The central dataset used is that of sun photometer measurements (Aerosol Robotic Network, 2020a) and derived (inversion) parameters (Aerosol Robotic Network, 2020b) from the AERONET (Holben et al., 1998) site at Manila Observatory in Quezon City, the Philippines (14.64∘ N, 121.08∘ E, ∼ 70 m a.s.l.). Direct sunlight extinction measurements were made at nominal wavelengths of 340, 380, 440, 500, 675, 870, 940, and 1020 nm, from which aerosol optical depth (AOD) was calculated (except for 940 nm, which is for water vapor) (Eck et al., 2013). AOD is a commonly used proxy for aerosol particle loading in the air column from the ground up (Holben et al., 2001); higher AOD translates to more aerosol particle extinction in the column above a location. The extinction Ångström exponent (EAE) and the fine-mode fraction (FMF) are also AERONET direct-sun products that are retrieved after the application of a spectral deconvolution algorithm (O'Neill et al., 2003). For the inversion products, it is through radiative retrievals that the volume size distribution (VSD) and complex refractive index (RI) are gathered and from which single-scattering albedo (SSA) and asymmetry factor (AF) are calculated. The AERONET observations were made during clear-sky conditions, which has been shown (Hong and Di Girolamo, 2022) to be able to represent all sky conditions.

For the inversions, four wavelengths (440, 670, 870, and 1020 nm) of the radiometer spectral channels were chosen for diffuse radiance measurements and to avoid gas absorption (Dubovik et al., 1998). Version 3 direct-sun and inversion algorithms (AERONET, 2019; Giles et al., 2019) were used with the almucantar sky scan scenario to derive the following parameters with level 2.0 (automatically cloud-cleared and quality-controlled datasets with pre- and post-field calibrations) data quality: column AOD (500 nm), fine-mode fraction (500 nm), extinction Ångström exponent (440–870 nm), precipitable water (940 nm), single-scattering albedo (440, 670, 870, and 1020 nm), asymmetry factor (440, 670, 870, and 1020 nm), refractive index (440, 670, 870, and 1020 nm), and VSD. The version 3 products are able to keep fine-mode aerosol particle data (haze and smoke) and remove optically thin cirrus clouds in order to retain more aerosol particle measurements in the database (Giles et al., 2019). Cloud screening in the version 3 product improves remote sensing measurements in Southeast Asia in general, where cirrus clouds are pervasive (Reid et al., 2013). At most, a total of 29 037 direct-sun and 1419 inversion AERONET daytime data points were available between January 2009 and October 2018.

Modern Era-Retrospective Analysis for Research and Applications, version 2 (MERRA-2: 0.5∘ × 0.625∘ approximate resolution), meteorological and aerosol particle composition reanalysis data (Bosilovich, 2016; Gelaro et al., 2017; Randles et al., 2017) were acquired for the area around Manila Observatory (14.25–14.75∘ N, 120.9375–121.5625∘ E). The aerosol reanalysis data include data assimilation of AOD from the Moderate Resolution Imaging Spectroradiometer (MODIS: Terra, 2000 to present, and Aqua, 2002 to present), Advanced Very High Resolution Radiometer (AVHRR, 1979–2002), and Multiangle Imaging SpectroRadiometer (MISR, 2000–2014) (Buchard et al., 2017; Rizza et al., 2019). The following products were used: M2I3NPASM assimilated meteorological fields (3-hourly) for 975 mb level winds, temperature, relative humidity, and sea level pressure (Global Modeling and Assimilation Office, 2015a); M2T1NXFLX surface flux diagnostics (1-hourly from 00:30 UTC time-averaged) 2D for planetary boundary layer height (Global Modeling and Assimilation Office, 2015b); M2T1NXCSP COSP satellite simulator (1-hourly from 00:30 UTC time-averaged) for MODIS mean low-cloud fraction (cloud top pressure > 680 hPa) (Global Modeling and Assimilation Office, 2015c); and M2T1NXAER aerosol diagnostics (1-hourly from 00:30 UTC time-averaged) for total AOD and speciated AOD (sulfate, black carbon (BC), organic carbon (OC), dust, and sea salt) (Global Modeling and Assimilation Office, 2015d).

MERRA-2 meteorological and aerosol particle composition monthly mean reanalysis data (Bosilovich, 2016; Gelaro et al., 2017; Randles et al., 2017) were also acquired for a larger region (30∘ × 30∘), i.e., the Southeast Asia region (0–30∘ N, 105–135∘ E), for the period from January 2009 to December 2018. This is within the spatial domain of the CAMP2Ex airborne measurement campaign, which, as mentioned earlier, targets the interaction between tropical meteorology and aerosol particles. The following datasets (0.5∘ latitude and 0.625∘ longitude resolution) were used: MERRA-2 tavgM_2d_aer_Nx: aerosol assimilation (M2TMNXAER) for total 500 nm AOD and speciated 500 nm AOD (sulfate, BC, OC, dust, and sea salt) (Global Modeling and Assimilation Office, 2015e) and MERRA-2 instM_3d_ana_Np: analyzed meteorological fields (M2IMNPANA) for 1000 and 725 hPa level U and V winds (Global Modeling and Assimilation Office, 2015f). The total MERRA-2 AOD (reanalysis) for the region (mean over 30∘ × 30∘region) was used along with MISR AOD (remote sensing) data (mean over 30∘ × 30∘region) to assess the influence of long-range sources on the aerosol column over Manila Observatory. The monthly meteorological and aerosol particle composition data for the region will be used for empirical orthogonal functions, which will be described later.

