Contributions of biomass-burning, urban, and biogenic emissions to the concentrations and light-absorbing properties of particulate matter in central Amazonia during the dry season

1 Urbanization and deforestation have important impacts on atmospheric particulate matter 2 (PM) over Amazonia. This study presents observations and analysis of submicron PM1 3 concentration, composition, and optical properties in central Amazonia during the dry season. 4 The focus is on delineating the anthropogenic impact on the observed quantities, especially as 5 related to the organic PM1. The primary study site was located 70 km to the west of Manaus, a 6 city of over two million people in Brazil. As part of the GoAmazon2014/5 experiment, datasets 7 from a large suite of instrumentation were employed. A high-resolution time-of-flight aerosol 8 mass spectrometer (AMS) provided data on PM1 composition, and aethalometer measurements 9 were used to derive the absorption coefficient babs,BrC of brown carbon (BrC) at 370 nm. The 10 relationships of babs,BrC with AMS-measured quantities showed that the absorption was 11 associated with less-oxidized, nitrogen-containing organic compounds. Atmospheric processing 12 appeared to bleach the BrC components. The organic PM1 was separated into different classes by 13 positive-matrix factorization (PMF). Estimates of the effective mass absorption efficiency 14 associated with each PMF factor were obtained. Biomass burning and urban emissions appeared 15 to contribute at least 80% of babs,BrC while accounting for 30 to 40 % of the organic PM1 mass 16 concentration. In addition, a comparison of organic PM1 composition between wet and dry 17 seasons revealed that only a fraction of the nine-fold increase in mass concentration between the 18 seasons was due to biomass burning. An eight-fold increase in biogenic secondary organic PM1 19 was observed. A combination of decreased wet deposition and increased emissions and oxidant 20 concentrations, as well as a positive feedback on larger mass concentrations are thought to play a 21 role in the observed increases. Fuzzy c-means clustering identified three clusters to represent 22 different pollution influences during the dry season, including “baseline” (dry season 23 background, which includes biomass burning), “event” (increased influence of biomass burning 24 and long-range transport of African volcanic emissions), and “urban” (Manaus influence on top 25 2 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-1309 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 9 January 2019 c © Author(s) 2019. CC BY 4.0 License.

of the background). The baseline cluster was associated with a mean mass concentration of 9 ± 3 26 μg m -3 . This concentration increased on average by 3 μg m -3 for both the urban and the event 27 clusters. The event cluster was characterized by remarkably high sulfate concentrations. 28 Differences in the organic PM1 composition for the urban cluster compared to the other two 29 clusters suggested a shift in oxidation pathways as well as an accelerated oxidation cycle due to 30 urban emissions, in agreement with findings for the wet season. 31

Introduction 32
The Amazon basin has undergone significant urbanization and deforestation in the past 33 decades (Davidson et al., 2012;Martin et al., 2017;van Marle et al., 2017). An understanding of 34 how the composition of atmospheric particulate matter (PM) changes due to anthropogenic 35 activities and how these changes affect PM optical properties is essential for quantifying the 36 global anthropogenic radiative forcing (IPCC, 2013;Sena et al., 2013). Light absorption 37 coefficients, babs, and their spectral dependence, commonly referred to as the Ångström 38 absorption exponent, åabs, are needed for accurate interpretation of satellite-retrieved aerosol 39 optical depth (AOD) for climate modeling. Estimates of the mass absorption efficiency Eabs for 40 PM subcomponents are useful for models to estimate optical effects based on PM composition 41 and mass concentrations . 42 Organic material that can efficiently absorb radiation in the near-ultraviolet through the 43 blue end of the visible spectrum, with decreasing absorption efficiency as wavelength increases, 44 is termed "brown carbon" (BrC) (Pöschl, 2003;Andreae and Gelencsér, 2006;Laskin et al., 45 2015). By comparison, black carbon (BC) absorbs light efficiently throughout the visible 46 spectrum. Although global climate models have typically treated organic PM as purely 47 scattering, several studies have shown that brown carbon can contribute substantially to light 48 absorption by PM, especially in regions affected by biomass burning and urban emissions 49 Ramanathan et al., 2007;Bond et al., 2011;Bahadur et al., 2012;50 Ma and Thompson, 2012;Feng et al., 2013). In addition to primary emissions of BrC, secondary 51 production of BrC can occur from the oxidation of volatile organic compounds (VOCs) present 52 in biomass smoke (Saleh et al., 2014) and from atmospheric multiphase reactions involving a 53 wide range of precursor VOCs (Nozière et al., 2007;De Haan et al., 2009;Nguyen et al., 2012;54 Additional information on the backtrajectory calculations and on the radar were described in de 146 Sá et al. (2018). 147

Comparison of PM concentration and composition across clusters for the T3 site 249
A second approach to investigate the changes in concentrations and compositions of the 250 PM with pollution influences employed a combination of positive-matrix factorization (PMF) 251 and Fuzzy c-means (FCM) clustering. The PMF analysis was applied to the organic mass spectra 252 to separate the organic PM1 into representative component classes (section 3.1.2.1). The FCM 253 clustering algorithm was applied to auxiliary measurements to identify times of urban and 254 biomass burning influences at the T3 site (section 3.1.2.2). The results of the FCM analysis were 255 crossed with the findings of the PMF analysis for further insights into pollution-related 256 variability of PM concentration and composition (section 3.1.2.3). 257

Classification of organic PM by positive-matrix factorization 258
The organic mass spectra recorded by the AMS at the T3 site were analyzed by PMF 259 . Details and diagnostics of the PMF analysis are presented in the 260 13 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-1309 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 9 January 2019 c Author(s) 2019. CC BY 4.0 License.
