The GoAmazon 2014/5 campaign was conducted in the vicinity of Manaus, Brazil, from 1 January 2014 through 31 December 2015 (Martin et al., 2016). During this period, DOE's G-1 research aircraft, based out of the Manaus International Airport, was deployed for two periods, during which time it sampled the Manaus urban plume as it was transported downwind over the Amazon rain forest. The first aircraft deployment period occurred during the wet season from 15 February to 26 March 2014, while the second occurred in the dry season from 1 September to 10 October 2014. Sixteen research flights were conducted in the wet season, and 19 were conducted in the dry season. Table 1 lists takeoff and landing times, the altitude of level flight legs, and meteorological parameters measured by the G-1 instrumentation shortly after takeoff on each flight. In general, flight plans were focused on successive perpendicular plume crossings spaced approximately equally at intervals of 24 km downwind of the city. A representative flight path for 13 March 2014, an exemplary day in the wet season with few clouds, is shown in Fig. 1. The paths for all flights during the two deployments are shown, segregated by altitude, in Martin et al. (2016) and therefore are not reproduced here. On 13 March, the first leg approximately bisected the city along a NW–SE line, the second leg passed near a highly instrumented ground site directly across the Amazon River from Manaus (T2), and the fourth leg passed over a second highly instrumented site located NE of Manacapuru (T3), with an additional leg between the T2 and T3 sites, and a final leg downwind of the T3 site. The first pass through the pattern was generally at an altitude of 500–700 m above ground level, with a second pass at a higher altitude and often multiple altitudes and often, though not always, focused on sampling clouds. The flight pattern discussed above and shown in Fig. 1 was rotated to align with the prevailing wind direction. Flights generally departed in the late morning or early afternoon and lasted between 3 and 3.5 h (Table 1).

An Aerodyne High-Resolution Time-of-Flight Aerosol Mass Spectrometer (abbreviated as AMS hereafter) was deployed on the G-1 to measure aerosol chemical composition (Jayne et al., 2000; DeCarlo et al., 2006). The AMS operated only in the standard “V” mass spectrometer (MS) mode (the particle sizing mode was not used) with a 13 s data averaging interval and equal chopper open and closed periods of 3 s. Before, during, and after flights, the AMS-sampled air was periodically diverted through a HEPA filter to remove particulates, and these filter periods were used to account for gas-phase interferences with isobaric particulate signals. Based on the standard deviation of these blank measurements (3σ) as described in the literature (DeCarlo et al., 2006), the detection limits of the AMS at the 13s sampling interval were 0.12, 0.01, 0.014, and 0.005 µg m-3 for organics, sulfate, nitrate, and ammonium, respectively. All aerosol instruments, including the AMS, sampled from a common, double-diffuser isokinetic inlet. Flow for the AMS was subsampled from the center of the isokinetic inlet and dried to RH <40 % by passage through a 1/2 in. ID Perma Pure Nafion membrane. A constant-pressure inlet operating at approximately 620 hPa was used to maintain a constant volumetric flow to the AMS up to altitudes of approximately 3900 m (Bahreini et al., 2008). Because the AMS was powered off between flights, a correction based on the real-time N2+ signal was applied to all data to account for drifting sensitivity as the instrument warmed up during flights. The AMS was regularly calibrated in the field using monodisperse ammonium nitrate particles quantified with a TSI condensation particle counter (CPC). Data were analyzed in Igor Pro (v6.37) using the high-resolution analysis package (Squirrel v1.55H, PIKA v1.44H) and techniques described in the literature (Canagaratna et al., 2015; Kroll et al., 2011; Aiken et al., 2007; Allan et al., 2003; Jimenez et al., 2003). All AMS data discussed in this paper have been processed using the high-resolution data analysis routine. The O : C and H : C values reported here use the updated calibrations described in Canagaratna et al. (2015). All AMS data are normalized to the laboratory calibration conditions of 23 ∘C and 1013 hPa. Positive matrix factorization (PMF) analysis was performed on the wet-season data set by combining all flight data into a single experiment using the PMF Evaluation Tool (v 2.06) and the PFM2 algorithm (v 4.2) (Ulbrich et al., 2009; Paatero, 1997; Paatero and Tapper, 1994).