Hourly precipitation data were obtained from the Precipitation Estimation from the Remotely Sensed Information using the artificial neural networks (PERSIANN) database of the Center for Hydrometeorology and Remote Sensing (CHRS) at the University of California, Irvine (UCI) (Nguyen et al., 2019a). Hourly data were accumulated for running 3 d totals, which were compared to AERONET data. The data were averaged between the four grids that included the area of interest, ensuring a similar spatial domain (14.5–15.0∘ N, 120.75–121.25∘ E) to the MERRA-2 dataset.

Monthly 500 nm AOD data (level 3 global aerosol: 0.5∘ × 0.5∘ spatial resolution in the region 0.25–30.25∘ N and 104.75–134.75∘ E) from January 2009 to December 2018 (EarthData, 2018) are used from the Multi-angle Imaging SpectroRadiometer (MISR) (Diner et al., 2007; Garay et al., 2018) as regional (Southeast Asia) baseline remote sensing data to support the Manila Observatory AERONET data. The regional (30∘ × 30∘) MISR data were used to confirm regional sources of aerosols that may be influencing the AOD over Metro Manila. Level 3 MISR products are global maps of parameters available in level 2 (measurements derived from the instrument data) products. MISR is ideal for remote sensing in the CAMP2Ex region because it has an overpass at 10:30 ECT (equatorial crossing time) (descending mode) (when cirrus is minimal) and its retrievals have been shown to not be impacted by small cumulus (Zhao et al., 2009), which are typical in the region. MISR has relatively more accurate AOD and agrees better with AERONET data compared to other satellite products due to its multi-angle measurements (Choi et al., 2019; Kuttippurath and Raj, 2021). The MISR sampling noise is relatively small due to the large domain and seasonal averages that are considered in this study. MISR is also the only passive sensor that speciates aerosol particle size and shape. All these factors led to the choice of using regional MISR data to associate long-range sources influencing AERONET data at Manila Observatory. Monthly mean AOD (bin 0) data were extracted for Southeast Asia (0.25–30.25∘ N, 104.75–134.75∘ E) within the CAMP2Ex region. Monthly mean AOD values were then calculated for each 0.5∘ grid point and then for the 30∘ × 30∘ region, where the standard error in the monthly mean for the region is less than 0.002. MISR monthly mean time series of size, shape, and absorption speciated 550 nm AOD and Ångström exponent in the CAMP2Ex domain (6.5–22.5∘ N, 116.5–128.5∘ E; March 2000 to December 2020) are also used to support the findings from the AERONET data.

Archived maps of total and speciated optical depths and surface concentrations of sulfate, dust, and smoke for Southeast Asia are used from the Navy Aerosol Analysis and Prediction System (NAAPS: 1∘ × 1∘ spatial resolution) (Lynch et al., 2016), and which are publicly available at https://www.nrlmry.navy.mil/aerosol/ (last access: 4 June 2021). This reanalysis product relies on the Navy Global Environmental Model (NAVGEM) for meteorological fields (Hogan et al., 2014). Hourly maps were downloaded for aerosol particle events of interest based on AERONET data. These maps help associate possible regional emission sources to extreme aerosol loading events in Manila Observatory. Previous studies have used NAAPS data for an overview of aerosol sources in specific regions of interest (Ross et al., 2018; Foth et al., 2019; Markowicz et al., 2021; Harenda et al., 2022; Mims III, 2022). More recent studies show the need to improve aerosol representation in NAAPS (Edwards et al., 2022), so we will use NAAPS qualitatively, together with MERRA-2 compositional AOD data and back trajectories, for an overview of aerosol sources that may contribute to extreme events with high AOD from AERONET.

Back trajectories from the National Oceanic and Atmospheric Administration's (NOAA) Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (Stein et al., 2015; Rolph et al., 2017) were used to provide support for the AERONET monthly aerosol characteristics and the chosen case studies. Both 3 and 7 d back trajectories with 6 h resolution were generated based on the NCEP/NCAR reanalysis meteorological dataset with a resolution of 1∘ and a vertical wind setting of “model vertical velocity”. The 3 d data were used to map the density of trajectories (The MathWorks Inc., 2022) reaching Manila Observatory in each month from 2008 to 2019. The 7 d data were used in the analysis of the case studies. Trajectories were computed for an end point with an altitude of 500 m above ground level at Manila Observatory. This altitude represents the mixed layer based on related surface air quality studies (Crosbie et al., 2014; Mora et al., 2017; Schlosser et al., 2017; Aldhaif et al., 2020), including a previous study for the same area (Stahl et al., 2020).

Archived maps of cloud fraction (Aqua MODIS and Terra MODIS) over Metro Manila and Southeast Asia were downloaded from NASA Worldview (https://worldview.earthdata.nasa.gov, last access: 17 January 2023) for events of interest based on AERONET data.