profile" and "factor loading" are their counterpart mathematical products obtained from the 263 PMF analysis. A six-factor solution was obtained, and the factor profiles, diel trends of the factor 264 loadings, and the time series of the factor loadings and other related measurements are plotted in 265 Figure 5. The correlations of factor loadings with co-located measurements of gas-and particle-266 phase species are presented in Figure 6. 267 The factors were interpreted considering the mass spectral characteristics of the factor 268 profiles and the correlations between factor loading and mass concentrations of co-located 269 measurements. Three resolved factors interpreted as secondary production and processing 270 closely matched the counterpart profiles of the wet season (R ≥ 0.99; Table 1) (de Sá et al., 271 2018). These three factors consisted of a more-oxidized oxygenated factor ("MO-OOA"), a less-272 oxidized oxygenated factor ("LO-OOA"), and an isoprene epoxydiols-derived factor ("IEPOX-273 SOA"). Temporal correlations with external tracers and oxidation characteristics were also 274 similar to those of the wet season, corresponding to IOP1 ( Figure 6; Table 1; de Sá et al., 2018). 275 Although a hydrocarbon-like factor ("HOA") was analogous to its counterpart in IOP1 (R = 276 0.94), it also had characteristics of an IOP1 anthropogenic-dominated factor ("ADOA") tied to 277 other urban sources including cooking. The HOA factor of IOP2 therefore represented a mix of 278 the HOA and ADOA factors of IOP1, which could not be separated by PMF in IOP2 due to their 279 lower relative contributions. The interpretation of the HOA, IEPOX-SOA,  OOA factors follows that of IOP1, as presented in de Sá et al. (2018). The following discussion 281 focuses on the two biomass burning factors of IOP2. 282 A less-oxidized factor ("LO-BBOA") and a more-oxidized factor ("MO-BBOA") were 283 resolved for IOP2. For IOP1, a single "BBOA" factor was resolved, and it accounted for 9% of 284 the organic PM1 mass concentration. For IOP2, there were enough differences in mass spectral 285 features and temporal contributions, as well as larger overall contributions of biomass burning, 286 that the PMF analysis identified two different factors. The MO-BBOA and LO-BBOA factors 287 respectively accounted for 18% and 12% of the mean organic PM1 mass concentration. 288 Therefore, the relative contribution of biomass burning to organic PM1 during the dry season was 289 at least a factor of three higher compared to the wet season (a more detailed discussion is 290 presented at the end of this section). 291 The LO-BBOA and MO-BBOA factor profiles had a distinct peak at nominal m/z 60 292 (C2H4O2 + ) ( Figure 5a). The fractional intensity f60 at m/z 60 was larger for LO-BBOA (0.051) 293 than for MO-BBOA (0.013). A peak at m/z 73 (C3H5O2 + ) was also present in both profiles, 294 although its intensity was three to four times smaller than that at m/z 60. The peaks at m/z 60 and 295 m/z 73 are attributed to fragments of levoglucosan and other anhydrous sugars that are produced 296 by the pyrolysis of biomass (Schneider et al., 2006;Cubison et al., 2011). Accordingly, the 297 loadings of both factors correlated with the concentrations of several biomass-burning tracers in 298 the particle phase, including levoglucosan, vanillin, 4-nitrocatechol, syringol, mannosan, 299 syringaldehyde, sinapaldehyde, and long-chain alkanoic acids (C20, C22, C24) and of tracers in the 300 gas phase (acetonitrile) ( Figure 6). The loadings also correlated with less-specific tracers, 301 including CO concentration and particle number concentration. The Pearson-R correlations were 302 typically higher for the LO-BBOA factor than for the MO-BBOA factor. 303 The LO-BBOA profile had the greatest ratio of signal intensity of the C2H3O + ion (m/z 304 43) to that of the CO2 + ion (m/z 44) compared to all other factors (Figure 5a). In comparison, the 305 15 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-1309 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 9 January 2019 c Author(s) 2019. CC BY 4.0 License.