An Ionicon quadrupole high-sensitivity Proton-Transfer-Reaction Mass Spectrometer (PTR-MS) was used to measure selected gas-phase VOC concentrations (Lindinger et al., 1998). The PTR-MS was run in the ion monitoring mode in which signals of a limited number of pre-selected of m/z values are sequentially measured, with one measurement cycle taking 3.5 s. Averaging time at each m/z in the series varied depending on the sensitivity of the instrument to that species, the expected concentration, and the background, but it was generally between 0.2 and 0.5 s. Isoprene at m/z69 was sampled multiple times during the cycle to enable flux analysis (Gu et al., 2017). Drift tube temperature, pressure, and voltage were held at 60 ∘C, 2.22 hPa, and 600 V, respectively, resulting in an electric-field-to-gas-density (E/N) ratio of 134 Td (1 Td =1×10-17 cm2 V-1 s-1). The PTR-MS sampled air through a dedicated forward-facing inlet that consisted of approximately 6′′ of 1/4 in. OD stainless steel followed by approximately 46 in. of 1/4 in. Teflon tubing and 36 in. of 1/16 in. OD PEEK tubing. The flow through the Teflon tubing was 600 cc min-1 with 300 cc min-1 subsampled through the PEEK tubing for introduction into the PTR-MS. To assess the PTR-MS background, air was periodically diverted through a stainless-steel tube filled with a Shimadzu platinum catalyst heated to 600 ∘C, which removes VOCs from the airstream without perturbing RH. The catalyst efficiency was tested during the campaign by comparing signal from air containing VOCs passed through the catalyst with signal from VOC-free air. The PTR-MS was calibrated by introducing known concentrations of calibration gases into the instrument with variable dilution by VOC-free air. The calibration tank VOC concentrations were determined gravimetrically and verified using gas chromatography analysis by the manufacturer (AiR Environmental, Inc).

Ozone was measured with a Thermo Scientific Model 49i ozone analyzer based on measurement of UV absorption at 254 nm. The instrument was regularly calibrated in flight by displacement of known quantities of ozone and zeroed in flight using ozone-scrubbed ambient air. CO was measured using a Los Gatos Research CO–N2O–H2O analyzer that is based on cavity-enhanced near-IR absorption and was also calibrated regularly in flight. Additional information on the instrumentation deployed on the G-1 can be found in the GoAmazon 2014/5 overview (Martin et al., 2016).

As discussed in the Introduction, the different precipitation and mesoscale transport patterns in the wet and the dry season are expected to produce considerably different concentrations of pollutants in the Amazon Basin, and measurements confirmed this expectation. Figure 2 and Table 2 summarize the AMS results for data collected on the G-1 flights in both the wet and the dry season. The top panel of Fig. 2 shows box-and-whisker plots summarizing AMS data for organic, sulfate, nitrate, and ammonium particulate concentrations, with one data point representing each flight (note the scale change between the wet- and dry-season plot). The middle panel shows the relative distribution of the chemical species on each flight. The bottom panel shows the distribution of the chemical species as a mass-weighted average of all data from the respective season. Several trends are readily apparent in the data. First, it is clear that aerosol loadings were significantly higher in the dry season than in the wet season. Median organic loadings for the wet- and dry-season data were 0.85 and 4.29 µg m-3, respectively. Median sulfate, nitrate, and ammonium loadings increased from 0.19, 0.02, and 0.06 µg m-3 in the wet season to 0.77, 0.05, and 0.25 µg m-3 in the dry season. Mean loadings of all species were somewhat higher than the median. Mean organic, sulfate, nitrate, and ammonium loadings were 0.91, 0.16, 0.02, and 0.05 µg m-3 in the wet season and 4.41, 0.80, 0.1, and 0.26 µg m-3 in the dry season. In the wet season, the mass loading of all species were lower than typically observed over continental regions in the Northern Hemisphere (Zhang et al., 2007).

While this paper will focus on the particulate data, measurements of the volatile organic compounds provide insights into the precursors that are oxidized to form OA and help to identify the source of an air parcel. Figure 3 and Table 3 summarize the concentrations of several relevant VOCs measured on board the G-1 with the PTR-MS. Similar to the trends seen in the aerosol mass loading data, concentrations of most VOCs measured by the PTR-MS are significantly higher in the dry season than in the wet season. Concentrations of isoprene (m/z69) and its oxidation products (m/z71), biogenic precursors of OA, are a factor of 2–3 higher in the dry season than the wet season. The average daily high temperatures in Manaus are 33.5 ∘C in September (dry season) and 30.9 ∘C in March (wet season) (INMET, 2018), and isoprene emissions have been shown to scale with temperature, among other variables (Guenther et al., 2012). Measurements of benzene (m/z79), which is primarily anthropogenic in origin, were also significantly higher in the dry season. While benzene itself is unlikely to significantly contribute to SOA formation over the timescales we observe on the G-1 flights (4–6.5 h) due to its ∼5-day atmospheric lifetime (Atkinson and Arey, 2003), other unmeasured anthropogenic VOC concentrations which would contribute to more immediate OA formation may have been higher as well. The higher VOC concentrations may contribute to the higher OA concentrations measured in the dry season. However, the PTR-MS data, in addition to satellite and remote-sensing measurements (Martin et al., 2016), also show that biomass burning significantly impacted the region. Measurements of acetonitrile (m/z42), whose major source is biomass burning (Yokelson et al., 2009, 2007), are 2–3 times higher in the dry season than the wet season, indicating a significant biomass burning impact in the region. Biomass burning is a known source of OA (Jolleys et al., 2012; Bond et al., 2004; Yokelson et al., 2009; Ferek et al., 1998) and would also contribute to the higher aerosol concentrations observed in the dry season.

