Measurement report: quantifying source contribution of fossil fuels and biomass-burning black carbon aerosol in the southeastern margin of the Tibetan Plateau

Anthropogenic emissions of black carbon (BC) aerosol are transported from Southeast Asia to the southwestern Tibetan Plateau (TP) during the pre-monsoon; however, the quantities of BC from different anthropogenic sources and the transport mechanisms are still not well constrained because there have been no high-time-resolution BC source apportionments. Intensive measurements were taken in a transport channel for pollutants from Southeast Asia to the southeastern margin of the TP during the premonsoon to investigate the influences of fossil fuels and biomass burning on BC. A receptor model that coupled multi-wavelength absorption with aerosol species concentrations was used to retrieve site-specific Ångström exponents (AAEs) and mass absorption cross sections (MACs) for BC. An “aethalometer model” that used those values showed that biomass burning had a larger contribution to BC mass than fossil fuels (BCbiomass = 57 % versus BCfossil = 43 %). The potential source contribution function indicated that BCbiomass was transported to the site from northeastern India and northern Burma. The Weather Research and Forecasting model coupled with chemistry (WRF-Chem) indicated that 40 % of BCbiomass originated from Southeast Asia, while the high BCfossil was transported from the southwest of the sampling site. A radiative transfer model indicated that the average atmospheric direct radiative effect (DRE) of BC was +4.6± 2.4 W m−2, with +2.5± 1.8 W m−2 from BCbiomass and +2.1± 0.9 W m−2 from BCfossil. The DRE of BCbiomass and BCfossil produced heating rates of 0.07± 0.05 and 0.06± 0.02 K d−1, respectively. This study provides insights into sources of BC over a transport channel to the southeastern TP and the influence of the cross-border transportation of biomass-burning emissions from Southeast Asia during the pre-monsoon.


Introduction
The Tibetan Plateau (TP) is an important regulator of climate change in the northern hemisphere and it plays a crucial role in the functions of global ecosystem and climate stability (Liu et al., 2019bLiu et al., 2020a. The TP is covered by one of the largest ice masses on Earth, and it has been called the water tower of Asia (Liu et al., 2020b). The glaciers on the TP are facing rapid retreat, however, and if 10 unchecked, that could result in adverse effects on Asian hydrological cycle and Asian monsoon (Luo et al., 2020;Hua et al., 2019). In spring, the glaciers on the TP begin to melt as part of the natural hydrological cycle, but the increasing quantities of black carbon (BC) aerosol transported to the TP has accelerated this process (Bond et al., 2013) by causing a warming effect in atmosphere over the TP and enhancing the absorption of radiation on the surface of the glaciers (Ming et al., 2009). information on the contributions of different sources of BC over the TP is lacking, but it is critically needed for a better understanding the influence of anthropogenic emissions on its environment and climate.
Several studies have assessed the contributions of different BC sources through model simulations or isotopic methods. For example, Zhang et al. (2015) investigated BC sources for different parts of the TP 5 by using a chemical transport model and a source tagging approach, and they found that the contributions of BC sources varied among regions and with the seasons. Li et al., (2016) used filter sampling and carbon isotopes (∆ 14 C and 13 C) to determine the BC from fossil fuels and biomass burning in several areas of the TP. A major disadvantage of filter-based measurements is they are constrained by low time resolution, and this makes it challenging to capture the detailed evolution of pollution events. On the other hand, the 10 accuracy of model simulations is dependent on many factors, including uncertainties associated with initial particle parameters, aging processes, the accuracy of emission inventory, meteorological fields over the complex terrain, and the modules for chemistry and planetary boundary layer (PBL) dynamics, etc. (Koch et al., 2009;Madala et al., 2014;Vignati et al., 2010). Nonetheless, these studies have been helpful for understanding the sources of BC over the TP. 15 To make up for the deficiencies of filter-based analysis, BC source apportionments based on high-time resolution online data has been conducted in many locations (e.g., Herich et al., 2011;Zhu et al., 2017;Rajesh and Ramachandran, 2018) but for the TP are limited. An 'aethalometer model' based on multiwavelength absorption data is one of efficient approaches for distinguishing between BC from fossil fuel and biomass burning sources (Sandradewi et. al., 2008). The accuracy of the 'aethalometer model' relies 20 on the input parameters, including absorption Ångström exponent (AAE) and BC mass absorption crosssection (MAC) of different sources (Zotter et al., 2017). Limited information on site specific AAEs and MACBCs lead most studies to rely on values taken from measurements made in other locations (e.g., Healy et al., 2017;Zhu et al., 2017). This results in unquantified uncertainties because the AAE and MACBCs can vary with specific fuel subtypes and combustion conditions (Wang et al., 2018;Tian et al., 2019). 25 Therefore site-dependent AAE and MACBC are essential for improving the reliability of BC source apportionment by the 'aethalometer model'.
