Forest plots of three ha each were inventoried in the igapó, the campinarana, the terra firme on ancient river terraces, and the terra firme on the plateau in order to provide a preliminarily description of the floristic composition and turnover as well as the aboveground wood biomass (AGWB). All trees with ≥ 10 cm DBH (diameter at breast height) were numbered, tagged with aluminum plates, and, when possible, identified in the field. Fertile and sterile vouchers were collected for later identification in the INPA herbarium. The AGWB was estimated by a pantropical allometric model (Feldpausch et al., 2012) considering DBH, tree height, and wood specific gravity. We measured tree height with a trigonometric measuring device (Blume-Leiss) and determined wood specific gravity by sampling cores from the tree trunk and calculating the ratio between dry mass (after drying the wood samples at 105 ∘C for 72 h) and fresh volume. Additionally we used data from the Global Wood Density Database DRYAD (Chave et al., 2009) for tree species in the terra firme forests and from Targhetta (2012) for tree species in the campina and igapó forests.

An RGB camera (StarDot NetCam XL 3MP) was installed in June 2013 at the top of the walk-up tower. The wide-angle view with 2048×1536 pixel resolution includes over 250 separable tree crowns within an area of ∼ 4 ha of the forest plateau. The camera aim is steeply oblique and toward the west so that the sun is behind the camera when images are recorded from mid-morning until noon. Illumination artifacts are minimized by selecting images with homogeneous overcast sky and a fixed narrow range of incident radiance, and by post-selection radiometric normalization. Leaf phenology change is most evident in individual crowns, so timelines of the green chromatic coordinate, gc, (Richardson et al., 2007) were made for each crown. A steep and sustained increase in the gc of a crown can only be caused by the flushing of a new leaf cohort. The number of crowns reaching a flush-caused peak of gc in their individual timelines were counted each month.

Soil sampling was performed on the ancient terraces (old floodplains) and terra firme plateaus at the ATTO site according to a standard protocol (Quesada et al., 2010). Five samples up to 2 m in depth were taken in each forest plot and one 2 m depth pit was dug close to each plot. We used the World Reference Base for soil resources to classify soil types (IUSS (International Union of Soil Science) Working Group WRB, 2006). Soil exchangeable cations were determined with the silver thiourea method (Pleysier and Juo, 1980), and soil carbon and nitrogen were analyzed using an automated elemental analyzer (Pella, 1990; Nelson and Sommers, 1996). Particle size was analyzed using the pipette method (Gee and Bauder, 1986). Soil physical properties were calculated for each plot using the “Quesada Index” (Quesada et al., 2010). This index is based on measurements of effective soil depth, soil structure, topography, and anoxia. To investigate the current soil weathering levels, a chemically based weathering index, total reserve bases (∑RB), was calculated. ∑RB is based on total soil cation concentration and is considered to give a chemical estimation of weatherable minerals (Quesada et al., 2010).

Aerosols are sampled above the canopy at 60 m height, without size cut-off, and transported in a laminar flow through a 2.5 cm diameter stainless steel tube into an air-conditioned container (aerosol lab at mast, see Sect. 2.4). The sample humidity is kept below 40 % using silica diffusion driers. Since January 2015, the aerosol sample air is being dried using a fully automatic silica diffusion dryer, developed by the Institute for Tropospheric Research, Leipzig, Germany (Tuch et al., 2009). Aerosol size distributions at 60 m are currently measured from 10 nm up to 10 µm using three instruments: a Scanning Mobility Particle Sizer (SMPS, TSI model 3080, St. Paul, MN, USA; size range: 10–430 nm), an Ultra-High Sensitivity Aerosol Spectrometer (UHSAS, DMT, Boulder, CO, USA; size range: 60–1000 nm), and an Optical Particle Sizer (OPS, TSI model 3330; size range: 0.3–10 µm). The SMPS provides an electromobility size distribution, whereas the UHSAS and OPS measure aerosol light scattering and derive the size distributions from the particle scattering intensity (Cai et al., 2008). In addition to these continuous above-canopy measurements, aerosol size distributions are measured with a Wide Range Aerosol Spectrometer (WRAS, Grimm Aerosol Technik, Ainring, Germany; size range: 6 nm–32 µm) from a separate inlet line below the canopy at 3 m height. The WRAS provides electromobility size distributions in the size range of 6–350 nm and uses particle light scattering for the size range above 300 nm. Details of the instrumentation setup are given in Table 2.

For measuring aerosol light scattering, we use a three-wavelength integrating nephelometer (until February 2014: TSI model 3563, wavelengths 450, 550, and 700 nm; after February 2014: Ecotech Aurora 3000, wavelengths 450, 525, and 635 nm) (Anderson et al., 1996; Anderson and Ogren, 1998). Calibration is carried out using CO2 as the high span gas and filtered air as the low span gas. The zero signals are measured once every 12 h using filtered ambient air. For the 300 s averages applied here, the detection limits, defined as a signal to noise ratio of 2, for scattering coefficients are 0.45, 0.17, and 0.26 Mm-1 for 450, 550/525, and 700/635 nm, respectively. Since sub-micrometer particles predominate in the particle number size distribution at our remote continental site, the sub-micron corrections given in Table 4 of Anderson and Ogren (1998) were used for the truncation corrections. Bond et al. (2009) suggested that this correction is accurate to within 2 % for a wide range of atmospheric particles, but that the error could be as high as 5 % for highly absorbing particles.

A Multi-Angle Absorption Photometer (MAAP, Model 5012, Thermo Electron Group, USA, λ=670 nm) and a 7-wavelength Aethalometer (until January 2015 model AE-31, since then model AE-33) (Magee Scientific Company, Berkeley, CA, USA, λ=370, 470, 520, 590, 660, 880, and 950 nm) are used for measuring the light absorption by particles. The MAAP and aethalometer have been deployed at ATTO since March 2012. In the MAAP instrument, the optical absorption coefficient of aerosol collected on a filter is determined by radiative transfer calculations, which include multiple scattering effects and absorption enhancement due to reflections from the filter. A mass absorption efficiency (αabs) of 6.6 m2 g-1 was used to convert the MAAP absorption data to equivalent BC (BCe). For the Aethalometer, an empirical correction method described by Rizzo et al. (2011) was used to correct the data for the scattering artifact.

Refractory black carbon (rBC) is measured by a four-channel Single Particle Soot Photometer (SP2). The instrument is calibrated every 6 months using monodisperse fullerene aerosol particles for rBC calibration, and polystyrene latex (PSL) spheres for scattering calibration. The instrument is sensitive to rBC in the size range between 70 and 280 nm. A recent instrumental upgrade provides a broader rBC dynamic range (70–480 nm).

Regular quality checks are performed with all aerosol sizing instruments and CPCs, including flow checks, zero tests, and intercomparisons with ambient aerosol and monodisperse PSL cells. Exemplary plots are already included in the manuscript (Fig. 26). The MAAP and aethalometer are subject to frequent intercomparisons with the other optical instruments. For example, two aethalometers and the MAAP were operated side-by-side during an intensive campaign in November/December 2014. The BCe concentrations from the individual instruments agreed well. The SP2 instrument was carefully intercalibrated with another SP2 during the GoAmazon-2014 campaign.

Fluorescent biological aerosol particles (FBAPs) are measured with the Wideband Integrated Bioaerosol Spectrometer (WIBS-4A, DMT). The WIBS utilizes light-induced fluorescence technology to detect biological materials in real-time based on the presence of fluorophores in the ambient particles (Kaye et al., 2005). A 2×2 excitation (280 and 370 nm) – emission (310–400 and 420–650 nm) matrix is recorded along with the particle optical size and shape factor. The FBAP concentrations reported in this study correspond to the FL3 channel (excitation at 370 nm and emission in the waveband of 420–650 nm) of the WIBS instrument (Healy et al., 2014).

The submicron non-refractory aerosol composition at a height of 60 m is measured using an Aerosol Chemical Speciation Monitor (ACSM, Aerodyne, USA) as described by Ng et al. (2011). The ACSM samples aerosol particles in the 75–650 nm size range. The non-refractory fraction flash vaporizes on a hot surface (600 ∘C), the evaporated gas phase compounds are ionized by 70 eV electron impact, and their spectra determined using a quadrupole mass spectrometer. The chemical speciation is determined via deconvolution of the mass spectra according to Allan et al. (2004). Mass concentrations of particulate organics, sulfate, nitrate, ammonium, and chloride are obtained with detection limits < 0.2 µg m-3 for 30 min of signal averaging. Mass calibration of the system is performed using size-selected ammonium nitrate and ammonium sulfate aerosol following the procedure described by Ng et al. (2011). A collection efficiency (CE) of 1.0 is applied (similar to Chen et al., 2015), yielding good agreement with other instruments.

