GRAEGOR (ECN, the Netherlands) is a semiautonomous wet-chemistry instrument capable of online quantification of the concentrations of the water-soluble inorganic trace gases NH3, HCl, HONO, HNO3 and SO2, as well as their associated aerosol counterparts NH4+, Cl-, NO2-, NO3- and SO42-, at hourly resolution at two separate heights . It consists of two sample boxes and a detector box at ground level. For this study, the sample boxes were set at two heights on the 80 m walk-up tower: z1= 42 m and z2=  60 m.

Each sample box consists of a horizontally aligned wet rotating annular denuder (WRD) and a steam jet aerosol collector (SJAC) connected in series. Air is simultaneously drawn through both sample boxes at a rate of 16.7 Lmin-1, which is kept constant through critical orifices located downstream of the SJACs. The inlets of the sample boxes are directly connected to the WRDs via a 0.3 m length high-density polyethylene (HDPE) tubing, which minimises losses of HNO3 and NH3. A HDPE insect gauze is attached to the filters, preventing insects or coarse debris entering the filter. The air streams first pass through the WRDs, which are coated in a continuously replenishing sorption solution of 18.2 MΩ double-deionised (DDI) water. Water-soluble trace gases contained within the laminar air flows diffuse into the liquid sorption solution, which is then fed to the detector box at ground level for analysis. Free of trace gases, the air streams then enter the SJACs and are mixed with water vapour fed from the DDI solution. This precipitates a supersaturation event, such that any particles contained in the air streams rapidly (0.1 s) grow to droplets of 2 µm diameter. The particle-containing droplets are then separated from the air steams by use of a cyclone, and they are fed as liquid samples to the detector box. To prevent biological contamination of the WRDs, the DDI solution includes 0.6 mL of 30 % hydrogen peroxide (H2O2) (9.8 M) per 10 L of DDI.

A series of liquid-pressure regulators were placed in the path of the liquid samples being fed to the detector box in order to prevent damage to it caused by the high hydrostatic pressures in the 42 and 60 m high sample columns. Liquid samples from the SJACs and WRDs are analysed for NH4+ and NH3, respectively, by a flow injection analysis (FIA) unit . A 761 compact ion chromatography (IC) unit (Metrohm, Switzerland), equipped with a Dionex AS12 column, determines the liquid concentrations of HCl / Cl-, HONO / NO2-, HNO3 / NO3- and SO2 / SO42- in the WRD/SJAC liquid streams, respectively, based on the measured anion conductivity of the samples compared to a 50 ppb Br- reference standard added to the sample solution, taking into account the specific conductivities of the various ions compared with Br-. A flow control scheme enables continuous analysis of liquid samples. Air concentrations relative to moist air, reported as mass concentrations at ambient temperature and pressure, are derived from the measured liquid concentrations according to 1ci=cliq⋅Qsample+QBrQsample⋅cIScBr⋅QsampleQair⋅Mwi(air)Mwi(liq), where cliq is the liquid concentration of the species measured; Qsample, QBr, and Qair are the sample, internal Br- standard, and air-mass flow rates, respectively; and cIS and cBr are the expected internal standard concentration and the detected concentration of the internal standard, respectively. The ratio of the molecular weights for air (Mwi(air)) and liquid (Mwi(liq)) are included to account for the mass differences between the measured ions in the liquid sample and the corresponding gas-phase species. For aerosol species, this ratio is equal to 1. GRAEGOR therefore provides a half-hourly-averaged measurement of trace gas and aerosol concentrations for each height and species.

The concentrations of the trace gases and aerosols measured by GRAEGOR are expressed in terms of mass per volume in units of µgm-3 at ambient temperature and pressure. Equivalent ambient molar mixing ratios (ri), with respect to moist air, were calculated using the following formulation: 2ri=RTpMi×ci, where R is the gas constant (8.314 JK-1mol-1), Mi is the molecular weight of the trace gas or aerosol, ci is the concentration in µgm-3 of the trace gas or aerosol, p is the air pressure in pascal (Pa), and T is the temperature in kelvin (K).

Calibration of the FIA unit is autonomous, conducted 24 h after GRAEGOR begins measurement after start up and every 72 h afterwards. The calibration uses three liquid NH4+ sample concentrations of 0, 50 and 500 ppb. For this study, a total of 10 autonomous internal calibrations took place. The IC unit is continuously calibrated by the addition of the 50 ppb Br- internal standard which is added to every liquid IC sample.

Sample box airflows were monitored continuously via the pressure drop across a flow restrictor, calibrated every 5 d using a model 4140 mass flowmeter (TSI, USA) measuring at ambient volumes (Lmin-1). Additional checks of the instrument performance were conducted daily, e.g. visual checks that the WRDs or SJACs were not contaminated.

Due to the short inlet length and absence of any size selection, measurements of aerosol taken by GRAEGOR are of water-soluble total suspended particulate (TSP). Furthermore, as the instrument measures any compound that dissociates to form the measured anion, GRAEGOR has a number of potential artefacts. These include interferences in HONO measurements from NO2 during periods of high SO2 concentrations (discussed in detail in Sect. ) and interference in HNO3 measurements at night from dinitrogen pentoxide (N2O5). Nevertheless, GRAEGOR has proven capable of time-resolved flux measurements in previous campaigns .