In this study, field measurements of BC were taken on the southeastern margin of the TP during the premonsoon. This region connects the high altitude TP with the low altitude Yungui Plateau and forms a transport channel for pollutants from Southeast Asia (Wang et al., 2019a), and it is an ideal region for investigating the impact of pollutant transport to the southeastern TP. A receptor model that combined with multi-wavelength absorption with aerosol species concentrations was used to retrieve site-dependent 5 AAEs and BC MACBCs. This was done to improve the 'aethalometer model' with the goal of obtaining a more accurate BC source apportionment. The primary objectives of this study were to (1) quantify the mass concentrations of BC from fossil fuel and biomass burning sources; (2) determine the impact of regional transport on source-specific BC, and (3) assess the radiative effects caused by BC from different sources. This study provides insights into the BC sources on the southeastern TP and an assessment of 10 their radiative effects during the pre-monsoon.

Sampling site
Intensive field measurements were made at the rooftop of a building (~10 m above the ground) at the Lijiang Astronomical Station, Chinese Academy of Sciences (3260 m above sea level, 100°1'48"E, During the campaign, westerly winds created a potential pathway for cross border transport from southeast Asia to southwest China. During the study, the average relative humidity and temperature were 80% ± 20% and 7.6 ± 3.2℃, respectively; the mean wind speed near surface was 5.4 ± 2.1 m s -1 , and the winds were mainly from the west and southwest. The sampling site is 3-5 km from Gaomeigu village, 20 which has 27 households and 110 residents. Villagers there rely on farming for their livelihoods, and biomass is the primarily residential fuel (Li et al, 2016). There are no large industries near the village and traffic is light. However, two highways (Hangzhou-Ruili Expressway and Dali-Nujiang Expressway) are located ~5.5 km to the west of the sampling site.

Online and offline measurements
Aerosol light absorption coefficients at multiple wavelengths (babs(λ), λ = 370, 470, 520, 590, 660, 880, and 950 nm) were retrieved with the use of a model AE33 aethalometer (Magee Scientific, Berkeley, CA, USA). The sampled particles were selected by a PM2.5 cuff-off inlet (SCC 1.829, BGI Inc. USA) and dried with a Nafion® dryer (MD-700-24S-3; Perma Pure, Inc., Lakewood, NJ, USA) , and the flowrate 5 of the sampler is 5 L min -1 . Detailed operating principles of the AE33 aethalometer can be found in Drinovec et al. (2015). Briefly, light at wavelengths (λ) = 370, 470, 520, 590, 660, 880, and 950 nm emitted from diodes is used to irradiate aerosol deposition spots on the filters. The light attenuation produced by the captured particles is measured with optical detectors. Non-linear loading and filter matrix scattering effects are common issues for filter-based absorption measurements (Coen et al., 2009). A dual-10 spot technique for the aethalometer measurements was used to compensate for non-linearity, while a factor of 2.14 was used to correct the artifacts caused by quartz filters (Drinovec et al., 2015).
The babs(532) was measured simultaneously with an intracavity photoacoustic technique in the acoustic chamber. Detailed description of PAX may be found in Carrico et al. (2018). During the campaign, the selected concentrations of ammonium sulphate and freshly-generated propane soot were used to calibrated the bscat(532) and babs(532) measurements, respectively. Details regarding the calibration 6 Technology, Santa Clara, CA). Detailed descriptions of the chemical analyses are given in Text S1 in the Supplementary Material.

