Interactive comment on “ Ozone production and transport over the Amazon Basin during the dry-to-wet and wet-to-dry transition seasons ”

Abstract. The Regional Carbon Balance in Amazonia (BARCA) campaign provided the first Amazon Basin-wide aircraft measurements of ozone (O3) during both the dry-to-wet (November and December 2008) and wet-to-dry (May 2009) transition seasons. Extremely low background values (


Introduction
In the Amazon Basin, trace gases from biomass-burning, urban, and biogenic emissions are important sources of ozone precursors, which are efficiently transported by 25 intense convective activity to the upper troposphere, where they can be dispersed over plants and damage leaf tissue, harming wild ecosystems and reducing crop productivity (Reich and Amundson, 1985). In the upper troposphere, O 3 acts as a greenhouse gas, increasing surface radiative forcing (IPCC, 2001). The Amazon Basin continues to rapidly urbanize, and urban emissions of O 3 precursors are also expected to grow (Gallardo et al., 2010). An improved understanding/quantification of O 3 temporal and 15 spatial variability in the tropical rainforest environment is important for projecting future impacts of land use and climate change in the Amazon Basin and other tropical rainforest regions worldwide on their expanding human populations and significant biodiversity.
Motivated by the impact of O 3 in the Amazon Basin on human and ecosystem health 20 and global climate, we collected aircraft observations of O 3 during BARCA and conducted regional chemistry simulations in order to answer the following scientific questions: how does O 3 vary spatially and seasonally over the Amazon Basin? What are the sources and sink of O 3 in this region? How well can state-of-the-art regional chemistry models reproduce O 3 distributions over the Amazon Basin? 25 Previous analyses of satellite ozone data have noted early-year O 3 maximums in the tropical Southern Hemisphere primarily associated with cross-Atlantic transport of biomass burning emissions from Africa (Thompson et al., 1996), Northern Hemisphere fires and lightning NO x (Edwards et al., 2003). In the Amazon region, ground-based and aircraft campaigns (e.g., Kirchhoff et al., 1990;Browell et al., 1996;Kaufman et al., 1998;Longo et al., 1999;Cordova et al., 2003;Andreae et al., 2001Andreae et al., , 2002Rummel et al., 2007;Kuhn et al., 2010;Martin et al., 2010;Toon et al., 2010) have observed both background O 3 levels of 10-20 ppb and elevated levels of 60-80 ppb due to production from regional fire emissions and recirculated urban pollution from SE Brazil, as well as 5 evidence of deep convective transport of boundary layer air to the middle and upper troposphere.
In-situ data on cloud properties and chemical species, as well as observations of land use changes, boundary layer dynamics and larger-scale cloud-aerosol interactions, are scant in this region. Therefore, models are essential tools for monitoring and predict- 10 ing atmospheric chemistry composition, weather, and climate at local, regional, and global scales. Uncertainties in the model representations of parameterized convection, turbulence, land surface and other subgrid scale processes lead to significant errors in simulated transport and chemical transformation of the atmospheric composition (Beck et al., 2013). The Regional Carbon Balance in Amazonia (BARCA) Large-Scale and WRF-Chem simulations. In Sect. 3, the results of the analysis of meteorological and O 3 observations and simulations are presented, with final discussions and conclusions in Sect. 4.

BARCA aircraft campaigns
The BARCA flights were conducted with the EMB 110 Bandeirante aircraft of the 15 Brazilian National Institute for Space Research (INPE). The flights consisted of quasi-Lagrangian measurements of carbon dioxide (CO 2 ), carbon monoxide (CO), methane (CH 4 ), ozone (O 3 ), and aerosols, and were designed to constrain basin-wide fluxes and understand distributions and sources of these species. The EMB 110 Bandeirante INPE aircraft had a ceiling of 4500 m, and flights usually consisted of ascending and 20 descending vertical profiles separated by short (5-30 min) horizontal legs. In-situ measurements were made of CO 2 , CH 4 , CO, O 3 , aerosol number concentration and optical properties. Flask samples were collected to determine CO 2 , CH 4 , sulfur hexafluoride (SF 6 ), CO, nitrous oxide (N 2 O), hydrogen, and the oxygen-nitrogen ratio (O 2 /N 2 ). A detailed description of the aircraft measurements can be found in Andreae et al. (2012). convective activity in that region during BARCA B precluded flying. During BARCA B, fire activity was low throughout the Amazon region due to heavy precipitation, while during BARCA A, intense fire activity occurred on the northern coast of Brazil and scattered fires were present throughout the southeastern Amazon. 5 Andreae et al. (2012) summarized the BARCA campaign, meteorological background, carbon monoxide and aerosol observations and CO results from several regional transport and chemistry models. These included the CCATT-BRAMS and WRF-Chem simulations analyzed in greater detail in this paper. Meteorological analysis showed that during BARCA A, when the Inter-Tropical Convergence Zone (ITCZ) was to the north of 10 the Amazon Basin, inflow to the Amazon was primarily from the Southern Hemisphere. During BARCA B, the ITCZ extended to 20 • S and air at low levels was of Northern