Events related to transported biomass burning and smoke were chosen from the AERONET VSD data that were clustered as urban and industrial (with a dominant submicrometer peak) (Eck et al., 1999) over Metro Manila. Cases with the highest black carbon contribution to total AOD from the MERRA-2 dataset were considered. Maps from NAAPS of high smoke contributions to AOD and surface smoke contributions in the direction of HYSPLIT back trajectories were used to provide support for the likely source and transport pathway for the smoke cases.

A dust transport case over Metro Manila was identified from the AERONET VSD dust cluster (with an enhanced coarse peak in the AERONET VSD compared to the submicrometer fraction) (Eck et al., 1999), the highest dust contribution to AOD from the MERRA-2 dataset, and high dust contributions to AOD from NAAPS. Surface dust concentrations from NAAPS along the HYSPLIT back trajectories improved the plausibility of dust for this case.

Cloud processing events were identified based on bimodal submicrometer VSDs (Eck et al., 2012) and a relatively large sulfate contribution to AOD over Metro Manila from the MERRA-2 dataset, since this species is predominantly produced via cloud processing (Barth et al., 2000; Faloona, 2009). The presence of clouds was verified qualitatively with MODIS (Aqua and Terra) imagery from NASA Worldview in the path of air parcels reaching Metro Manila based on HYSPLIT back trajectories.

Monthly median AOD (AERONET, 500 nm) (Fig. 3a) over Manila Observatory was highest from August (0.21) to October (0.23) around the time of the summer monsoon when winds were coming from the southwest (Fig. 2h to i) (Holben et al., 2001). This is the same time of year when biomass burning activities occur in the Indonesian region southwest of Metro Manila (Glover and Jessup, 1998; Kiely et al., 2019; Cahyono et al., 2022). Studies have shown that AOD in the Philippines increases during the biomass burning season in Indonesia (Nguyen et al., 2019b; Caido et al., 2022). Regional AOD (550 nm) over the larger Southeast Asian domain from MISR and MERRA-2 (Fig. 4) had a similarly large peak around the same time beginning in September until October, which was second in magnitude only to a March peak influenced by biomass burning in peninsular Southeast Asia (PSEA) (Gautam et al., 2013; Hyer et al., 2013; Dong and Fu, 2015; Wang et al., 2015; Yang et al., 2022). This is consistent with the peak in speciated AOD due to fine (radii < 0.7 µm), spherical, and absorbing aerosols that were observed by MISR from March to April (Fig. S1 in the Supplement). This larger peak in March, attributed to PSEA (which is ∼ 2000 km west of the Philippines), was not as prevalent in the AERONET AOD data over Manila Observatory in Metro Manila due to the dominant easterly winds in the Philippines in March (Fig. 2c) and more localized sources.

There is a notable dip in the monthly median AERONET AOD over Manila Observatory from the peak in October to the lowest monthly median AOD (0.11) in November (Fig. 3a), which is just slightly above defined background levels (< 0.1) (Holben et al., 2001), when the wind speeds were picking up and were coming from the east to northeast directions (Fig. 2k) in the direction of the Philippine Sea and the western Pacific Ocean. This dip was also observed in the regional (30∘ × 30∘) AOD data (MISR and MERRA-2, Fig. 4). This is most probably due to the decrease in the AOD contribution from fine (radii < 0.7 µm) and spherical particles based on size-speciated MISR AOD (Fig. S1). Larger and non-spherical particle contributions to AOD increase in November in the Southeast Asia region. The MERRA-2 AOD is relatively high compared to the MISR AOD, which is probably due to assimilation of MODIS data into MERRA-2. Studies in Asia (Xiao et al., 2009; Qi et al., 2013; Choi et al., 2019) have observed relatively high MODIS AOD compared to MISR AOD.

There were 338 instances (∼ 1.2 % of the time based on the total number of 28 538 valid AERONET AOD data points) of AOD values exceeding 1, indicative of heavy aerosol particle loading (Huang et al., 2021). Because AOD is extrinsic (it depends on mass), AOD describes total aerosol particle loading, and we examine other aerosol particle parameters from AERONET to make more informed inferences about size and composition.

The extinction Ångström exponent (EAE) relates the extinction of light at specific wavelengths and is indicative of aerosol particle size (Ångström, 1929). The EAE is usually greater for smaller particles (∼ 4 for very small particles that undergo Rayleigh scattering, > 2 for small particles, <1 for large particles like sea salt and dust, and 0 for particles as large as cloud drops) (Schuster et al., 2006; Bergstrom et al., 2007). The highest monthly median EAE (Fig. 3b) from 2009 to 2018 over Manila Observatory was observed from July (∼ 1.4) to September (∼ 1.3), during the southwest monsoon. This period is associated with the biomass burning southwest of the Philippines (Oanh et al., 2018; Stahl et al., 2021; Crosbie et al., 2022). The median (per month) EAE ranged from ∼ 0.9 in November to ∼ 1.4 in August, a range that is within the values from previous studies collected from mixed sites and urban and industrial areas with both fine and coarse particles (Eck et al., 2005; Giles et al., 2012). The high EAE over Manila Observatory from July to September is consistent with the regional (30∘ latitude × 30∘ longitude) MISR data that show increased AOD from fine, spherical, and absorptive particles (Fig. S1) in Southeast Asia during the same months. This suggests that the high EAE observed at Manila Observatory during these months is not necessarily from local sources.