PM2.5 sampling was carried out from 7 March to 21 April 2012 on Nuclepore^® polycarbonate filters at 80 m on the walk-up tower using a Harvard Impactor; samples were collected over 48 hour periods. They were analyzed by energy-dispersive X-ray fluorescence (EDXRF) (MiniPal 4, PANalytical) at 1 mA and 9 kV for low-Z (Na to Cl) elements, and 0.3 mA, 30 kV, and internal Al filter for the other elements. Soluble species were determined by ion chromatography (Dionex, ICS-5000) using conductivity detection for cations and anions and UV-VIS for soluble transition metals. For cation separation, a capillary column CS12A was used, for anions, an AS19 column, and for transition metals, a CS5A column (calibrated to quantify traces of Fe2+ and Fe3+).

Size-resolved cloud condensation nuclei (CCN) measurements are performed using a continuous-flow streamwise thermal gradient CCN counter (CCNC; model CCN-100, DMT, Boulder, CO, USA), a differential mobility analyzer (DMA, Grimm Aerosol Technik, Ainring, Germany) and a condensation particle counter (CPC model 5412, Grimm Aerosol Technik). By changing the temperature gradient, the supersaturation of the CCNC is set to values between 0.1 and 1.1 %. The completion of a full measurement cycle comprising CCN efficiency spectra at 10 different supersaturation levels takes ∼ 4 h. The CCNC is calibrated frequently as part of the maintenance routines with size selected monodisperse ammonium sulfate particles (Rose et al., 2008; Gunthe et al., 2009).

Aerosol samples for scanning electron microscopy with electron probe micro-analysis (EPMA) were collected on top of the 80 m tower in April 2012. For the collection of size-segregated samples for single particle analysis, we used a Battelle impactor with aerodynamic diameter cut-offs at 4, 2, 1 and 0.5 µm. The particles were collected on TEM (transmission electron microscope) grids covered with a thin carbon film (15–25 nm). Aerosol samples for x-ray microspectroscopy were collected using a single stage impactor, operated at a flow rate of 1-1.5 L min-1 and a corresponding 50 % size cut-off of about 500 nm. Particles below this nominal cut-off are not deposited quantitatively; however, a certain fraction is still collected via diffusive deposition. Aerosol particles were collected onto silicon nitride substrates (membrane width 500 µm, membrane thickness 100 nm, Silson Ltd., Northampton, UK) for short sampling periods (∼ 20 min), which ensures a thin particle coverage on the substrate appropriate for single particle analysis.

Scanning transmission X-ray microscopy with near-edge X-ray absorption fine structure analysis (STXM-NEXAFS) measurements were made at the Advanced Light Source (ALS, Berkeley, CA, USA) and the Berliner Elektronenspeicherring-Gesellschaft für Synchrotronstrahlung (BESSY II, Helmholtz-Zentrum Berlin für Materialien und Energie (HZB), Germany). A detailed description of the instrumentation and techniques can be found elsewhere (Kilcoyne et al., 2003; Follath et al., 2010; Pöhlker et al., 2012, 2014). Scanning Electron Microscopy with Energy Dispersive X-ray spectroscopy (SEM/EDX) analysis was carried out using a Jeol JSM-6390 SEM equipped with an Oxford Link SATW ultrathin window EDX detector. For EPMA, quantitative and qualitative calculations of the particle composition were performed using iterative Monte Carlo simulations and hierarchical cluster analysis (Ro et al., 2003) to obtain average relative concentrations for each different cluster of similar particle types.

Filter sampling for secondary organic aerosol (SOA) analysis was performed on the walk-up tower at a height of 42 m above ground level. Fine aerosol (PM2.5) was sampled at a flow rate of 2.3 m3 h-1 on TFE-coated borosilicate glass fiber filters (Pallflex, T60A20, Pall Life Science, USA). The sampling times were 6, 12, or 24 h. After sampling the filters were stored at 255 K until extraction.

The extraction of the filters was performed with acetonitrile (≥ 99.9 %; Sigma Aldrich) in a sonication bath at room temperature. The filter extracts were evaporated with a gentle nitrogen flow at room temperature in an evaporation unit (Reacti Vap 1; Fisher Scientific), and the residue was re-dissolved in 100 µL HPLC grade water (Milli-Q water system, Millipore, Bedford, USA) / acetonitrile (≥ 99.9 %; Sigma Aldrich) mixture (8:2).

The separation and analysis was performed with an UHPLC (ultrahigh performance liquid chromatography) system (Dionex UltiMate 3000) coupled to a Q Exactive electrospray ionization Orbitrap mass spectrometer (Thermo Scientific). A Hypersil Gold column (50 mm × 2.1 mm, 1.9 µm particle size, 175 Å pore size; Thermo Scientific) was used. The eluents were HPLC grade water (Milli-Q water system, Millipore, Bedford, USA) with 0.01 % formic acid and 2 % acetonitrile (eluent A) and acetonitrile with 2 % HPLC grade water (eluent B). The flow rate of the mobile phase was 0.5 mL min-1. The column was held at a constant temperature of 298 K in the column oven. The MS was operated with an auxiliary gas flow rate of 15 (instrument specific arbitrary units, AU), a sheath gas flow rate of 30 AU, a capillary temperature of 623 K, and a spray voltage of 3000 V. The MS was operated in the negative ion mode, the resolution was 70 000, and the measured mass range was m/z 80–350.

In total, 7293 trees ≥ 10 cm DBH were recorded in the 12 1-ha inventoried plots, which included 60 families, 206 genera, and 417 species. Tree species richness was highest in the terra firme forest on the plateau, followed by the terra firme forest on the fluvial terrace, the campinarana, and the seasonally flooded igapó (Table 3). Floristic similarity (Bray–Curtis index) within plots of the same forest types ranged from 45–65 %, but was highly variable between different forest types (2–54 %). Accordingly, the species turnover across the investigated forest types was high, especially when seasonally inundated forest plots were compared to their non-flooded counterparts (Fig. 4). AGWB varied considerably between the studied forest ecosystems as a result of varying tree heights, DBH, and basal area (Table 3). Carbon stocks in the AGWB increased from 74 ± 12 Mg ha-1 in the igapó forest to 79 ± 26 Mg ha-1 in the campina/campinarana, and 101 ± 13 Mg ha1on the ancient fluvial terrace, reaching maximum values of 170 ± 13 Mg ha-1 in the terra firme forests. Tree species richness correlated significantly with carbon stocks in AGWB (n=12; r2=0.61; p<0.01).

The floristic data indicate that the rain forests at the ATTO site combine high alpha diversity with high beta diversity at a small geographic scale, where tree species segregate mainly due to contrasting local edaphic conditions (e.g., Tuomisto et al., 2003; ter Steege et al., 2013; Wittmann et al., 2013). Biomass and carbon stocks vary considerably between habitats, and show low values on flooded and nutrient-poor soils and high values on well-drained upland soils, as previously reported elsewhere for other Amazonian regions (e.g., Chave et al., 2005; Malhi et al., 2006; Schöngart et al., 2010).

We are investigating the potential of cryptogamic covers to serve as a source of bioaerosol particles and chemical compounds. Cryptogamic covers comprise photoautotrophic communities of cyanobacteria, algae, lichens, and bryophytes in varying proportions, which may also host fungi, other bacteria, and archaea (Elbert et al., 2012). A common feature of all these organism groups is their poikilohydric nature, meaning that their moisture status follows the external water conditions. Thus the organisms dry out under dry conditions, being reactivated again upon rain, fog, or condensation.

Since September 2014, we have been conducting long-term measurements to monitor the activity patterns of cryptogamic covers at four different canopy heights at 10 min intervals, during which we measure temperature and water content within and light intensities directly on top of bio-crusts growing on the trunk of a tree. First analyses of the microclimate data indicate that microorganisms in the upper stem region of the trees are activated by fog or dewfall in the early morning hours, often coinciding with an aerosol particle burst in the accumulation mode. Particle measurements conducted on isolated organisms also show a significant release of accumulation mode particles by wet and thus active organisms, e.g., fungi belonging to the phylum of Basidiomycota. Thus, we have the first clear indications that cryptogamic covers may play a key role in the enigmatic bioaerosol occurrence frequently observed at the ATTO site.

A single annual leaf flush was seen in most upper canopy crowns, concentrated in the five driest months (July to November) (Fig. 5). Consequently, mature leaves with high light-use efficiency will be most abundant in the late dry season and early wet season. Massive leaf renewal in the dry season on the ATTO plateau may drive seasonality of photosynthesis and of photosynthetic capacity at the landscape scale, as has been indicated at the Santarém and the LBA km34 eddy flux tower sites in the central Amazon (Doughty and Goulden, 2008; Restrepo-Coupe et al., 2013).