The ATTO site is equipped with an extensive suite of other instruments that provide long-term observations of meteorology, gases and particle properties. Wind speed, wind direction, sensible heat (H), air pressure (p) and frictional velocity (u*) were measured at 46 m on the 80 m walk-up tower using an ultrasonic anemometer (Gill WindMaster). Continuous measurements of relative humidity and air temperature (both measured using a Vaisala HMP45C-L), rainfall (HS Hyquist TB4-L rain gauge) and net radiation (Kipp & Zonnen net radiometer) were also available. Concentrations of equivalent black carbon (BCe) were measured by an Aethalometer (Magee Scientific AE33) at 325 m on the ATTO tall tower, and concentrations of carbon monoxide (CO) were measured at 52 m by a Picarro CKADS18. Also presented in this study are concentrations of NH4+, Cl-, NO3- and SO42- recorded by a time-of-flight aerosol chemical species monitor (ToF-ACSM, Aerodyne Inc) at 321 m on the ATTO tall tower.

The aerodynamic gradient method (AGM) is based upon flux-gradient similarity theory, which assumes that the flux of a tracer c (such as a gas or particle) can be determined if its vertical concentration gradient and its diffusion coefficient are known . In this study, a modified hybrid form of the AGM is used, whereby the flux of a trace gas or aerosol species can be determined from the vertical concentration difference of the species (Δc) and a series of stability parameters and the friction velocity (u*) derived by eddy-covariance from fast-response ultrasonic anemometry : 3Fc=-u*κΔcln⁡z2-dz1-d-ΨHz2-dL+ΨHz1-dL.

Here, κ is the dimensionless von Kármán constant (κ=  0.41); z2 and z1 are the heights at which the concentrations were measured (60 and 42 m, respectively, in this study); d is the zero-plane displacement height in metres; ΨH is the integrated form of the heat stability correction term, included to account for deviations from the log-linear profile; and ζ=(z-d)/L is a dimensionless atmospheric stability parameter based on L, the Obukhov length. By convention, a negative flux value denotes deposition to the surface, while a positive flux denotes an emission from the surface.

The zero-plane displacement height, d, is a critical parameter for calculation of the flux, and for a closed canopy is related to the canopy height, hc (d=(0.66 to 0.9) × hc). The analysis of this campaign uses a value of d=33.4 m as determined by from measurements of the logarithmic wind profile at the same tower.

The dry deposition velocity (Vd) is the negative ratio of the flux of the species to its concentration at a reference height (z) with consideration to the zero-plane displacement height: 4Vd(z-d)=-Fccz(z-d).

For gases, the deposition velocity can also be determined from the resistance analogy for dry deposition . Here, Vd is the reciprocal of the sum of the aerodynamic resistance Ra, the quasi-laminar boundary layer resistance Rb and the canopy resistance Rc: 5Vd(z-d)=1Ra(z-d)+Rb+Rc Ra and Rb can be calculated from Eqs. () and () :

6Ra(z-d)=u(z-d)u*2-ΨH(ζ)-ΨM(ζ)κu*,7Rb=Bu*-1, where ΨM is the integrated form of the momentum stability correction term; B is the sub-layer Stanton number , which is the product of the turbulent Reynolds number and the Schmidt number.

If the Vd of a trace gas is known from its flux via Eq. () and Ra and Rb are calculated using micrometeorological data, the canopy resistance Rc can be inferred from rearranging Eq. (). Similarly, a theoretical maximum deposition velocity (Vmax) for a trace gas can be determined if Ra and Rb are known, by setting Rc=0, which is equivalent to assuming perfect absorption of the gas by the canopy: 8Vmax⁡(z-d)=1Ra(z-d)+Rb.

The deposition of particles is more difficult to parameterise using the dry deposition resistance analogy, due to the different behaviour of particles compared to gases. In particular, the physical transport of particles through the quasi-laminar boundary layer is dependent on processes other than Brownian diffusion, such as impaction and interception. Consequently, although aerosol deposition velocities can be calculated as per Eq. (), the associated theoretical Vmax – which depends on measurements of Rb – cannot. Furthermore, due to the complexity in modelling the deposition process for larger particles, the deposition velocity for a particle is often replaced by an associated surface deposition velocity (Vds) value, parameterised by : 9Vds=11Vd-Ra.

The aerodynamic gradient method is ultimately based on Monin–Obukhov similarity theory (MOST). One of its assumptions is that fluxes are measured in the inertial sub-layer, where fluxes deviate little with height. For this reason, the inertial sub-layer is often termed the “constant flux layer” (CFL). However, in the roughness sub-layer (RSL), which extends over the individual roughness elements of the surface, MOST does not strictly hold . As a result, one of the underlying assumptions of the AGM is invalid, and consequently flux measurements using AGM can be erroneous .

