Long-range transport patterns into the tropical northwest Pacific

size distributions, and the impact of convection 4 Miguel Ricardo A. Hilario, Ewan Crosbie, Michael Shook, Jeffrey S. Reid, Maria Obiminda 5 L. Cambaliza, James Bernard B. Simpas, Luke Ziemba, Joshua P. DiGangi, Glenn S. 6 Diskin, Phu Nguyen, Joseph Turk, Edward Winstead, Claire E. Robinson, Jian Wang, 7 Jiaoshi Zhang, Yang Wang, Subin Yoon, James Flynn, Sergio L. Alvarez, Ali Behrangi, 8 Armin Sorooshian 9 10 1 Manila Observatory, Quezon City 1108, Philippines 11 2 NASA Langley Research Center, Hampton, VA, USA 12 3 Science Systems and Applications, Inc., Hampton, VA, USA 13 4 Marine Meteorology Division, Naval Research Laboratory, Monterey, CA, USA 14 5 Department of Physics, Ateneo de Manila University, Quezon City 1108, Philippines 15 6 Department of Civil & Environmental Engineering, University of California Irvine, Irvine, CA 92697, USA 16 7 NASA Jet Propulsion Laboratory, Pasadena, CA, USA 17 8 Center for Aerosol Science and Engineering, Department of Energy, Environmental and Chemical Engineering, 18


Introduction
As pollution emissions from Asian countries have surpassed those of countries in Europe and North America (Akimoto, 2003;Smith et al., 2011), Asia becomes increasingly important from a global climate and health 50 perspective.The tropical Western North Pacific (TWNP), situated adjacent to Southeast Asia (Fig. 1), is a receptor for multiple sources of aerosol particles throughout the region (Bagtasa et al., 2018;Hilario et al., 2020a;Huang et al., 2019;Reid et al., 2015) and is one of the most susceptible regions to global climate change (IPCC, 2014;Reid et al., 2013;Yusuf and Francisco, 2009).Amidst several multi-scale meteorological phenomena ranging from the Asian monsoon system (Akasaka et al., 2007;Chang et al., 2005), the El Niño Southern Oscillation (Cruz et al., 2013;Jose 55 et al., 1996), the Madden-Julian Oscillation (Maloney & Hartmann, 2001;Pullen et al., 2015), and intermittent typhoons (Bagtasa, 2017;Maloney and Dickinson, 2003), the TWNP hosts arguably one of the most complex meteorological environments in the world with likewise intricate relationships to aerosol lifecycle and climate impacts (Reid et al., 2012;Ross et al., 2018).
Owing to atmospheric residence times ranging from days to weeks (Balkanski et al., 1993;Kritz and Rancher, 1980) 60 and enabled by the surrounding meteorology, aerosol particles from multiple sources can undergo long-range transport into the TWNP (Lin et al., 2007;Xian et al., 2013).These sources include biomass burning from the Maritime Continent (MC) (Hilario et al., 2020a(Hilario et al., , 2020b;;Reid et al., 2015), anthropogenic and dust outflow from East Asia (EA) (Bagtasa et al., 2019;Braun et al., 2020;Geng et al., 2019;Miyazaki, 2003;Oshima et al., 2012;Tan et al., 2012), emissions from Peninsular Southeast Asia (PSEA) (Bagtasa et al., 2019;Geng et al., 2019;Huang et al., 2020;Lin et 65 al., 2009;Nguyen et al., 2020), and marine aerosol particles from the western Pacific (WP).Such transport is controlled by the interplay of several factors such as topography, sea breeze, monsoon flows, and typhoons (Reid et al., 2012;Wang et al., 2013b).Aside from the risk posed by transported anthropogenic aerosol on public health (Lelieveld et al., 2015;Zhang et al., 2017), such a diverse set of aerosol sources and types can result in variable aerosol-cloud-climate interactions (Hamid et al., 2001;Heald et al., 2014;Rosenfeld, 1999;Ross et al., 2018; 70 Sorooshian et al., 2009;Yu et al., 2006;Yuan et al., 2011), which are complicated further by the spatial inhomogeneity of transported aerosol particles (Akimoto, 2003).As the influence of aerosol particles on climate remains one of the largest uncertainties in our understanding of the atmosphere (IPCC, 2014), investigating the composition and transport mechanisms of air masses from different source regions will aid in the future development of transport models and lead to a better understanding of the transport pathways that modulate aerosol particles in this part of the world.

75
Previous aircraft campaigns in Asia and the Pacific include the Transport and Chemical Evolution Over the Pacific (TRACE-P) campaign (Jacob et al., 2003), the Aerosol Radiative Forcing in East Asia (A-FORCE) campaign (Oshima et al., 2012), the Pacific Exploratory Mission -West A and B (PEM-West) (Hoell et al., 1996(Hoell et al., , 1997)), and the Oxidant and Particulate Photochemical Processes Above a South East Asian Rainforest (OP3) project (Hewitt et al., 2010).These campaigns examined springtime outflow from the Asian continent (e.g., Koike et al., 2003;Kondo et al., 2004; 80 Park, 2005) and early-summertime characteristics of local and transported aerosol over Borneo (e.g., Robinson et al., 2011Robinson et al., , 2012)); however, no study to our knowledge has utilized aircraft data to characterize long-range transport patterns over the TWNP coinciding with the peak agricultural burning period for Indonesia and Malaysia.Limited ship observations in association with the 7 Southeast Asian Studies (7SEAS) program (e.g., Reid et al., 2015Reid et al., , 2016aReid et al., , 2016b) ) found a highly dynamic aerosol environment (Atwood et al., 2017;Hilario et al., 2020c;Reid et al., 2015).