The lack of a near-infrared (NIR) band in our camera precludes the direct measurement of leaf amount, but the RGB band space discriminates crown phenostages whose relative NIR reflectances are known. Gradual leaf attrition over the wet season, when leaf replacement is low, followed by early dry season pre-flush abscission and the emergence of young unexpanded leaves, should all lead to a lower landscape-scale amount of fully expanded leaves around June or July. Completion of leaf flushing in most crowns by the late dry season should lead to a maximum amount of fully expanded leaves in the late dry and early wet seasons. This is consistent with the seasonal pattern of central Amazon leaf amount detected with the Enhanced Vegetation Index from the MODIS orbital sensor (e.g., Huete et al., 2006) and counters recent critiques of detectability of seasonal change in Amazon forest greenness (Galvão et al., 2011; Morton et al., 2014).

Soils in the terra firme plateaus were classified as ferralsols, which are ancient, highly weathered, and well-drained soils frequently occurring in geologically ancient surfaces (Chauvel et al., 1987). Soils at the fluvial terraces were classified as alisol, which show a more recent pedogenetic status when compared to the highly weathered Ferralsols at the plateaus. Due to their lower weathering degree, soils from the terrace have a greater capacity to supply nutrients, with higher total P and higher total reserve bases. The soil carbon stocks varied from 129 ± 7 Mg ha-1 on the terrace to 164 ± 7 Mg ha-1 on the plateau, indicating that belowground C stocks are of similar magnitude to the aboveground carbon stocks in the forest (Table 4). Differences of belowground carbon stocks between terrace and plateau are mainly associated with a higher clay content of the plateau soils.

Soil physical constraints are more frequent on the terraces, which show higher bulk density values (Fig. 6) and therefore increased soil compaction. Some of these terrace soils also show signs of anoxia (mottling) in deeper layers. Such impeditive conditions may have an influence on forest structure (Quesada et al., 2012; Emilio et al., 2014) and dynamics (Cintra et al., 2013), thereby possibly restricting tree height or even tree individual biomass storage (Martins et al., 2015).

The wind roses for the dry season (15 June–30 November) and the wet season (1 December–14 June) (based on half-hourly averages of wind speed and direction measured at 81 m a.g.l. for the period from 18 October 2012 to 23 July 2014; Fig. 7) indicate the dominance of easterly trade wind flows at the measurement site. A slight shift of the major wind direction towards ENE is observed during the wet season, whereas flows are mainly from the east during the dry season. This seasonality can be explained by the inter-annual north–south migration of the Intertropical Convergence Zone (ITCZ), which also governs the amount of rainfall (see Poveda et al., 2006). The wind roses show a slight diurnal variation with small contributions from the north, west and south during nighttime, when the nocturnal boundary layer is decoupled, in both seasons. In contrast, during daytime the wind blows nearly all the time from the east (dry season) and northeast (wet season), with much higher wind speeds. Maximal wind speeds observed at the site are about 9 m s-1. The influence of river and/or lake breeze systems caused by the Rio Uatumã (∼ 12 km distance) is of minor importance and an effect from Lake Balbina (∼ 50 km distance) or other thermally driven mesoscale circulations could not be detected. This shows that the sampled air masses mainly have their origin within the fetch of the green ocean extending several hundred kilometers to the east of the site.

As is typical for the central Amazon Basin, the mean air temperature does not show strong variations at seasonal timescales due to the high incident solar radiation throughout the year (Nobre et al., 2009). Climatologically in the Manaus region, the highest temperatures are observed during the dry season, with a September monthly mean of 27.5 ∘C, whereas the lowest temperatures prevail in the rainy season, with a monthly mean of 25.9 ∘C in March.

Vertical profiles of temperature show clear diurnal cycles driven by radiative heating of the canopy during the day and cooling of the canopy and the forest floor during the night (Fig. 8). Therefore, both temperature minima and maxima are observed at the canopy top during both seasons. A second temperature minimum during night can be observed at the forest floor during the dry and wet season. During the day warm air from above the canopy is transported into the forest. Minimum temperatures at the canopy top are around 22.5 ∘C during both seasons, whereas daytime maxima are around 28 ∘C during the wet season and may reach slightly above 30 ∘C in the dry season.

Rainfall in the Manaus region shows a pronounced seasonal variation, reaching the highest amounts in March (335.4 mm) and the lowest amounts in August (47.3 mm), for an average annual total of 2307.4 mm at the INMET station in Manaus for the standard reference period 1961 to 1990 (www.inmet.gov.br). Precipitation at the ATTO site follows this seasonal cycle with maximum values around March and minimum values in August and September (Fig. 9). The interannual variability appears to be high at all times of the year, but especially in the transition to the rainy season, a fact that has also been evident in the data from the years 1981 to 2010 at the Manaus station (Fernandes, 2014). Therefore, the large deviations from the regional mean during October to January and also in April, when the ATTO values from the years 2012–2014 differ substantially from the long term mean of Manaus, are likely the result of interannual variability.

Overall, however, the precipitation patterns at the ATTO site are in good agreement with its position in the central Amazon, where the months between February and May are the wettest ones. In this period, the ITCZ reaches its southernmost position and acts as a strong driver of convective cloud formation at the equatorial trough. Due to the interaction of trade winds and sea breeze at the northeast Brazilian coastline, the ITCZ also takes part in the formation of instability lines that enter the continent and regenerate during their westerly propagation (Greco et al., 1990). In this way, they account for substantial amounts of precipitation. After this period, the ITCZ shifts to the Northern Hemisphere, accompanying the movement of the zenith position of the sun. This leads to less precipitation at the ATTO site, with the driest months being between July and September, when precipitation is formed mostly by local convection. In the following months, the amount of precipitation increases again, which coincides with the formation of a cloud band in a NW/SE direction that is linked to convection in the Amazon due to the South Atlantic Convergence Zone (SACZ) (Figueroa and Nobre, 1990; Rocha et al., 2009; Santos and Buchmann, 2010).

The radiation balance at ATTO as well as the albedo presents a clear difference between the wet and the dry seasons. Some episodes when the incident solar radiation exceeds the top-of-atmosphere radiation have been observed for the ATTO data. They were more frequent during the wet season, probably due to the effect of cloud gap modulation that intensifies the radiation received at the surface by reflection and scattering.

The measurement of turbulent fluxes over tall forest canopies very often implies that these measurements are made in the so-called roughness sublayer (RSL). It is usually assumed that the RSL extends to 2 or 3 times the height of the roughness obstacles, h0 (Williams et al., 2007). The roughness sublayer is considered to be a part of the surface sublayer of the atmospheric boundary layer, but it is too close to the roughness elements for Monin–Obukhov similarity theory (MOST) to hold. Some progress in the parameterization of the RSL has been made in terms of applying correction factors to the traditional similarity functions of the surface layer (see for example, Mölder et al., 1999, and references therein). However, the universality of such procedures remains unknown.

In this section, we briefly show strong evidence that a simple adjustment factor that depends on the factor z/z∗ (where z is the height of measurement and z∗ is the height of the RSL), as employed by Mölder et al. (1999), is not able to collapse the “variance method” dimensionless variables ϕw(ζ)≡σwu∗ and ϕa(ζ)≡σaa∗, where σw is the standard deviation of the vertical velocity, u∗ is the friction velocity, σa is the standard deviation of a scalar, and a∗ is its turbulent scale (see Eqs. 3 and 4 below). In Eqs. (1) and (2), ζ is the Obukhov length with a zero-plane displacement height calculated as d0=2h0/3, h0=40 m.

We analyzed measurements collected during April 2012 at the 39.5 m level, which is right at the height of the tree tops, in terms of the turbulent scales u′w′‾≡-u∗2 and ∣w′a′‾∣≡u∗a∗.

We only analyzed measurements under unstable conditions, and considered only cases where the sensible and latent heat fluxes are both positive (directed upwards) and the CO2 flux is negative (directed downwards). In Eq. (4), the absolute value is used so that a∗ is always positive. The scalar a represents virtual temperature θv (measured by the sonic anemometer), specific humidity q, and CO2 mixing ratio c.

The analysis is made in terms of the dimensionless standard deviation functions, ϕw(ζ) and ϕa(ζ), defined above. The overall results for vertical velocity, virtual temperature, and CO2 concentration are shown in Fig. 10. The solid lines in the figure give representative functions found in the literature for the surface layer well above the roughness sublayer (see, for example, Dias et al., 2009).

Similar figures were drawn for specific times of day, namely 07:00–09:00, 09:00–11:00, 11:00–13:00, 13:00–15:00 and 15:00–17:00 LT, in an attempt to identify periods of the day when better agreement (or even a systematic departure, for example by a constant vertical shift) with the surface-layer curves could be identified. Temperature and humidity are somewhat better behaved in this case, but not CO2, for reasons that are not clear. Because no conclusive explanation can be found, we do not show these analyses here.

Finally, we tried to apply some concepts recently developed by Cancelli et al. (2012) to relate the applicability of MOST to the strength of the surface forcing. Cancelli et al. (2012) found that the applicability of MOST can be well predicted by their “surface flux number”, Sfa=∣w′a′‾∣(z-d0)νaΔa‾, where νa is the molecular diffusivity of scalar a in the air, and Δa‾ is the gradient of its mean concentration between the surface and the measurement height.