BC source apportionment
The 'aethalometer model' proposed by Sandradewi et. al. (2008) was optimized by excluding the babs (370) contributed by the secondary aerosols and soil dust (babs(370)secondary and babs(370)dust, respectively). The 5 The formulae used for the 'aethalometer model' were as follows: where AAEfossil and AAEbiomass are the AAEs for emissions of fossil fuel contribution and biomass burning.
These were retrieved through an optical source apportionment and discussed in section 3.
where MACBC(880)fossil and MACBC(880)biomass are the MACBCs at λ = 880 nm generated from fossil fuel combustion and biomass burning, respectively. The MACBC(880)s for the two sources were retrieved from a source apportionment of the optical data discussed in section 3.1.

Optical source apportionment
Optical source apportionments were obtained using a positive matrix factorization (PMF) model. The 5 fundamental objective of PMF for applications such as ours is to resolve the chemical mass balance by separating data matrix into factor contributions and factor profiles as follows: where ij represents the input matrix elements; p is the number of sources; ik is the source contribution of the th factor to the th sample; kj is the factor profile of th species in the th factor; and ij is 10 representative of the residual. ik and kj are non-negative. The two matrices are resolved by minimizing the sum of squares of the normalized residuals as follows: where represents the object function; and ij denotes the uncertainties of ij . The PMF version 5.0 (PMF5.0, from USEPA) was used for the analyses, and the optical parameters (primary babs at different 15 wavelengths) and chemical species concentrations (including carbonaceous aerosols, inorganic elements, K + , levoglucosan and organic markers) were used as model inputs for the optical source apportionment.

Trajectory-related analysis
To determine the influences of regional transport on BC at Gaomeigu, trajectory clusters were produced from hourly three-day backward-in-time air mass trajectories at 500 m above the ground level. The 20 trajectories were calculated with the Hybrid Single-Particle Lagrangian Integrated Trajectory model (Draxler and Hess, 1998). The meteorological data were from Global Data Assimilation System (GDAS, https://www.ready.noaa.gov/gdas1.php), which take into account the influences of terrain. As we focused on differentiating and clustering the main spatial features of the incoming trajectories, an angle-oriented distance definition was adopted in the cluster analysis. Details regarding the trajectory clustering methods can be found in Wang et al. (2018). For the investigations of the effects of transport on the chemical and optical properties of the BC aerosol, trajectories with BC mass concentration greater than 75th percentile were considered as polluted.
The potential source contribution function (PSCF) was used to identify the likely pollution regions that 5 influenced BC loadings at Gaomeigu based on the back trajectories. The geographic region covered by the trajectories was overlaid by a of 0.5 × 0.5°(i, j) grid. The PSCF value of each grid was calculated as follows: where ij is the number of endpoints associated with BC mass concentration higher than the set criterion; 10 and ij is the total endpoints of the th cell. To improve the resolution of PSCF source identifications, the 75th percentile of each source's BC mass concentration was set as the criterion for a polluted sample (i.e., 0.6 μg m -3 for BCbiomass and 0.45 μg m -3 for BCfossil) (Cheng and Lin, 2001). Furtherrmore, arbitrary weighting factors ( ij ) were applied to different ij ranges to reduce the uncertainty caused by the small ij (Polissar et al., 1999), and they were defined by following the approach of Polissar et al. (2001)

Regional chemical dynamical model
The Weather Research and Forecasting model coupled with chemistry (WRF-Chem) model was used to quantify the contribution of biomass-burning from Southeast Asian to BC mass at Gaomeigu. Detailed descriptions of the model configurations have been described in our previous publication (Xing et al., 20 2020). Briefly, the model resolution was 3 km × 3 km, and there were 320 grid cells. The domain included the southwest of China, southern and southeastern Asia, with a central at 100.03°E, 26.70°N. Thirty-five vertical layers has been set in the model from the ground surface to 50hPa. The BC emission inventory used for the model was based on the Asian anthropogenic emission inventory (that is MIX) for the year of 2010 . The inventory has a spatial resolution of 0.25° × 0.25° and it included industry, power, transportation, and residential sources (e.g., fossil fuel and biofuel). The FINN fire inventory (Wiedinmyer et al., 2011) was used for the biomass-burning emission during the simulation. 5 The direct radiative effects (DRE) of source-specific BC were estimated with the widely used Santa Barbara DISORT Atmospheric Radiative Transfer (SBDART) model-a detailed description of which may be found in Ricchiazzi and Yang (1998). The important input parameters include aerosol optical depth (AOD), light extinction coefficient (bscat+babs), SSA, asymmetric parameter (ASP), and visibility.

