The ground-based aerosol data presented here were collected at the Amazon Tall Tower Observatory (ATTO) site (called T0a in the GoAmazon2014/5 experiment; Martin et al., 2016) located in the Uatumã Sustainable Development Reserve, Amazonas, Brazil. Details about the ATTO site infrastructure, instrumentation, and an overview of running measurements can be found in Andreae et al. (2015). Figure 1 shows the ATTO site location and the location of the Nyamuragira volcano in the DRC (1.41∘ S, 29.2∘ E; 3058 m a.s.l). The long-term measurements at ATTO started in 2012. A systematic backward trajectory (BT) analysis of air masses arriving at ATTO can be found in C. Pöhlker et al. (2018). The ATTO aerosol measurements were taken on a triangular mast (02∘08.602′ S, 59∘00.033′ W; 130 m above sea level, a.s.l.) using a 25 mm diameter, 60 m high, stainless steel tube with a laminar sampling flow rate of 30 L min-1. The instruments were installed inside an air-conditioned container where the temperature was kept between 29 and 31 ∘C. All aerosol data were aggregated to hourly mean values. More details on the aerosol sampling setup can be found elsewhere (Moran-Zuloaga et al., 2017; Pöhlker et al., 2016; Saturno et al., 2017).

Equivalent black carbon (BCe) mass concentrations, MBCe, were calculated from absorption measurements by a multi-angle absorption photometer (MAAP; model 5012; Thermo Electron Group, Waltham, USA). The details of the instrument are described elsewhere (Petzold and Schönlinner, 2004). The BC mass absorption cross section (MAC) was retrieved from fitting MAAP absorption coefficients at 637 nm wavelength and refractory black carbon (rBC) mass concentrations were measured by using a single-particle soot photometer (SP2) revision D (Droplet Measurement Technologies, Longmont, USA). Details of the technique can be found in Stephens et al. (2003). The MAC calculations are described in Saturno et al. (2017). Light-scattering coefficients were measured using a three-wavelength integrating nephelometer (Aurora 3000; Ecotech Pty Ltd., Knoxfield, Australia). For details of the instrument, see Müller et al. (2011). Absorption and interpolated scattering measurements at a 637 nm wavelength were used to calculate the single scattering albedo of dry aerosol particles, ω0, at this wavelength, which is defined as the ratio of scattering to extinction coefficients (extinction = scattering + absorption). Strong BB- and FF-influenced observations were defined as periods when MBCe exceeded the 75th percentile (0.27 µg m-3) and the absorption wavelength dependence was above or within the BC-only regime, respectively. A detailed discussion on the absorption wavelength dependence can be found in Saturno et al. (2017).

An aerosol chemical speciation monitor (ACSM; Aerodyne Research Inc., Billerica, USA) was used to measure online non-refractory aerosol chemical composition (Carbone et al., 2018). These measurements started in February 2014 at the ATTO site. The technique resolves the submicron aerosol chemical species in the following categories: organics, sulfate, nitrate, ammonium, and chloride (Ng et al., 2011). In this study, we only use organics and sulfate mass concentration data, Morg and Msulfate, respectively.

Cloud condensation nuclei (CCN) number concentrations, NCCN, were measured with a CCN counter (CCNC; model CCN-100; Droplet Measurement Technologies, Longmont, USA), which was deployed at the ATTO site starting in March 2014. The instrument scanned over a range of different supersaturations and particle diameters; more details can be found elsewhere (Pöhlker et al., 2016). The hygroscopicity parameter, κ, retrieved for a CCN activation ratio of 50 % is used in this study. Condensation nuclei number concentrations (> 5 nm), NCN, were measured with a condensation particle counter (CPC; model 5412; Grimm Aerosol Technik, Germany).

Chemical species of submicron aerosol particles were measured using a compact time-of-flight aerosol mass spectrometer (C-ToF-AMS) installed onboard the German High-Altitude and Long Range Research Aircraft (HALO; http://www.halo.dlr.de, last access: 13 September 2017), a modified business jet G550 (Gulfstream, Savannah, USA). The C-ToF-MS details are presented in Drewnick et al. (2005) and Schulz et al. (2018). A single-particle soot photometer (SP2; Droplet Measurement Technologies, Longmont, USA) was used to measure rBC mass concentration. The measurements took place between 6 September and 1 October 2014 during the “Aerosol, Cloud, Precipitation, and Radiation Interactions and Dynamics of Convective Cloud Systems” (ACRIDICON) – “Cloud Processes of the Main Precipitation Systems in Brazil: A Contribution to Cloud Resolving Modeling and to the GPM (Global Precipitation Measurement)” (CHUVA) campaign over the Amazon rain forest (Machado et al., 2014). More details on the flight trajectories and instrumentation can be found in Wendisch et al. (2016). In this study, only data up to 7 km of altitude have been used.