In our case, there is no easy way to obtain Δa‾, so instead we use Sfa=∣w′a′‾∣(z-d0)νaσa.

As a measure of the applicability of MOST, we use the absolute value of the difference between the observed value of ϕa(ζ) and its reference value for the surface layer, as used by Dias et al. (2009), and shown by the solid lines in Fig. 10. The results are shown in Fig. 11. A relatively stronger forcing is clearly related to a behavior that is closer to that expected by MOST for both temperature and humidity, but not for CO2. This suggests that CO2 presents even greater challenges for our proper understanding of its turbulent transport in the roughness sublayer over the Amazon Forest.

Ultimately, the lack of conformity to MOST found in these investigations (a fact that has been generally observed in the roughness sublayer over other forests) implies that scalar fluxes over the Amazon forest derived from standard models, which use MOST, are bound to have larger errors here than over lower vegetation, such as grass or crops. We can expect this to affect any chemical species, and therefore the implications for ATTO are quite wide-ranging. On the other hand, once the 325 m tall tower is instrumented and operational, a much better picture will emerge on the extent of the roughness sublayer and the best strategies to model scalar fluxes over the forest.

During daytime, intense turbulent activity provides an effective and vigorous coupling between the canopy layer and the atmosphere above it. As a consequence, vertical profiles of chemical species do not commonly show abrupt variations induced by episodes of intense vertical flux divergence. Accordingly, scalar fluxes between the canopy and the atmosphere are relatively well-behaved during daytime so that their inference from the vertical profiles of mean quantities can be achieved using established similarity relationships. At night, on the other hand, the reduced turbulence intensity often causes the canopy to decouple from the air above it (Fitzjarrald and Moore, 1990; Betts et al., 2009; van Gorsel et al., 2011; Oliveira et al., 2013). In these circumstances, vertical fluxes converge to shallow layers in which the scalars may accumulate intensely over short time periods. In the absence of convective turbulence, which is the main factor for daytime transport, other physical processes become relevant in the stable boundary layer (SBL), such as drainage flow (Sun et al., 2004; Xu et al., 2015), vertical divergence of radiation (Drüe and Heinemann, 2007; Hoch et al., 2007), global intermittency (Mahrt, 1999), atmosphere-surface interactions (Steeneveld et al., 2008), and gravity waves (Nappo, 1991; Brown and Wood, 2003; Zeri and Sa, 2011).

In this section, we discuss the role of intermittent turbulent events of variable intensity and periodicity, which provide episodic connection between the canopy and the atmosphere and can induce oscillatory behavior in the nocturnal boundary layer (Van de Wiel et al., 2002). They are characterized by brief episodes of turbulence with intervening periods of relatively weak or unmeasurably small fluctuations (Mahrt, 1999). In some cases, such events may comprise almost the entirety of the scalar fluxes during a given night. The effects of gravity waves are discussed in the next section.

Nocturnal decoupling occurs rather frequently at the ATTO site, usually punctuated by intermittent mixing episodes, in agreement with previous studies made over the Amazon forest (Fitzjarrald and Moore, 1990; Ramos et al., 2004). During a typical decoupled, intermittent night, the horizontal wind components are weak in magnitude and highly variable temporally, often switching signs in an unpredictable manner (Fig. 12). As a consequence, it is common that winds from all possible directions occur in such a night. The example from the ATTO site indicates that despite such a large variability, both horizontal wind components are generally in phase above the canopy, from the 42 m to the 80 m level. Vertical velocity at the 42 m level is highly intermittent, with various turbulent events of variable intensity scattered throughout the night. While being less turbulent, the 80 m level is also less intermittent, presenting a more continuous behavior. The relevance of the intermittent events to characterize canopy-atmosphere exchange becomes clear when one looks at the fluxes of the scalars, such as CO2 (Fig. 12, bottom panel). During this night, the majority of the exchange just above the canopy (42 m) happened during two specific events, at around 02:00 and 03:30 LT.

A proper understanding of nocturnal vertical profiles and fluxes of scalars above any forest canopy depends, therefore, on explaining the atmospheric controls on intermittent turbulence at canopy level. In the Amazon forest, this necessity is enhanced, as there are indications that turbulence is more intermittent here, possibly as a consequence of flow instabilities generated by the wind profile at canopy level (Ramos et al., 2004). This is corroborated by early observations at the ATTO site, which indicated decoupling and intermittency occurring during more than half of the nights.

It is not yet clear what triggers these intermittent events. In general, previous studies indicate that the more intense events are generated above the nocturnal boundary layer, propagating from above (Sun et al., 2002, 2004). On the other hand, less intense events that occur in the decoupled state have been characterized as natural modes of turbulence variability generated near the surface (Costa et al., 2011). At ATTO, the occurrence of the highest intensity at 42 m indicates that intermittency is generated at the canopy level. Is it possible, then, to identify the mechanisms that trigger their occurrence?

Some evidence can be gathered from a spectral decomposition of the turbulent flow at the different observation levels. Although the horizontal velocities in Fig. 12 are highly in phase between 42 and 80 m, it is clear from this plot that the wind speed is generally higher at 80 m, while there are more turbulent fluctuations at 42 m. When these signals are decomposed in terms of their time scale to provide a turbulent kinetic energy (TKE) spectrum (Acevedo et al., 2014), the more intense turbulence at 42 m appears as a peak at time scales just greater than 10 s (Fig. 13). At longer time scales, on the other hand, there is a sharp energy increase at 80 m, making this the most energetic level for scales larger than 100 s. This low-frequency flow at 80 m is characterized by the large wind direction variability apparent in Fig. 12. These are non-turbulent flow patterns that have been recently classified as “submeso” (Mahrt, 2009). Submeso flow has low intensity, with large and apparently unpredictable temporal variability. It is usually present in the atmospheric boundary layer, becoming dominant in conditions when the turbulent scales are highly reduced, such as in the decoupled nocturnal boundary layer.

Evidence from ATTO indicates that it is possible to associate the intermittent events at canopy level with the mean wind shear above the canopy. In Fig. 14, it is evident that the two intense events at 42 m, around 21:30 and 02:00 LT, are triggered by episodes of intense wind shear between 42 and 80 m. In conditions where the 80 m wind field is dominated by submeso processes, such as in the examples in Figs. 12 and 14, it is this portion of the flow that determines the occurrence of intense wind shear episodes. Furthermore, it is clear from these examples that flow patterns at levels as high as 80 m exert important controls on the exchange of scalars at canopy level. Questions such as the height variation of submeso flow have yet to be answered. Tall tower observations, such as those planned to be carried on at ATTO, are very important to provide the data for this kind of analysis and to deepen the understanding of exchange processes between the canopy and the atmosphere during the calm nights that are common in the Amazon forest.

Gravity waves (GWs) may occur in the forest boundary layer during relatively calm nights. Depending on the magnitude of the turbulent drag, they influence the exchange processes that take place in the stable boundary layer of the atmosphere (Steeneveld et al., 2009). Internal gravity waves can be generated by several forcing mechanisms, including sudden changes of surface roughness, topography, convection, terrain undulations, etc. (Nappo, 2002). These features can reallocate energy and momentum and are significant in determining atmospheric vertical structure and the coupling of mesoscale to microscale phenomena (Steeneveld et al., 2008, 2009). Chimonas and Nappo (1989) showed that under typical conditions of the planetary boundary layer, GWs can interact with the mean flow resulting in turbulence at unexpected altitudes.

Fast response data of vertical wind velocity, w, and temperature, T, measured in the nocturnal boundary layer (NBL) at the ATTO site were analyzed to detect the occurrence of GWs, and to identify under which situations they would be generated by terrain undulations, using the methodology proposed by Steeneveld et al. (2009). One of the goals of this study is to investigate the structure of turbulence associated with the conditions under which GWs would be forced by ground undulations (class I) in contrast to those under which GWs would be expected to be forced by other mechanisms (class II). To reach this goal, the methodology of Steeneveld et al. (2009), based on Chimonas and Nappo (1989), has been used to define whether a specific measurement belongs to class I or class II, based on the condition: LS2=N2/U2-U′′/U>k2, where k is the wave number associated with the ground undulations, L is the Scorer parameter, U is the mean wind speed, and U′′ is second derivative of the wind speed in relation to the height, z, computed as U′′=∂2U∂2U∂z2∂z2. N is the Brunt–Väisälä frequency, defined as follows: N=gΔzθ/θ, where g is the gravity acceleration and Δzθ/θ is the dimensionless gradient of the virtual potential temperature.