Estimations of direct radiative effects and heating rate
Here, the input optical parameters were estimated by Optical Property of Aerosol and Cloud (OPAC)  Table S1) and used to obtain the optical parameters at the nearest observed relative humidity. The underlying assumption was that when the modelled bscat(λ), babs(λ) were very close to their measured counterparts, the derived optical parameters were a reasonable representation of the measured aerosols. This assumption has been widely used in previous studies (Dumka et al., 2018;Panicker et al., 2010;Rajesh et al., 2018). Finally, the DREs 20 attributable to source-specific BC (or PM2.5) at the surface atmosphere (SUF) and the top of the atmosphere (TOA) were estimated as the difference in the net flux with and without BC (or PM2.5) under cloud-free conditions. The solar heating rate change induced by atmospheric DRE (DRE at TOA subtracts DRE at SUF) was calculated as follows (Ramachandran and Kedia, 2010): where DRE ATM (W m -2 ) is the atmospheric DRE; DRE TOP is the DRE at top atmosphere; DRE SUF is the DRE at surface atmosphere; is the heating rate (K day -1 ); is the acceleration due to gravity (9.8 m s -2 ); is the specific heat capacity of air at constant pressure; and ∆ is the difference in atmospheric pressure between the ground and 3 km above.
3 Results and discussion 5

Source-dependent AAEs and MACs
Four sources were identified as the main contributors to primary babs(λ) based on the optical source apportionment (Fig. 2). The simulated primary babs (λ) values at different wavelengths all correlated well (r = 0.96 -0.97, p < 0.01, Fig. S1) with the model inputs, indicating that the PMF5.0 performed well. As shown in Fig. 2, the first source factor exhibited high contributions of K + (90%), levoglucosan (60%), 10 and primary babs(λ) (45 -64%) as well as moderate loadings of OC (38%) and EC (47%). The K + and levoglucosan are widely used markers for biomass burning (Urban et al., 2012), and these chemical markers are strong indications that this factor resulted from biomass burning emissions. Furthermore, we note that the presence of BrC in this factor made higher absorption at shorter wavelengths, which is consistent with absorption feature of biomass burning emissions (Forello et al., 2019). 15 Based on the contributions of biomass burning to babs(370) and babs (880), the AAEbiomass was estimated to be 1.7, which is within a relative boarder range of AAEbiomass (1.2 -3.5) determined by other methods (e.g., △ 14 C and organic tracers) in previous studies (Sandradewi et al., 2008;Helin et al., 2018;Harrison et al., 2012;Zotter et al., 2017). The estimated average MACBC(880)biomass was 10.4 m 2 g -1 ; this is more than twice the value for uncoated BC particles suggested by Bond and Bergstrom (2006) 20 (MACBC(880)uncoated = 4.7 m 2 g -1 , extrapolated from 550nm to 880 nm by assuming AAEBC = 1.0). The large MACBC (880) indicates that BC particles from biomass burning experienced substantial aging processes during their transport because numerous studies have confirmed that aged BC could result in MAC increases by a factor of 1.5-3.5 relative to uncoated particles Ma et al., 2020) due to the 'lensing effect' (Lack and Cappa, 2010). 25 The second source factor was characterized by large loadings of benzothiazolone (54%), Pb (46%), Br (40%), Cu (35%), Zn (27%), EC (36%), and OC (30%). Benzothiazolone is released from the breakdown of the antioxidant in motor vehicle tires (Cheng et al., 2006) while Br is another tracer of motor vehicle emission (Guo et al., 2009). Similarly, Zn and Cu are associated with the combustion of lubricating fluids and the wear of brakes and tires (Lough et al., 2005;Song et al., 2006). Finally, EC and OC also are 5 components of motor vehicle emissions (Cao et al., 2013). Although unleaded gasoline has been used extensively used in China since 2005, a considerable portion of Pb in the environment