EAE increases with AOD (Fig. S2), which means that the greater particle loading is contributed by smaller particles (Smirnov et al., 2002). Of the high-loading cases (AOD > 1) over Manila Observatory, the EAE values were mostly greater than 0.8, indicating fine-mode particles (Che et al., 2015). The EAE values in August were the highest compared to other months and also had the highest minimum value of any month (0.71) (Fig. S2) due to smaller particles (∼ EAE > 1 for fine particles, Table 2). The lowest EAE values (0.08) and thus the largest particles were observed in December, which again may be regional in nature with MISR EAE also being at its lowest during this time and increased AOD from larger and non-spherical particles (Fig. S1).

The fine-mode fraction (FMF) describes the prevalence of fine-mode particles in the column of air above the surface. The fine-mode fraction (Fig. 3c) from 2009 to 2018 was highest in August (monthly median of 0.75) and lowest in November (monthly median of 0.45). This is consistent with the EAE values discussed earlier, with a prevalence of smaller particles in August and larger particles in November. In August (Fig. 2h), the southwest monsoon is known to coincide with the transport of fine smoke particles to Luzon. In November (Fig. 2k), the prevalent winds may have already shifted to easterly (Matsumoto et al., 2020) implying more marine-related sources associated with coarser particles.

The single-scattering albedo (SSA) is the most important aerosol particle parameter determining whether aerosol particles will have a warming or cooling effect (Reid et al., 1998). SSA is the ratio of the scattering coefficient to the total extinction (scattering and absorption) coefficient (Bohren and Clothiaux, 2006) of aerosol particles. Higher SSAs are related to more reflective aerosol particles, while more absorbing aerosol particles will have lower SSA values; values range from 1 (reflective) to 0 (absorbing). Monthly median SSA values were largest in June (0.94 at 440 nm), suggesting the presence of more reflective aerosol particles, and smallest in August (0.88 at 440 nm and 0.78 at 1020 nm), suggesting more absorptive particles that are similar in range to the SSA of biomass burning particles (Table 2). August is when biomass burning is prevalent to the southwest of the Philippines and associated with soot particles that are absorptive.

The sensitivity of SSA to different wavelengths depends on the type of aerosol particles present. More specifically, aerosol particle size and refractive index (which is related to aerosol particle composition) both affect the SSA (Dubovik and King, 2000; Bergstrom et al., 2007; Moosmüller and Sorensen, 2018). For dust-type particles, SSA increases with wavelength because of lower dust absorption in the higher visible to infrared wavelengths (Dubovik et al., 2002), while for urban particles (including black carbon), which absorb light at longer wavelengths, SSA decreases with wavelength (Reid et al., 1998; Bergstrom et al., 2002). The presence of organic carbon may affect this spectral dependence; however, because organic particles absorb in the UV portion of the spectrum, this lowers SSA at wavelengths shorter than 440 nm (Kirchstetter et al., 2004). Monthly median SSA generally decreased with increasing wavelength for all months with available data (Fig. 3d), presumably due to the influence of more urban particles in contrast to dust. However, the monsoon transition months of April, September, and October (Fig. 3d) are noteworthy, as they had increased SSA from 440 to 670 nm, possibly from organics and black carbon due to transported smoke. The back trajectories for these months (Fig. 2d, i, and j) suggest sources from the northeast that are closer to Luzon during these months compared to other months. This indicates the possibility of more local sources. Increasing the certainty of sources associated with aerosol particles necessitates looking at other available aerosol particle parameters, which will be discussed below.

The asymmetry factor quantifies the direction of scattering of light due to aerosol particles, with values ranging from -1 (back scatter) to 0 (uniform scattering) to 1 (forward scatter). It is important in modeling climate forcing because it affects the vertical distribution of the radiation in the atmosphere (Kudo et al., 2016; Zhao et al., 2018). The asymmetry factor is dependent on particle size, shape, and composition, and a value of 0.7 is used in radiative models (Pandolfi et al., 2018).

Lower asymmetry factors are related to smaller particles (at constant AOD) (Bi et al., 2014). Measured values due to biomass burning are, for example, 0.54 (550 nm) in Brazil (Ross et al., 1998) and 0.45–0.53 (550 nm and including dust) over central India (Jose et al., 2016). There have been relatively high values observed for western, central, and eastern Europe (0.57–0.61 at 520–550 nm) (Pandolfi et al., 2018) and the US East Coast (0.7 at 550 nm) (Hartley and Hobbs, 2001). In Norway, the asymmetry factor for background summer conditions was 0.62 and was higher in the springtime at 0.81 (862 nm) during Arctic haze events (Herber et al., 2002). The highest values are associated with dust, such as those measured in the Sahara being 0.72–0.73 (500 nm) (Formenti et al., 2000). Over Metro Manila, the asymmetry factors from the AERONET data at the 675, 870, and 1020 nm were similar across months (Fig. 3e). The monthly median asymmetry factors at 440 nm ranged from 0.70 (April and May) to 0.74 (October), while for 670, 870, and 1020 nm the monthly median asymmetry factors were smaller and ranged from 0.62–0.69. These values were closely related to those observed over the US East Coast as mentioned earlier, perhaps due to the proximity of the location to the coast (10 km east of Manila Bay and 100 km west of the Philippine Sea) and its location in Manila, which is a large local source (mostly due to vehicles) (Cruz et al., 2019).