Two kinds of data were used: topographic and meteorological. A digital topographic image of the region surrounding the experimental site (Fig. 15a) was used to analyze the features of surface undulations and their scales of occurrence, as well as the space-scale analysis by complex Morlet wavelet transforms (Farge, 1992; Thomas and Foken, 2005). Local geomorphometric variables were derived from the Shuttle Radar Topographic Mission (SRTM) data (Valeriano, 2008). These data were refined to 1 arcsecond (∼ 30 m) from the original spatial resolution of 3 arcsecond (∼ 90 m) and are available on the site www.dsr.inpe.br/topodata/dados.php.

Time series of the vertical wind velocity and of the fast response temperature data provided by a sonic anemometer and thermometer were used to detect GW events at a height of 81 m above the ground. The sampling rate of the measured turbulence data was 10 Hz. Wind speed and temperature vertical profiles were provided by cup anemometer and thermometer measurements, respectively, with a sampling rate of 60 Hz for both, making it possible to compute the Brunt–Väisälä frequency, the vertical gradients of wind velocity, and the Scorer parameter for GW classification (Steeneveld et al., 2009). Data from five nights were analyzed, consisting of 120 files of 30 min each between Julian days 42 and 46 of the year 2012, representing the first observational data available from the ATTO site. The analyses were carried out for the time between 18:00 and 06:00 LT for each night with available data (Fig. 15).

The black points on the arrows in Fig. 15b represent the GW events that were induced by the topography of the terrain, whereas the gray points represent GW events that were not generated by terrain orography. The results show that a considerable fraction of the analyzed situations represent GWs induced by terrain undulations. This finding is very important for the environmental studies that are being carried out at the ATTO site, as it indicates that some mixing characteristics of the nocturnal boundary layer depend on the characteristics of terrain undulations and therefore change with the wind direction.

Coherent structures (CSs) are a ubiquitous phenomenon in the turbulent atmospheric flow, particularly over forests (Hussain, 1986). They occur in the roughness sub-layer immediately above the plant canopy, where the CSs of the scalar signals show a “ramp-like” shape associated with the two-phase movement of sweep and ejection of the flow interacting with the canopy. Coherent structures play an important role in biosphere–atmosphere exchange processes (Gao and Li, 1993; Serafimovich et al., 2011). There is some consensus that CSs are associated with turbulent flows, although there is no full agreement on the percentage of the turbulent fluxes associated with them (Lu and Fitzjarrald, 1994; Thomas and Foken, 2007; Foken et al., 2012). There has been much research on the dominant scale of occurrence of CSs (Collineau and Brunet, 1993; Thomas and Foken, 2005) and the physical mechanisms responsible for their generation (Paw U et al., 1992; Raupach et al., 1996; McNaughton and Brunet, 2002; Campanharo et al., 2008; Dias Júnior et al., 2013). Considerable research has also been devoted to the detection of CSs (Collineau and Brunet, 1993; Krusche and Oliveira, 2004) and the dissimilarity between CSs associated with the transport of momentum and scalars (Li and Bou-Zeid, 2011). However, many aspects of their occurrence are still poorly known, particularly: (i) their vertical variability (Lohou et al., 2000); (ii) the manifestations of their interaction with gravity waves (Sorbjan and Czerwinska, 2013); (iii) the influence of surface heterogeneity on their features; (iv) aspects of their numerical simulation (Patton, 1997; Bou-Zeid et al., 2004; Dupont and Brunet, 2009; Wan and Porte-Agel, 2011), particularly in the nocturnal boundary layer (Durden et al., 2013; Zilitinkevich et al., 2013), and (v) implications of the existence of CSs for the chemistry of the atmosphere (Steiner et al., 2011; Foken et al., 2012).

A study on the structure of atmospheric turbulence was performed at the ATTO site under daytime conditions, with the aim of contributing to the detection of CSs and developing a better understanding of their vertical and temporal variability over a very uneven terrain covered by primary forest in central Amazonia. Wind, temperature, and humidity data were obtained using sonic anemometers and gas analyzers, installed at 42 and 81 m above ground, as specified in the methods section. The scales of coherent structures were determined following the methodology proposed by Thomas and Foken (2005). Figure 16 shows the average duration of CSs for horizontal and vertical wind velocities (u,w), temperature (T), and humidity (q). For the data at 81 m height, the CS of u and w exhibit temporal scales around 46 s and 29 s, respectively. For the two scalars, T and q, the time scales of the CS are about 44 and 55 s, respectively. For the height of 42 m the coherent structure time scales of u, w, T, and q were approximately equal to 33, 26, 30, and 31 s, respectively.

The results revealed that the CS time scale of the vertical wind velocity is often smaller than the scales of the horizontal velocity and the scalar properties, for both levels. This can be explained by the fact that the scalar spectra exhibit greater similarity to the spectra of the horizontal velocity than to the vertical velocity for low frequencies. Another interesting feature is that the temporal scale of the CSs for both the wind velocity and scalars are considerably shorter for the data measured at 81 m compared with those at 42 m, i.e., the region immediately above the forest canopy appears to be under the influence of a high-pass filter that removes the lower frequency oscillations of the turbulent signals (Krusche and Oliveira, 2004; Thomas and Foken, 2005).

The characteristics of the nocturnal boundary layer (NBL) at the ATTO site near the Uatumã River were analyzed for the wet and the dry seasons, based on two methodologies: (i) the thermodynamic classes of the NBL proposed by Cava et al. (2004) and (ii) the turbulence regimes proposed by Sun et al. (2012).

Cava et al.'s (2004) classification of nocturnal time series is based on the existence of a dominant pattern in scalar data, such as CO2 concentration, temperature, or specific humidity. It also takes into account the variability of nocturnal net radiation (Rn), measured at a sufficiently high sampling rate, which allows cloud detection (with passage of clouds being identified by rapid Rn changes greater than 10 W m-2). Classes (I), (II), (III), are defined by atmospheric conditions free of the influence of clouds, which can disturb the stable boundary layer above the forest. The classes are defined as followed by Cava et al. (2004): (I) the occurrence of coherent structures in the form of “ramps” in scalar time series; (II) the presence of sinusoidal signals (“ripples”) that simultaneously occur in the time series of scalars above the canopy and that are typical for gravity waves; (III) the existence of turbulence fine structure (i.e., according to Cava et al. (2004), “periods that lack any geometric structure or periodicity in the time series data”). The last two categories, (IV) and (V), of Cava et al.'s classification refer to the simultaneous occurrence of clouds and organized movements with variations in Rn>10 W m-2. They are (IV) cases where the net radiation induces organized movements, and (V) those where the change in net radiation is not correlated with changes in organized movements.

The search of parameters to characterize the turbulence regimes of the nocturnal boundary layer is based on Sun et al. (2012). The three turbulence regimes in the NBL are defined as follows: Regime 1 shows weak turbulence generated by local shear instability and modulated by the vertical gradient of potential temperature. Regime 2 shows strong turbulence and wind speed exceeding a threshold value (Vλ), above which turbulence increases systematically with increasing wind speed. This describes the turbulence generated by bulk shear instability, defined as the mean wind speed divided by the measuring height. In Regime 3, the turbulence occurs at wind speeds lower than Vλ, but is associated with occasional bursts of top-down turbulence. In Regimes 1 and 2, the scale of turbulent velocity (VTKE) is related to the mean wind velocity, V. The turbulent velocity, VTKE, is defined as follows: VTKE=1/2σu2+σv2+σw21/2, where u,v, and w are the components of the zonal, meridional and vertical winds, respectively, and σ represents the standard deviation of each variable.

We analyzed 53 data files from the wet season and 79 data files collected during the dry season at the ATTO site. Our results show that the prevailing conditions in the NBL are represented by Cava's classes I, II, and V for both wet and dry seasons (Table 5). Furthermore, during the wet season the classes I and V show their highest percentage of occurrence associated with turbulence Regime 3. Class IV is more frequent when turbulence Regime 1 prevails. For the dry season we observe that turbulence classes I, IV and V occur most frequently in situations associated with Regime 1 (Table 6).

Figure 17a–c shows the diurnal cycles of the vertical distributions of CO2, CH4, and CO at the ATTO site. CO2 and CO show a nighttime accumulation in the sub-canopy space and a corresponding steepening of the vertical concentration gradient, which is greatly reduced during daytime due to the enhanced vertical mixing throughout the canopy. In addition, CO2 exhibits a clear minimum during daytime at mid-canopy level induced by photosynthesis. Interestingly, the build-up of the nighttime maximum of CH4 proceeds from above the canopy (Fig. 17b). The origin of this behavior, which seems to be linked to multiple processes, is under investigation. During daytime, CO2, CH4, and CO still exhibit a small vertical gradient below the canopy, indicating a local source near the ground.

Additional evidence for local surface sources are sporadic concurrent increases of CH4 and CO, predominantly at the lowest measurement level. Examples are shown in Fig. 18. The origin of this local CH4 and CO source is not known. A remote source (e.g., from the large water reservoir behind the Balbina Dam, 60 km northwest of ATTO) seems unlikely, as such a signal would be vertically diluted before reaching the ATTO site. A combustion source also appears unlikely, as the observed CH4/ CO ratios are several orders of magnitude higher than the values typical of combustion emissions.