The MO-BBOA and LO-BBOA factors had O:C ratios of 0.70 ± 0.07 and 0.53 ± 0.04, 307 respectively. In addition, the LO-BBOA factor loading had higher correlation with the estimated 308 inorganic nitrate concentrations than with the total nitrate concentrations whereas the MO-309 BBOA factor did not ( Figure 6; Supplementary Material, Section S1 describes the nitrate 310 estimates). Taken together, these results point to a less-oxidized, higher-volatility character of 311 the LO-BBOA factor and a more-oxidized, lower-volatility character of the MO-BBOA factor, 312 both with biomass-burning characteristics Cubison et al., 2011;Gilardoni 313 et al., 2016;Zhou et al., 2017). 314 The extent of the biomass burning influence and atmospheric oxidation on the 315 composition of organic PM1 can be visualized in a scatter plot of f44 and f60 (Figure 7a) (Cubison 316 et al., 2011). A background f60 value of 0.3% ± 0.06% (vertical black dashed line) indicates a 317 threshold for negligible or completely oxidized biomass-burning PM1. Points in the lower right 318 of the f44-f60 representation usually characterize PM1 tied to recent biomass burning emissions. 319 For IOP1 (blue markers), all points lie on or close to the background value suggested by Cubison 320 et al. (2011), indicating the absence of a strong influence from biomass burning. During the wet 321 season, biomass burning was limited to local sources or to sources far enough away such as 322 Africa that the PM1 was extensively oxidized by arrival in central Amazonia . 323 For IOP2 (red markers), the f60 values are greater for most observations, showing that for most 324 times T3 was influenced to some extent by biomass burning (see Section 3.1.2.3). This finding is 325 in line with the widespread occurrence of fires during the dry season ( Figure 3 The LO-BBOA factor of high f60/f44 and low O:C thus appears associated with primary 331 PM1 emitted by biomass burning. The MO-BBOA factor, characterized by low f60/f44 and high 332 O:C, may represent a combination of primary PM1 of higher oxygen content as well as secondary 333 PM1 tied to biomass burning in its early stages of atmospheric processing (Cubison et al., 2011;334 Gilardoni et al., 2016). These secondary pathways could include (i) the heterogeneous oxidation 335 of primary PM1, such as that represented by the LO-BBOA factor, and (ii) the oxidation of gas-336 phase biomass-burning emissions or of species evaporated from primary PM1, followed by the 337 condensation of the gas-phase products onto the PM1. 338 The LO-BBOA and MO-BBOA factor loadings had greater magnitude and variability at 339 night compared to during day ( Figure 5b). Their summed loading, represented as "BBOAT", 340 accounted for 40% and 13% of the organic PM1 during night and day, respectively. Overall, they 341 accounted for 30% of the organic PM1. This result reflects the importance of fire activity during 342 all times of day and during the entirety of IOP2 ( Figure 3). The surface concentrations were 343 lower during the day because biomass burning emissions are diluted with the development of the 344 planetary boundary layer (PBL) and with the increased wind speeds as compared to the stagnant 345 air and shallower PBL at night. The occurrence of significant dilution indicates that the emission 346 sources were at least in part within a day of transport, meaning a distance on the order of a few 347 hundred kilometers. The fractional contribution of the MO-BBOA factor to BBOAT shifted from 348 0.7 to 0.5 from day to nigh, while that of LO-BBOA correspondingly shifted from 0.3 to 0.5 349 ( Figure 7b). This result is consistent with an additional secondary contribution to the MO-BBOA 350 loading during daytime, including from LO-BBOA oxidation and possibly tied to photochemical 351 processing, on top of a primary source from biomass burning. 352 Although the footprint of biomass burning is geographically more widespread throughout 353 the basin compared to the urban footprint of nearby Manaus, fire incidence and large-scale 354 emissions have historically concentrated in a region known as the arc of deforestation along the 355 southern rim of the forest (Fuzzi et al., 2007;Artaxo et al., 2013). Several campaigns have 356 focused on the effects of biomass burning during the dry season at locations that are highly 357 affected by fires, usually in the states of Rondônia or Mato Grosso, within the arc of 358 deforestation (SCAR-B, Kaufman et al., 1998;LBA-SMOCC, Fuzzi et al., 2007;LBA-359 EUSTACH, Andreae et al., 2002;TROFEE, Yokelson et al., 2007; SAMBBA, Morgan et al., 360 2013). At a ground site in Porto Velho, Rondônia, a PMF analysis of ACSM data showed that 361 70% of the organic PM1 could be attributed to biomass burning . Compared to 362 the present study, in which at least 30% of the organic PM1 can be directly attributed to biomass 363 burning, the contributions of fires to PM1 in the arc of deforestation region are considerably 364 larger. 