The aerosol mass loadings we measure in the dry season are consistent with aircraft-based AMS measurements made during the SAMBBA campaign at the same time of year, though several hundred kilometers to the SW in the state of Rondônia (Allan et al., 2014). Chen et al. (2009) reported AMS-measured wet-season campaign (7 February to 13 March 2008) average organic and sulfate loadings of 0.7 and 0.15 µg m-3 from a ground site that was within the GoAmazon 2014/5 flight domain, though they excluded data impacted by Manaus emissions from their analysis. When comparing aircraft to ground site data, it is important to acknowledge differences in the data sets. First, the aircraft samples a different spatial domain in all three spatial dimensions (latitude, longitude, and altitude) than a ground site. In addition, the G-1 typically sampled the boundary layer in the late morning or early afternoon (see Table 1 for flight times) and thus may have missed peak concentrations of secondary species. Finally, the goal of the flights was often to sample the Manaus plume for the first half of each flight and clouds for the second half. The mean aerosol loadings reported here agree well with those reported by Chen et al. (2009), considering the inherent differences in data sets discussed above. We also found good agreement when comparing the G-1-measured aerosol loadings to those measured at the T3 site when the aircraft passed overhead and at 500–600 m altitude (de Sá et al., 2018). Our loadings are at the low end of the range (1–2 µg m-3) reported from a ground site in Rondônia, approximately 825 km to the southwest of Manaus, though those data were more heavily influenced by biomass burning and relied on a combination of offline techniques to speciate the aerosol (Fuzzi et al., 2007). To understand potential bias in the data set due to altitude, we calculated statistics for data collected below 700 m, which captures the lowest altitude portion of each flight. Giangrande et al. (2017) analyzed radiosonde data from balloons launched at the T3 site and report average mixed-layer heights of greater than 1000 m above ground level at 10:00 local time in both the wet and dry seasons of GoAmazon 2014/5. Thus, the G-1 was typically sampling in the boundary layer on the lowest flight legs, and the data collected below 700 m should be representative of boundary layer conditions. In the wet season, the low-altitude AMS loading statistics are generally either slightly higher or unchanged relative to the full data set. In the dry season, the low-altitude AMS loading statistics are either slightly lower relative or unchanged relative to the full data set, likely due to transport of biomass burning from the south in elevated layers. Both the median and mean concentrations of most VOCs measured are slightly higher for the data at <700 m altitude relative to the full data set (Table 3). This is the expected behavior for most VOCs measured here as they either are directly emitted at the surface (e.g., isoprene, benzene, monoterpenes) or form from oxidation of VOCs emitted at the surface (e.g., isoprene oxidation products).

The fractional contribution of each species to the total loading is nearly identical when comparing seasons, despite the large differences in aerosol absolute mass concentrations and the larger influence of biomass burning in the dry season (Martin et al., 2016). Organics dominate the chemical composition, comprising 80 % of the total, with sulfate comprising 14 %, nitrate 2 %, and ammonium 4 %, regardless of the season. The relative distribution of the products changes only modestly among flights (Fig. 2, middle panel). Organics were the largest fraction of the total aerosol mass on each flight, varying between 69 and 90 %. Sulfate was the second-largest fraction of the total mass (6–20 %), and the sum of sulfate and organic was nearly constant. The fractional contributions of each chemical species described herein are nearly identical to the previous reports from Chen et al. (2009, 2015). The aircraft-measured chemical composition is also nearly identical to that measured at the T3 ground site (de Sá et al., 2018) during GoAmazon 2014/5. The agreement between the T3 and G-1 product distributions suggests that the T3 ground site is sampling air that is representative of the regional air and is not unduly biased by local emissions, at least in the aggregate.

The observation that the total concentrations of all aerosol components significantly increase yet remain in the same proportions when comparing the wet and the dry season is unexpected. The more frequent and widespread precipitation events in the wet season and the lower concentrations of precursor VOCs contribute to the lower aerosol concentrations in the wet season. If the aerosol particles were internally mixed, it would be reasonable to assume that rainfall would remove aerosol in approximately equal proportions. However, as discussed previously, aerosol sources are different between seasons, and there was a much larger biomass burning influence in the dry season (Martin et al., 2016), which is a known and significant source of OA (e.g., Jolleys et al., 2012; Bond et al., 2004; Yokelson et al., 2009; Ferek et al., 1998). Biomass burning was previously shown to emit particulate sulfate and nitrate directly, as well as NOx, sulfuric acid, and methane sulfonic acid (MSA), precursors of particulate nitrate and sulfate, but not in the same ratio as OA and its precursors (Yokelson et al., 2009). Thus, given the significant influence of biomass burning in the dry season, it is surprising that the chemical composition of the aerosol does not change.