Apart from these CH4 and CO peaks, we occasionally observe, mostly during nighttime, short CH4 peaks of up to more than 100 ppb amplitude. These peaks last a few hours, they do not always concur with increases in CO concentrations, and often coincide with “bursts” of particles with a diameter of a few tens of nanometers.

Figure 19 shows the statistics of monthly daytime (defined as between 13:00–16:00 LT, or 17:00–20:00 UT) 30 min measurements of CO2 from three levels (4, 38, and 79 m). The values at the 4 m level are consistently higher than at the upper levels, while the 38 m level consistently shows lower values during daytime than the top level (79 m). This indicates that photosynthesis is active throughout the year. The record is still too short to reveal a clear seasonality. Nevertheless, it appears that CO2 from June to August is about 5 ppm above the values during the months from December to February.

Statistics of monthly daytime 30 min measurements of CH4 and CO are shown in Fig. 20 (from the 79 m level only). Because of a large data gap due to a malfunctioning of the analyzer, a seasonal cycle is not discernible in the present CH4 record. CO does show a seasonal cycle at ATTO with concentrations higher by about 50 ppb during the dry months with a significant fraction of air coming from the southeast (see Fig. 3), where vegetation fires are very active at this time.

Monthly daytime concentrations of CO2, CH4, and CO are compared in Fig. 21 with measurements upstream of ATTO: Cape Verde (green symbols) reflecting the subtropics of the Northern Hemisphere, and Ascension Island (brown symbols) representing conditions in the Southern Hemisphere. At least during the period of July to December, CO2 concentrations clearly reach lower values than at both upstream locations, reflecting the regional carbon sink in the Amazon domain. In contrast, CH4 levels at ATTO lie almost on Northern Hemisphere levels throughout the year, even when the ITCZ is north of ATTO in austral winter, and the site is in the atmospheric Southern Hemisphere with its lower background CH4 concentrations. This suggests the presence of regional CH4 emissions in the airshed of ATTO. The CO concentrations at ATTO during the wet season are close to those at Cape Verde, reflecting the absence of significant combustion sources in the South American part of the fetch during this season. In contrast, dry-season CO mixing ratios at ATTO are about 80 ppb higher than those at Ascension Island, reflecting biomass burning emissions in the southeastern Amazon (Andreae et al., 2012).

The first successful vertical gradients of biogenic VOCs and total OH reactivity were measured in November 2012 at the walk-up tower using the gradient system as described in Sect. 3.4. Diurnal fluctuations of isoprene are apparent at all heights (Fig. 22). Under daylight conditions, isoprene mixing ratios were always highest at the 24 m level, reaching up to 19.9 ± 2.0 ppb (average ± standard deviation) and indicating a source at the canopy top. During nighttime, the light-driven emissions of isoprene cease and the in-canopy mixing ratio fell to 1.1 ± 0.5 ppb, which was lower than observed above the forest at 79 m (2.3 ± 0.3 ppb). Measurements in the canopy (24 m) vary by a factor of 10 from day to night, while measurements close to the ground (0.05 m) vary only by a factor of two. This clearly demonstrates a canopy emission of isoprene, with a peak around noon, when light and temperature are at their maximum. Isoprene mixing ratios at the ground level were always the lowest, indicating a potential sink at the soil/litter level or relatively slow downward mixing. A detailed discussion of measurements of isoprene and other biogenic VOC at ATTO was published recently (Yáñez-Serrano et al., 2015).

In November 2012, the high levels of isoprene measured above the canopy contributed significantly (on average about 85 %) to the total OH reactivity. From Fig. 23, it can be seen that median isoprene mixing ratios of between 0.5 ppb at 06:00 LT and 8 ppb in the late afternoon above the canopy give an OH reactivity of about 1–20 s-1. The gap between the two curves is the fraction of total OH reactivity that is not due to isoprene. For most of the time this gap is small and within the uncertainty of the measurements. On two occasions, however, the total OH reactivity was significantly higher than the isoprene contribution, these being in the early morning (09:00 LT) coincident with a drop in light levels, and in the afternoon just after sunset (17:00 LT). For all other times in the course of the day, isoprene was the major sink for OH above the canopy. Overall, a distinct diel variability in total OH reactivity can be observed, similar to that of its major contributor, isoprene. The median lifetime of OH radicals during the dry-to-wet transition season above the forest canopy at 80 m varied from about 50 ms by day to 100 ms at night. Ongoing measurements will determine the seasonal variability in total OH reactivity and the relative contribution of isoprene.

The O3 mixing ratios (Fig. 24) show typical diurnal cycles for both seasons, with values increasing from the morning to the afternoon and subsequently decreasing due to deposition and chemical reactions. The afternoon O3 maxima at the uppermost height (79 m) are about a factor of 1.4 higher during the dry season than during the wet season, averaging about ∼ 11 and ∼ 8 ppb, respectively. As found in previous studies, its deposition to surfaces causes O3 to exhibit pronounced vertical gradients (Fig. 24), which makes a direct intercomparison to measurements at other sites difficult. However, the mixing ratios above the canopy from different studies in the Amazonian rain forest during the wet season are within a narrow range of 7 to 12 ppb, and the value measured at ATTO falls near the lower end of this range (Kirchhoff et al., 1990; Andreae et al., 2002; Rummel et al., 2007; Artaxo et al., 2013). A budget study by Jacob and Wofsy (1990) revealed that downward transport of O3 mainly controlled the losses near the surface, with only a minor contribution from photochemical formation above the canopy. This may explain the similar mixing ratios in the different studies. Furthermore, only small O3 differences were measured between 38 m (just above the canopy) and the top of the tower at 79 m during the wet season.

A different picture is observed during the dry season, with much higher O3 mixing ratios at more polluted sites (∼ 40 ppb in Rondônia: Kirchhoff et al., 1989; Andreae et al., 2002; Rummel et al., 2007; Artaxo et al., 2013), which can be related to biomass burning emissions causing photochemical O3 formation (Crutzen and Andreae, 1990). A site comparable to the ATTO site is the ZF2 site, located about 60 km northwest of Manaus, which has been used extensively in the past (Artaxo et al., 2013), but which is occasionally affected by the Manaus urban plume (Kuhn et al., 2010; Trebs et al., 2012). At the ZF2 site, mean maximum O3 mixing ratios measured at 39 m from 2009–2012 (Artaxo et al., 2013) match exactly those measured at the ATTO site for the wet season, but are about a factor of 1.5 higher during the dry season. This may be attributed to the more pristine character of the ATTO site, but could also be related to the different measurement periods or different biogenic emissions at the sites. In order to distinguish these different influences, high-quality long-term measurements are required, which are now being generated within the ATTO project.

During the wet season, the amplitude of the mean diurnal cycle at 79 m is only about 2 ppb, whereas it is 3–4 ppb during the dry season. The highest amplitudes are observed within the canopy and the understory with up to 5 ppb (24 m) in the dry season. These variations can be attributed to downward mixing of O3, which is “stored” within the canopy (so called storage flux, see Rummel et al., 2007). It is subsequently depleted by chemical reactions, mostly with soil biogenic NO, and deposition after the forest canopy becomes decoupled from the atmosphere above at nighttime. During the wet season, the largest decrease in O3 mixing ratios occurs at the canopy top. This might be attributed to a lower canopy resistance to O3 deposition due to enhanced stomatal aperture during the wet season as proposed by Rummel et al. (2007) and will be the subject of future work. Further investigations will also focus on the interactions between turbulence (supply of O3) and trace gases that react with O3, especially nitric oxide (NO).

The aerosol optical properties measured at the ATTO site are shown as a time series in Fig. 25 and summarized in Table 7. The averages were calculated for the dry season, August–October, and the wet season, February–May (2012–2013 for the absorption measurements and 2013 for the scattering measurements). The transition periods between these two seasons are not included in the summary, in order to show the contrast between the cleanest and “more polluted” periods. The scattering coefficients are similar to those reported by Rizzo et al. (2013) from measurements performed at the ZF2 site (60 km N of Manaus). The regional transport of biomass burning emissions and fossil-fuel derived pollution is the main source of particles during the dry season. Its influence is pronounced, as can be seen by comparing the scattering and absorption coefficients from both seasons, which average about 3–6 times higher during the dry than during the wet season. During the wet season, ATTO is meteorologically located in the NH and the scattering and absorption coefficients reach their minimum values; however, episodes of long-range transport of aerosols from the Atlantic Ocean and Africa still lead to episodically elevated values.

The contrast between the wet and dry seasons can be attributed to a combination of higher removal rates by wet deposition during the wet season and the dominant influence from biomass burning and fossil fuel emissions during the dry season, which are the main sources of submicron particles at that time. The scattering Ångström exponent (ås) averages 1.25 during the wet season, lower than the 1.62 obtained for the dry season. This behavior results from the high relative proportion of larger particles (mostly primary biogenic particles, but also dust and seaspray) during the wet season, because in contrast to the large seasonal variability of the submicron particles, the supermicron fraction shows less intense seasonal changes.