The monthly median asymmetry factor in Metro Manila was greatest towards the end of the year (October to December) for all of the wavelengths, suggesting larger particles when winds (Fig. 2j to l) come from the Philippine Sea in the northeast. It was in March and April that the monthly median asymmetry factor was minimal for 440 nm and was minimal in August for 670, 870, and 1020 nm. These were the times when aerosol particles were smallest. March to April represents the driest time of the year in Manila (Fig. 1b and h), perhaps preventing particle growth, and where the local sources may be dominant, even as back trajectories (Fig. 2c and d) extend all the way from the Philippine Sea to the east. This is corroborated by results from other studies showing that the asymmetry factor seems to be enhanced by relative humidity (Zhao et al., 2018). The unexpected low asymmetry factor values in August, however, are probably because of the source of the particles. August had the highest relative humidity and precipitable water (Fig. 1b and h) but is also when the back trajectories (Fig. 2h) were from the southwest, possibly affected by the Indonesian fires, which could have transported more non-hygroscopic fine particles.

Fine particles have been observed to exhibit decreasing asymmetry factors with increasing wavelength (Bergstrom et al., 2003). This trend is observed in all months for the monthly median asymmetry factors (Fig. 3e), suggesting the predominance of smaller aerosol particles. The greatest decrease in the asymmetry factor (all wavelengths) was in August, consistent with the lowest observed values of the year (670, 870, and 1020 nm). Transported biomass burning particles are the probable dominant particles during this time. They are usually composed of hygroscopic inorganics, non-hygroscopic soot, and relative non-hygroscopic organic fractions (Petters et al., 2009). Knowing the composition of biomass burning particles over the study region will help in the understanding of hygroscopicity and its impacts on radiation.

Refractive index is an intrinsic parameter as it does not depend on the mass or the size of particles, and thus can be used to infer aerosol particle composition (Schuster et al., 2016). For the case of the AERONET data, which include refractive index values that are insensitive to coarse particles (Sinyuk et al., 2020), the focus of the discussion will be for fine-mode particles and may be limited when coarse particles are involved. Refractive index measurements are complex since they include real and imaginary parts related to light scattering and absorption, respectively. All aerosol particles scatter light, but only certain types absorb light significantly. The most prominent particle absorbers in the atmosphere are soot carbon, brown carbon (organic carbon that absorbs light), and free iron from dust (hematite and goethite in the ultraviolet to mid-visible) (Schuster et al., 2016). For this study, we examine refractive index values at 440 nm wavelength. Pure sources of soot carbon have the highest real refractive index values (∼ 1.85) and the highest imaginary refractive index (∼ 0.71), which are both independent of wavelength (Koven and Fung, 2006; Van Beelen et al., 2014). Brown carbon and dust have relatively low real refractive index values at 440 nm (∼ 1.57 and ∼ 1.54) and imaginary refractive index values (∼ 0.063 and ∼ 0.008) that decrease with increasing wavelength (Xie et al., 2017).

In this study the range of the monthly median real refractive index values (440 nm) was from 1.33 (December and January) to 1.43 (March) (Fig. 3f). Water uptake by aerosol particles decreases the real refractive index values (Xie et al., 2017), and thus the lowered real refractive indices over Manila Observatory could be due to the presence of more water in the atmosphere in general and/or the increased presence of more hygroscopic particles. December and January are not necessarily the months that have the highest moisture content, but they are months when back trajectories reaching the column over Manila Observatory are from the Philippine Sea to the northeast, presumably transporting hygroscopic particles. As reported in previous sections, relatively large particles are observed around this time of the year, and thus sea salt can be an important contributor. The greatest change in the monthly median real refractive index with increasing wavelength was also observed in December (Fig. 3h), possibly due to the increased fractional contribution of constituents other than soot carbon (because the real refractive index of soot carbon is invariant with wavelength). The month of August (Fig. 3f) is also noteworthy as it has the smallest range of real refractive index values, possibly indicating a more homogenous aerosol particle source compared to other months. August is the month with the highest relative humidity (Fig. 1b) and highest precipitable water (Fig. 1h), and this is also the month when long-range biomass burning emissions are observed to be at their highest and real refractive index values would otherwise be expected to be at their highest.

Water content seems to play a significant role in the real refractive index values in Manila. March, when the monthly median real refractive index values are highest (Fig. 3f), is when precipitable water vapor (Fig. 1h) is among the lowest in the year. The months around March are also when maximum real refractive indices (1.57 in February, 1.59 in March, and 1.60 in April) were observed (Fig. 3f). March was when there was a relatively small change in real refractive index value, with wavelength perhaps being related to greater soot carbon fractions during this time, possibly due to the contribution of biomass burning from peninsular Southeast Asia (Shen et al., 2014). Looking more closely at the imaginary refractive index values will help elucidate this issue.