365 The combined contribution of 30% by MO-BBOA and LO-BBOA at T3 represents a 366 lower bound of biomass burning influence because more-oxidized material from biomass 367 burning could be accounted for by the MO-OOA factor. In the limiting assumption that all MO-368 OOA loadings originated from BBOA loadings, an upper limit of 50% can be established for the 369 mean contribution of biomass burning to organic PM1 concentrations at T3. Considering that all 370 organic PM1 components have been observed to age into MO-OOA at similar rates (Jimenez et 371 al., 2009) although PM1 concentrations increase on average by a factor of 8.5 between seasons, not all of 375 the increase is due to biomass burning, which has been a common assumption in previous studies 376 (Artaxo et al., 1994;Holben et al., 1996;Echalar et al., 1998;Maenhaut et al., 1999;Andreae et 377 al., 2002;Artaxo et al., 2002;Mace et al., 2003;Martin et al., 2010b;Artaxo et al., 2013;Rizzo 378 et al., 2013;Brito et al., 2014;Pöhlker et al., 2016). In absolute mass concentrations, the 379 contribution from biomass burning increased from 0.12 µg m -3 in the wet season to 3.4 µg m -3 in 380 the dry season, which represents a 30-fold increase. This result corresponds to a change in 381 percentage contribution to organic PM1 from 9% to 30% (not counting with the mass presumably 382 present in MO-OOA). Nevertheless, the contribution from secondary biogenic sources (and their 383 anthropogenically affected processes), as represented by the LO-OOA and IEPOX-SOA factors, 384 also increased by around 8-fold from 0.6 µg m -3 to 4.8 µg m -3 . In absolute terms, this mass 385 increase (of 4.2 µg m -3 ) is comparable to the one associated with biomass burning (3.3 µg m -3 ). 386 Because the 8-fold mass increase of LO-OOA and IEPOX-SOA was similar to the 8.5-fold 387 increase in total organic PM1, these factors show a similar mass percentage contribution of 42% 388 to organic PM1 for both seasons. The MO-OOA factor loadings increased by 6-fold from 0.4 µg 389 m -3 to 2.3 µg m -3 . Because this relative increase was smaller than that of the total organic PM1, 390 the MO-OOA factor had a decrease from 30% to 20% of contribution to organic PM1. The 391 contribution from urban sources, as represented by the HOA and ADOA factors, increased by 392 three-fold between seasons, from 0.24 µg m -3 to 0.76 µg m -3 , representing a decrease in mass 393 percentage contribution from 18% to 7%. 394 Therefore, reasons other than increased biomass burning in the dry season must have 395 played a role in increasing organic PM1 concentrations. One aspect is that BVOC emissions are 396 19 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-1309 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 9 January 2019 c Author(s) 2019. CC BY 4.0 License. typically higher in the dry season Alves et al., 2016), which might 397 partly explain the increases in LO-OOA, IEPOX-SOA, and MO-OOA factors. In addition, the 398 directly-measured biogenic (total) secondary organic PM1 formation potential of ambient air 399 increased by a factor of 2.4 (1.7) between seasons . Increased organic mass 400 available for partitioning may account for another factor of 2 . As a 401 consequence of increased PM1 mass concentrations, the lifetime of semi-volatile gases may also 402 be increased, since lifetime against dry deposition is much larger for particles than for gases 403 (Knote et al., 2015). Increased oxidant levels during the dry season could also be a contributing 404 factor (Rummel et al., 2007;Artaxo et al., 2013;Andreae et al., 2015;Yáñez-Serrano et al., 405 2015;Fuentes et al., 2016). Importantly, the mass concentrations of sulfate and ammonium also 406 increased by six-fold between seasons (Figure S10), indicating that atmospheric physical 407 processes governing particle mass concentrations possibly played an important role. In this 408 context, reduced wet deposition due to reduced convection in the dry season may be another 409 appreciable contributor to the organic PM1 increases (Machado et al., 2004;Nunes et al., 2016410 Chakraborty et al., 2018. 411