85
The NASA Cloud, Aerosol, and Monsoon Processes-Philippines Experiment (CAMP 2 Ex) aircraft campaign examined the influence of meteorology, convection, and radiative effects on gas and aerosol species in the TNWP.Based at Clark, Luzon, Philippines, from 24 August to 5 October 2019, CAMP 2 Ex obtained a wide array of measurements between 0 -9 km above ground level (AGL) across 19 research flights (RF).Some RFs were conducted in coordination with the seaborne research vessel R/V Sally Ride as part of the Office of Naval Research Propagation of 90 InterSeasonal Tropical OscillatioNs (PISTON) project (https://onrpiston.colostate.edu/).The CAMP 2 Ex campaign is unique in that it began during the peak of the Asian southwest monsoon (SWM) and coincided with an early monsoon transition (MT), which occurred in late-September instead of the more common time in October (Chang et al., 2005;Matsumoto et al., 2020).The early MT led to diverse transport patterns (Fig. 2) that offered an opportunity to examine long-range transport into the TWNP.Aircraft campaigns allow for vertically-resolved measurements of air mass properties, which are essential to better understand the atmosphere, as aerosol-cloud-climate interactions vary by altitude (Dahutia et al., 2019;Dong et al., 2017;Hansen, 2005;Mishra et al., 2015) and vertical transport can influence air mass composition (Matsui et al., 2011;Moteki et al., 2012;Oshima et al., 2012Oshima et al., , 2013)).Furthermore, as the main route of aerosol removal from the atmosphere, wet scavenging is a crucially important aspect of aerosol vertical profiles and are linked to significant 100 uncertainties in climate models (Neu & Prather, 2012;Wang et al., 2013).Vertically-resolved in situ observations in field campaigns targeting aerosol-cloud-meteorology interactions are necessary to advance understanding of scavenging processes to inform the spectrum of models ranging from smaller-scale process models to larger-scale climate models (MacDonald et al., 2018;Sorooshian et al., 2019).
As Asian emissions continue to increase as a consequence of rapid economic development, it is imperative to 105 understand the influence of long-range transport on air quality and aerosol-cloud-climate feedbacks in this region.In this study, we focus on characterizing transported air masses from four key regions: the Maritime Continent (MC: 5° S -6.8° N, 94.9° E -119.5°E), Peninsular Southeast Asia (PSEA: 10° N -23° N, 95° E -109.5°E), East Asia (EA: 22° N -44° N, 100° E -122° E and 30° N -44° N, 122° E -145° E), and the West Pacific Ocean (WP: 3° N -25° N, 130° E -145° E).Using air mass back trajectories to complement the CAMP 2 Ex data, this study aims to (1) identify 110 regional transport pathways into the TWNP and their associated synoptic conditions, (2) characterize air masses coming from different source regions in terms of composition and aerosol size distribution, and (3) estimate the influence of convection and precipitation on transported air masses.By examining how transport and scavenging mechanisms impact air mass composition, our results have implications for improving modeling of aerosol lifecycles in this meteorologically complex region of the world and guiding policymaking related to public health and climate.115

CAMP 2 Ex observations
A major goal of the 2019 CAMP 2 Ex aircraft campaign was to understand aerosol-cloud-climate feedbacks in the TWNP (Di Girolamo et al., 2018).Although multiple aircraft were deployed, this study focused on measurements made onboard the NASA P-3B Orion (N426NA) aircraft.Aircraft altitude (m AGL hereafter) was calculated as the 120 difference between GPS altitude and ground elevation provided by the Google Maps API, with an uncertainty of ±5 m.Dry optical size distribution data were collected by the Laser Aerosol Spectrometer (LAS; TSI Model 3340) and are presented as an integrated particle number concentration for diameters between 100 and 1000 nm (N100-1000nm; cm - 3 ).Uncertainty of N100-1000nm is estimated at 20%.Carbon monoxide (CO; ppm) and methane (CH4; ppm) mixing ratios were measured by a dried-airstream near-infrared cavity ringdown absorption spectrometer (G2401-m; PICARRO, 125 Inc.), with uncertainties of 2% and 1% and precisions of 0.005 ppm and 0.001 ppm, respectively.Ozone (O3; ppbv) measurements were conducted with a dual beam UV absorption sensor (Model 205; 2B Technologies) with an uncertainty of 6 ppbv.Non-refractory aerosol composition in the submicrometer range was measured with a High-Resolution Time-of-Flight Aerosol Mass Spectrometer (AMS; Aerodyne) (DeCarlo et al., 2008).The species of relevance to this study include sulfate (SO4 2-), ammonium (NH4 + ), nitrate (NO3 -), and organic aerosol (OA), all of 130 which are reported in units of µg m -3 with uncertainties up to 50%.The AMS was operated in 1 Hz Fast-MS mode and averaged to 30-second time resolution for this study, with campaign-averaged 1-sigma detection limits (in ug m - 3 ) as follows: 0.169 (OA), 0.039 (SO4 2-), 0.035 (NO3 -), 0.169 (NH4 + ).Mass concentrations below these detection limit values, which are sometimes negative due to the AMS difference method, are statistically equal to zero.Black carbon (BC; ng m -3 ) was measured with a Single-Particle Soot Photometer (SP2) (Moteki & Kondo, 2007, 2010), including 135 an uncertainty of 10%.We note that BC data were unavailable during one flight covering a major Borneo smoke event (RF9); thus, the BC value presented in Table 1 is likely under-represented compared to the AMS data.A Fast Integrated Mobility Spectrometer (FIMS) measured aerosol size distribution between 10 nm and ~600 nm with a concentration uncertainty of ~ 15% and size uncertainty of ~ 3% (Wang et al., 2017a(Wang et al., , 2017b(Wang et al., , 2018a)).
All aerosol data are reported at standard temperature and pressure (STP; 273 K, 1013 hPa).Only data collected from 140 outside of clouds via isokinetic sampling (McNaughton et al., 2007) were used to eliminate sampling artifacts from the shattering of large water and ice particles (Murphy et al., 2004).Background concentrations of each species were defined as the lowest 10th percentile of all CAMP 2 Ex data along vertical profiles for every 5 K range of potential temperature (Koike et al., 2003;Matsui et al., 2011a).Enhancements above these background concentrations are https://doi.org/10.5194/acp-2020-961Preprint.Discussion started: 28 September 2020 c Author(s) 2020.CC BY 4.0 License.denoted by Δ.For species ratios, only data with ΔCO > 0.02 ppm were included similar to past work (Kleinman et al., 145 2007;Kondo et al., 2011;Matsui et al., 2011b).
Only data along profiles extending vertically more than 2 km were considered for analysis as they provide a "snapshot" of the atmosphere with which we can compare more directly air mass characteristics across different altitudes.Data collected when the P-3B aircraft sampled directly over urbanized Luzon (13° N -15.8°N, 120° E -122° E) were excluded from analysis to minimize the impact of local emissions.Flight tracks and identified vertical profiles are 150 shown in Fig. S1 of the Supplementary Information (SI).