Monthly median imaginary refractive index values (440 nm) ranged from 0.007 in June to 0.015 in September and December (Fig. 3g). These are low compared to those of the pure soot carbon mentioned earlier because of the mixed nature of the sampling site with contributions from brown carbon and dust. The highest imaginary refractive index values in September and December suggest the greatest fractional contribution of soot because the highest imaginary refractive index values are associated with soot. These are also similar in magnitude to biomass burning particles in the Amazon (0.013) (Guyon et al., 2003). The key distinction between soot carbon and other major absorbers (brown carbon and dust) is that its imaginary refractive index is invariant with wavelength. Both brown carbon and dust exhibit a decrease in the imaginary refractive index with increasing wavelength (Xie et al., 2017). The ratios of imaginary refractive index values (440 nm to average of 670–1020 nm) (Fig. 3h) show a relative invariance with wavelength (ranging from 0.88 to 1.4), which indicates the dominance of soot as the major absorber in the region (Eck et al., 2003). While observed wavelength invariance points to high soot contributions, the size of the particles can help distinguish between brown carbon sources, which reside mainly in the fine mode, and dust sources, which yield more coarse particles (Schuster et al., 2016). September is during the southwest monsoon, which is when, as noted in earlier sections, fine particles were most prevalent. This is also the time when the imaginary refractive index varied most with wavelength (1.4 ratio of the imaginary refractive index at 440 nm and the imaginary refractive index average for 670 to 1020 nm in Fig. 3h), possibly with greater absolute contributions from brown carbon, even with the highest soot carbon fractional contributions. Brown carbon has been observed both from primary and aged aerosol particle emissions from biomass burning (Saleh et al., 2013). As noted earlier, December also had the highest imaginary refractive index values and relatively coarse particles, possibly due to larger dust absolute contributions even with the highest soot carbon fraction contributions. In contrast, the lowest monthly median imaginary refractive index values are in June, when fine-mode particles prevail suggest highest fractional contributions of brown carbon relative to other months (Fig. 3h).

The volume size distribution (VSD) is another way to be able to more deeply characterize aerosol particles, specifically related to their effect on climate, weather, and clouds (Haywood and Boucher, 2000; Feingold, 2003). In the Manila Observatory dataset, there was a bimodal VSD for the entire dataset (Fig. 5a). The fine-mode median values peaked in the accumulation mode at 0.148 µm particle radius, while the coarse-mode median values peaked at 3.857 µm (Fig. 5a and Table S1). The median coarse-mode amplitudes and volume concentrations were higher than the fine-mode amplitudes and volume concentrations for most of the year (DJF, MAM, and SON, Fig. 5b and Table S1), except during the southwest monsoon (JJA) when the fine-mode amplitude and volume concentration was higher. This is consistent with observations earlier of fine-mode prevalence during the southwest monsoon. Median VSD amplitudes (Fig. 5c) were greater in the afternoon, with higher peaks and volume concentrations for both the fine and coarse modes compared to the morning. There was a slightly larger coarse median amplitude and volume concentration compared to the accumulation mode median amplitude and volume concentration for both the morning and afternoon size distributions. While the VSDs confirm several observations based on the analysis of the aerosol particle parameters presented earlier, not much further information is gained, especially regarding chemical composition. Size distributions are a result of contributions from multiple sources, and thus being able to discriminate the sources based on their characteristic size distributions will help identify relevant sources.

Five clusters were identified to best represent the VSD (Fig. 6a). The average of the VSDs in each cluster varied depending on the height of the peaks in the accumulation mode and the coarse mode. In Metro Manila, the accumulation mode is associated with aged aerosol particles and combustion (Cruz et al., 2019). The majority of the data (830 count out of 1419 total VSD profiles) were clustered together in a profile (cluster 1) that had relatively low average magnitudes of volume concentration for both the accumulation (0.01 µm3 µm-2) and coarse (0.02 µm3 µm-2) modes, with the volume concentration magnitude of the coarse mode peaking slightly higher than the volume concentration magnitude of the accumulation mode peak. The next most prevalent cluster profile (284 counts, cluster 2) had an average fine-mode peak for the volume concentration (0.04 µm3 µm-2) that was more than twice as much as the previous profile but with a similar coarse-mode peak for the volume concentration (0.02 µm3 µm-2). The average coarse-mode peak for the volume concentration (0.04 µm3 µm-2) was the highest (compared to the four other cluster profiles) for the third most prevalent cluster profile (166 counts, cluster 3); cluster 3 also had a volume concentration peak in the coarse mode that was slightly shifted to a higher radius (5.06 µm) compared to other clusters. The coarse mode dominated this VSD compared to other profiles (lower magnitude for the accumulation mode peak for the volume concentration, 0.02 µm3 µm-2). The two remaining cluster profiles exhibited high average magnitudes of volume concentration in both the accumulation and coarse modes. The fourth most prevalent cluster profile (74 counts, cluster 4) had the highest average absolute magnitude for the volume concentration in the accumulation mode (0.11 µm3 µm-2), while the fifth most prevalent cluster profile (65 counts, cluster 5) had a slightly smaller accumulation mode peak for the volume concentration (0.07 µm3 µm-2) that was shifted to a slightly higher radius (0.19 µm compared to 0.15 µm). Both clusters 4 and 5 had similar average coarse-mode peak volume concentration magnitudes (0.04 µm3 µm-2).

The clusters were distributed across seasons (Fig. 6b), with clusters 1 and 2 being the most evenly distributed among the clusters. Cluster 3, which had the highest coarse-mode peak, had the greatest contribution from September to November compared to other clusters. Cluster 4, which had the highest accumulated mode peak compared to other clusters, had the greatest contribution from March to May and for afternoon VSDs compared to other clusters (Fig. 6b and c). Relative contributions of VSDs from June to August were highest for cluster 5, which had the shifted accumulated mode peak.