Over forests, the roughness sub-layer can extend to almost 3 times the height of the canopy. Indeed, it is virtually impossible to make gradient flux measurements that avoid measuring within the roughness sub-layer, both for logistical reasons but also because gradients become increasingly weak at higher height and because of the limitations of the CFL . As with other studies, the flux measurements presented here were made at least partially within the RSL of the rainforest, where the height of the canopy was 37.1 m and the roughness sub-layer height therefore extended to an estimated 111 m.

As the profiles of concentrations and turbulence deviate from the logarithmic shape assumed by Eq. () within the RSL, fluxes calculated with the standard approach are likely to be underestimated compared to the true flux value . However, the overall flux-gradient relationship within the roughness sub-layer can still hold and be used to determine fluxes, but correction factors (also termed enhancement factors) must be implemented to account for measuring within the roughness sub-layer.

Work by at the ATTO site has led to development of such a correction factor, hereafter termed γF, that can be applied to flux measurements made using AGM above tropical rainforest. The γF value is dependent upon atmospheric stability, with a larger correction factor applied during stable atmospheric conditions compared to unstable conditions. This reflects the findings made by over tropical rainforest, that the solar zenith angle alters the predictions of scalars by MOST in the roughness sub-layer, with best agreement between observations and predictions at noon. Using measurements of L as a parameter for stable and unstable atmospheric stability, the values of γF developed by were applied to AGM flux calculations throughout this study; after, it was verified that they provide good agreement between measured and theoretically derived deposition velocities for HCl and HNO3 (see Sect. ).

The concentration limit of detection (LOD) (defined as 3σ above the background signal, where σ is the standard deviation) is of critical importance when measuring in regions of very low concentrations such as the Amazon rainforest. The LOD for each species measured by GRAEGOR was determined from a field blank test, which was conducted during the campaign over a 22 h period from 18:00 on 23 October to 16:00 on 24 October 2017 (local time). As detailed by , the field blank test to determine concentration LODs involves switching off the sample box air pump and sealing the air inlets of the samples boxes while leaving the rest of the system operating under measurement conditions. LODs are then determined as 3σ from the resulting background signal. Concentration LODs determined during this campaign are presented in Table  for individual trace gases and associated aerosol species.

The overall error in concentration measurements (σm) for the trace gases and aerosol components can be expressed as the product of the mixing ratio (m) with the individual error measurements, estimated by using a Gaussian error propagation approach : 10σm=mσmliqmliq2+σBr(SD)Br(SD)2+σQBrQBr2+σmBrmBr2+σQairQair2‾.

Each term in the propagation product denotes a measurement parameter and its associated standard deviation (σχ). In order, these are the mixing ratio of the compounds found in the liquid sample (mliq), the mixing ratio of the Br- standard (Br(SD)), the flow rate of the internal Br- standard (QBr), the mixing ratio (as analysed by the IC system) of the Br- standard (mBr) and the air-mass flow through the system (Qair). This formulation applies strictly for calculating the error in concentration measurement of species measured using IC. For NH3 and NH4+, which were analysed using FIA, the error in concentration measurement can also be determined by using Eq. (), omitting the terms for Br(SD) and m(Br), and replacing the factor QBr with QS, which is the flow rate of the NH3/NH4+ liquid sample. Calculated uncertainties ranged from 9 % to 19 %, with QS, QBr and mBr being the largest contributors to total measurement uncertainty.

As outlined by and , the flux measurement error (σF) for a trace gas or aerosol is composed of two terms: the product of the error in the concentration difference (Δc) and its associated standard deviation (σΔc) with the error in the flux-gradient relationship (here, expressed as a transfer velocity), which is dominated by the error in u* (σu*); and the flux (F) of the trace gas or aerosol measured: 11σF=Fσu*u*2+σΔcΔc2.

The error in the concentration difference can be determined through extended side-by-side measurements, where both sample boxes are placed at the same height and are supplied with a common air inlet. The instrument is then allowed to operate normally. The concentrations measured by both sampling boxes during this side-by-side sampling period are plotted against each other and fit with orthogonal regression. Using the orthogonal fit equation, the concentrations for the side-by-side sampling period and the wider campaign can then be corrected to account for systematic errors between each sample box. After correction, the remaining scatter in the side-by-side sampling concentrations (the residuals) is used to determine the error in the concentration difference. For the ATTO campaign, extended side-by-side measurements were conducted on 6 November at the end of the measurement period, with both sample boxes placed at 60 m.

The value of σu* is dependent upon the sonic anemometer used to measure u* and the atmospheric stability at the time of measurement . For this campaign, a value of 10 % for σu* was used during non-neutral conditions and 12 % for neutral conditions.

The median error values in flux calculations, as a percentage of flux values, are presented for trace gases and aerosol components in Tables  and , respectively. These values are in line with those calculated for previous studies .