Back trajectory classification
The National Oceanic and Atmospheric Administration (NOAA) Hybrid Single Particle Lagrangian Integrated Trajectory Model (HYSPLIT) (Rolph et al., 2017;Stein et al., 2015) was used to generate 120-hour back trajectories along vertical profiles with one minute temporal resolution.Input meteorological data was from the National Centers 155 for Environmental Prediction (NCEP) Global Forecast System (GFS) at a horizontal resolution of 0.25° × 0.25°.
Our analysis focused on transport from key source regions (MC, PSEA, EA, WP).We note here that "source region" refers to the attributed origin of an air mass as identified by our trajectory classification scheme and does not preclude the possibility of entrainment from emission sources during transport (e.g., shipping).Our classification scheme considered two important environmental factors: (1) the synoptic shift that occurred around 20 Sep 2019, dividing the 160 CAMP 2 Ex period into the SWM (before 20 Sep) and MT (after 20 Sep); and (2) the vertical wind shear across the region (Fig. 2).To better capture the pronounced effect of the monsoon shift, air masses were only classified as MC or PSEA (EA or WP) if sampled during the SWM (MT).For example, instances of EA air sampled during the SWM were classified as "Other" while air from EA sampled during MT were classified as EA.The inclusion of a monsoon phase filter more explicitly highlights the temporal aspect of meteorology in the TWNP; however, without this 165 monsoon phase criterion, resulting air mass classifications remain largely unchanged (Section 3.2).Furthermore, to account for regional vertical inhomogeneity (Atwood et al., 2013;Sarkar et al., 2018), our analysis of air mass characteristics differentiates between boundary layer (BL; < 2 km) and free troposphere (FT; > 2 km) (Section 3.3).
For an air mass to be classified into a source region, its back trajectory must pass within a source region's bounding box for more than 6 h at an altitude below 2 km, which is the typical summertime BL height in the region (Chien et 170 al., 2019).In addition to excluding data collected over urbanized Luzon (Section 2.1), trajectories passing through the Philippines (12° N -18.25°N, 120.5°E -122.5°E and 5.1° N -14.5°N, 122.5°E -126.7°E) under our trajectory classification scheme were excluded to further focus our analysis on long-range transport and associated processes.
Most air masses came from only one of the four source regions: WP (117 occurrences), MC (174 occurrences), PSEA (88 occurrences), EA (130 occurrences).Air masses that passed through both EA and WP (12 occurrences) were 175 considered as EA air due to the considerable influence of EA outflow on air mass composition (Talbot et al., 1997).Other transport permutations (e.g., air that passed through both MC and PSEA) did not meet the requirements of our classification scheme and were omitted.Thus, we focus on the four major transport pathways (MC, PSEA, WP, EA).Focusing on these major pathways adds robustness to the analysis by partly compensating for the limits of the trajectory model in capturing more complex meteorological phenomena such as wind shear (Freitag et al., 2014), 180 which have been shown to contribute to trajectory uncertainty (Siems et al., 2000;Stohl et al., 2002).In addition to requiring that the back trajectories pass through the source regions, the additional criteria imposed (e.g., altitude < 2 km over the source region) increase our confidence that the remaining cases represent instances of long-range transport.Resulting source region contributions per RF are shown in Fig. S2.We emphasize that these source region contributions represent frequencies of observation rather than frequencies of occurrence, as the sampling location of 185 the aircraft introduces a bias inherent in aircraft campaigns (Section 3.2).
As a consequence of our filtering scheme, a large portion of trajectories were tagged as "Other" (66.8%).This is attributable to several scenarios: (1) air masses that passed over source regions, but above our BL height threshold of 2 km; (2) air masses influenced by the Philippines (i.e., air masses that stayed over the Philippines at < 2 km for more than 6 hours); (3) stagnant air masses that did not reach any source region; and (4) other transport permutations that 190 occurred too infrequently to provide robust statistics.
Trajectory clustering was performed using two well-established methods: k-means and Ward linkage (Govender and Sivakumar, 2020) in order to confirm the robustness of our predefined source regions.K-means clustering classifies data into k clusters such that the sum of squares per cluster is minimized (Hartigan and Wong, 1979), with the drawback that k must be specified beforehand.Ward linkage is a hierarchical clustering method that merges clusters 195 such that the increase in intra-cluster Ward's distance is minimized (Ward, 1963) and has been described as the method that most closely accomplishes the goals of clustering (Tufféry, 2011).More comprehensive descriptions of these clustering methods can be found elsewhere (Govender and Sivakumar, 2020;Pérez et al., 2017).Prior to clustering, a weighted distance matrix was calculated, similar to Taubman et al. ( 2006): (1) normalized trajectory coordinates to give equal weighting to both horizontal and vertical transport; (2) weighted time steps linearly back in time; and (3) 200 assigned nearby points (time step < 6 h) zero weighting on the clustering to remove the influence of aircraft position on the clustering.