To investigate the probability of the volcanic sulfate plume reaching the ATTO site, trajectories were calculated using the National Oceanic and Atmospheric Administration (NOAA) hybrid single-particle Lagrangian integrated trajectory HYSPLIT model (Draxler and Hess, 1997, 1998; Stein et al., 2015). NOAA Global Data Assimilation System (GDAS; Kleist et al., 2009) data at 1∘ × 1∘ resolution were used as meteorological input for HYSPLIT.

As one of the most abundant gases in a volcanic plume, SO2 is often used as a tracer for volcanic emissions by a variety of spectroscopic remote sensing techniques. The strong characteristic absorption features in the UV spectral range allow for the quantification of SO2 using differential optical absorption spectroscopy (DOAS; see, e.g., Platt and Stutz, 2008 and Richter and Wagner, 2011) both from the ground (e.g., Bobrowski and Platt, 2007; Galle et al., 2003) and from space (e.g., Eisinger and Burrows, 1998; Khokhar et al., 2005 and Krueger, 1985).

The ozone monitoring instrument (OMI) onboard the National Aeronautics and Space Administration (NASA) Aura satellite, launched in 2004, detects backscattered solar radiation in the UV–Vis range (Levelt et al., 2006). The polar-orbiting instrument crosses the Equator at 13:30 LT (local time). DOAS analysis of OMI spectra yields column densities of trace gases such as NO2, SO2, and HCHO with a spatial resolution of about 13 × 24 km2 away from the swath edges. The OMI wide swath of 2600 km allowed for daily global coverage until the first occurrence of the so-called row anomaly in June 2007, an instrumental problem that causes grievous radiance errors in up to half of the OMI ground pixels (Van Hoek and Claas, 2010). The row anomaly strongly affects the reliability of observations; therefore all affected pixels were removed from the data set prior to analysis.

The OMI SO2 vertical column density (VCD) data presented in this paper were retrieved using the NASA principal-component-based algorithm with an a priori vertical profile representative of a volcanic plume in the middle troposphere (TRM; Li et al., 2013, 2017). The assumption that the volcanic plume is located in the mid-troposphere is justified by the elevation of the volcano (3058 m), the strength of the eruption, and, particularly, the HYSPLIT trajectory analysis presented later in this paper. It is, however, important to note that the sensitivity of the satellite measurements depends systematically on plume altitude. Thus, the absolute values of the SO2 VCD derived from the satellite observations overestimate or underestimate the true values if the plume is located at a higher or lower altitude, respectively. Fortunately, this does not influence our study, as the focus of this paper is on the spatial pattern of the SO2 plumes and not on the SO2 amount. The level-2 data were downloaded from https://mirador.gsfc.nasa.gov/ (last access: 27 October 2017) and gridded to a regular 0.1∘ × 0.1∘ grid for easily handling.

Given that the Nyamuragira and Nyiragongo volcanoes are so close to each other (within ∼ 15 km), their emissions detected by remote sensors are often treated as a paired source (Carn et al., 2017). Hereafter, the term “Nyamuragira” refers to the couple Nyamuragira–Nyiragongo in this text. Time series of area-averaged OMI SO2 observations are shown in Fig. 3a. The area where the average was calculated (Fig. 3b) corresponds to approximately 200 000 km2 around the volcanoes. Nyamuragira produced frequent intensive SO2 emission events in 2014 with a mean emission of 14.4 × 106 kg SO2 day-1 (Barrière et al., 2017). Multiyear OMI SO2 VCD daily averages from 11∘ S to 17∘ N are summarized in Fig. S1 in the Supplement as a function of time and longitude. The figure shows observations during the ATTO measurement period (March 2012 to July 2017) and a snapshot of September 2014. According to Barrière et al. (2017), the emissions from June to October 2014 were mostly due to lava fountaining activity in Nyamuragira, characterized by strong tremors. The high emission activity of Nyamuragira at the end of 2014 was also observed by in situ measurements from November 2014 to October 2015, when SO2 emission fluxes reached up to 12.7 × 106 kg day-1 (Bobrowski et al., 2017). These emissions were often transported westward, as can be observed in HYSPLIT forward trajectories calculated for 2014 (Fig. S2).