Despite the similarity of the distribution of organic, sulfate, and nitrate in the aerosol particles between the wet and dry seasons, the chemical composition of the organic aerosol is quite different between seasons. Figure 4 shows normalized probability distributions for three metrics of the organic aerosol composition – the hydrogen-to-carbon ratio (H : C), the oxygen-to-carbon ratio (O : C), and the mean carbon oxidation state (OS‾c) – segregated by season. It is clear that the organic particles are more oxidized in the dry season than the wet season. The median O : C, H : C, and OS‾c in the dry season were 0.78, 1.58, and -0.02, respectively, while they were 0.6, 1.65, and -0.45 in the wet season. While the H : C ratios change only slightly between season, the O : C and OS‾c both show significant increases in the dry season. Thus, the aerosol in the dry season is significantly more aged, consistent with aged biomass burning aerosol contributing to the organic aerosol mass in the dry season and relatively fresh, locally generated organic aerosol dominating in the wet season (Martin et al., 2010b). The probability distributions during the wet season are also wider than in the dry season. We postulate that small contributions from a wider range of sources were able to influence the data to a larger degree in the wet season, when the total organic aerosol concentrations were much smaller. In contrast, during the dry season, small sources of organic mass would have a smaller impact on the organic composition due to the significant biomass burning background.

Figure 1 shows the flight path on 13 March 2014, which is an exemplary day for observing the evolution of the Manaus plume in the wet season. Flight legs were perpendicular to the prevailing wind direction and spaced approximately 24 km apart at successively increasing distance from Manaus. The length of the legs was chosen such that measurements (e.g., CO concentration) returned to near background levels at the ends of the leg. Due to dispersion of the plume, the length of the legs generally increased downwind of Manaus, ranging from 56 km close to the city to 80 km downwind of the T3 site. The pattern was flown first at an altitude of 500 m, with a second pass at 1000 m overlapping the first. This flight was marked by mostly sunny skies, with no clouds intercepted by the G-1 at 500 m and only brief passage through spatially small clouds at 1000 m. There was little local or regional biomass burning during this time period. Thus, this flight represents a case with coherent transport of the Manaus plume downwind into regions of high biogenic emissions with few complicating factors, such as biomass burning or cloud processing.

Figure 5 shows a time series of several important gas- and particle-phase species for the 13 March 2014 flight. Passage through the Manaus plume is indicated by clear and significant increases in CO, ozone, and PTR-MS m/z79 (benzene) above the background levels. While CO may be produced from the oxidation of biogenic VOCs (BVOCs; Slowik et al., 2010), the intense peaks are primarily from urban emissions, given the proximity to Manaus and the correlation of CO with other known anthropogenic species, such as ozone, benzene, and toluene (not shown). Concentrations of CO (80 ppbv) and ozone (10 ppbv) are lowest at the edges of the flight legs; although we did not sample extensively in regions that were far removed from Manaus, these concentrations are consistent with the background concentrations that have been previously reported in the region (Andreae et al., 2015; Martin et al., 2010b; Harriss et al., 1990). Isoprene concentrations (PTR-MS m/z69) typically reach their highest levels outside of the plume and are lower inside, though isoprene concentrations of approximately 1 ppbv are still observed in the plume. As seen in Fig. 1, much of the land cover outside of Manaus is tropical forest, with some pasture land interspersed. Significant isoprene emissions are expected and indeed measured from the forest (Gu et al., 2017; Guenther et al., 2006). Significant concentrations of methacrolein, methyl vinyl ketone, and isoprene hydroxyhydroperoxides (PTR-MS m/z71), all first-generation oxidation products of isoprene, are also observed and have a more complicated structure (Liu et al., 2016). Concentrations of these products tend to be highest near the edges of the plume, and they do not correlate strongly with either isoprene (r2=0.24) or anthropogenic plume markers such as CO (r2=0.24). Production of these oxidation products requires OH, the concentration of which depends on NOx, to oxidize isoprene, and thus they are expected to have a more complex relationship with the urban plume and the biogenic emissions from the forest. Taken together, the observations of both isoprene and its oxidation products suggest that photooxidation of BVOCs is enhanced in the plume relative to the background, as would be expected (Martin et al., 2017). Monoterpene concentrations (not shown) on this flight are approximately a factor of 10 lower than isoprene concentrations, consistent with previous measurements (Kesselmeier et al., 2000), and are often near or below the instrument limit of detection (∼200 pptv). Monoterpene concentrations correlate with isoprene concentrations during the flight, when they are above the instrument detection limit.