Cluster Analysis 412
The time series of the afternoon concentrations of particle number, NOy, ozone, rBC, 413 carbon monoxide, and sulfate were analyzed by Fuzzy c-means clustering at the time resolution 414 of the AMS measurements. The algorithm attributed degrees of cluster membership to each data 415 point based on similarity in the sets of input concentrations (Section S2). The scope was 416 restricted to data sets for which ten-hour air mass back trajectories did not intersect precipitation. 417 The scope also excluded data sets tied to the lowest 10% of solar irradiance averaged over the 418  (Table S1). In particular, the sulfate concentrations (2.3 µg m -3 at the centroid) were the highest 448 among the three clusters. Results for September 23, September 27, and September 28 illustrate 449 these findings for T3, with sulfate concentrations reaching 4 µg m -3 ( Figure 8). This trend in 450 sulfate concentrations was consistent across all three sites ( Figure 2). The backtrajectories 451 associated with the event cluster were variable, passing to the north, directly over, and to the 452 south of Manaus, although always with an east component ( Figure 9b). The long-range transport 453 and increased regional fire count during the event period thus appeared more important in 454 defining this cluster than did the directions of the backtrajectories in a smaller scale, making 455 Manaus emissions of secondary importance. 456 The urban cluster had the highest centroid concentrations of NOy (2.6 ppb), ozone (56.4 457 ppb), and particle number (4600 cm -3 ) among the three clusters (Table S1)

Comparison of PM1 composition among clusters 465
Species mass concentrations and PMF factor loadings associated with the cluster 466 centroids were determined (Section S2). The resulting organic, sulfate, ammonium, nitrate, and 467 chloride mass concentrations associated with each cluster are represented in Figure 10a. The 468 PMF factor loadings associated with each cluster are likewise represented in Figure 10b. 469 The summed NR-PM1 mass concentrations for the centroids of the event and urban 470 clusters were both 12.3 µg m -3 . This concentration was 33% higher than that representing the 471 baseline cluster (9.2 µg m -3 ). This result thus agrees with that based on direct comparison of PM1 472 mass concentrations between the T3 and the T0a sites (Section 3.1.1). Therefore, the overall 473 effect of Manaus pollution was to add 1 to 3 µg m -3 on top of the upwind concentrations. The relationship between clusters and PMF factors is represented in Figure 10b. All three 487 clusters were associated with an organic PM1 composition dominated by secondary production. 488 The baseline cluster was largely dominated by the LO-OOA factor (40%). By comparison, the 489 event cluster had significant increases in the LO-BBOA, MO-BBOA, and IEPOX-SOA factor 490 loadings. The increase in LO-BBOA and MO-BBOA loadings (40%) can be associated with the 491 increased contributions of primary and secondary particle components from biomass burning, 492 respectively. The LO-BBOA factor had the highest loading (0.5 µg m -3 ) for the event cluster, 493 consistent with the high incidence of fires during the period represented by this cluster. The 494 increase of 65% in IEPOX-SOA loading can be explained by the disproportionally higher 495 increase of 65% in the sulfate concentration (which favors higher IEPOX-SOA loadings), 496 accompanied by the relatively moderate increase of 34% in NOy concentration, (which 497 suppresses IEPOX-SOA loadings), leading to a net increase in IEPOX-SOA loadings (Table S1;  498 de . 499 The composition of the organic PM1 associated with the urban cluster differed from that 500 of the two other clusters, as indicated by the factor contributions ( Figure  The similarity in IEPOX-SOA factor loading for the baseline and the urban clusters may 509 be explained by the following aspects. First, the lifetime of IEPOX-derived PM in the boundary 510 layer is thought to be around 2 weeks (Hu et al., 2016). Therefore, a substantial fraction of this 511 component observed at T3 will be formed upwind of the Manaus plume. Second, favored 512 conditions for IEPOX production and uptake are low NO concentrations (i.e., HO2-dominant 513 pathway for the ISOPOO radical) and high sulfate concentrations . Sulfate 514 concentrations increased by 31%, and NOy concentrations, used as an indicator for exposure of 515 the airmass to NO concentrations, increased by 100% for the urban compared to the baseline 516 cluster. These two changes work against one another with respect to IEPOX production and 517 uptake. For the wet season, de  reported that the IEPOX-SOA factor loading was 518 more sensitive to changes in NOy concentration for 1 ppb and less. By comparison, NOy 519 concentrations in the dry season were consistently greater than this value. Due to this lower 520 sensitivity, large increases in NOy may not be tied to large decreases in IEPOX-SOA factor 521 loading in the dry season. In sum, the opposite roles of sulfate and NOy concentrations can 522 explain the net zero change in IEPOX-SOA factor loadings between baseline and urban clusters. 