Fortunately, during the high emission period mentioned above, specifically in September–October 2014, ground and airborne measurement campaigns took place in the Amazon Basin (Andreae et al., 2015; Wendisch et al., 2016; Martin et al., 2017). Satellite SO2 VCD observations over central Africa and the Atlantic Ocean were examined during this period in order to precisely identify the plume trajectory. A map of gridded OMI SO2 TRM VCD observations from 7 to 17 September 2014 is available as a Supplement (Fig. S3). Two emission events were observed at the Nyamuragira location, one on 7 September and another on 12 September. The latter exhibits a clear westward transport of the SO2 plume starting from 13 to 17 September. Figure 4 shows SO2 VCD observations during this period within the region between 20∘ W–30∘ E and 15∘ S–5∘ N with SO2 VCD larger than 2.5 × 1014 molecules cm-2. Several sets of trajectory calculations were performed. First, three to seven starting locations were selected within the SO2 plumes detected by OMI on 12–17 September 2014. At each location, 15-day (360 h) forward trajectories were started at the time of the satellite overpass (11:00–14:00 UTC) at seven altitudes spaced equally between 1 and 7 km. The resulting trajectories initialized at 4 km of altitude on 13 September are in best agreement with the satellite data and are shown in Fig. 4. All starting parameters were systematically varied and very consistent patterns were found (see Fig. S4). The trajectories are superimposed on the map presenting all SO2 plumes detected by OMI between 12 and 17 September. Trajectories started within the leading edge of the plume are in good agreement with the OMI data, as after 24 h (second marker) both trajectories are located within the plume detected on 14 September, and after 48 h (third marker) both trajectories are located within the plume detected on 15 September.

This approach of using the OMI data to evaluate the trajectories was used also with the purpose of minimizing uncertainties by calculating shorter trajectories instead of initializing them from the volcano location. It should be noted here that the trajectory calculations by HYSPLIT yield a line that can be understood as the center line of a propagating plume that widens both by stochastic uncertainty in the model calculations and by dispersive processes in the atmosphere. Consequently, the location of the plume becomes more uncertain the longer the model is run, and the physical size of the plume increases as well. Given the 9000 km distance between the volcanoes and ATTO, the uncertainty in the calculated plume trajectory position would become very large. To mitigate this problem, we use a multistep approach in which we follow the emissions using the satellite-observed locations of the SO2 plume for the initial days and then calculate forward trajectories from the observed location of the plume.

The two southernmost trajectories in Fig. 4 make a sharp turn after 15 September, which is in agreement with the observed pattern, although there is no longer an exact match with the respective OMI observations (in red and maroon). This discrepancy may be due to inaccuracy in the individual trajectories, or the SO2 plumes might have been below the OMI detection limit. The southernmost trajectories reach South America and come within several hundred kilometers of ATTO within 15 days, which is well within the HYSPLIT uncertainty estimated at 15–30 % of the trajectory length (https://www.arl.noaa.gov/hysplit/hysplit-frequently-asked-questions-faqs/faq-hg11/, last access: 19 March 2018). One trajectory reached the direct vicinity of ATTO on 25 September at 1.8 km of altitude, whereas another trajectory passed an area south of ATTO at an altitude of 1.5 km on 24 September. As discussed above, the locations along the trajectory should be interpreted as the centers of a plume, which had dispersed both vertically and horizontally. This dispersion is not considered in the trajectory mode of HYSPLIT (Stein et al., 2015), but of course takes place in reality by turbulent and convective processes. Consequently, a trajectory “parcel” moving at 1.8 km of altitude will certainly be dispersed to the surface in the highly convective Amazonian lower troposphere (Andreae et al., 1988).

In addition to the plume forward trajectory analysis, backward trajectories initiated at the ATTO site at an altitude of 300 m (approximately 170 m aboveground) were calculated for 360 h. These trajectories were initiated every hour from 20 September 00:00 UTC to 30 September 23:00 UTC. The results, presented as a trajectory density plot in Fig. 5, indicate that during this time period essentially all air masses come from southern Africa and a significant number of trajectories passes the area west of the volcano (e.g., 0–10∘ E, 0–10∘ S), where the volcanic plume was detected.