Figure  shows the average diel cycles of the deposition velocities in comparison with those of Vmax for HCl and HNO3. Two sets of values are presented: values calculated using the standard modified aerodynamic gradient method (Sect. ) without the application of a correction factor for measuring within the roughness sub-layer, termed “pre-correction values”, and values calculated with the application of a flux correction factor developed by , γF, discussed in Sect. , which adjusts values derived from the aerodynamic gradient method when measuring in the roughness sub-layer, termed “post-correction values”. Due to their high water solubility (and resulting large effective Henry coefficient), HCl and HNO3 are expected to deposit at Vmax , unless chemical conversions affect their fluxes . The correction brings the Vd for these gases in close agreement with Vmax, within the measurement error. The correction increases the average Vd of HNO3 from 10.2 to 12.4 mms-1 (average Vmax= 12.3 mms-1) and that of HCl from 12.5 to 15.2 mms-1 (average Vmax= 15.3 mms-1). This suggests that, overall, the γF correction works well, and the remainder of the paper discusses post-correction values only. With this consideration in mind, Fig.  shows the average diurnal cycles of the post-γF-corrected deposition velocity in comparison with that of Vmax for the remaining trace gases measured: NH3, HONO and SO2 .

Table  presents a statistical summary of the calculations for fluxes, deposition velocities (Vd), theoretical maximum deposition velocities (Vmax) and canopy resistances (Rc) for the inorganic trace gases measured during the campaign. As discussed above, with the roughness sub-layer correction of , both HCl and HNO3 are observed to deposit at Vmax within the error of the measurement, with a canopy resistance < 3 sm-1, although the results would be sensitive to the Rb parameterisation used, which forests can vary significantly depending on the ecosystem and climatic conditions .

Time series for the post-filtered fluxes of the inorganic trace gases measured are shown in Fig. . The inorganic trace gases HNO3, SO2 and HCl were nearly always deposited to the surface. Any upward fluxes calculated for these gases lay within their respective error ranges. Fluxes which exceed the median values for these gases and the maximum calculated fluxes for these species, were recorded during the drier, warmer polluted conditions that prevailed from 18 to 26 October 2017. For example, the maximum calculated flux for SO2 and the largest flux of any species measured during the campaign, was -33 ngm-2s-1, which occurred on 21 October at 11:00. Conversely, while increased deposition fluxes are observed for NH3 and HONO during this same period, multiple periods of emission were recorded for these gases throughout the campaign. Although the predominant pattern of surface–atmosphere exchange throughout the campaign for HONO and NH3 was deposition to the surface, as reflected in their respective median flux and Vd values, periods of emission are a significant proportion of overall surface–atmosphere exchange. For HONO and NH3, respectively, 26 % and 19 % of calculated fluxes were positive, i.e. emissions. The median diel pattern of trace gas emissions is highlighted in Fig. . HONO emissions were concentrated in the early morning, with positive median values indicating a prevalent pattern of emission present at 07:00 and 08:00. In contrast, NH3 emissions were observed in the afternoon, from 14:00 to 16:00. The other trace gases – HCl, HNO3 and SO2 – showed maximum deposition fluxes in the afternoon, with decreased fluxes during the night and early morning hours.

A statistical summary of fluxes and deposition velocities for the aerosol species is presented in Table . Also included for each species is the minimum detectable flux (FLOD) and the percentages of calculated fluxes which exceed this value (fLOD %).

Median Vd values for NH4+ and SO42- were 2.64 and 2.81 mms-1, respectively. In the comparison of GRAEGOR and ToF-ACSM concentration measurements outlined in Sect. , we found a reasonable agreement for NH4+ and SO42-, considering the difference in measurement height and instrumentation. Given that the ToF-ACSM measures only the submicron (< 1 µm particle diameter) range, this suggests that the NH4+ and SO42- quantified by GRAEGOR were also dominated by the submicron range. From process-orientated modelling of aerosol Vd, it has been suggested that particle Vd increases over increasingly rough surfaces . In a meta-analysis of field flux data, parameterised this relationship as a function of the surface deposition velocity, Vds, and the surface roughness (given as the surface roughness length, z0, in metres): 12Vds=0.581log⁡z0+1.86.

Using the median value of the surface roughness lengths calculated at the site (and including only lengths with a valid calculated value of aerosol Vd) yields a value of 2.86 m for z0. Substituting this into the Eq. () parameterisation suggests a Vds of 2.1 mms-1 for submicron particles. Values of NH4+ and SO42- Vd converted to Vds values using Eq. () result in median Vds values for NH4+ and SO42- of 2.9 and 3.3 mms-1, respectively. Although these values are higher than the parameterised value, Eq. () was derived specifically for particles in the range 0.1–0.2 µm. Larger particle sizes would have higher Vds values for a given value of u* . Thus, if the particle size range for NH4+ and SO42- exceeds 0.2 µm but remains in the submicron range the measured median Vd would exceed the parameterised value.

In contrast to Vds values for NH4+ and SO42-, which are in the range for parameterised values for the site, the median Vds values for Cl- and NO3- are 3 to 4 times greater than the parameterised value of 2.1 mms-1. The median Vds value for Cl- is 10.2 mms-1, while for NO3- it is 7.6 mms-1. As the parameterised value holds only for particle diameters between 0.1 and 0.2 µm and considering that modelling indicates an increase in Vds with increasing particle size, the larger median Vds values for Cl- and NO3- are consistent with GRAEGOR vs. ACSM comparison, which suggests that these aerosol counterparts were present in the supermicron (> PM1) fraction.