is still associated with vehicle-related particles, especially from the wear of metal alloys (Hao et al., 2019). Therefore, this second source factor was identified as traffic-related emissions. This source constitutes a moderate percentage of primary babs(λ) (15-30%). The estimated traffic emission-related AAE (AAEtraffic) was 0.8, 10 consistent with the finding that BC is the dominant light-absorbing carbonaceous aerosol species for traffic emissions (Kirchstetter et al., 2004). The AAEtraffic found here also was close to a value obtained using the △ 14 C approach (Zotter et al., 2017). The estimated MACBC(880) of traffic emissions (MACBC(880)traffic = 9.1 m 2 g -1 ) was similar with MACBC(880)biomass, indicating that traffic emissionrelated BC particles were also subjected to substantial aging. 15 The third factor was dominated by high loadings of As (70%), S (37%), and Cu (47%), which are typically associated with coal combustion (Hsu et al., 2016;Kim and Hopke, 2008). Although coal is not used extensively near the site on the TP, emissions from coal combustion may have been transported to site from surrounding areas (e.g., East Asia, Li et al., 2016). This source contributed 12-17% of primary babs(λ), which is less than that from biomass burning or traffic emissions. The obtained AAE of coal 12 The fourth source factor had high loadings of Ca (35%), Ti (66%), Mn (47%), and Fe (61%), consistent with the characteristics of crustal elements (Guo et al., 2009); thus, it was assigned to soil dust. The light absorption of soil dust is mainly due to the presence of iron oxides and varies with the types and relative concentrations of iron oxide species (Alfaro, 2004;Valenzuela et al., 2015). The mineral dust here contributed a relative small amount to the primary babs(λ) (6-9%), presumably due to the low levels of 5 iron oxides, and this is consistent with the other results obtained on the southeastern TP . The estimated AAE of mineral dust (AAEdust) was 1.5, which is within the range of 1.2-3.0 obtained from multi-non-oceanic sites (Dubovik et al., 2002).

BC source apportionment
As the results above showed that fossil fuel-related BC aerosol mainly originated from traffic and coal 10 combustion, the AAEfossil (0.9) and MACBC (880) (880)biomass), the mass concentrations of BCfossil and BCbiomass were then estimated using the improved 'aethalometer model' (Eqs. 1-6). As shown in Fig. 3a, no correlation (r = 0.01, p = 0.02) was 15 found between BCbiomass and BCfossil, implying that BC from these two sources was effectively separated by the improved aethalometer model. The data for the biomass-burning and traffic-related tracers (levoglucosan and benzothiazolone, respectively) further support the results of BC source apportionment.
That is, BCbiomass was significantly correlated with levoglucosan (r = 0.75, p < 0.01, Fig. 3b) and the same was true for BCfossil benzothiazolone (r = 0.67, p < 0.01, Fig. 3c), respectively. These results indicate that 20 the source-specific AAEs and MACs(880) obtained from optical source apportionment were appropriate Fig. 3d shows a time series plot of hourly averaged mass concentrations of total BC, BCbiomass, and BCfossil during the campaign. The hourly total BC mass concentration varied by ~50-fold from 0.1 to 4.9 μg m -3 , with an arithmetic mean (± standard deviation) of 0.7 (± 0.5) μg m -3 , which was lower than what has been reported for the western TP but higher than in the northern TP (Wang et al., 2018, and references therein). 25 The larger BC loading in the western TP can be explained by relatively strong influences from Southeast Asia, where anthropogenic actives are intensive (Kurokawa et al., 2013). With reference to BC sources, the concentrations and contributions as listed in Table 1, show higher mass concentrations of BCbiomass (0.4 ± 0.3 μg m -3 , 57% of total BC) compared with BCfossil (0.3 ± 0.2 μg m -3 , 43% of total BC) on average.