Enhanced sulfate aerosol mass concentrations were observed above 3 km of altitude over the Amazon Basin during the ACRIDICON-CHUVA campaign, which were substantially greater than the concentrations measured at lower altitudes for some flights. A map including all airborne observations on the different flights can be found in the Supplement (Fig. S5). However, given the multiple sulfate aerosol sources, sulfate itself cannot be used as a tracer of volcanic emissions. In order to distinguish the volcanogenic sulfate from other aerosol sources like BB, which is important during this time of the year, we examined the Msulfate vertical profiles together with their sulfate-to-OA mass ratio (Msulfate/Morg). A list of the ACRIDICON-CHUVA flights and Msulfate vertical profiles are presented as Supplement Table S1 and Fig. S6, respectively.

From the different airborne observations, the elevated plumes observed on flights AC14 (21 September 2014) and AC17 (27 September 2014) exhibited median sulfate-to-OA mass ratios larger than 1.0, indicating strong volcanogenic influence. The Msulfate vertical profile measured on 21 September 2014 (AC14) is presented in Fig. 6. The observations show a sulfate-enhanced layer between 4 and 5 km of height, which was also observed during flight AC17 on 27 September 2014. The average Msulfate observed during flight AC14 was 1.1 ± 0.5 µg m-3 between 3 and 6 km of height. This sulfate-enhanced layer exhibits the highest sulfate-to-OA ratio and can be distinguished from lower layers below 3 km of height, which are characteristically rich in OA due to BB and biogenic emissions. Usually, BB aerosol particles have been shown to have higher OA mass concentrations than other aerosol particles (McNaughton et al., 2011; Saturno et al., 2017). The concentrations of rBC in the region between 4 and 5 km were also very low (9 ng m-3) compared to the values below 3 km of altitude (270 ng m-3; Holanda et al., 2018), ruling out a combustion source for the sulfate. In light of all the observations (i.e., enhanced sulfate layer above 4 km of height, high sulfate-to-OA ratio, very low rBC mass concentration) the most plausible explanation for the source of this sulfate plume is the LRT of volcanogenic aerosols.

Backward trajectories were calculated from several points along the flight paths. Figure 7 shows backward trajectories started at nine points along the AC14 (21 September 2014) flight track for which sulfate-to-OA ratios larger (colored lines) or smaller than 1 (gray lines) were detected; the flight and aerosol data measured at each point are presented in Table 1. Figure 7 clearly shows that the colored trajectories, with one exception, initialized within the sulfate plume originate from central Africa, whereas the gray trajectories, started outside of the sulfate plume, appear to originate from South America or from more southern regions over or across the Atlantic Ocean. The air mass trajectory analysis supports our hypothesis that the AC14 observations were likely the result of probing the volcanic plume emitted on 7 September, the first one detected by OMI (see Fig. S3). For flight AC17 a similar pattern is observed, with three out of four (colored) trajectories started within the sulfate plume originating from central Africa and half of the other trajectories clearly coming from regions more to the south (see Fig. S7 and Table S2).

Figures 4, 5, and 7 show that the trajectories agree well but not perfectly with the ground-based, airborne, and satellite measurements, which is mainly caused by the uncertainty of such long trajectories. Nevertheless, the fact that forward and backward trajectories calculated from various starting points and times agree on the sulfate source is a further indication that the sulfate plumes observed at and near ATTO originate from the Nyamuragira volcano. Combined with the westward transport pattern derived from SO2 satellite data and the lack of an alternative strong sulfate source, this makes quite a convincing case.

The arrival of African volcanic emissions over the Amazon rain forest is reflected in the aerosol physical and chemical properties measured at the ATTO site. The most evident effect was the significant increase in Msulfate. The 90th percentile of Msulfate measured at the ATTO site during the dry season 2014 was used as a threshold to define the volcanic influence event (Nya2014) as the time when this threshold was exceeded. By this criterion, the Nya2014 event spanned from 21 September 2014 at 02:00 UTC to 1 October 2014 at 01:00 UTC. Figure 8 shows different aerosol parameters measured before, during, and after the Nya2014 event. The NCN, shown in Fig. 8a, did not vary greatly from the values typical of the season (Pöhlker et al., 2016). The average NCN during the dry season in 2014 was (1.3 ± 0.6) × 103 particles cm-3, whereas during the Nya2014 event, there were three peaks lasting for a few hours with particle number concentrations higher than 3000 particles cm-3 on 27, 29, and 30 September 2014, likely related to BB plumes (details in M. L. Pöhlker et al., 2018).