OA mass increases from background values of 0.5 µg m-3 to as high as 2.8 µg m-3 in the Manaus plume, and on this day it is correlated relatively well with both ozone (r2=0.65) and CO (r2=0.76). While OA concentrations do not reach the levels that are typically seen in the outflow of large North American cities, they are significantly larger than background concentrations (up to 6×), indicating a clear impact of the plume on OA concentrations. Sulfate also clearly increase in the plume, though its peak concentrations occur toward the southern edge of the plume, relative to organics, CO, and ozone, suggesting sulfur emissions may not be co-located with other anthropogenic emissions. However, there are also occasional peaks in the sulfate concentration (for example at 10:47 LT) that coincide with minima in other plume indicators, such as CO and ozone. Nitrate concentrations are significantly lower than sulfate concentrations but tend to correlate more strongly with organic concentrations than sulfate concentrations. There are indications that a significant portion of the nitrate mass exists as organic nitrates, consistent with measurements at the T3 site and with aircraft measurements (Allan et al., 2014; de Sá et al., 2018; Schulz et al., 2018). The NO+ : NO2+ ion ratio is 5–7 in the plume, compared to a NO+ : NO2+ ion ratio of 1.2–1.5 measured during the campaign for ammonium nitrate calibration aerosol. NO+ : NO2+ ion ratios significantly above that of ammonium nitrate are indicative of the presence of organic nitrates, though quantifying the organic fraction is challenging (Farmer et al., 2010). Ammonium concentrations tend to closely mirror sulfate concentrations. The bottom panel of Fig. 5 shows the fractional composition of the aerosol, which helps to illustrate the relative differences among the aerosol species. Organic species clearly dominate at all points, with sulfate the next-most-important species.

At 1000 m altitude, the plume can still be observed in the G-1 data, though the concentrations of plume markers such as CO, ozone, and benzene are lower. Increases in organic concentration are less sharp and defined than observed at 500 m, though enhancements are still clear. The organic mass fraction is somewhat smaller at 1000 m, and sulfate and ammonium mass fractions are larger. Based on radiosonde measurements, the boundary layer grew from 900 to 1200 m between the beginning and end of the flight on this date. Data from the G-1 suggest that the 1000 m flight path was near the top of the boundary layer. Thus, at 1000 m altitude, the data are influenced both by the local conditions and from air that has been transported over longer distances.

Figure 6 shows a PMF analysis of the organic species for the 13 March flight along with independently measured tracers that might be expected to correlate with the PMF factors. We were able to resolve the organic aerosol into three factors: an OOA factor that is a proxy for SOA; an HOA factor that is a proxy for primary organic particulate emissions; and an IEPOX SOA factor, which forms from the heterogeneous uptake of IEPOX, a product of isoprene oxidation under low-NOx conditions (Hu et al., 2015; Robinson et al., 2011; Lin et al., 2012; Zhang et al., 2011;, 2005a; de Sá et al., 2017). Details of the PMF analysis, including mass spectra of the factors, can be found in the Appendix and Supplement. As expected, both the OOA and HOA factor mass loadings are significantly higher in the plume relative to the background. The HOA factor correlates strongly with CO (r=0.79 for all data at 500 m), while the OOA factor correlates strongly with ozone (r=0.81 for all data at 500 m), though we will show below that the slope of the correlation changes with plume age. The changing slope has the effect of degrading the HOA–CO and OOA–ozone correlations, which are higher for individual plume legs. The HOA factor is a larger fraction of the organic mass on the legs nearest to Manaus and becomes a smaller fraction on the downwind legs. The OOA mass fraction displays the opposite behavior, increasing downwind as the plume ages. The transformation of HOA to SOA and/or OOA has been observed previously, both in laboratory experiments and in the field (Sage et al., 2008; Robinson et al., 2007; DeCarlo et al., 2008; Zhang et al., 2007; Ng et al., 2010). Taken together, these observations are consistent with the literature assignment of HOA as a marker of primary organic emissions and OOA as a marker of secondary organic aerosol. Outside of the plume, HOA concentrations approach zero. OOA concentrations, on the other hand, remain significant and approach the organic aerosol background concentrations. The IEPOX SOA factor does not appear to change with passage through the plume and does not correlate with sulfate aerosol on this day as may be expected based on the chemical mechanism, which requires heterogeneous uptake of gas-phase IEPOX onto inorganic aerosol to generate IEPOX SOA (Hu et al., 2015; Surratt et al., 2010; Gaston et al., 2014; Nguyen et al., 2014). We note that the IEPOX SOA factor is noisy, is near or below the limit of detection of organic aerosol (0.13 µg m-3) for much of this flight, and is shown largely for the sake of completeness. Because of the low concentrations, it is difficult to draw definitive conclusions about IEPOX SOA formation on this flight. However, given the focus of this flight was on tracking the Manaus plume, which contains significant NOx, our findings are consistent with de Sá et al. (2017), who report that NOx from the Manaus plume suppresses IEPOX SOA formation in the plume and that the IEPOX SOA factor was less than 5 % of the total OA at T3 at the time of the G-1 overflight.