Accumulated precipitation along individual trajectories
Accumulated precipitation along individual trajectories (APT) was calculated using data from satellite precipitation products (SPPs): (1) the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks 205 -Climate Data Record (PERSIANN-CDR) (Ashouri et al., 2015;Nguyen et al., 2018); (2) the Integrated Multi-satellitE Retrievals for the Global Precipitation Measurement (GPM) mission (IMERG) (Huffman et al., 2019); and (3) the Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA) 3B42-V7 (Huffman et al., 2007).The purpose of utilizing multiple SPPs is to account for the uncertainties inherent in satellite retrievals, particularly during very light or heavy rainfall conditions, providing us with an ensemble of estimates rather 210 than relying on a single SPP (Chen et al., 2020;Liu, 2016;Maggioni et al., 2016;Mahmud et al., 2017;Tan & Santo, 2018).Furthermore, although these SPPs measure surface precipitation and do not fully capture scavenging effects aloft, we use APT as an indicator of whether an air mass passed through a convectively active region.
PERSIANN-CDR (0.25° × 0.25°, daily resolution) uses a modified PERSIANN algorithm utilizing NCEP Stage IV hourly precipitation and monthly precipitation from the Global Precipitation Climatology Project (GPCP) to maintain 215 monthly amounts consistent with GPCP (Ashouri et al., 2015).PERSIANN-CDR data are available at the Center for Hydrometeorology and Remote Sensing (CHRS) Data Portal (http://chrsdata.eng.uci.edu)(Nguyen et al., 2019).IMERG (0.1° × 0.1°, 30-min resolution) integrates multiple satellite retrievals from the passive microwave (MW) precipitation retrievals provided by the suite of GPM constellation passive microwave radiometers (Kummerow et al., 2015), the Climate Prediction Center MORPHing technique (CMORPH) from NOAA, and PERSIANN-Cloud 220 Classification System (PERSIANN-CCS; Hong et al., 2004) from the University of California, Irvine.These data are available from the NASA Precipitation Processing System (Skofronick-Jackson et al., 2018).For inter-calibrating various MW precipitation products, IMERG uses the GPM_2BCMB product (Olson et al., 2018) that utilizes the GPM Microwave Imager (GMI) and Dual-frequency Precipitation Radar (DPR) instruments on the GPM core satellite IMERG (Hou et al., 2014;Kidd et al., 2020).For this study, we use IMERG Final Run data, available at 225 https://pmm.nasa.gov/data-access/downloads/gpm.TMPA (0.25° × 0.25°, 3-h resolution) similarly combines data from multiple satellites such as TRMM (pre-2015), NASA's Aqua, and the NOAA satellite series, involving calibration with gauge data when feasible (Huffman et al., 2007).Though TRMM ended its service in 2015, the TMPA 3B42 algorithm was continued in parallel with IMERG through December 2019 and had been based on a climatological calibration since 2014.As TMPA is climatologically 230 calibrated, TMPA may be less sensitive to interannual variability in precipitation; thus, including TMPA in this study may provide a better idea of the spread among SPPs.TMPA data are accessible through https://pmm.nasa.gov/dataaccess/downloads/trmm.Precipitation along each trajectory was integrated from the trajectory spawn point (i.e., P3-B sampling location) to the point when it reaches the boundary of a source region.An additional 24 h along the trajectory after reaching a source 235 region was included in the APT integration to account for precipitation effects within the source region (Matsui et al., 2011a, b).No significant APT differences were found between using 0, 24, or 48 h for the APT calculation, suggesting that our results are robust with regard to the added duration.Furthermore, APT comparisons demonstrate that our results are independent of chosen SPP in terms of APT ranking (i.e., all SPPs agreed on which source regions are associated with the highest or lowest APTs).

255
Figure 1 provides an overview of the general source regions impacting the TWNP.The TWNP is surrounded by areas of high population density in EA, MC, and PSEA (Fig. 1b).Burning was prevalent mainly in the MC (Fig. 1c); though, fires were also detected along the eastern PSEA coast.Satellite retrievals of PBL SO2 reveal possible point sources (Fig. 1d), perhaps from volcanoes, shipping, burning, and industry (Fioletov et al., 2016;Guttikunda et al., 2001;Zhang et al., 2019); however, we caution that cloud contamination may influence the SO2 retrievals and are used here 260 to demonstrate the variety of sources in Asia.
Trajectories from each source region show distinct pathways (Fig. 3; left column), indicative of differences in accompanying synoptic-scale circulation.These pathways are generally corroborated by both k-means and Ward linkage clustering methods (Figs.S3 and S4), confirming the robustness of our predefined source regions (Fig. 1a).Prior to further discussion, we emphasize the temporal aspect of these observed transport patterns (Figs. 2 and 3), in 265 particular, their dependence on both synoptic (e.g., SWM) and mesoscale meteorology (e.g., typhoons), which varied during CAMP 2 Ex in terms of phase and frequency, respectively.Consequently, a specific transport pattern may be more dominant in one monsoon phase and less so in another while being enhanced (or suppressed) by intermittent mesoscale phenomena.