On the other hand, during the Nya2014 period, Msulfate averaged 1.7 ± 0.6 µg m-3, which was significantly above the dry season 2014 average of 0.7 ± 0.3 µg m-3; see Fig. 8b. The highest Msulfate value observed at the ATTO site was 3.6 µg m-3 (hourly mean) on 26 September 2014. For comparison, during the SAMBBA campaign in southern Amazonia, Msulfate barely exceeded 1.0 µg m-3 despite organics nearly reaching 100 µg m-3, MBCe of 5 µg m-3, and NCN above 25 × 103 particles cm-3 during the peak of biomass burning (Brito et al., 2014). A long-term measurement study also conducted in southern Amazonia reported Msulfate of 1.1 ± 0.7 µg m-3 during the dry season, with a maximum of 3.3 µg m-3 for aerosol particles with diameters smaller than 2 µm (Artaxo et al., 2002). It is important to note that the observations by Artaxo et al. (2002) were made under strong BB influence with average elemental carbon (EC) mass concentrations, MEC, of 3.8 ± 4.2 µg m-3, with a maximum of 25 µg m-3. In contrast, the BCe measurements at the ATTO site during the Nya2014 event had an average of 0.4 ± 0.1 µg m-3, with a maximum of 0.8 µg m-3, indicating that the BB influence was relatively weak during the period of interest, with some short (a few hours) spikes due to the influence of nearby fire events; see Fig. 8b. At a sampling site impacted by Manaus emissions, the submicron Msulfate was about 0.2 µg m-3 during the wet season, rarely exceeding 0.6 µg m-3 (de Sá et al., 2017). Therefore, even considering a range of pollution sources, our measurements at ATTO during the Nya2014 event are the highest submicron sulfate concentration ever reported in the Amazon Basin; see Martin et al. (2010) for a summary of wet and dry season aerosol observations. For comparison, the ACRIDICON-CHUVA airborne measurements are also included in Fig. 8b. The Msulfate measured on flight AC14 was significantly enhanced between 3 and 6 km of altitude, reaching a median of 1.0 µg m-3 and a 75th percentile of 1.6 µg m-3. Previous aircraft measurements during the SAMBBA campaign reported a Msulfate flight average of 0.48 µg m-3 (Allan et al., 2014).

The increased Msulfate period was accompanied by an enhanced sulfate-to-OA mass ratio according to the ATTO observations (Fig. 8c). The Nya2014 sulfate-to-OA average over about 10 days was 0.38 ± 0.09, significantly higher than the dry season average of 0.18 ± 0.15. During some BB pulses, decreased sulfate-to-OA ratios were observed, but the whole Nya2014 period was exceptionally high compared to typical dry season conditions. The sulfate-to-OA values measured at ground level were usually lower than the airborne values observed between 3 and 6 km of height because the OA sources (BB and biogenic emissions) are located at ground level and the LRT sulfate that arrives at higher altitudes is diluted upon vertical mixing. The possibility of FF burning was ruled out as an important sulfur source during the event discussed here because of the particularly high dry aerosol ω0 measured during the event (0.89 ± 0.04), as can be observed in the color-coded data in Fig. 8c. Usually, FF emissions rich in BC present characteristically low ω0 (0.2–0.3; Bond and Bergstrom, 2006) and its addition would have lowered the value of ω0. Instead, an increase in ω0 was observed to values higher than 0.90 during the period of maximum Msulfate (26–27 September 2014).

The effects of the volcanic sulfur plume on aerosol hygroscopicity were explored by analyzing the κ values measured at different supersaturations. Higher κ values were measured for the accumulation-mode aerosol (particles with a diameter greater than 100 nm; Fig. 8d; note the color-coded particle activation diameter, Da). During the Nya2014 event, the κ values increased significantly, especially when the maximum Msulfate was observed. For example, the average κ for a supersaturation of 0.10 % (Da= 167–179 nm, 25th and 75th percentile, respectively) was 0.26 ± 0.04 during the Nya2014 event, with a maximum of 0.36. Similarly, the particle hygroscopicity was also increased at Jungfraujoch (Switzerland) under the influence of the Eyjafjallajökull volcanic emissions in 2010 (Bukowiecki et al., 2011). In that case the Aitken-mode particles (10 to 50 nm diameter) were more strongly affected than the accumulation-mode particles because of the shorter aging period of the sulfate particles. The Nya2014 κ average was slightly higher than the 2014 dry season average of 0.21 ± 0.03 for 0.10 % supersaturation (excluding the volcanic event) and significantly higher than a strong BB event average of 0.18 ± 0.01 for 0.10 % supersaturation, whose high OA content (sulfate-to-OA ratio < 0.04) caused a significant drop in κ (M. L. Pöhlker et al., 2018).