A time series of the aerosol counterpart fluxes is presented in Fig. . The predominant direction of surface–atmosphere exchange for all aerosol species was deposition, as reflected in the median flux values in Table . However, individual emission fluxes were recorded for all species, with the maximum emission values for Cl- and SO42- (+3.6 ngm-2s-1 and +4.3 ngm-2s-1, respectively) being particularly large. The time series of values is filtered for identifiable errors in measurement and for micrometeorological values that fall outside specified limits (Sect. ). These emission fluxes are therefore unlikely to be caused by instrumentation faults or calculation errors. They are, however, limited in duration and overall extent – positive particle emissions are never observed consecutively, occurring exclusively within 1 h periods, and constitute only between < 1 % (NH4+) and 5 % (Cl-) of total fluxes. While particle emission fluxes have previously been observed with GRAEGOR , these previous observations have occurred during periods of known flux divergence.

The relative proportions of inorganic trace gases over the ATTO site during the campaign can give important insight into the overall atmospheric chemistry. As the primary basic gas in the atmosphere, NH3 can react with the acidic gases HCl, HNO3 and H2SO4 (produced by the oxidation of SO2) to form ammonium salts whose lifetime and behaviour are dependent upon the associated gas. To investigate the importance of the various acidic gases for total acidity at this remote Amazon site, the fractional contribution to total inorganic acid loading for HCl, HONO, HNO3 and SO2 as measured by GRAEGOR was determined (Fig. ). Taken as an arithmetic mean value, the fractional contributions of SO2, HNO3 and HCl are similar. While not as significant a contributor in comparison, HONO also contributes at an average fraction of 0.13, which remains consistent throughout the duration of the campaign. The contributions of SO2 and HNO3 average at 0.31 and 0.30, respectively, whilst the contribution of HCl averages at 0.26 but fluctuates throughout the campaign, varying between ∼ 0.05–0.10 during near-pristine conditions to almost 0.40 during the polluted period from 19 to 25 October.

Fossil fuel combustion is the primary anthropogenic (and overall predominant) source for NO2 in the troposphere. The increase in HNO3 concentrations on 25 October (also resulting in increased deposition fluxes) could be due to air masses that picked up emissions of NOx (NO2+NO) from the urban areas of Manaus and Santarém. Measurements of NO2 downwind and west of the Manaus urban area showed elevated NO2 concentrations in remote areas affected by emission plumes from the city . With air masses arriving at the site from the south and south-east, which had travelled over the eastern suburbs of Manaus and the city of Santarém, respectively, it is likely that NO2 plumes are responsible for the elevated HNO3 observed on 25 October.

While a moderate, positive monotonic relation exists between concentrations of HCl and BCe, it is unlikely that the presence of HCl above the detection limit of GRAEGOR could be sustained throughout the campaign solely through anthropogenic emissions. HCl is highly reactive and water soluble, with a mean lifetime of ≈ 36 h . Consequently, it is unlikely that regional or global biomass burning could contribute meaningfully to the HCl concentrations observed at this remote site. The peak in HCl concentrations observed during the relatively polluted periods of the campaign could be a result of biomass burning from local sources in close proximity, but an alternative explanation must be considered for the background concentrations of HCl. Globally, much of the HCl derives from the displacement reaction of HNO3 with aerosol Cl- compounds; this typically happens with NaCl sea salt but potentially other Cl- compounds at this site (see Sect.  below). A further potential contributor is oxidation of methyl chloride (CH3Cl), whose predominant natural source is tropical forest . The emissions are driven principally by dipterocarps and ferns , whose emission rates are unaffected by abiotic conditions . measured an average emission for CH3Cl of 9.5 µgm-2h-1 over Guyanese and Surinamese rainforest, while reported CH3Cl concentrations above a rainforest canopy in Rondônia, Brazil, confirming that the Amazon rainforest region is a net regional source for CH3Cl.

proposed an OH-driven oxidation pathway for CH3Cl that terminates with stoichiometric production of HCl. It is thus possible that the tropical forest emissions of CH3Cl, combined with the local high oxidative capacity, could yield the background HCl concentrations observed in this study. However, to confirm this idea, simultaneous measurements of CH3Cl and HCl concentrations would be required, together with confirmation of Sanhueza’s postulated CH3Cl oxidation pathway.