523 Because all of the loadings for other factors increased, the fractional loading of IEPOX-SOA 524 decreased from 26% to 15%. 525

Brown carbon light absorption 527
The diel trends of babs, babs,BrC, babs,BrC/babs, and åabs are shown in Figure 11. Both babs and 528 babs,BrC were larger and had greater variability at night compared to day. The variability of the 529 fractional contribution of BrC to the total absorption, represented by babs,BrC/babs, was smaller 530 than the variability of its components babs and babs,BrC (i.e., Figure 11c  properties can be modified through atmospheric processing, which may involve reactions at the 551 gas-particle interface, reactions in the aqueous phase of particle and cloud droplets, and 552 photolysis driven by sunlight Zhao et al., 2015;Sumlin et al., 2017;Lee et 553 al., 2014;Romonosky et al., 2015). In addition, Saleh et al. (2014)  An important contribution of nitrogen-containing organic molecules to babs,BrC is 562 suggested by the relationship in Figure 12b. The percent contribution of the CxHyOzNp + family to 563 each PMF factor profile is listed in Table 2 and is highest for the HOA and LO-BBOA factors. 564 The correlations of factor loadings with the CxHyOzNp + mass concentrations as well as with the 565 babs,BrC values are highest for these two factors (R > 0.8 and R > 0.6, respectively) ( Table 2). The 566 correlations of the MO-BBOA factor loading with these two parameters are lower but still 567 significant. By comparison, the corresponding correlations for the IEPOX-SOA, LO-OOA, and 568 MO-OOA factor loadings are all lower than 0.5. These results further support that the HOA and 569 LO-BBOA factors to a larger extent and the MO-BBOA factor to a lesser extent were tightly 570 associated with nitrogen-containing, light-absorbing organic molecules. 571 In contrast to the CxHyOzNp + family, the correlations between PMF factor loadings and 572 mass concentrations of organic nitrates are low (R < 0.4, Table 2; Figure S12 Regarding the further atmospheric processing of these nitrogen-containing organic compounds, 593 laboratory studies have shown that hydroxy radical oxidation of nitro-aromatic species in 594 aqueous solutions leads to fragmentation into smaller organic acids (e.g., oxalic, glycolic, 595 malonic, and isocyanic) or, in general, reduce the size of the conjugated molecular systems, 596 leading to a decrease in light absorption at visible wavelengths (Sumlin et al., 2017;Hems and 597 Abbatt, 2018). These findings may help to explain the bleaching of BrC as the material becomes 598 more oxidized. In the context of the PMF factors, these smaller later-generation products may 599

Contribution of organic PM components to BrC absorption 615
Herein, advantage is taken of the representation of the organic PM in its subcomponents 616 provided by the PMF factors to estimate a mass absorption efficiency for each of them. The 617 absorption coefficient is the sum of the absorption coefficient of the n parts of the organic PM 618  Table 3. 640 A scatter plot of the predicted babs,BrC,pred against the observed babs,BrC is shown in Figure  641  alternatively have co-variability also captured in the PMF factor loadings. 646 The highest values of Eabs at 370 nm were associated with the HOA and LO-BBOA 647 factors (2.04 ± 0.14 and 1.50 ± 0.07 m 2 g -1 , respectively). These results support the interpretation 648 presented in the previous section about the association of the HOA and LO-BBOA factors with 649 light absorption. As a point of comparison, Eabs of 2 to 3 m 2 g -1 at 300 nm was reported for 650 HULIS extracts from PM2.5 filter samples collected under biomass burning conditions during the 651 Amazon dry season in Rondônia, Brazil . HULIS have been recognized as 652 important components of BrC from biomass burning (Mukai and Ambe, 1986;Andreae and 653 Gelencsér, 2006;Graber and Rudich, 2006). The Eabs value of the MO-BBOA factor was 0.82 ± 654 0.04 m 2 g -1 . The result of Eabs,MO-BBOA < Eabs,LO-BBOA is consistent with an interpretation of 655 photochemically driven oxidation and bleaching during the atmospheric transport of biomass 656 burning emissions. 