The mass fraction of BCbiomass increased with the BC loadings while BCfossil mass fraction showed an inverse relationship to the loadings (Fig. S2). One can infer from this that biomass-burning emissions were responsible for the high BC loading episode during the campaign. 5 Distinct diurnal variations in the mass concentrations of BCbiomass and BCfossil were observed as shown in Fig. 4a. The BCbiomass started to increase after midnight, reached a small peak at ~05:00 and then remained at a constant level before sunrise (~08:00). This may be attributed to effects associated with changes in the height of the planetary boundary layer (PBL) height (https://rda.ucar.edu/datasets/ds083.2) (Fig. 4b).
Thereafter, the BCbiomass increased again and reached the maximum value at midday. This enhancement 10 was accompanied by an increase in PBL height and higher wind speed (https://rda.ucar.edu/datasets/ds083.2) (Fig. 4b). Generally, higher PBLs and stronger wind cause local pollutants to disperse and as a result lower their loadings (Wang et al., 2015). However, thus the build up of BCbiomass in daytime at Gaomeigu was more likely influenced by the transport of BCbiomass from regions upwind. After sunrise, the PBL began to deepen and that was accompanied with west/southwest winds 15 from 08:00 to 12:00 (Fig. S3). These meteorological conditions are favourable for pollutants (including BC) transport from high-density biomass-burning emission areas to the sampling site (Chan et al., 2017).
After the midday peak, BCbiomass decreased sharply until midnight. The initial portion of this decrease (13:00 -18:00) occurred as the PBL height and wind speed increased, which promoted the dispersion of BCbiomass. Subsequent reduction occurred at night even though the PBL height and wind speed decreased, 20 and that was likely due to the curtailment of local biomass-burning activities.
As shown in Fig. 4a, the BCfossil showed diurnal trend that was roughly opposite trend that of the unimodal pattern seen for BCbiomass, and that may be explained as follows. Increases in the PBL height and wind speed from 09:00 to 15:00 were associated with a decrease in BCfossil, unlike the increasing trend seen for BCbiomass; this presumably reflects minor effects from the regional transport on BCfossil. Further, because 25 of the small contribution of coal combustion to EC (12%, Fig. 2c), the BCfossil was best explained by motor vehicle emissions from areas upwind of the site. The subsequent increase in BCfossil from 17:00 to 20:00 was attributed to the reduction of PBL height and as a result the build up of pollutants in the near surface air. As there were minimal impacts from traffic at night, the BCfossil loadings remained steady from 21:00 to 08:00. The stable nocturnal BCfossil may reflect the impact of fossil fuel emissions on BC in the southeastern margin of the TP due to the accumulation resulting from the low PBL heights.

Regional influences of BC biomass and BC fossil
To investigate the regional impacts on BC, three groups of air masses were identified based on their 5 transport pathways (Fig. 5a). Cluster 1 originated from northeastern India and then passed over Bangladesh before arriving Gaomeigu. The average BC mass concentration of this cluster was the highest (0.8 ± 0.4 μg m -3 ) of the three clusters. About 74% of total trajectories were associated with Cluster 1, of which 22% was identified as polluted and had an average BC loading of 1.3 ± 0.5 μg m -3 . Cluster 2 originated over Burma had an average BC loading of 0.7 ± 0.7 μg m -3 . This cluster c accounted for only 10 24% of total trajectories, but among them, about 37% was referred to pollution with BC reaching as high as 1.6 ± 0.9 μg m -3 . The air masses associated with Cluster 3 originated from the interior of China, and this group had the lowest BC mass concentrations of the three clusters, 0.4 ± 0.1 μg m -3 . This third cluster composed small fraction of total trajectories (2%), and none of them were identified as polluted, suggesting minor influences from the mainland China during the campaign. 15 The diurnal patterns of BCbiomass and BCfossil mass loadings from the three clusters were used to investigate the impacts of regional transport. As shown in Fig. 4c and e, similar diurnal variations in BCbiomass were found for Clusters 1 and 2-they had both larger values during daytime (8:00-12:00) compared with night. This pattern of higher daytime BCbiomass was associated with regional transport from northeastern India (Cluster 1) and Burma (Cluster 2). For Cluster 3, BCbiomass decreased during the day and increased 20 at night (Fig. 4g), and that pattern tracked the daily variations in PBL height. Unlike Clusters 1 and 2, the diurnal variation of BCbiomass for Cluster 3 were more likely due to influences of biomass-burning activities from areas surrounding the sampling site than regional transport. However, it should be noted that these cases were uncommon because of only 2% of air-masses were associated with Cluster3.