This campaign presents the first tower measurements of time-resolved SO2 fluxes over tropical rainforest. Standard commercial SO2 monitors struggle to resolve such low concentrations. Although aircraft , denuder tube and filter pack measurements of SO2 over rainforest exist, they lack the time resolution of the measurements during this campaign or do not measure fluxes. This study has shown that LRT pollution episodes can significantly enhance SO2 deposition fluxes (a maximum deposition flux of -33.2 ngm-2s-1 was recorded during the most polluted period of the campaign) and that even during relatively pristine conditions SO2 concentrations remained above the LOD. As Fig.  demonstrates, the close correlation between SO2 and BCe suggests that long-term measurements of SO2 over tropical rainforest may be worthwhile as a further method to identify episodes of increased pollution or biomass burning. Long-term measurements would also show whether concentrations of SO2 remain above detection limits during the pristine conditions of the wet season and help determine potential sources during these periods. It is possible that a biogenic source may have contributed to SO2 measured during the relatively pristine conditions. For example, SO2 could derive from the oxidation by OH of dimethyl sulfide emitted from the rainforest .

The comparison between ACSM and GRAEGOR water-soluble aerosol concentrations in Sect.  indicates good agreement between them for SO42- and NH4+ but significant divergence for NO3- and, in particular, Cl-.

Long-term measurements of aerosol chemical composition at the ATTO site using an ACSM have been conducted since 2014, and the first publication of data from 2015 suggested that aerosol chemical speciation varied surprisingly little across the wet and dry seasons . As recorded by the ACSM during this campaign, organic aerosols are always the dominant mass fraction (compromising ∼ 70 % of aerosol), followed by SO42- (10 %–15 %), BCe (5 %–11 %), NH4+ (∼ 5 %), NO3- (∼ 4 %) and finally Cl- as the smallest contributor. Focusing only on the aerosol species measured by both GRAEGOR and ACSM during this dry season campaign, the average ACSM mass fractions are 55 % SO42-, 22 % NH4+, 18 % NO3- and 5 % Cl-. As Fig.  demonstrates, the total mass fraction contribution to the inorganic particulate as measured by GRAEGOR suggests that the contribution of NO3- and Cl- is more significant than suggested by previous measurements. The relative contribution of each species to TSP as measured by GRAEGOR in this campaign (in descending order) is SO42- = 34.4 %, NO3- = 30.8 %, NH4+ = 19.0 % and Cl- = 15.3 %. In comparison to ACSM measurements, the relative proportion of SO42- is reduced, NO3- becomes the second most abundant species with an almost equal contribution to SO42-, and Cl- – while remaining the smallest contributor to total mass – has a greater relative contribution to the mass of TSP. measured a similar contribution order for the dry season using ion chromatography, with SO42- contributing the most to the total mass fraction and Cl- the least, but with a differing proportion (SO42- = 51 %, NO3- = 26 %, NH4+ = 19 %, and Cl- = 4 %). Variations in the ion proportions may be attributable to differences in the number and intensity of long-range transport episodes, which contribute Cl- and SO42-, during a given field campaign.

The ACSM samples only the submicron (PM1) aerosol size range, while GRAEGOR samples TSP (< 50–100 µm particle diameter). Furthermore, the ACSM only detects non-refractory aerosol compounds and is therefore insensitive to refractory sea salt and crustal material . The close similarity in SO42- and NH4+ measurements between the two instruments suggests that the majority of SO42- and NH4+ particles during the campaign were contained within submicron aerosol and that the SO42- represented semivolatile ammonium compounds. Conversely, the difference between ACSM and GRAEGOR NO3- measurements suggests that most of the NO3- was contained within the coarse mode and/or represented non-volatile compounds such as NaNO3 and Ca(NO3)2, and almost all of the Cl- measured by GRAEGOR in this campaign was found in the coarse-mode and/or as NaCl. Previous work had found Cl- to be exclusively associated with the coarse fraction . This is consistent with thermodynamic considerations which would suggest that volatile NH4NO3 aerosol, the NO3- compound typically measured by the ACSM, should not exist at the high temperature and relatively low gas-phase concentrations of NH3 and HNO3 at this site. This was confirmed using the ISORROPIA-2 () thermodynamic modelling framework.

Consistent with the insensitivity of the ACSM to refractory particles, a possible source for coarse Cl- aerosols could be sea salt. Although a continental site, intrusions of sea salt through long-range transport have been noted previously at ATTO . The presence of sea salt could also account for a source of coarse NO3-, as the reaction between HNO3 and NaCl would result in the formation of the coarse aerosol NaNO3 , a refractory aerosol component that would not be detected by the ACSM. The reaction of HNO3 with sea salt would also form HCl, the measured concentrations of which are closely linked to those of Cl- in this campaign. Alternatively, the strong link between HCl and Cl- concentrations could be accounted for by biomass burning emissions arriving at the ATTO site, whereby Cl- particulate from biomass burning is principally in the form of fine KCl . Other crustal material, such as dust and soil particles which are recorded in elevated amounts at ATTO during the dry season , could provide a source of coarse NO3-. These can include a variety of NO3--containing mineral species, such as NaNO3, Ca(NO3)2 and Mg(NO3)2 . The surface of dust and suspended soil particles could also act as a sink for HCl in the marine boundary layer , allowing the heterogeneous formation of coarse Cl- particulate.