657 The Eabs value of the IEPOX-SOA factor was 0.40 ± 0.05 m 2 g -1 , and the Eabs values of 658 the MO-OOA and LO-OOA factors (0.01 ± 0.02 m 2 g -1 ) were not statistically different from 659 zero. Laboratory studies suggest that biogenic PM does not appreciably absorb light in the near-660 UV and visible range although this result may change with atmospheric exposure to ammonia 661 and amines, changes in particle acidity, and other factors (Nakayama et al., 2012;Liu et al., 662 2013;Flores et al., 2014;Lin et al., 2014;Laskin et al., 2015). Biogenic PM is typically 663 characterized by carbonyls, carboxyls, and hydroxyls without substantial conjugation; this 664 composition does not have the low-energy electronic transitions relevant for brown-carbon light 665 absorption . By contrast, PM produced by the photo-oxidation of aromatic 666 VOCs, such as toluene, m-xylene, naphthalene, and trimethylbenzene, tends to absorb 667 31 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-1309 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 9 January 2019 c Author(s) 2019. CC BY 4.0 License. significantly, and the light absorption is greater for PM produced under conditions of higher NOx 668 concentrations because of the production of nitro-aromatic compounds (Zhong and Jang, 2011;669 Liu et al., 2012;Lee et al., 2014;Liu et al., 2015). This absorption, however, may decrease with 670 atmospheric processing as previously discussed for the case of biomass burning emissions, 671 which is also reflected in the negligible value of Eabs for MO-OOA. In central Amazonia, the 672 organic PM is dominated by biogenic forest precursors even in the pollution plume of Manaus, 673 which helps to explain the negligible Eabs value for LO-OOA. It may also be that some aromatic 674 PM is associated with the HOA factor, which has a high Eabs value. 675 A comparison of the relative contributions of PMF factor loadings to organic PM1 mass 676 concentration and to light absorption is presented in Figure 15 Table 4. Biomass burning and urban emissions, as represented by the 681 BBOA and HOA factors, appeared to contribute 80% of babs,BrC while accounting for at least 682 30% of the organic PM1 mass concentration. The IEPOX-SOA factor was associated with the 683 balance of babs,BrC while representing 16% of the organic PM1 mass concentration. Studies with 684 further information on black carbon size distribution, particle mixing state, and the effect of RH 685 on particle absorption are warranted to refine the estimates of Eabs for the components of organic 686 PM1 and therefore their contributions to BrC light absorption. A similar attribution analysis as 687 the right panel of Figure 15 was carried out for the baseline, event, and urban clusters separately 688 and is discussed in the Supplementary Material ( Figure S15). 31%, and HOA for the remaining 7%. An important conclusion is that the 8.5-fold increase in 696 organic PM1 concentrations between the wet and dry seasons is not all due to biomass burning, 697 but also to a concurrent increase of biogenic secondary organic PM1 of eight-fold and smaller 698 increases in urban PM1. Reasons that possibly played a role in such increases for the dry season 699 are: increased BVOC emissions, increased formation potential of biogenic secondary organic 700 PM1, reduced wet and dry deposition and PBL ventilation of PM1 particles, and increased 701 partitioning due to larger organic PM1 mass concentrations in the dry season. 702 The FCM clustering analysis identified the baseline, event, and urban clusters. Relative to 703 the baseline cluster (9.2 μg m -3 ), both the event and the urban cluster had an increase of 3 μg m -3 . 