For BCfossil, similar diurnal patterns were found for Clusters 1 and 2 ( Fig. 4c and e), most likely due to 25 the influences of traffic emissions from surrounding areas as well as daily cycles of PBL height as discussed in section 3.2. The BCfossil loadings of Cluster 3 (Fig.4g) were relatively stable, showing only sporadic fluctuations. Unlike the declining trend of BCfossil during the daytime found for Clusters 1 and 2, the relative stable BCfossil loadings in Cluster 3 indicate that there were emissions from fossil fuel sources that offset the effect of the changes in PBL height. The transportation sector has grown rapidly in mainland China (Liu, 2019), and the regional transport of motor vehicle emissions may have been the cause for the observed diurnal variations in BCfossil for Cluster 3. 5 The PSCF model was applied to further investigate the likely spatial distribution of pollution source regions for BCbiomass and BCfossil. As shown in Fig. 5b, a low PSCF value of BCbiomass was found near Gaomeigu while high values were concentrated in the northeastern India and northern Burma, consistent with intensive fire activities in these areas (Fig. S4). This indicates that large BCbiomass loadings at Gaomeigu were more likely influenced by cross-border transport of biomass burning rather than local 10 emissions. For BCfossil (Fig. 5c), the most likely impact region was located to the southweast of Gaomeigu, near where two highways are located (e.g., Hangzhou-Ruili Expressway and Dali-Nujiang Expressway).
Owing to the low consumption of coal in the southeastern TP (Li et al., 2016), the high PSCF values of BCfossil were more likely from traffic emissions than coal combustion. Moreover, sporadic high PSCF values of BCfossil also were found in the northern Burma, indicating possible influences of fossil fuel 15 emissions here.
To further quantify the contributions of the BC transported from Southeast Asia to Gaomeigu, we studied a high BC episode (23 -27 March) using a simulation with WRF-Chem model. Two scenarios of emissions were simulated: one involved all BC emission sources, and the other turned off biomassburning emissions in Southeast Asia. The variation of modelled BC mass concentration shows an 20 acceptable degree of consistency with the measured values (r = 0.63, p < 0.01, Fig. S5), Furthermore, the index of agreement was estimated to be 0.77, indicating that the development of this BC episode was effectively captured by the WRF-Chem model. Nonetheless, the normalized mean bias between the measured and modelled BC values was estimated to be 24%, suggesting that simulation was biased towards high values. This discrepancy is best attributed to the uncertainties in the simulation associated 25 with the emission inventory and meteorological conditions. Fig. 6a shows the spatial distributions of BC loadings in Gaomeigu and surrounding areas. The mass concentrations of BC at times exceeded 15 μg m -southeastern margin of TP (0.7 μg m -3 ). After turning off the biomass burning emission source in Southeast Asia, the BC loading at southeastern TP dropped over 40% (Fig. 6b), suggesting a substantial impact of biomass-burning activities in Southeast Asia countries, which is consistent with results of the trajectory cluster analysis and PSCF.

Radiative effects and heating rate
5 Fig. 7 shows the average atmospheric direct radiative effects of PM2.5 and BC (including BCbiomass and BCfossil) at the TOA and SUF during the campaign. The average PM2.5 DRE at the TOA was +0.03 ± 1.1 W m -2 , implying that the positive effect of light-absorbing carbon on the temperature of the atmosphere outweighed the negative effect of scattering aerosols. In fact, BC produced +1.6 ± 0.8 W m -2 at the TOA on average. At the SUF, BC DRE (-3.0 ± 1.5 W m -2 ) contributed nearly half of the PM2.5 DRE (-6.3 ± 4.5 10 W m -2 ). The difference in the DRE between the TOA and SUF was +4.6 ± 2.4 W m -2 for BC aerosol, and that accounted for 73% of DRE attributed to PM2.5 (6.3 ± 4 W m -2 ), this suggests substantial radiative effect caused by BC over the southeastern margin of TP even though its mass fraction is small in PM2.5 (3.3%).