It has been shown previously that primary biological aerosol particles (PBAPs) contribute the majority of the mass fraction of measured coarse aerosol in the Amazon . The PBAPs over the rainforest consist of a variety of different biological materials, such as plant and animal matter fragments, algae, pollen and fungal spores. The latter contributor is particularly important, as fungi which actively discharge their spores through liquid jets have been identified by to be a source of inorganic ions in particulate matter. Fungi that actively discharge their spores do so via a liquid jet, whereby spores are forcibly discharged from a spore sac (ascus) along with a liquid mix of sugars and ions, of which Cl- forms a significant fraction . The spore itself can rupture under conditions of high relative humidity, resulting in the formation of fragments containing inorganic ions . In a chemical imaging analysis of such spore fragments above the Amazon rainforest, found that almost 40 %–60 % of these fragments contain Na+ and Cl- associated as a salt, which appeared “morphologically similar to dry sea salt” and which grew to supermicron sizes in conditions of high relative humidity. The contribution of fungal spores to total Na+ mass during the wet season over the rainforest was estimated as ∼ 69 % by the same study, with the conclusion that measured concentrations of coarse Na+ and Cl- could mistakenly be ascribed to marine sources rather than to locally originating fungal spore emissions. As discussed in Sect. , emission fluxes for Cl- are recorded throughout the campaign and occurred during cooler, wetter periods at night. As noted by , fungal spore emissions also predominantly occur under the same conditions. The possibility that Cl- concentrations measured during this campaign are biogenically driven through the active discharge or rupturing of localised fungal spore emissions should therefore not be discounted.

As detailed in Sect. , HCl, HNO3 and SO2 were always deposited with no instances of emissions. The surface canopy resistance (Rc) for these gases was calculated for the campaign using a rearranged form of Eq. (). As expected on the basis of their high water solubility, HNO3 and HCl deposited with a very small average canopy resistance of 1.42 and 2.92 sm-1, respectively, which is not significantly different from zero given the typical uncertainty in the Rb parameterisation used to infer this value. By contrast, the average canopy resistance for SO2 in this campaign was considerable, with a mean value of 86 sm-1 throughout the entirety of the campaign, and a potentially more robust mean value of 28 sm-1 for measurements during daytime. Using his widely used dry deposition parameterisation, derives a typical Rc value of 120 sm-1 for SO2 for deciduous forests with “lush vegetation” during “midsummer”, evaluated at an incoming solar irradiance of 800 Wm-2. In the absence of tropical flux measurements, the appropriateness of the value for tropical forest has never been tested. While the observed average in this work is 3 times less than Wesely’s parameterisation, the daytime average value from this campaign covers a wider set of meteorological conditions than used for the calculation of the modelled SO2 Rc.

elaborated upon Wesely's dry deposition parameterisation through the development of a new formulation for the non-stomatal resistance component of the model. Modelled vd values for a variety of chemical species, including SO2, were developed for different land use classifications (LUCs), including broadleaf tropical forest. While notes good agreement between modelled and observed vd values for LUCs such as short grasses and crops, the mean measured vd for SO2 during this campaign deviates significantly from its corresponding modelled value for a tropical broadleaf LUC. This study measured a mean vd of 10.4 mms-1 for SO2, while suggests values between 1.5 and 3.8 mms-1, with the limits for dry and wet canopies, respectively. As with the comparison with , the appropriateness of modelled values have not been tested due to the lack of corresponding measurements. Similarly, the values for this campaign cover a wide range of meteorological conditions.

Both HONO and NH3 fluxes revealed periods of emission from the rainforest, with 26 % of all HONO fluxes and 19% of NH3 fluxes recorded as emissions. Due to the complexities of the chemical and physiological parameters controlling NH3 emissions from the canopy surface to the atmosphere, discussion of the NH3 fluxes measured in this study are considered in a separate paper , which investigates inter alia the influence of leaf wetness and modelled canopy compensation points upon NH3 bidirectional exchange with reference to established models of NH3 surface–atmosphere exchange. It demonstrates that the observed NH3 emissions are consistent with stomatal emission during the warmest part of the day and shows that measured leaf wetness is a more successful parameter in describing the cuticular deposition process than relative humidity and vapour pressure deficit. The present paper therefore focuses on discussion of the observed emissions of HONO at this site.

The median diel fluxes of HONO in Fig.  show emission in the early morning after dawn (from 07:00 to 09:00), with deposition dominating throughout the rest of the day. Three possible explanations are considered here. The first considers the influence of soil emissions below the forest canopy. HONO emissions from soil have been observed in a number of studies , with possible sources including the volatilisation of HONO from soil nitrite , the temperature-dependent activity of ammonia-oxidising bacteria or the oxidation of hydroxylamine released from soil microorganisms . During night-time, radiative cooling above the forest causes stable stratification, generating a nocturnal boundary layer that prevents mixing between the air below and above the canopy . Consequently, HONO emissions from the soil would accumulate below the canopy. At dawn, turbulent mixing starts to break up the nocturnal boundary layer, generating unstable conditions and a mixed layer. This creates a “venting” effect where the below-canopy accumulated HONO is transported upwards and appears as an early morning emission flux. Such venting episodes, representing negative storage fluxes, are commonly observed for CO2 over tall vegetation and have been noted previously also in tower measurements above rainforests for CO2 , methane and particles , with recording maximum median diel CO2 and CH4 fluxes between 06:00 and 10:00, similar to the period of maximum median diel HONO emissions here. CO2 flux measurements taken at the ATTO site concurrently with this study also showed a characteristic early morning flux, supporting the explanation of a venting effect for the HONO emissions.