The seasonality of the absorption coefficient, σa, is comparable to that of the scattering coefficient. The regional transport of biomass burning emissions, most important between August and October, produces a rise in the σa values, reaching an average of 3.46 Mm-1 during this period. In contrast, during the wet season, σa is very low, around 0.52 Mm-1 on average.

The absorption Ångström exponent (åa) is often used to estimate the composition of light absorbing aerosols. An åa ∼ 1 indicates the aerosol is in the Rayleigh regime, and the absorption is dominated by soot-like carbon and is therefore wavelength independent (Moosmüller et al., 2011). Higher åa values indicate the presence of additional light absorbing material, like brown carbon (BrC) (Andreae and Gelencsér, 2006). This kind of yellowish or brown organic material, abundant in biomass burning aerosols, usually has an åa ∼ 2.0 or greater (Bond et al., 1999). Our measurements show only relatively minor seasonal differences in åa, with somewhat higher values during the wet season (1.53) than in the dry season (1.40), suggesting that soot carbon is the most important contributor to aerosol light absorption throughout the year. The contribution of the different light absorbing components of the aerosol to the total observed aerosol absorption is currently being investigated.

We conducted the first long-term refractory black carbon (rBC) measurements by an SP2 instrument at a remote Amazonian site. The mass absorption cross section (αa) has been calculated by applying an orthogonal regression to the MAAP absorption coefficient measurements at 637 nm vs. the rBC mass concentrations measured by the SP2. The average αa obtained for the 2013–2014 wet season measurements was 13.5 m2 g-1, which is much higher than the 4.7 m2 g-1 reported previously for another Amazonian forest site (Gilardoni et al., 2011), who used a thermal–optical method to determine the apparent elemental carbon content. The high apparent αa could be partly explained by the fact that the SP2 size dynamic range was 70–280 nm and thus the technique did not account for rBC particles larger than 280 nm. However, it is also likely related to an enhancement of light absorption by coatings on the rBC particles, or to the presence of additional light-absorbing substances besides rBC (Bueno et al., 2011; McMeeking et al., 2014). Single-particle studies (Sect. 4.3.7) on aerosols from the ATTO site consistently show thick coatings on the soot carbon particles. Our preliminary results indicate that the constant αa (6.6 m2 g-1), implemented by the MAAP in order to retrieve the BC mass concentration, is not representative of the true optical properties of Amazonian aerosol particles.

Continuous measurements of aerosol particle size and concentration have been conducted at the ATTO site since March 2012. Over the last years, the extent of the sizing instrumentation has been increased stepwise to provide uninterrupted and redundant aerosol size and concentration time series. Figure 26 shows one of the frequent instrument intercomparisons, including four different instruments that are based on optical and electromobility sizing. It confirms the overall consistency and comparability of the different sizing techniques. Integrated particle number concentrations agree within 15 % with measurements of total particle number concentrations by a CPC. The sample air for this intercomparison was collected through the main aerosol inlet at 60 m height, which is also used for instruments measuring aerosol scattering, absorptivity, hygroscopicity, and chemical composition.

At the ATTO site, the atmospheric aerosol burden shows remarkable differences in terms of size distribution and concentration depending on the seasons. Figure 27 displays the average particle number and volume size distributions for typical wet (6–13 May 2014) and dry season (13–20 September 2014) conditions, covering an aerosol size range from 10 nm to 10 µm. The wet season is characterized by clean air masses from NE directions (Fig. 3), which result in a near-pristine atmospheric state at the ATTO site. Total particle concentrations typically range from 100–400 cm-3 and aerosol size spectra reveal the characteristic “wet season shape”. A representative example is shown in Fig. 27. The size spectrum is characterized by a three-modal shape with pronounced Aitken and accumulation modes as well as a noticeable coarse mode. Aitken (maximum at ∼ 70 nm) and accumulation (maximum at ∼ 150 nm) modes are separated by the so called Hoppel minimum (at ∼ 110 nm), which is thought to be caused by cloud processing (e.g., Zhou et al., 2002; Rissler et al., 2004; Artaxo et al., 2013).

The near-pristine conditions that prevail at the ATTO site during the wet season, when the aerosol concentrations are remarkably low and dominated by local and/or regional biogenic sources, are episodically interrupted by long-range transport of sea spray, Saharan dust, and/or African biomass burning and fossil fuel combustion aerosol (e.g., Talbot et al., 1990; Martin et al., 2010a; Martin et al., 2010b; Baars et al., 2011). Figure 28 displays characteristic changes in the wet season size distribution during selected episodes with long-range transport intrusions. Typically, the aerosol abundance in the accumulation and coarse mode size range is substantially increased, and the Hoppel minimum almost completely disappears. The aerosol volume distribution clearly indicates a pronounced enhancement of coarse particles, which increases the integrated particle volume concentration by almost 1 order of magnitude (Fig. 28b).

During the dry season, the dominant wind direction is E to SE (Fig. 3), which brings polluted air from urban sources and deforestation and pasture fires in the southeastern Amazon and the Brazilian Nordeste to the ATTO site. Dry season aerosol number concentrations typically range from 500–2000 cm-3. A characteristic dry season size spectrum is illustrated in Fig. 27, which shows increased particle concentrations across the entire size range. Typically, the accumulation mode (maximum at ∼ 140 nm) shows the highest relative increase and therefore partly “swamps” the Aitken mode (shoulder at ∼ 70 nm).

Besides the Aitken and accumulation modes, which dominate the total aerosol number concentration, a persistent coarse mode is observed at about 3 µm, which accounts for a significant fraction of the total aerosol mass (Fig. 27). The coarse mode peak occurs throughout the year, with higher abundance in the dry season. In the absence of long-range transport, primary biological aerosol particles (PBAP) are assumed to dominate the coarse mode (Pöschl et al., 2010; Huffman et al., 2012).

Autofluorescence-based techniques such as the Wideband Integrated Bioaerosol Sensor, WIBS-4A) are an efficient approach to probe fluorescent biological aerosol particles (FBAP) in online measurements (Kaye et al., 2005; Healy et al., 2014). Figure 29 shows the first measurements of the FBAP number and volume size distributions from the WIBS instrument at the ATTO site. The FBAP size distributions are dominated in number by a narrow peak at 2.7 µ m and in volume by a broad peak from 2 to 5 µm (Fig. 29). For particles larger than 1 µm, the mean integral FBAP number concentration is 0.22 cm-3 (40 % of the concentration of supermicron particles), and the corresponding volume concentration is calculated to be 3.0 µm3 cm-3 (62 %). The ratio of FBAP to total particles (number concentration) shows a clear size dependence, starting from 10 % at 1 µm and rising to a peak value ∼ 70–80 % in the size range of 3–10 µm. These observations are consistent with FBAP measurements made with an alternative instrument (UVAPS) during the AMAZE-08 campaign at the ZF2 rainforest site north of Manaus (Pöschl et al., 2010; Huffman et al., 2012).

CCN size/supersaturation spectra have been measured since 2014 and are being continued. The long-term data set provides unique information on the size dependent hygroscopicity of Amazonian aerosol particles throughout the seasons. The results will complement and extend the results from previous campaigns (e.g., Gunthe et al., 2009; Rose et al., 2011; Levin et al., 2014). The measurements of CCN and other aerosol properties at the ATTO site will also be an important reference for the analysis of the results from the ACRIDICON-CHUVA aircraft campaign, which took place in central Amazonia in September 2014 (Wendisch et al., 2015).

For the continuous determination of aerosol composition, an Aerosol Chemical Speciation Monitor (ACSM) was installed at the ATTO site in February 2014 with the objective of making long-term measurements. The data reported here summarize the annual cycle of aerosol concentrations and composition from May 2014 to April 2015 (Fig. 30).

During the middle of the rainy season (March to May), the aerosol concentrations at ATTO reach their annual minimum and are in relatively good agreement with previous wet season studies, including those conducted at the ZF2 site, ca. 140 km SW of ATTO (Chen et al., 2009; Pöschl et al., 2010; Artaxo et al., 2013). With the onset of the dry season, the shift of air mass origins to the southeast, and the transition of ATTO into the atmospheric Southern Hemisphere (Fig. 3), aerosol concentrations increase sharply and remain high until the end of December, well into the rainy season. Trajectory analyses suggest that burning in Africa may contribute significantly to pollution levels at ATTO during this part of the year. Only when the rainy conditions in the Amazon combine again with dominant air mass origins in the tropical and subtropical North Atlantic and with the waning of biomass burning in West Africa can aerosol concentrations at ATTO drop again to their seasonal lows.

