period had an overall IQR of 482 to 661 cm-3, with generally higher ultrafine number concentrations than other periods with similar total concentrations. The accumulation mode was similar in both size and hygroscopicity (κ=0.65) to the accumulation mode of the background marine type, while the smaller Aitken mode showed larger modal fractions and overall number concentrations, and slightly higher hygroscopicities (κ=0.50) than the background marine measurements. However, we note that the 0.38 % supersaturation hygroscopicity measurement would likely not have been sensitive to these below 30 nm particles and therefore was likely not representative of this smallest third mode. Additionally, while total number concentration was slightly higher than the background marine population, measured CCN concentrations and activated fractions were generally lower, indicating many of the additional particles would not be expected to influence CCN concentrations until higher environmental supersaturations were reached. While not enough information is available to verify the nature of differences between ultrafine particles in these types, the results are consistent with an influx of smaller particles and VOCs into a background marine air mass, and they were sufficiently distinct to be identified as a coherent period by the unsupervised k-means analysis.

Transit: this type was associated with measurements taken during a transit away from the port of Puerto Princesa, a city with a population of over 200 000. During this period light, westerly winds advected anthropogenic pollution out over the Sulu Sea and along the path of the Vasco, allowing for sampling of the urban plume as it diluted and mixed with aerosol from other sources. Size distributions were dominated by an Aitken mode with a number median diameter around 80–90 nm, unique in measurements from this study, mixed with an accumulation mode with a smaller modal fraction than other types. The population had an IQR of 738 to 1029 cm-3, while the generally decreasing number concentrations were consistent with an urban plume diluting and mixing with other aerosol populations. Modal hygroscopicity values of 0.58 and 0.62 for the accumulation and Aitken modes, respectively, were closer than those of most of the other population types and consistent with high levels of sulfate aerosol in typical urban plumes.

Port: this type was assigned to the measurements taken during a short period in the port of Puerto Princesa. Local anthropogenic emissions were dominant during this period, with number concentrations that fluctuated between 4000 and 10 000 cm-3. Ultrafine particles (Dp<100 nm) dominated number concentrations during this period, although large number concentrations of accumulation mode particles with diameters between 100 and 300 nm were also observed. As measurements were fluctuating rapidly and only one CCN scan at each supersaturation setting could be completed before instrumentation was shut down, hygroscopicity results were inconclusive and uncertain. This type is considered separate from the other types as it was not measured in a remote marine area away from the immediate influence of a nearby terrestrial source.

Finally, throughout the study coarse-mode particles with diameters larger than about 800 nm were consistently observed in the PCASP volume distributions (Fig. 2b). Concentrations of particles in this size range increased with increasing wind speed (Fig. 5), consistent with generation of sea spray aerosol due to bubble breaking and wave action (O'Dowd and Leeuw, 2007). While the total number concentration of coarse particles is small compared to typical CCN concentrations (Fig. 2e, f), in the cleanest conditions we measured they represented non-trivial fractions of CCN active at 0.14 and 0.38 % supersaturations. The large diameter of these particles makes them likely to activate at very low supersaturations, and they are present in more than sufficient number concentration to impact the microphysical structure and processes in stratocumulus clouds by serving as “giant CCN” (Feingold et al., 1999).

No significant relationship between wind speed and fine-mode aerosol population type was noted. However, particles in the coarse-mode range are not measured or accounted for in our cluster analysis (CCN system range: 17–500 nm), while submicron aerosol was often dominated by aerosol from other sources. Modini et al. (2015) utilized a dedicated size distribution fitting analysis that included size-resolved observations of particles above 500 nm to examine primary submicron marine aerosol production. They found a primary mode with a median diameter around 200 nm and tail that extended to sizes well above 500 nm, with number concentrations of 12 ± 2 cm-3 during a period of low wind speeds that increased to 71 ± 2 cm-3 as winds increased. Concentrations differences of around 50–60 cm-3 due to wind speed changes may not have resulted in large enough changes to concentrations or size distributions to alter the clustering of the observed population types, which often included number concentrations that were larger by an order of magnitude or more.