However, morning HONO emissions have also been observed at short vegetation sites , where storage effects are much smaller and which therefore must have resulted from a different mechanism. This is that early morning HONO emissions are a consequence of the photolysis of HNO3 . Accumulation of HNO3 on leaf surfaces during night-time results in a reservoir of HNO3 within the canopy. At dawn, incoming solar radiation photolyses this reservoir, resulting in the formation of exited NO2 radicals that – in the presence of photosensitising organics such as humic acid – are reduced to HONO. The concurrent breakdown of the nocturnal boundary layer again results in an upward emission flux of HONO. However, while recorded emissions of HONO from forests between the hours just after dawn until late afternoon, with maximum fluxes recorded around solar noon, in this study emissions occurred predominately during the hours immediately after dawn. While emissions were recorded at noon and during the afternoon on certain days, medial diel emissions were confined to 07:00 to 09:00. Furthermore, has shown that this pathway would have a negligible effect on HONO formation based on the kinetic values for the pathway. Future work should measure the gradients of HONO above and below the canopy, preferably by taking a concentration gradient extending from below canopy to above canopy, to determine whether HONO accumulation below canopy during stable night-time conditions is occurring, followed by venting during morning hours due to turbulent mixing.

Finally, transient emission blips following sunrise have been observed for NH3 during several studies, where they were attributed to desorption of NH3 that had been dissolved in dew and microscopic water layers overnight. As these water layers evaporate in the morning, concentrations increase to a point where they get driven into the gas phase. Studies have postulated that the same process occurs for HONO and contributes to the bidirectional exchange seen during some of the aforementioned observations. They show that timing is indeed consistent with the temporal dynamics of the emission at a UK grassland site. At ATTO, the temporal dynamics of the NH3 flux were different, with emission peaks occurring later in the day than for HONO, and it was therefore concluded that desorption did not contribute to the NH3 emission fluxes . It therefore remains unclear why desorption would have been more important for HONO than for NH3.

It is important to note that measurements of HONO by the GRAEGOR system are not artefact free. As detailed by , the presence of SO2 and NO2 on wet denuder walls can introduce a positive artefact that results in an overestimate of HONO concentrations, which – if using a gradient system with two or more wet denuders set at different heights – can result in erroneous concentration gradient profiles. Correction algorithms exist for general application and specifically for GRAEGOR that allow the influence of the artefact to be quantified using concentrations of SO2 and NO2. However, for this campaign, no correction was necessary as the SO2 concentration recorded during the campaign was 5 to 10 times lower than those relevant to artefact formation.

The recorded deposition velocities of the aerosol species are consistent with the GRAEGOR/ACSM intercomparison: NO3- and Cl- aerosols were predominantly contained in the coarse fraction, while NH4+ and SO42- were contained within the submicron aerosol. From a process-orientated approach , the deposition velocity of a particle is dependent upon its size. For particles > 0.1 µm, deposition velocity (normalised against u*) increases with increasing particle diameter. As outlined in Sect. , the close agreement between measured SO42- and NH4+ deposition velocities (and parameterised values for 0.1–0.2 µm size range aerosols) above tropical rainforest suggest that these aerosols were contained in the fine mode. These observed deposition velocities also agree well with modelled deposition velocities for < 1 µm diameter particles above forest with similar mean roughness lengths and u* values as recorded at ATTO . Conversely, the larger observed deposition velocities for NO3- (5.8 mms-1) and Cl- (7.3 mms-1) exceed the parameterised values obtained using the formulation of and fit within the modelled values given by for particles in the 2–10 µm range above surfaces with a similar roughness length.

As detailed in Sect. ., occasional periods of apparent particle emissions from the rainforest were recorded throughout the campaign for all aerosol species measured. Deviations from near-exclusive deposition were rare (between 1 %–3 % of all measured fluxes), confined to 1 h periods, and are unlikely to be due to measurement error. Similar to the emissions of HONO recorded during this campaign, upward particle fluxes may be caused by early morning turbulent mixing generating upward entrainment fluxes into the growing mixing layer. recorded a similar pattern of particle emissions at a tropical rainforest site in North Borneo, as did at a rainforest site in the Amazon Basin located 120 km south-west of the ATTO site. However, both studies recorded a more predominant pattern of early morning emissions than here. recorded particle emissions for almost all mornings, while reported 40 % of all particle fluxes as emissions. Both studies record later (08:00–09:00) emission periods. As both studies measured total particle number which was not chemically speciated, it is possible that the flux behaviour of the organic fraction of aerosol – which dominates the total aerosol mass fraction over tropical rainforest – is a more important driver for observed particle emissions than the aerosol species measured during this campaign.