Comprehensive characterizations of aerosols and trace gases were carried out onboard the US Department of Energy Gulfstream-1 (G-1) research aircraft and at several surface sites, during the Observations and Modelling of the Green Ocean Amazon (GoAmazon2014/5) experiment near Manaus, Amazonas, Brazil (3.1° S, 60.0° W), in the central Amazon basin9. The pollution plume of Manaus, a city of about two million people, generally follows the north-easterly trade winds. The present analysis focuses on measurements under clean conditions during the wet season in March 2014 onboard G-1 and from March to May 2014 at the T0a surface site (Amazon Tall Tower Observatory, ATTO10) located 150 km northeast of Manaus (Extended Data Fig. 1). Air masses arriving at T0a during the wet season are typically brought by the north-easterly trade winds and travel across at least 1,000 km of undeveloped tropical rainforest, and are therefore generally clean (Extended Data Fig. 2). Some analysis was also conducted using datasets from the T3 site (70 km west of Manaus). Air masses arriving at T3 were clean or polluted depending on the direction of the Manaus pollution plume.
We observed a strong vertical gradient in the particle size spectrum above central Amazonia under clean conditions. Figure 1a shows the size spectra measured upwind of Manaus at five altitudes between 650 m and 5,800 m (above mean sea level) from 13:18 to 14:42 (all times are in UTC) on 7 March 2014. Altitudes of 3,200 m, 4,500 m and 5800 m were within the free troposphere, as indicated by the equivalent potential temperature (θe). At 5,800 m, the spectrum was dominated by an Aitken mode of 40 nm diameter. These particles can arise from new particle formation in the outflow of deep convective systems11, 12, 13, in which the particle surface area is low owing to wet scavenging of existing particles and the ambient temperature is low, facilitating the formation of particles of a few nanometres from gas-phase precursors through several potential mechanisms12, 14, 15, 16, 17, 18. Condensational and coagulational growth of the nucleated particles between the points of formation and airborne observations can lead to the observed size spectrum13.
Figure 1: Measurements made onboard the G-1 aircraft at five different altitudes upwind of Manaus on 7 March 2014.
a, Particle size spectra normalized to standard temperature and pressure (273.15 K and 101.325 kPa; STP). b, Particle number concentration (N) normalized to STP and volume-average particle diameter (Dp,v) derived from the size spectrum. c, Water vapour mixing ratio and equivalent potential temperature (θe). The measurements are averaged over horizontal legs that were outside cloud or precipitation regions to avoid potential sampling artefacts. The quantities are averages of 324 to 360 1-s measurements that correspond to spatial scales ranging from 34 km to 40 km.
The elevated concentration of Aitken-mode particles at high altitude led to a strong vertical gradient in the particle number concentration between the free troposphere and the atmospheric boundary layer. As altitude decreased, the spectrum shifted towards larger particle sizes, as evidenced by the increase in the volume-average particle diameter (Fig. 1b). At the lowest altitude (650 m), which was inside the boundary layer, we found a pronounced increase in the concentration of particles with diameters greater than 100 nm—at this size the particles are sufficiently large to serve as cloud condensation nuclei (CCN) under typical atmospheric conditions. The G-1 sampled particles in the free troposphere during four flights, all of which showed trends similar to that represented in Fig. 1 (Extended Data Fig. 3). The size spectra observed in both the lower free troposphere and the boundary layer are very similar to those reported earlier over the coastal region of the Amazon rainforest in Suriname19, suggesting that the observed vertical gradient is generally representative over Amazonia.
In many other environments, new particle formation in the boundary layer is one of the main processes producing small particles that subsequently grow and become CCN. In Amazonia, however, new particle formation has almost never been observed in the boundary layer under natural conditions3, 20, 21. Given the high number concentrations of Aitken particles at high altitudes, transport from the free troposphere can serve as an important source of small particles into the boundary layer. Once in the boundary layer, the atmospheric oxidation of biogenic volatile organic compounds emitted by the forest contributes the mass required for particle growth to CCN sizes6, 7, 8, 22, 23, 24. The general mechanism of particle transport from the free troposphere to the boundary layer has been previously reported over marine regions (that is, ‘blue oceans’)25, 26, 27. For blue oceans, vertical transport is believed to take place by continuous but slow entrainment through a strong inversion at the top of the boundary layer on a timescale of days25, 26, 27. However, such steady and slow entrainment could not explain the rapid increases in Aitken-mode particle concentrations observed in an earlier study19 and reported here. We found that vertical transport over the rainforest of continental Amazonia (the so-called Green Ocean28) often occurs through rapid, sporadic downdrafts associated with precipitation, instead of the slow and continuous entrainment reported over marine regions.