Southwest monsoon 270
Beginning with transport during the SWM prior to 20 Sep 2019, PSEA air is advected by uniform westerlies (Fig. 3a) associated with cyclonic activity over the northern South China Sea (SCS) (Fig. 3b) (Cheng et al., 2013;Huang et al., 2020;Lin et al., 2009).In comparison, although MC transport also occurs during the SWM, the mechanism behind MC transport is driven instead by southwesterlies originating across the MC (Fig. 3d) (Ge et al., 2014;Wang et al., 2013;Xian et al., 2013).Transport from the MC is further promoted by well-developed cyclones entering PSEA (Fig. 275 3e), as previously highlighted by observational (Hilario et al., 2020;Reid et al., 2015) and modeling studies (Wang et al., 2013).The similar cyclonic activity over northern SCS/PSEA may explain the confluence of air masses from PSEA and MC (e.g., RF6), indicated by the frequent sampling of MC and PSEA air in the same RF (Fig. S2).
A key difference between PSEA and MC air is that PSEA air passed through convective areas over the PSEA (Takahashi et al., 2010), the SCS (Fig. 3a, c) (Chen et al., 2017), and along the western coast of the Philippines 280 (Akasaka et al., 2007;Chen et al., 2017;Cruz et al., 2013;Hilario et al., 2020b) while MC air passed through areas with less precipitation (Figs.3d, f).As a result, PSEA air showed much higher APT than MC air (Table 1) and was more likely to have been processed by clouds.The transport pathway of PSEA through convective regions may lead to wet scavenging and aqueous-phase processing (MacDonald et al., 2018;Moteki et al., 2012;Sorooshian et al., 2006Sorooshian et al., , 2007;;Wonaschuetz et al., 2012), affecting both air mass composition and particle size distributions (Section 3.3).

285
In terms of sampled air masses, PSEA and MC showed contributions of 5.7% and 11.3%, respectively (Fig. 4a), and differ in terms of vertical distribution (Fig. 4b).PSEA air was sampled across a wide range of altitudes with the https://doi.org/10.5194/acp-2020-961Preprint.Discussion started: 28 September 2020 c Author(s) 2020.CC BY 4.0 License.majority of observations over 900 hPa, similar to Kondo et al. (2004), explainable by convection-related lofting (Fig. 5a).Very few observations of PSEA air were made near the surface.The lofting of PSEA air can occur over the PSEA itself (Fig. 3c) (Kondo et al., 2004), through mechanisms like orographic effects (Lin et al., 2009).Convection over 290 the SCS trough likely also contributes to lofting (Fig. 3c) (Chen et al., 2017).Lofting of PSEA air into the FT has important downstream ramifications as it modulates both aerosol composition and size distribution (Matsui et al., 2011a;Moteki et al., 2012;Oshima et al., 2013).However, we note that vertical motion is an important source of uncertainty in trajectory models (Harris et al., 2005) and should be interpreted with caution.
In contrast to PSEA air, sampled MC air was well-mixed within the BL (Fig. 5b), but the observation frequency of 295 MC air dropped sharply above 750 hPa, consistent with previous modeling studies (e.g., Xian et al., 2013).Such distinct vertical distributions between MC and PSEA air are perhaps due to highly sheared environment during the SWM, generally contributing to distinct air mass sources across different altitudes (Atwood et al., 2013;Sarkar et al., 2018) and varying degrees to which these air masses are processed (Section 3.3).

Monsoon transition 300
The arrival of the MT period after 20 Sep 2019 led to a synoptic-scale shift (Fig. 2), allowing the sampling of air from EA and WP (Fig. 3g, j).Transport from EA was observed across several MT flights (Fig. S2) and originated mainly from southeastern China, Korea, and Japan (Fig. 3g), suggesting the entrainment of anthropogenic emissions (Section 3.3) (Cheng et al., 2013;Hatakeyama et al., 2001Hatakeyama et al., , 2004;;Kim et al., 2009;Wang et al., 2016).Depicted in Fig. 3h, Asian outflow was promoted by the pairing of a well-developed cyclone passing over the East China Sea (Hatakeyama 305 et al., 2001(Hatakeyama 305 et al., , 2004;;Uno et al., 1998) and an anticyclone over the Asian continent (Honomichl and Pan, 2020).In comparison, WP transport was observed mainly towards the end of the campaign (Fig. S2), likely a consequence of sampling location, and was driven by Pacific northeasterlies (Figs.3jk).Transport from the WP, similar to that of EA, coincided with an anticyclone over the Asian continent (Fig. 3k); however, WP transport is marked by the absence of the East China Sea cyclone that promoted southward transport of EA air (Fig. 3h).This difference may explain why 310 EA and WP air were usually sampled in separate RFs (Fig. S2), in contrast to PSEA and MC air, which tended to be sampled together.
Air from EA and WP show similarly low APT (Table 1), explainable by the generally lower precipitation in MT (Figs. 3i, l) compared to SWM (Figs. 3c, f) (Matsumoto et al., 2020), as well as the lower number of cyclone occurrences after 20 Sep 2019.Although EA transport was driven by a well-developed cyclone (Fig. 3h), trajectories suggest that 315 EA air traveled through the outer bands of the cyclone (Fig. 3g), largely avoiding high precipitation areas (Fig. 3i).This suggests that anthropogenic emissions entrained in EA air experienced low levels of scavenging and were more likely to be sampled, in contrast with high APT urban source regions like PSEA (Section 3.3).
Transport from EA and WP were quite similar in terms of relative contribution (8.5% and 7.6%, respectively; Fig. 4a) and vertical sampling distribution (Fig. 4b).Sampling of EA and WP air were largely constrained to the BL, though 320 sampled EA air was unimodal while WP air was more evenly sampled.In terms of vertical motion during transport, some EA trajectories exhibited downward motion (Fig. 5c), likely due to subsidence from the continental anticyclone (Fig. 3h), contrasting the vertical motion of PSEA air, which generally experienced upward motion associated with convection (Fig. 5a).
In summary, important transport features over the TWNP include the following: (1) SWM transport from the MC and 325 PSEA was driven by southwesterlies and cyclonic activity over northern SCS/PSEA while MT transport from EA and the WP was driven by Pacific northeasterlies, anticyclones over the Asian continent, and well-developed cyclones over the East China Sea; (2) EA and MC air were sampled largely within the BL, did not exhibit significant upward motion, and experienced low APT, suggesting that they likely carry urban/continental or biomass burning emissions; in contrast, (3) PSEA air may have undergone a high degree of aerosol scavenging over convective regions (e.g., 330 SCS), indicated by high APT and upward motion of trajectories.