Figure 7 summarizes several relevant plume quantities at 500 m altitude as a function of the approximate distance downwind from Manaus, which serves as a proxy for the photochemical age of the air mass. The ratio of excess organic aerosol to excess CO (Δorg/ΔCO) is a metric that is often used to quantify organic aerosol formation in a source plume as it ages. The utility of the Δorg/ΔCO metric rests mainly on the assumptions that CO is conserved on the timescale of the measurements and that urban emissions scale linearly with CO. Our calculation also assumes that Manaus emissions are the dominant source of CO in the plume. Using this ratio rather than absolute concentrations of organic aerosol can normalize for dilution due to mixing of the plume with background air. The Δorg/ΔCO values shown in Fig. 7 were calculated as the slope of a linear regression line between the AMS-measured organic mass loadings and CO concentrations. All data for each flight leg perpendicular to the wind direction were included. Background values of OA and CO were not subtracted from the data, and the regression was not forced through the origin. We acknowledge that calculation of Δorg/ΔCO values can be sensitive to the calculation method and choice of background values for both OA and CO; therefore, we performed calculations using different methods and assuming different background concentrations for both OA and CO and find general agreement between these methods (see Supplement for more details).

Two aspects of the Δorg/ΔCO distinguish these measurements from most previous measurements in the literature. First, Δorg/ΔCO is ∼31 µg m-3 ppmv-1, a value that is lower than the value of 70±20 µg m-3 ppmv-1 observed in the outflow of cities in North America (e.g., de Gouw and Jimenez, 2009, and references therein). The Δorg/ΔCO values observed here are also significantly smaller than the values of 77–157 µg m-3 ppmv-1 we previously observed when fresh urban anthropogenic emissions evolved in the presence of strong biogenic emissions, as is the case in the present data set (Shilling et al., 2013; Setyan et al., 2012). We measure a ΔHOA/ΔCO value of 17.6 µg m-3 ppmv-1 near Manaus (Fig. 7), which should represent fresh emissions. This value is at the upper end of what is reported in the literature for North American cities, indicating that lower primary particulate emissions from Manaus are not responsible for the lower Δorg/ΔCO values (DeCarlo et al., 2010; de Gouw and Jimenez, 2009).

The second new aspect of these observations is that we do not observe a corresponding increase in Δorg/ΔCO as the plume photochemically ages, contrary to our expectations. We note that, as described above, organic aerosol concentrations are significantly higher in the plume than in the background, and the constant Δorg/ΔCO values do not indicate that no OA formation occurs in the plume. Rather, the Δorg/ΔCO analysis that follows focuses on understanding the aging of the Manaus plume. Based on literature observations (de Gouw and Jimenez, 2009; Kleinman et al., 2008; DeCarlo et al., 2010; Takegawa et al., 2006; Sullivan et al., 2006; Freney et al., 2014), we hypothesized that Δorg/ΔCO would rapidly increase in the Manaus plume due to OA formation from sources not associated with urban CO emissions (Setyan et al., 2012; Shilling et al., 2013). Specifically, we expected that enhanced OH concentrations in the plume would lead to rapid oxidation of BVOCs emitted from the surrounding forest, which would produce OA with little concomitant production of CO and lead to a rapid increase in Δorg/ΔCO. As seen in Fig. 7, the data did not support this hypothesis. As described in the Supplement, we performed the Δorg/ΔCO calculation using several methods and assuming different background CO and OA values and did not find significant increases in Δorg/ΔCO in any of these calculations. Furthermore, though we focus on Δorg values based on the AMS data, we also calculated Δvolume/ΔCO using aerosol size distribution data from two independent instruments, the Ultra-High Sensitivity Aerosol Spectrometer (UHSAS) and the Fast Integrated Mobility Spectrometer (FIMS), that were also on board the G-1. Neither of these calculations indicates an increase in Δvolume/ΔCO with plume age.