Median total (AERONET) AOD values (Fig. 7b) were lowest (0.12) for cluster 1, though it had the second-highest sea salt fractional contributions (31 %) (Fig. 7a) to total AOD (MERRA-2) among all the clusters. Cluster 2 had relatively mid-range median total AOD values (0.27) that, along with clusters 4 and 5, were dominated by sulfate and organic carbon (46 % and 20 %). Cluster 3 had similar but slightly lower median total AOD (0.25) compared to cluster 2. Cluster 3 was distinct because it had the largest total (0.04) and fractional contribution (37 %) from sea salt among all clusters. Clusters 4 and 5 had the highest median total AOD values (0.47 and 0.56), with cluster 5 having the highest absolute and fractional sulfate contributions (0.14 and 64 %) among the clusters. Integrating the above results with their corresponding aerosol particle properties can help associate the clusters with air masses.

Air masses have been classified in previous studies based on their AOD, EAE, FMF, and SSA values (e.g., Lee et al., 2010; Aldhaif et al., 2021). The criteria from different studies (Table 2) were applied per cluster. The median total AOD of cluster 1 (0.12) was less than 0.2 (Fig. 7b), which is the threshold for sea salt sources. Half of the data points in cluster 1 also fall below the threshold for clean environments (AOD < 0.1) (Sorooshian et al., 2013). Based on its median EAE (1.07, where EAE < 1 is coarse and EAE > 1 is fine) and FMF (0.60) values (Fig. 7d and e), cluster 1 is a mixture of fine and coarse particles. The fine cluster 1 is the only cluster with a median that meets that threshold value for clean marine sources (AOD < 0.2), and we know from Sect. 3.3.1 that its average VSD magnitude was greater for the coarse fraction and that its sea salt contribution to total AOD was second greatest among the clusters. Thus cluster 1 is most probably a background clean marine source since it is also predominant throughout the seasons (Fig. 6b). This makes sense given the proximity of the ocean to Metro Manila from both the east and west. The median SSA (0.90 at 440 nm) for cluster 1 (Fig. 7c), however, suggests the presence of absorbing particles, which is most probably due to high black carbon in the local source (Cruz et al., 2019) that is mixed in with this generally clean marine source.

Most of the data from the other clusters all fall in the polluted category (Table 2) based on their median total AODs (> 0.1) (Fig. 7b). Cluster 2 has a median FMF value of 0.78 (Fig. 7e), which suggests that most of the particles in this air mass are in the fine fraction. They are, however, not sufficiently dominant in the aerosol for them to be typical of urban and industrial sources. The average VSDs (Fig. 6a) of cluster 2 similarly suggest that their relative accumulation mode magnitude is higher than the coarse magnitude but not by much. Like cluster 1, cluster 2 is also more evenly distributed across the seasons (Fig. 6b). The median SSA for cluster 2 (0.90 at 440 nm) is also similar to the SSA of cluster 1 (Fig. 7c) where the local and background particles are mixed. Cluster 2 could be a fine polluted background source superimposed onto the dominant marine source. Metro Manila is a megacity with continuous and large numbers of sources that could be, due to its proximity to the ocean, interacting with the background.

Based on its median EAE value (1.04) (Fig. 7d), cluster 3 is mixed but mostly in the coarse fraction, consistent with its VSD profile (Fig. 6a), which has the highest coarse magnitude (FMF = 0.60) compared to the other clusters. The contribution of data from September to February is greatest in cluster 3 (Fig. 6b), consistent with expected coarser particles during this period when the winds are initially shifting from the southwest before becoming more northeasterly, as previously noted. Median SSA (0.89 at 440 nm) was lowest for cluster 3 (Fig. 7c); this and the relatively high coarse particle contribution suggests cluster 3 as a possible dust source based on past studies (Lee et al., 2010). This air mass can be a mixture of local sources and transported dust air masses. The large sea salt contribution (∼ 37 %) to total AOD in cluster 3 (Sect. 3.3.1) could be related to long-range transport.

Both clusters 4 and 5 have median total FMF (0.83 and 0.91) (Fig. 7e) values exceeding the mark (> 0.8, Table 2) for urban and industrial air masses. Combining this and results from the previous sections confirms that cluster 4 could be an urban and industrial source given that it had the highest median accumulated mode peak and organic carbon contribution (∼ 20 %) to total AOD among the clusters. The median SSA for cluster 4 (0.90 at 440 nm) was similar to the median SSA of clusters 1 and 2 (Fig. 7c), but the maximum SSA value for this cluster was lowest in general among all the clusters, suggesting cluster 4 has the net most absorptive effect. The cluster 4 air mass is probably from local sources and transported biomass burning emissions. The high median EAE (1.40, Fig. 7d) may be associated with aerosol particles due to biomass burning (Deep et al., 2021).

Cluster 5 had the highest median total AOD (0.56) and FMF (0.91) values (Fig. 7b and e). It also had the highest sulfate contribution (∼ 64 %) to total AOD (Fig. 7a), the highest median SSA (0.92 at 440 nm, thus most reflective particles among the clusters) (Fig. 7c), and a shifted accumulation mode peak (Fig. 6a). These characteristics suggest that cluster 5 is a possible cloud-processing air mass (Eck et al., 2012). The larger peak in the accumulation mode is possibly the cloud signature. Previous studies have attributed this larger mode to cloud processing due to the conversion of SO2 to sulfate (Hoppel et al., 1994). Cloud processing is a major source of sulfate (Barth et al., 2000).