Sensitivity analysis
In order to assess the uncertainty associated with our trajectory classification, we evaluated the effect of the following variables on source classification: (1) trajectory height threshold over source regions; (2) back trajectory run time; (3) vertical profile filtering; (4) monsoon phase; and (5) aircraft sampling location.Results are provided in Table S1 and https://doi.org/10.5194/acp-2020-961Preprint.Discussion started: 28 September 2020 c Author(s) 2020.CC BY 4.0 License.summarized below.Except for aircraft sampling location, independently changing any of these variables had little effect on the resulting source-region distribution.The relative contributions of source regions did vary significantly with sampling location, though areas surrounding Luzon (e.g., East of Luzon, North of Luzon) showed some degree of similarity.Thus, we emphasize that, as with any aircraft campaign, observed transport is to some degree dependent on aircraft location.

340
In order to reduce the effect of local emissions, we excluded trajectories classified as influenced by the Philippines (PH).To evaluate our filter for Philippine-influenced trajectories (hereafter, PH filter), air mass characteristics were compared between transported air unaffected by PH (No-PH air; e.g., MC) and transported air mixed with PH air (With PH; e.g., MC-PH).A local signal was observed for N100-1000nm, suggested by differences in the histograms of N100-1000nm between non-PH and PH-mixed air (Fig. S5), particularly for MC and PSEA air.Differences in the species 345 concentration histograms of non-PH and PH-mixed air were also observed for other anthropogenic species (BC, OA, SO4 2-; not shown), confirming the effectiveness of the PH filter.

Chemical composition of transported air masses
A convenient opportunity afforded by having conducted the air mass classification presented above was to examine how gas and aerosol properties vary for each source region based on vertically-resolved in situ aircraft measurements.

350
To account for regional vertical wind shear (Fig. 2) while considering the generally lower classification frequency at higher altitudes (Fig. 4b), air mass characterization was resolved into BL and FT subsets and composited by source region (Table 2).The delineation between BL and FT composition is demonstrated by selected species (Fig. 6), which generally dropped in concentration above the BL (< 850 hPa).Prior to further discussion, we note here that shipping is a major regional source (Streets et al., 1997(Streets et al., , 2000) ) and may contribute appreciably to all air masses.355 Significant differences in composition were observed in the same monsoon season (e.g., SWM) depending on air mass origin.Air from PSEA had much lower species concentrations than MC (Table 2) due to decreased emissions and increased potential for wet scavenging.Sampled MC air showed statistically significant differences between BL and FT concentrations for both gas and aerosol species (Table S2), indicative of emissions constrained to the BL, and exhibited strong biomass burning signals in its composition profile (e.g., N100-1000nm, CO, NO3 -, OA, and BC) (Maudlin 360 et al., 2015;Pósfai et al., 2003;Reid et al., 1998Reid et al., , 2005;;Theodoritsi et al., 2020;Yadav et al., 2017).We note that peaks of CO (Fig. 6b) and CO2 (not shown) were observed in MC samples at around 650 hPa, suggestive of MC burning emissions lofted into the FT; however, this feature consisted of few samples and did not appear in other gases (e.g., SO2) and thus warrants caution in its interpretation.
In contrast, PSEA air was generally characterized by concentration magnitudes between those of MC and WP.Aerosol 365 concentrations of PSEA air in the FT were lower by at least an order of magnitude than those in the BL (SO4 2-, NH4 + , OA, BC) while trace gases (CO, CH4) showed more similar BL and FT concentrations (Tables 2 and S2; Fig. S6).These aerosol-gas differences point to: (1) the lofting of PSEA air into the FT, as suggested by the similarity of trace gas concentrations between BL and FT, and (2) the consequent scavenging of aerosol particles, explaining the much lower aerosol concentrations in the FT (Oshima et al., 2012(Oshima et al., , 2013;;Sievering et al., 1984).For comparison, MC air 370 showed large BL and FT differences in both aerosol and gas species, the latter of which indicates the infrequent lofting of MC air (Figs.4b and 5b).Since PSEA air came from a populated region (Fig. 1b) and likely originally contained anthropogenic aerosol particles, these unique characteristics of PSEA air compared to MC and EA air support the likelihood of aerosol scavenging in PSEA air.These observations are robust due to the relatively even sampling frequency of PSEA across altitudes (Fig. 4b).

Species ratios
Composition profiles between regions (Table 2) reveal clear differences as a function of (1) emission source and (2) passage through convective regions indicated by APT (Table 1).The role of emission source was most evident when comparing air masses of low APT (EA, MC).Though EA and MC had similar BL values for N100-1000nm (Table 2), 385 they showed distinct chemical differences: MC air was characterized by higher concentrations of biomass burning tracers (e.g., CO, CH4, NO3 -, OA) while EA air showed influence from urban/continental sources and secondary formation (e.g., O3, CH4, NH4 + , SO4 2-).Such differences in composition are corroborated by species ratios derived with linear regression (Fig. 7).Prior to a discussion on the species ratios, we note that the reported species ratios in this study are difficult to compare directly to at-source measurements of the same quantity because the composition 390 of air masses was likely influenced by sources and sinks during transport (e.g., Choi et al., 2019;Conte et al., 2019;Gruber et al., 2019;Yang et al., 2009); however, differences in species ratios can still aid in air mass characterization and point to possible emission sources.
In Fig. 7a, the lower ΔCO/ΔCO2 ratio of EA air versus MC air (Fig. 7a) is indicative of an inefficient combustion signature in MC air (Halliday et al., 2019), attributable to the predominantly smoldering phase of MC fires (Gras et 395 al., 1999;Reid et al., 2013).We note that the poor ΔCO-ΔCO2 correlation for MC air indicates that our reported ratio does not reflect expected emission ratios (Andreae, 2019;Hurst et al., 1994).This further suggests (1) our source region classification (i.e., MC, EA) may not perfectly capture air mass differences, and (2) additional sources of CO or CO2 during transport.Thus, it is necessary to use multiple species ratios to supplement air mass chemical characterization.