Previous studies have observed increases in Δorg/ΔCO with plume age, with the largest increases occurring for the first ∼1 day of aging and changes in Δorg/ΔCO gradually levelling off beyond approximately 1–2 days (DeCarlo et al., 2010; de Gouw and Jimenez, 2009; Kleinman et al., 2008; Freney et al., 2014). Based on the mean wind speeds observed along the flight track (7.3 m s-1) and the transport distance (up to 100 km), we estimate that the plume was 4–5 h old at the farthest leg and freshly emitted over the city. Unfortunately, photochemical clocks could not be used to more precisely calculate the photochemical age of the plume (de Gouw et al., 2005; Kleinman et al., 2003; Parrish et al., 1992). NOy measurements were not available on this flight. Benzene and toluene concentrations were low and noisy, particularly at increasingly downwind distances from Manaus. Though our observations are limited to shorter aging timescales (4–5 h) than many literature studies (1–2 days), the literature studies report measurable changes in Δorg/ΔCO at short aging timescales (Kleinman et al., 2008; DeCarlo et al., 2010). Furthermore, we clearly observe other indicators of photochemical aging in the plume. Ozone concentrations are 30–50 ppbv in the plume, compared to background levels of 10–15 ppbv, indicating active photochemistry. Concentrations of isoprene are lower in the plume than the background values observed outside of it. Concentrations of isoprene inside the plume do not show a monotonic dependence on plume age. The concentration of isoprene oxidation products measured in the plume by the PTR-MS at m/z71 monotonically increases with plume aging, from 0.96 ppbv directly over the city to 1.27 ppbv at the farthest leg. The average toluene : benzene ratio in the plume also monotonically decreases with plume aging as would be expected, though attempts to convert these ratios into a photochemical age resulted in unrealistically high estimates of the plume age, likely due to noise in the data as mentioned above (de Gouw et al., 2005). The mean particle oxidation state (OS‾c) of plume OA increases from -0.6 to -0.44 as it ages. It is expected that photochemical aging would produce progressively more oxygenated species downwind of Manaus that subsequently partition to the aerosol phase, increasing the mean carbon oxidation state. Other studies have observed a similar phenomenon as particles were transported downwind of urban centers (DeCarlo et al., 2010). The particles size distributions measured by both the FIMS and the UHSAS also indicate that particles size increases downwind of Manaus. As discussed in the previous section, we also observe that the mass fraction of HOA decreases and simultaneously the mass fraction of OOA increases as the plume ages. As seen in Fig. 7, ΔHOA/ΔCO is 17.6 µg m-3 ppmv-1 on the leg nearest to the city and decreases to 10.6 µg m-3 ppmv-1. At the same time, ΔOOA/ΔCO increases from 9.2 to 23.1 µg m-3 ppmv-1. Thus, some fraction of HOA appears to be lost – through volatilization, deposition, or a combination of both – with the lost HOA mass balanced by an increase in OOA mass. All of these factors indicate that active photochemistry is occurring and is transforming VOCs into SOA and aging the particles. Thus, we would expect to see significant increases in Δorg/ΔCO for these aging times if similar observations of outflow from North American cities were representative of the Manaus plume.

The sum of the HOA and OOA factors explains >95 % of the total OA mass within the plume, as the IEPOX SOA mass is small. Though calculating the ΔHOA/ΔCO and ΔOOA/ΔCO ratios introduces some noise (particularly ΔOOA/ΔCO close to Manaus), the sum of these ratios is approximately constant with aging and equal to the Δorg/ΔCO values. Thus, the total OA mass is conserved with loss and volatilization of HOA balanced out by formation of OOA. This process appears to explain the lack of an increase in Δorg/ΔCO with aging. It is difficult to determine whether the OOA carbon mass is anthropogenic in origin, biogenic, or (most likely) a combination of both. Volatilized HOA and co-emitted anthropogenic VOCs would undergo oxidation in the plume, and the products may condense and be detected as OOA (Sage et al., 2008; Robinson et al., 2007). In addition, we see evidence that biogenic VOCs are oxidized in the plume, likely more rapidly than in the background due to enhanced in-plume OH concentrations, and these oxidation products may also condense as OOA. Either mechanism or a combination of both would explain the observed increase in the carbon oxidation state. It is surprising that the volatilized HOA mass is compensated by oxidation and recondensation of SOA for these relatively fresh particles.