The distribution of the air masses based on the abundance of the VSD profiles per cluster suggest prevalent clean marine (58 % of the total VSD counts) and background fine polluted (20 %) air masses over Metro Manila. The mixed dust (12 %), urban and industrial (5 %), and cloud processing (5 %) air masses contribute 22 % altogether. We can investigate more deeply and look at specific case studies that can better describe the air masses identified here.

Both cases of long-range transport of smoke discussed below have similar VSDs (Figs. 8a and 9a) to the urban and industrial cluster VSD (cluster 4, Fig. 6a). Organic carbon was the dominant contributor to AOD (Figs. 8b and 9b) for both long-range transport cases. The first of two events occurred around 1 April 2020 with smoke presumed to come from East Asia. The VSD of this specific case (Fig. 8a) is most like the urban and industrial cluster (cluster 4 in 3.3.2, Fig. 6a) because of the high magnitude of its accumulated mode peak, its timing (April), and the enhanced organic carbon contribution to AOD in the area (Fig. 8b). Though the absolute black carbon contribution to AOD was highest here compared to the other case studies, and in general for the AERONET data, it was organic carbon that was more prevalent in terms of contribution to total AOD. Smoke is comprised of both soot carbon and organic carbon, among other constituents (Reid et al., 2005).

The smoke contribution to AOD from NAAPS (Fig. 8c) for the first smoke case was visible in the Philippines (0.2) and seemed to come from East Asia where the smoke contribution to AOD was greater (reaching 0.8) (especially in peninsular Southeast Asia). Smoke surface concentrations were also widespread (Fig. 8c), with the greatest concentrations that reached the western Philippines coming from East Asia despite being seemingly disconnected over the sea. There were observed biomass burning emissions in the peninsular Southeast Asia region (southern China, Burma, and Thailand) at this time (Shen et al., 2014). The direction of the air mass coming into Metro Manila was from the northeast, which curved from the west in the direction of East Asia based on HYSPLIT back trajectories (Fig. 8d).

The second smoke case was on 15 September 2009, with the source being Southeast Asia. The back trajectories of this case study (Fig. 9d) are from the southwest of the Philippines and in the direction of Malaysia and Indonesia. NAAPS maps likewise show elevated AOD, specifically smoke contribution to AOD (Fig. 9c), as well as enhanced smoke surface contributions in the area around Metro Manila for this second smoke case study. The observed AOD and smoke surface concentration increased specifically from the southwest of the Philippines in the same direction of the back trajectories. There were fires in the lowland (peat) forests of Borneo around this time (NASA, 2009). MERRA-2 AOD contributions for this case were greatest due to organic carbon and sulfate (Fig. 9b), and the absolute black carbon contributions were greatest compared to other cases. The VSD of this smoke case from Southeast Asia (Fig. 9a) resembled that from long-range transported smoke from East Asia (Fig. 8a) and the urban and industrial air mass (cluster 4, Fig. 6a). This case occurred in the afternoon, which was the prevalent time that the urban and industrial air mass was observed (Fig. 6c).

The VSD of this specific case on 24 March 2018 (Fig. 10a) was most similar to the mixed dust cluster (cluster 3), which had a mixed size distribution but a more dominant coarse contribution. This is consistent with the most dominant contribution to AOD in the area, which was sea salt and dust (Fig. 10b). The back trajectories were from East Asia around the same latitude as Taiwan (Fig. 10d). At that time, the area in question had increased AOD in general from sulfate and dust (Fig. 10c). The AOD values from both AERONET and MERRA-2 (Fig. 10b) are lower than 0.3 (the AOD threshold for dust in other studies, Table 2) because of the long distance from the source (thousands of kilometers). The dust and sulfate seemed to have been transported to Metro Manila from East Asia based on the NAAPS sulfate and dust surface concentrations (Fig. 10c).

Sulfate dominated the AOD (Fig. 11b) for this case on 26 August 2009 in the area around Metro Manila. This along with its VSD exhibiting a second peak (Fig. 11a) in the accumulation mode make it very similar to the cloud processing cluster (cluster 5). Sulfate has been known to be enhanced through chemical productions in clouds and is used as a signature for cloud processing (Barth et al., 2000; Ervens et al., 2018). Aqueous production of sulfate is significant in areas with sources and clouds (Barth et al., 2000), and this case study has both. Aside from the high sulfate contribution to AOD, the cloud fraction (Aqua/MODIS, Terra/MODIS, Fig. S3) is very high (∼ 100 %) in the area of the back trajectories (Fig. 11d). Interestingly, there is no regional AOD elevation observed in the NAAPS maps (Fig. 11c) for this time. There are increased surface smoke and sulfate levels in East Asia as well as southwest of the Philippines, and though the back trajectories do show a northeastward direction, they do not reach far enough into mainland East Asia. It is possible that even while there are known regional sources of sulfate in Southeast Asia (Smith et al., 2011; Li et al., 2017), this case could be local to the Philippines. There is in fact a large power plant northwest of Metro Manila (Jamora et al., 2020).