An example of vertical transport by downdrafts and its effect on boundary layer aerosols is shown in Fig. 2 using data collected at T0a (ATTO) and onboard G-1 on 19 March 2014. On this day, a storm travelled southwest and reached T0a at 04:00. The trailing stratiform region of this same convective system was sampled by G-1 near Manaus (19 March, 14:00–17:00). Strong convective precipitation at T0a peaked at 04:10. The water vapour mixing ratio measured at the surface started to decrease around the onset of precipitation, as explained by the downward transport of drier air from the free troposphere (as shown in Extended Data Fig. 3i) to the surface. The mixing of free tropospheric air is also evident from the sharp decrease in θe. (The use of θeas a tracer for the downward transport of free tropospheric air is detailed in Methods.) Values of vertical velocity at T3 (for example, Extended Data Fig. 4) further confirm the transport of free tropospheric air into the boundary layer during precipitation, both by strong convective downdrafts and by weaker downward motion in the trailing stratiform region.
Figure 2: The contribution of vertical transport of free tropospheric small particles to the particle concentration in the atmospheric boundary layer at T0a during a precipitation event on 19 March 2014.
a, Water vapour mixing ratio, precipitation rate and equivalent potential temperature (θe). b, The total particle number concentration (N), the concentration of small particles with diameters (Dp) less than 50 nm (N<50) and the concentration of CCN-sized particles with diameters larger than 100 nm (N>100). c, Particle size spectra (dN/dlogDp) at ground level. d, Particle size spectra at ground level and those measured onboard G-1 on the same day.
At T0a, substantial changes in the particle size spectrum started during the precipitation event, and were coincident with decreases in the water vapour mixing ratio and θe, both indicators of the downward transport of free tropospheric air. Before the precipitation event, the size spectrum was bimodal and dominated by an accumulation mode at 200 nm. Following the event, the size spectrum was dominated by an Aitken mode at 50 nm. Forward trajectory analysis indicates that air masses starting at altitudes ranging from 625 m to 4,850 m above T0a around the precipitation event (that is, 04:00–06:30) later reached similar locations and altitudes to those of G-1 at the time of sampling (Extended Data Fig. 5). Although later in time, the measurements onboard G-1 suggest what the vertical gradient in the particle size spectrum had been when the system passed over T0a earlier. The particle size spectrum measured by G-1 in the free troposphere exhibited an Aitken mode with essentially the same mode diameter as observed at T0a following the precipitation event (Fig. 2d, Extended Data Fig. 3g). Moreover, the size spectrum measured at 625 m altitude within the boundary layer was consistent with that observed at T0a following the event (Fig. 2d).
In the following quantitative analysis, we focus on three particle classes: all particles, small particles with diameters less than 50 nm, and particles with diameters greater than 100 nm that can act as CCN. The respective surface concentrations of these particle classes are denoted by N, N<50 and N>100. There was a strong increase in N<50 following the strong convective precipitation and the stratiform rain, whereas N>100 was substantially lower. The lack of nucleation-mode particles (that is, particles with diameters smaller than 20 nm) and the sudden increase in the Aitken-mode particle concentration immediately following the precipitation indicate that the increase in N<50 cannot be explained by new particle formation within the boundary layer. The decrease in N>100 is attributed to low concentrations in the downward-transported free tropospheric air (Fig. 2d, Extended Data Fig. 3g) as well as efficient removal by precipitation scavenging. The total particle concentration following the precipitation event remained nearly the same as before, indicating that the removal of CCN by precipitation was compensated on a number basis by the increase in small particles from the free troposphere. An elevated N<50 was observed in the majority of cases following major precipitation events during the wet season. Another example of the vertical transport observed at T0a on 4 May is shown in Extended Data Fig. 6.