Thus far, we have focused on a detailed analysis of the 13 March 2014 plume because there is a clear contrast between the background and plume data through the portion of the flight in the boundary layer, there is little interference from biomass burning and cloud processing, the flight pattern was extended far enough downwind to investigate a significant aging time, and the data from all key instruments are available and complete. In an effort to understand whether the Δorg/ΔCO observations on 13 March are representative of plume aging in the wet season, we have screened the rest of the wet-season flight data and, based on the criteria mentioned above, identified two additional flights, 11 and 16 March 2014, for analysis. The Δorg/ΔCO values for these data as a function of distance from Manaus are shown in Fig. 8. On 11 March, Δorg/ΔCO values average 29.9 µg m-3 ppmv-1, show little change with distance from Manaus, and show similar variability between legs to the 13 March flight data. The mean wind speed for the analyzed portion of the flight was 5.3 m s-1, translating to an approximate maximum plume age of 5–5.5 h. On 16 March, Δorg/ΔCO values average 42.0 µg m-3 ppmv-1, show little significant increase with distance from Manaus, and exhibit a larger range of values than the other two data sets. The mean wind speed for the analyzed portion of the 16 March flight was 4.6 m s-1, translating to an approximate maximum plume age of 6–6.5 h. We estimate that the error in the Δorg/ΔCO measurement is approximately 10 µg m-3 ppmv-1; thus the difference in average Δorg/ΔCO values between the 16 March and the 11 and 13 March data sets is at the edge of what we consider significant. The average Δorg/ΔCO is 34.3 µg m-3 ppmv-1 for all flights, and the combined data set collectively represents the first 4–6.5 h of aging. In addition to the aircraft data, Cirino et al. (2018) report that the median Δorg/ΔCO values for the plume as measured at the T2 and T3 sites are nearly identical to one another in the wet season and similar to measurements from the G-1 (Cirino et al., 2018). Thus, the data from both G-1 and the ground sites suggest that Δorg/ΔCO values of 30–40 µg m-3 ppmv-1, which change little in the first 4–6.5 h of plume aging, are representative of the Manaus plume behavior in the wet season. As previously discussed, most studies examining outflow of North American cities have shown changes in Δorg/ΔCO that are largest at the shortest aging times, and we are unaware of reports of constant Δorg/ΔCO in an aging urban plume. Both the identity and the distribution of Manaus's emissions may very well be distinct from those of a typical North American city. In addition, background concentrations of anthropogenic pollutants are lower and biogenic VOC emissions are higher in the unperturbed rain forest surrounding Manaus relative to most North American cities. Finally, background OA concentrations near Manaus are lower in the wet season than in most other continental regions, and SOA yield is known to increase with OA mass loading (Odum et al., 1996). All of these factors may contribute of the difference in the Δorg/ΔCO observations.

A similar phenomenon, clear chemical aging with little change of Δorg/ΔCO, has been observed in field studies of biomass burning plumes by several researchers (Jolleys et al., 2012; Cubison et al., 2011; Hecobian et al., 2011; Akagi et al., 2012; Forrister et al., 2015). Clearly, the mix of organic compounds emitted from a forest fire and from the Manaus urban region will be significantly different, and this is not to imply that the detailed chemical mechanisms are the same. However, the similarity of the observations is suggestive that a similar process could occur in both types of plumes.

Sulfate was the second-largest contributor to the non-refractory aerosol mass during the GoAmazon 2014/5 campaign (accounting for 13 % of the PM1 mass in both the wet and dry season), so it is also instructive to examine the source of sulfate in the area. In 2014 Manaus generated much of its power from fuel oil and diesel, which are known to emit SO2 that eventually oxidizes to form condensable sulfate (Medeiros et al., 2017). However, several observations suggest that the major source of sulfate in the plume may not come from local SO2 emissions. First, on many flights, particulate sulfate levels clearly increases in the plume (e.g., Fig. 5), including on the legs closest to Manaus and thus very near the source. The lifetime of SO2 with respect to homogeneous oxidation by OH is expected to be about 1 week, and only a small fraction (∼5 % on the farthest leg) of SO2 emissions would have time to oxidize to H2SO4 in the gas phase. In addition, there is no indication of dramatically enhanced sulfate loading downwind of Manaus on the 13 March flight, when the ΔSO4/ΔCO ratio averages 2.6 µg m-3 ppmv-1 and does not increase with plume age. On the 13 March flight, there was no precipitation and there were relatively few clouds along the flight track, so heterogeneous oxidation of sulfate in cloud water cannot explain the sulfate enhancement seen in the plume.

We can possibly attribute these observations to two mechanisms. First, it is possible that natural sulfur emissions originating downwind of Manaus undergo oxidation by enhanced concentrations of OH in the Manaus plume and subsequently condense (Andreae et al., 1990; Chen et al., 2009). Andreae et al. (1990) estimate that in-basin sulfur emissions are responsible for approximately 0.05 µg m-3 of the total sulfate loading. Measurements during the AMAZE-08 campaign also suggested natural in-basin emissions and transport of natural out-of-basin emission were a significant source of particulate sulfate (Chen et al., 2009). A 0.05 µg m-3 sulfate loading is roughly consistent with our measured sulfate concentrations outside of the plume and would account for 15–30 % of the sulfate in the plume on the 13 March flight. Higher OH concentrations in the plume would speed oxidation of natural sulfur compounds and further increase the fraction of in-plume sulfate originating from natural emissions. Second, power plant plumes may directly emit sulfuric acid, which then may condenses on pre-existing particles. A previous aircraft study of the Manaus outflow observed high concentrations of particles smaller than 40 nm in power plant plumes, which they attribute to sulfuric acid (Kuhn et al., 2010). Though particles in this size range are unlikely to be detected by the AMS, condensation of sulfuric acid on larger, pre-existing particles would also contribute to the in-plume sulfate observations.