With respect to the BC sources (Fig. 7), the average BCbiomass (BCfossil) DRE was +0.8 ± 0.6 W m -2 (+0.7 15 ± 0.4 W m -2 ) at the TOA, and -1.7 ± 1.2 W m -2 (-1.4 ± 0.6 W m -2 ) at the SUF. This is equivalent to an average atmospheric DRE of +2.5 ± 1.8 W m -2 (+2.1 ± 0.9 W m -2 ). Presumably, the influences of regional transport caused the atmospheric DRE of BCbiomass to be more variable compared with that from BCfossil ( Fig. S6). For example, the atmospheric DRE of BCbiomass can be as high as +6.4 W m -2 when the air masses passed over the biomass-burning regions in Southeast Asia, while it was only 1.1 ± 0.2 W m -2 on 20 average when the air masses passed from the mainland of China.
The calculations showing a positive atmospheric DRE imply that energy was trapped in the atmosphere, which would lead to atmospheric heating over the study region. The heating rate calculated for BC varied from 0.02 to 0.3 K day -1 and an average of 0.13 ± 0.07 K day -1 . In terms of the DRE efficiency, the heating rate caused by a unit mass concentration of BC in this region was (0.19 (K day -1 ) (μg m -3 ) -1 ) , which is 25 roughly comparable with that reported for Qinghai Lake, on the northeastern TP (0.13 (K day -1 ) (μg m -3 ) - (Fig. S7). Moreover, the heating rate caused by BCbiomass may have been slightly higher (0.07 ± 0.05 K day -1 ) compared with BCfossil (0.06 ± 0.02 K day -1 ). Finally, the heating rate of BCbiomass increased to 0.16 K day -1 when the BC mass concentration was heavily influenced by the polluted air from Southeast Asia.

Conclusions
This study quantified the source contributions of BC aerosol from fossil fuel and biomass burning at a 5 site on the southeastern margin of the TP that represents a regional transport channel for air pollution during pre-monsoon. The study was conducted in pre-monsoon when the southeastern TP was heavily influenced by the air mass from southeast Asia. To reduce the uncertainties caused by interferences in absorption measurements (i.e. secondary absorption and dust) and assumptions relative to AAEs and MACBCs, the traditional 'aethalometer model' was optimized in two aspects. First, a BC-tracer method 10 coupled with a minimum R-squared approach was applied to separate secondary absorption from the total absorption, and as a result, the interferences of absorption from secondary aerosols have been eliminated.
Then, an optical source apportionment model that used primary multi-wavelength absorption and chemical species as inputs was used to derive site-dependent AAE and MACBC values-these minimize the uncertainties associated with prior assumptions on these parameters. The AAE (MACBC) calculated 15 in this way was 0.9 (12.3 m 2 g -1 ) for fossil fuel source and 1.7 (10.4 m 2 g -1 ) for biomass burning. The results of 'aethalometer model' that used these values showed that the average mass concentration of BC was 0.7 ± 0.5 μg m -3 of which 43% of BC from fossil fuel and 57% from biomass burning. Trajectory analysis showed that the BCbiomass over the site was mainly driven by regional transport from northeastern India and Burma, while BCfossil was primarily influenced by traffic emissions from areas surrounding the 20 sampling site. Moreover, the WRF-Chem model indicates that biomass burning in Southeast Asia contribute 40% of the BC loading over the southeastern margin of the TP. The SBDART model showed that a DRE of +4.6 ± 2.4 W m -2 for the total PM2.5 BC, of which +2.5 ± 1.8 W m -2 was from BCbiomass and +2.1 ± 0.9 W m -2 from BCfossil. The results of this study provide useful information concerning the sources of BC over an atmospheric transport channel to the southeastern TP, and they highlight the importance of 25 the cross-border transport of biomass burning emissions from Southeast Asia on the region during the pre-monsoon.
Data availability. All data described in this study are available upon request from the corresponding authors.
Supplement. The supplement related to this article is available online. 5 Author contributions. QW and JC designed the study. WR conducted the field measurements. LX provided the results of WRF-Chem model. YZ and TZ performed the chemical analysis of filters. HL and QW wrote the article. All the authors reviewed and commented on the paper.