400
The strong relationship between CH4 and CO in MC air is a good indication of biomass burning influence (Andreae, 2019;Hecobian et al., 2011).The ratio of ΔCH4/ΔCO (Fig. 7b) was much higher in EA air compared to MC air, indicating the dominance of CH4 from residential and industrial activity (Geng et al., 2019;He et al., 2019;Tohjima et al., 2014) as well as from rice cultivation in EA (Xia et al., 2020).
Although MC was calculated to have low APT (Table 1), a comparison of BL and FT air from MC (Figs.S6 and S7) allows for speculation on a possible scavenging mechanism acting on FT air.Linear regressions of ΔSO4 2-/ΔCO (Fig.

415
S7a) suggested the removal of SO4 2-from FT air while ΔOA/ΔCO (Fig. S7b) indicated no such effect on OA.Considering the higher hygroscopicity and therefore scavenging susceptibility of SO4 2-compared to OA (e.g., Kreidenweis and Asa-Awuku, 2014), we speculate the removal of SO4 2-to be related to wet scavenging.Comparisons of BL and FT species concentrations (Fig. S6) further supports the possibility of this hypothesized scavenging, as aerosol species have significantly lower concentrations in the FT compared to BL, a difference not observed for trace 420 gases.The potential for scavenging was also supported by number size distributions (Section 3.4).
We note that this wet scavenging mechanism is not apparent from APT, which suggested dry conditions for MC air (Table 1).This disagreement with APT stems from the usage of SPPs which typically describe surface precipitation and, consequently, our APT may not fully capture potential wet scavenging effects aloft.Thus, the speculated scavenging mechanism of MC air in the FT may occur at higher levels and may not be strictly linked to surface 425 precipitation.Perhaps, this mechanism is related to processes such as in-cloud scavenging (Sievering et al., 1984;Yang et al., 2020;Yu et al., 2020) but, indeed, this requires a deeper investigation in future work.
Due to BC's lack of secondary sources, the ratio of ΔBC/ΔCO has been used to gauge transport efficiency as affected by physical removal processes on air masses (Matsui et al., 2011b;Moteki et al., 2012;Oshima et al., 2012) and as an https://doi.org/10.5194/acp-2020-961Preprint.Discussion started: 28 September 2020 c Author(s) 2020.CC BY 4.0 License.indicator of combustion efficiency, which increases with ΔBC/ΔCO (Kondo et al., 2011).The ratio of ΔBC/ΔCO was 430 much higher in EA air than in MC air (Fig. 7f), similar to observations by Pani et al. (2019) in Taiwan.This difference is explainable by burning in industrial and residential areas in EA (Bond et al., 2004;Geng et al., 2019) and the predominance of smoldering fires in the MC (Gras et al., 1999;Reid et al., 2013), which yield a lower ΔBC/ΔCO than flaming fires (Kondo et al., 2011).

Size distributions of transported air masses 435
To more deeply characterize the air masses from different source regions, we examine the differences in normalized FIMS number (Fig. 8) and volume (Fig. S8) size distributions between BL and FT, which can also offer insights into the convection-related removal of PSEA air.In-cloud processing during transport may influence particle sizes in these air masses, whereby a combination of the following processes can occur (e.g., Ervens, 2015;Sorooshian et al., 2007;Wonaschuetz et al., 2012), followed by detrainment from the cloud or wet removal: (1) collisions between interstitial 440 aerosol and droplets; (2) coalescence among droplets; and (3) aqueous-phase processing in droplets.However, comparisons between size distributions between regions and between BL and FT may still lend valuable insights into transport-related processes (e.g., Moteki et al., 2012).
Firstly, comparing source regions of low APT and high aerosol concentration, EA air in the BL (Fig. 8a) showed a wider peak in its size distribution (40 -200 nm) than that of MC (Fig. 8b), which showed a clear unimodal peak (100 445 nm).This was perhaps an effect of multiple sources contributing to EA air masses, ranging from industrial activities to rice cultivation (Chen et al., 2020b;Geng et al., 2019;Wang et al., 2016;Xia et al., 2020).In comparison, biomass burning emissions from the MC have been shown to greatly influence air mass composition (Engling et al., 2014;Fujii et al., 2015;Santoso et al., 2011) and, by extension, such a dominant emission source can explain the large unimodal peak in MC's BL size distribution (Figs.8b and S8b).

450
A comparison of the FT and BL size distributions of MC air may point to a potential scavenging mechanism of MC air lofted into the FT, signaled by significant BL and FT differences above 50 nm (Fig. 8b).This potential scavenging mechanism of MC air was previously proposed using species ratios (Section 3.3.1),and the size distribution here provides further evidence for the hypothesized removal process; however, we emphasize that the hypothesized scavenging mechanism is for now speculative and warrants future examination.

455
In contrast, the size distribution of PSEA air in the BL shows two peaks at around 50 nm and 200 nm.Comparing PSEA and MC air in the BL reveals much smaller particle sizes in PSEA air (Fig. 8c), explainable by differences in source emissions as well as in APT.A comparison of BL and FT air from PSEA pointed to scavenging during lofting into the FT: the FT size distribution of PSEA air showed a sharp drop in particle number concentrations above 50 nm while the BL size distribution of PSEA air was much broader.In fact, the size distribution of FT air from the PSEA is 460 more similar in shape to that of WP air (Fig. 8d), which is representative of background FT air.Due to this similarity, the peaks at 30 nm in FT air from PSEA and WP may originate from new particle formation (Williamson et al., 2019) that has been shown to be connected to APT in marine environments (Ueda et al., 2016), such as the convectively active SCS.