704 For the event cluster, the increased sulfate concentrations together with only moderate increases 705 in NOy, resulted in remarkable increases of almost 1 μg m -3 (65%) in IEPOX-SOA factor 706 loadings relative to the baseline cluster. Regarding the urban cluster, increases in the factor 707 loadings of MO-BBOA (40 to 90%) and LO-OOA (20 to 25%) were observed in comparison to 708 the other two clusters. At the same time, the IEPOX-SOA contribution was either the same or 709 lower (by 40%) in absolute loadings, and always lower in relative contribution to organic PM 710 (15% of organic PM compared to 20-30% for the other clusters). These changes in the make-up 711 of organic PM were consistent with the changes observed for the wet season (de Sá et  contrast to models that assume organic PM as a purely scattering component (Ramanathan and 732 Carmichael, 2008;Myhre et al., 2013). Recent models have estimated the global BrC 733 contribution to DRF to be in the range of 0.1 to 0.25 W m -2 , corresponding to 10 to 25% of the 734 DRF by BC (Feng et al., 2013). In addition, BrC in cloud water can absorb light and thereby 735 34 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-1309 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 9 January 2019 c Author(s) 2019. CC BY 4.0 License. facilitate water evaporation and cloud dispersion (Hansen et al., 1997). This effect may 736 compensate the cooling that aerosol particles offer by serving as seeds for cloud droplet 737 formation and may also provide a positive feedback as increased fire activity may provoke more 738 fire-prone conditions by suppressing precipitation (Nepstad et al., 1999;Bevan et al., 2009;739 Gonçalves et al., 2015;Laskin et al., 2015). Another implication is that light absorption by BrC 740 in the ultraviolet may significantly decrease photolysis rates, thereby affecting the concentrations 741 of precursors and oxidants such as ozone and OH radicals in the atmosphere (Li et al., 2011;742 Jiang et al., 2012;Laskin et al., 2015).

Data availability
The data sets used in this publication are available at the ARM Climate Research Facility database for the GoAmazon2014/5 experiment (https://www.arm.gov/research/ campaigns/amf2014goamazon, last access: 1 August 2018).

Author contributions
SSdS, LVR, and STM defined the scientific questions and scope of this study. STM, JLJ, MLA, AHG, and PA designed, planned, and supervised the broader GoAmazon2014/5 field experiment. SSdS, BBP, PCJ, and DAD carried out the AMS measurements and data processing. AS collected and quality-checked the aethalometer data. LVR performed the BrC calculations based on the aethalometer data. LDY, RW, GYV, JB, SC, YJL, SS, and HMJB performed auxiliary data collection/processing and simulations. SSdS carried out the scientific analysis involving PMF and FCM. SSdS prepared the paper with contributions from all co-authors.  Tables   Table 1. Characteristics of the PMF factor profiles. Listed are f44 and f60, corresponding to the organic signal fraction at m/z 44 and m/z 60, respectively, as well as the oxygen-tocarbon (O:C) and hydrogen-to-carbon (H:C) ratios. Values and uncertainties were calculated by running the PMF analysis in "bootstrap mode" ).

List of
The Pearson-R correlations between the factor profiles of IOP2 and their counterparts in IOP1 are also listed (i.e., dry season compared to wet season). "N/A" means "not applicable". Elemental ratios were calibrated by the "improved-ambient" method, which has an estimated uncertainty of 12% for O:C and 4% for H:C (Canagaratna et al., 2015).    Figure S6, and the most important ion fits are shown in Figure S7. The AMS method characterizes organic nitrates through the NO + and NO2 + fragments, which remain distinct from the larger fragments of the CxHyOzNp + family (Section S1 and discussion therein).  Figure 3. T0a is the Amazonian Tall Tower Observatory (Andreae et al., 2015). T2 is a site 8 km downwind of Manaus, just across the Black River ("Rio Negro") . Measurements at T0a and T2 were made by an ACSM. Concentrations in both panels were adjusted to standard temperature (273.15 K) and pressure (10 5 Pa) (STP).     refractory black carbon (rBC). SV-TAG measurements refer to particle-phase concentrations, except for sesquiterpenes which refer to total concentrations and mostly occurred in the gas phase. The C8 and C9 aromatics include the xylene and trimethylbenzene isomers, respectively. The C20, C22, and C24 acids include eicosanoic, docosanoic, and tetracosanoic acids, respectively.   Trajectories were calculated using HYSPLIT4 in steps of 12 min for 10 h (Draxler and Hess, 1998). Twenty trajectories are plotted for each cluster, corresponding to the times of highest degree of membership to that cluster.   In complement, Figure S14 shows the relationships between the brown-carbon absorption coefficient and the fractional contributions of the CxHyOz + and CxHyOzNp + families to organic PM1. The åabs value corresponds to 370 to 430 nm. In panel a, the slope and intercept are 3.2 ± 0.1 and 6.8 ± 0.1, respectively. In panel b, they are 5.2 ± 0.1 and 1.1 ± 0.1.