The generalized role of vertical transport by downdrafts as a source of small particles into the atmospheric boundary layer was examined statistically using three months of measurements at T0a. As shown in Fig. 2b, the vertical transport of dry air leads to a pronounced decrease in θe in the boundary layer. Although θe was also influenced by other processes, such as solar heating, these processes had a strong and recognizable diel pattern (Extended Data Fig. 7a). The quantity was used as a surrogate for the extent of vertical transport of free tropospheric air, where is the diel average at the same time of day as , which represents seasonally detrended θe (Extended Data Fig. 7b). A negative Δθe value corresponds to a decrease in θedue to mixing with free tropospheric air and is typically associated with precipitation (Extended Data Fig. 8). Positive Δθe values can arise from high instability (that is, convective available potential energy) as well as strong mass and humidity convergence before rainfall events29.
For measurements at T0a under near-natural conditions from 1 March to 31 May 2014, the data points were grouped into ten Δθe bins. The statistics for N<50, N>100 and N for each bin were examined as a function of Δθe (Fig. 3). As Δθe decreased, N<50 increased from 35 cm−3 to 125 cm−3 (Fig. 3b), indicating that mixing with free tropospheric air increased the concentration of small particles. In contrast, N>100 decreased with decreasing Δθe (Fig. 3c). N varied little over a wide range of Δθe values (Fig. 3d), suggesting statistically that the removal of CCN is largely compensated by the influx of small particles transported from the free troposphere during precipitation. Given that precipitation scavenging is the main mechanism for the removal of CCN from the Amazon boundary layer, the similar N values over the wide range of Δθe (that is, before and after precipitation) suggest that, in terms of particle number, the vertical transport of free tropospheric air is an important source that replenishes particles in the boundary layer and thereby maintains the climate-relevant CCN population, at least under natural conditions. In contrast, a similar analysis of measurements at the T3 site shows a much lower particle number concentration following precipitation events (that is, at lower Δθe values) because, under polluted conditions, anthropogenic emissions are the main source replenishing boundary layer particles following precipitation (Extended Data Fig. 9). Complementary analysis of the T0a data was also carried out by removing the variations in θe over timescales longer than 20 h, including diel and seasonal variations, and using the filtered θe as a proxy for the extent of vertical transport of free tropospheric air. The statistics of N<50, N>100 and N were examined as a function of the filtered θe values, and the results exhibit essentially the same features as shown in Fig. 3 (Extended Data Fig. 10). The particle number fluxes due to precipitation scavenging and vertical transport are estimated to be 3.9 × 1011 m−2 d−1 and 3.1 × 1011 m−2 d−1, respectively (details in Methods). Although future work is needed for a more accurate quantification, the estimate supports the finding that downward transport can largely compensate the precipitation scavenging of particles on a number basis.
Figure 3: Variations in the particle number concentrations with Δθe.
a, Number of data points for each Δθe bin. b, Statistics of the small particle (Dp < 50 nm) concentration for each Δθe bin. c, Statistics of the CCN-sized particle concentration (Dp > 100 nm). d, Statistics of the total particle number concentration. The box and whisker plots are drawn for the 10th, 25th, 50th, 75th and 90th percentiles. The black circles represent the mean values.
In summary, the observations reported here show high concentrations of small particles in the lower free troposphere over the Amazon rainforest. We find that these particles arise from new particle formation in the outflow regions of deep convective systems, followed by condensational and coagulational growth. Under natural conditions, the small particles in the free troposphere become an important source of particles to the boundary layer through downward vertical transport, which occurs through rapid, sporadic downdrafts during convective precipitation events instead of the slow and continuous entrainment reported over marine regions. In the boundary layer, these particles grow and subsequently maintain the CCN population. An improved understanding of aerosol processes under natural conditions may help to reduce uncertainties in simulated climate change5, 30. The vertical transport becomes a net sink of particles in the boundary layer under polluted conditions—such as those prevailing over most other tropical continental regions—when the boundary layer particle concentration exceeds that in the free troposphere.