465
The relationship between composition and convection was further investigated through scatterplots of ΔBC/ΔCO ratio, an indicator of physical removal processes (Moteki et al., 2012), as a function of APT, an indicator of convection.The decrease in ΔBC/ΔCO ratio with APT (Fig. 9a) indicates that convection during transport is one of the main scavenging mechanisms in the TWNP.Both EA and MC air showed very high ΔBC/ΔCO ratios compared to PSEA, indicative of more efficient transport, while having low APT, which allowed for a clear transport signal as shown by 470 the high concentrations of anthropogenic or burning species in these air masses (Section 3.3).In contrast, PSEA air is characterized by a lower ΔBC/ΔCO ratio coinciding with high APT (Fig. 3c; Table 1), pointing to particle scavenging over convective areas.To further demonstrate the impact of convection on transport efficiency, Fig. 9b reveals ΔBC/ΔCO distributions resolved by low and high APT.The shift towards higher ΔBC/ΔCO ratios under low APT implies that dry and non-convective conditions are conducive to transport, suggesting the higher BL concentrations 475 of anthropogenic species in MC, EA air were likely enabled by lower levels of wet removal.

Summary and Conclusions
Utilizing airborne CAMP 2 Ex measurements between 24 Aug and 5 Oct 2019 and HYSPLIT back trajectories, we examined transport patterns into TWNP from key source regions (PSEA, MC, EA, WP).Key conclusions from this study include the following: 480 1. Meteorological phenomena driving transport as well as the origins of transported air masses shifted significantly with the monsoon phase.During the SWM, MC and PSEA transport were associated with monsoon-driven southwesterly winds and cyclonic activity over the northern SCS.Wind shear was associated with predominantly BL (FT) sampling of MC (PSEA) air, implying distinct aerosol processing between these two source regions.In comparison, transport during the MT period from EA and WP was driven by 485 northeasterly winds from the Pacific, anticyclones over the Asian continent, and well-developed cyclones passing through the East China Sea.These transport differences led to varying degrees of convection experienced by transported air masses.PSEA air generally passed through convective regions (SCS, west of Luzon, and over the PSEA itself) and was lofted into the FT, which led to scavenging of aerosols.In contrast, air masses from the MC and EA underwent relatively little convection, indicated by low APT, and mainly 490 were confined to the BL, enabling the transport of anthropogenic emissions.2. Characteristics of transported air masses differed primarily by emission source and passage through convective regions.Due to low APT and high ΔBC/ΔCO, transported air from MC and EA exhibited characteristic emissions: MC air from biomass burning (CO, well-correlated CO and CH4, NO3 -, OA) and EA air from anthropogenic outflow and secondary formation (O3, CH4, NH4 + , SO4 2-).Key species ratios 495 corroborated distinct sources between MC and EA air.Aerosol size distributions in EA air suggested multiple primary sources (industry, residential emissions, rice cultivation) as well as secondary formation, indicated by its relatively broad peak; in contrast, the narrower peak in the size distribution of MC air pointed to the predominance of biomass burning emissions.3. Air from the PSEA showed strong evidence of particle scavenging: passage over high precipitation areas, 500 convective lofting, high APT, low ΔBC/ΔCO, relatively low levels of anthropogenic species, and a size distribution shifted towards smaller particle sizes.Aerosol concentrations of PSEA air in the FT were lower by at least an order of magnitude than those in the BL, a difference that was not observed for trace gases, which pointed to scavenging of aerosol particles in the FT.Furthermore, PSEA air in the FT lacked the larger peak (Dp = 200 nm) observed in BL air and instead peaked at much smaller sizes (Dp = 30 nm), suggesting 505 large particle removal during convective lofting.The fine mode peak (Dp = 30 nm) for PSEA FT air also resembled that of WP air, suggestive of new particle formation during transport from the PSEA, perhaps occurring over the convective SCS. 4. A possible wet scavenging mechanism for MC FT air was inferred from ΔSO4 2-/ΔCO and ΔOA/ΔCO ratios between BL and FT, and corroborated by size distributions, which showed significant BL and FT differences 510 for larger particles (> 50 nm).The disagreement with APT was attributed to SPP limitations in capturing scavenging effects aloft, hinting that the scavenging mechanism acts at higher layers and may not be linked to surface precipitation.However, we emphasize that the exact scavenging mechanism is for now speculative and warrants its own investigation in the future.
Recommendations for future work include: (1) investigating the hypothesized scavenging mechanism of MC air aloft

530
Author contributions.MRAH performed the analysis and prepared the manuscript.All authors provided input for the manuscript and/or participated in data collection and processing.

Figure 1 :
Figure 1: Maps of (a) ground elevation from the Global Multi-resolution Terrain Elevation Data 2010 (GMTED2010), flight tracks in red, and approximate source regions in labeled boxes: Peninsular Southeast Asia (PSEA), Maritime Continent (MC), East Asia (EA), and West Pacific (WP), (b) 2020 population density from the Center for International Earth Science Information Network (CIESIN) Gridded Population of the World (GPW) v4, (c) MODIS active fire hotspot density (only fires tagged with > 80% confidence) averaged at 0.5° × 0.5° resolution from 1 Aug to 15 Oct 2019, and (d) OMI-retrieved PBL SO2 averaged over the same period.

Figure 3 :
Figure 3: Classified trajectories and synoptic conditions associated with transport from (ac) Peninsular Southeast Asia