The composition of the aerosol at ATTO shows surprisingly little variation throughout the year in spite of the huge change in total concentrations between seasons. Organic aerosol is always the dominant mass fraction at about 70 %, sulfate comprises about 10–15 %, followed by BCe (5–11 %), ammonium (∼ 5 %), nitrate (∼ 4 %) and chloride. Elevated concentrations of chloride were observed during a few episodes, when this species represented up to 13 % of the total submicron particulate mass, which is consistent with earlier observations of long-range transport of sea salt, going back to the ABLE-2B campaign (Talbot et al., 1990).

The ionic mass balance indicates that the aerosol was approximately acid-base neutral. While sulfate is mostly in the form of ammonium sulfate, there is some indication that part of the nitrate could be present in the form of organic nitrate. This is because the ratio between the fragments NO+ and NO2+ (main nitrate fragments measured by the ACSM at mass-to-charge ratios 30 and 46) is expected to be large (∼ 10) when this ion is in organic forms, and low (2–3) when in inorganic forms, such as ammonium nitrate (Alfarra et al., 2006; Fry et al., 2009). Large values for this ratio were often observed during this period and may indicate the presence of organic nitrate.

The bulk composition of PM2.5 was measured for up to 10 elements by EDXRF analysis on a set of samples obtained in March/April 2012. The analysis showed a high abundance of crustal elements, illustrating one exemplary episode of long-range dust transport from Africa (Fig. 31). Back trajectories indicate that this period was indeed influenced by dust transport from Africa, which is a phenomenon observed annually and particularly pronounced in February, March, and April (Prospero et al., 1981; Swap et al., 1992; Ben-Ami et al., 2012). Local sources of mineral dust aerosol can be excluded, especially during the wet season, because of the wetness of the soils. The prevalence of mineral dust aerosols during the wet season, when airmass trajectories reach from the North African deserts to the Amazon Basin, in combination with observations of transatlantic dust plumes by lidar, is strong evidence for the long-range origin of the observed crustal elements.

To explore the bioavailability of important trace elements, the oxidation state and solubility of iron (Fe) in the PM2.5 aerosols were analyzed. The soluble (and therefore bioavailable) fraction of Fe is an important parameter in the overall biogeochemical cycles, with impact on the phosphorus cycle and biomass production (Liptzin and Silver, 2009). A soluble fraction of only 1.5 % (1.8 ng m-3 Fe(III) of 120 ng m-3 of total Fe) was found, suggesting that aeolian transport of Fe is not likely to make a significant contribution of bioavailable Fe to the ecosystem at ATTO.

The extended measurements of aerosol composition at the ATTO site, now reaching well over a full year, suggest the need for a reassessment of the relative contributions of biogenic and anthropogenic sources even in this very remote region. Black carbon, a unique tracer of combustion, is present in a roughly equal fraction throughout the year. Sulfate, which has a more complex mixture of sources, also contributes a fairly constant fraction. In the rainy season, much of this sulfate could come from biogenic or marine sources (Andreae et al., 1990), but the high concentrations during the August to December period suggest substantial contributions from fossil fuel burning. Periods with aerosol compositions suggesting pristine conditions (low BCe and sulfate, dominant organic matter) occur more as episodes at ATTO than as seasonal characteristics, similar to what is observed at the remote ZOTTO site in Siberia (Chi et al., 2013).

The microspectroscopic analysis of aerosol samples can be seen as a “snapshot” of the aerosol population at a given time. In combination with the long-term aerosol measurements at the ATTO site, single particle characterization provides detailed insights into the highly variable aerosol cycling in the rain forest ecosystem. In the soft X-ray regime, STXM-NEXAFS is a powerful microscopic tool with high spectroscopic sensitivity for the light elements carbon (C), nitrogen (N), and oxygen (O) as well as a variety of other atmospherically relevant elements (e.g., K, Ca, Fe, S, and Na). The technique allows analyzing the microstructure, mixing state, and the chemical composition of individual aerosol particles. As an example, Fig. 32 displays the STXM-NEXAFS analysis of an aerosol sample with substantial anthropogenic pollution, collected at the ATTO site during the dry season. X-ray microspectroscopy reveals a substantial fraction of internally mixed particles with soot cores (strong π-bond signals) and organic coatings of variable thickness. The spectral signature of the organic coating is characteristic for secondary organic material (SOM) (Pöhlker et al., 2014). These observations underline the dominance of aged pyrogenic aerosols at the ATTO site during the dry season. During the rainy season, when biomass burning is absent and undisturbed biosphere-atmosphere interactions prevail in the region, the aerosol population is dominated by biogenic aerosol, such as primary biological aerosol particles (PBAP), biogenic SOA, and biogenic salts (Pöhlker et al., 2012). Figure 33 displays STXM elemental maps of this typical rainy season aerosol population.

As mentioned in the previous section, the biogenic background aerosol in the wet season (i.e., February to April) is episodically superimposed by transatlantic dust and smoke events. Statistical analysis of the electron microscope (EPMA) results by hierarchical clustering reveals the abundance of the various particle types observed at the ATTO tower in this season (Table 8). In order to determine the sources and possible chemical interactions, particles were classified into representative groups according to their chemical composition. They are classified as “mineral” when Al, Si, O, and Ca are dominant, and also contain minor elements like K, Na, Mg, and Fe. Particles are identified as being “organic”, when the concentrations of C and O in the particles are similar and when they also contain some P and S (< 10 wt %). “Biogenic” particles occur in the larger size classes; they have smooth boundaries and always contain C, O, S, N, P, and K. Irregular crystallized particles with Na, Mg, S, O and C are classified as “salt” particles. Soot particles can be distinguished by their morphology, and always contain the elements C and O.

With single particle analysis, important information was obtained concerning the contribution from organic aerosol particles and the agglomeration of various types of particles. The majority of particles in the fine fraction consist of organic matter with traces of S and K. This observation corroborates that small biogenic potassium and sulfur-containing particles from primary emissions can act as seeds for the condensation of organic material (Pöhlker et al., 2012).

Measurements of the organic chemical composition of the aerosol over the Amazon rainforest are rare. Levoglucosan, several mono- di- and polycarboxylic acids, as well as isoprene tracer compounds have been identified in the aerosol phase (Mayol-Bracero et al., 2002; Claeys et al., 2004, 2010; Schkolnik et al., 2005), whereas the contribution of highly reactive compounds, such as monoterpenes and sesquiterpenes, to SOA in this region is still largely unknown.

The concentrations of monoterpene and sesquiterpene oxidation products in ambient aerosol collected in November 2012 at the ATTO research site are shown in Fig. 34. A median concentration of 102 ng m-3 was measured for the sum of terpene oxidation products in the aerosol sampled over the Amazon rain forest. As can be seen in Fig. 34, monoterpene oxidation products accounted for the major part of the terpene oxidation products. Their concentration showed a high variance during November ranging between 23 and 146 ng m-3. The oxidation products derived from the pinene skeleton (α- and β-pinene) were most abundant in the sampled aerosols, followed by limonene oxidation products. Among the pinene derived oxidation products, MBTCA (3-methyl-1,2,3-butanetricarboxylic acid) was observed to be the most abundant individual monoterpene oxidation product, with concentrations of up to 73 ng m-3, followed by pinonic acid with a maximum concentration of 46 ng m-3 (van Eijck, 2013). Interestingly, these concentrations are in the same range as monoterpene oxidation products measured during summertime in boreal forest environments (Vestenius et al., 2014), ecosystems which are known to strongly emit monoterpenes. However, these observations of high monoterpene product concentrations match with the high monoterpene mixing ratios measured at the same site (Yáñez-Serrano et al., 2015).

The products from sesquiterpene oxidation showed much lower concentrations (Fig. 34). On average the sesquiterpene oxidation products reached about 10 % of the monoterpene oxidation product concentration; however, on some days they were as high as 26 % of the total monoterpene oxidation product concentration. Overall, 23 individual oxidation products of sesquiterpenes were identified in the aerosol collected in the Amazonian rainforest. The total concentration of these sesquiterpene oxidation products ranged from 6 to 12 ng m-3. The oxidation products could be assigned to four sesquiterpene precursors: β-caryophyllene, aromadendrene, cedrene, and isolongifolene. Among them, the products from the oxidation of β-caryophyllene were the most abundant (van Eijck, 2013). Very few measurements exist of sesquiterpene oxidation product concentrations and actually none in tropical forests, which complicates a comparison. Measurements in boreal ecosystems (Vestenius et al., 2014) showed mean summertime concentrations of caryophyllinic acid (one of the β-caryophyllene oxidation products) of about 8 ng m-3, which is a high concentration for a single compound compared to the concentrations measured at ATTO, where the measured concentration range of caryophyllinic acid is 0.26–1.38 ng m-3.

In summary, the contribution of monoterpene oxidation products to SOA at ATTO is relatively high and essentially comparable with their contribution to boreal forest SOA, whereas the contribution of sesquiterpene products is much less (about 1/10) than in boreal forest ecosystems.