Transition season meteorology and emissions
Hemisphere origin, including some smoke from west African fires. On the other hand, the mid tropospheric air was of mixed origin. The highest CO levels were observed on the flights on [25][26][27] November in the south- 15 eastern Amazon, influenced by regional biomass burning, as maximum values were observed from 1-3 km. These are typical of injection heights of smoke plumes from savanna fires (Freitas et al., 2007). The excess CO from biomass burning was between about 30 and 200 ppb, increasing from north to south across the Basin. The mean contribution from biomass burning to total CO during BARCA A was about 31 %, with 20 a contribution from background (110 ppb) of about 61 %. Biomass burning influence was indicated by CO mixing ratios up to 300 ppb, Condensation Nuclei (CN) approaching 10 000 cm −3 and a low CN to CO ratio (∆CN/∆CO) indicating aged smoke. This influence was highest in the southern Amazon from 1-3 km. Manaus back trajectories at 500 and 4000 m came from eastern Amazon fires rather than the intense African 25 fires occurring at the same time. During BARCA B, little biomass burning influence was observed. CN counts were 300-500 cm −3 and a CO enhancement of ∼ 10 ppb above the mixing ratios in air entering the Basin from the Atlantic was seen. layer enhancements were attributed to a source from the oxidation of biogenic VOCs (Andreae et al., 2012). Andreae et al. (2012) also showed simulated vertical CO profiles from CCATT-BRAMS and WRF-Chem simulations, as well as the Stochastic Time Inverted Lagrangian Transport (STILT) model with two different meteorological field inputs and 5 the WRF Greenhouse Gas Module (WRF-GHG). The simulated CO profiles matched mean observed values, but were overly vertical (too low near the surface and too high above 3 km). This suggested that the models had too much convective transport or vertical mixing from the PBL schemes. However, the probability densities were consistent with observations in the boundary layer, indicating that horizontal dispersion was rea-10 sonable. Beck et al. (2013) evaluated different CH 4 wetland emissions schemes and maps using WRF-GHG. They found the best agreement with BARCA CH 4 data for days where convective transport, as evaluated by comparison of upstream TRMM and WRF precipitation amounts, was well represented in the model. This indicates that proper representation of convective transport in models is essential for prediction of vertical 15 distributions of pollutants in the Amazon Basin.

Previous studies of O 3 in the Amazon
Analyses of satellite, aircraft and ground-based observations of O 3 over Amazônia since the 1980s have demonstrated the influence of long-range transport of African biomass burning and Northern Hemisphere inputs, local fire sources, NO soil and bio-20 genic VOC emissions, and convective transport on spatial and seasonal variability in O 3 . In particular, data from the ABLE-2B aircraft and ground campaign during the 1987 wet-to-dry transition season and the BARCA observations offer the opportunity to compare the regional O 3 distribution across decades.
Several studies of satellite data have reported a seasonal O 3 maximum in the trop- 25 ical Southern Hemisphere, largely associated with long-range transport of African fire emissions or lightning NO x sources. Fishman and Larsen (1987) Thompson et al. (1996)  The earliest O 3 measurements over the Amazon Basin were made during aircraft campaigns in the dry seasons 1979 and 1980 (Crutzen et al., 1985). Mixing ratios of 20-30 ppb and 40-50 ppb were observed in the boundary layer and the free troposphere, respectively, and elevated O 3 was attributed to photochemical reactions and Introduction  -August 1985), the Amazon Boundary Layer Experiment (ABLE-2A) integrated aircraft, ground-based and satellite observations to study the processes affecting the chemical composition in mixed layer over Amazonia . Jacob and Wofsy (1988) (Torres and Buchan, 1988 NO measured by Kaplan et al. (1988). Lidar observations during ABLE-2A showed 15 highly variable O 3 levels, with some small regions with up to 30-40 ppb, attributed to variable NO flux from the canopy . ABLE-2B was conducted during the wet-to-dry transition season (April-May 1987) (Harriss et al., 1990). Periodic inputs from the Northern Hemisphere were found to be a pollution source over Amazonia, and dry deposition in the region provides a significant sink in the global O 3 budget. 20 Aircraft measurements from TRACE-A over the south Atlantic in the 1992 dry season attributed high O 3 (> 100 ppb) in the upper troposphere to photochemical production from convectively lofted Brazilian biomass burning emissions. Elevated O 3 (> 75 ppb) originated in lower altitude (< 6 km) plumes from African fires (Browell et al., 1996) aboard the ER-2 aircraft during two flights between Cuiabá and Vilhena confirmed the co-occurrence of layers of elevated O 3 with smoke (Longo et al., 1999). As part of ABLE-2, near-continuous O 3 surface measurements (1.5 m above the surface) showed daytime maximum of 5.7 and 3.7 ppb in a clearing and forest, respectively, and measurements in a tower in a clearing showed an increasing gradient of O 3 5 with height, up to 6.9 ppb at 15 m above the surface (Kirchhoff et al., 1990). Furthermore, 20 ozonesondes launched in the clearing showed typical mixing ratios of 40 ppb from 500-300 hPa, with values about 10 ppb lower in the wet than dry season.
Observations of O 3 , NO x and CO at a pasture site in the state of Rondônia and forest sites in the states of Pará and Amazonas showed elevated (3×) O 3 and NO 2 10 levels in the dry-to-wet transition season at the pasture site due to the influence of biomass burning. This was shown by correlations with black carbon and aerosol number concentrations at the surface. On the other hand, NO levels were much lower in the dry-to-wet transition season due to the conversion of NO 2 to NO favored by elevated levels of VOCs, O 3 , and radicals, and by higher temperatures. In addition, nighttime 15 ozone was increased in the wet season by transport of ozone-rich cold air from the mid-and upper-troposphere by convective downdrafts, as shown by an anti-correlation of O 3 with equivalent potential temperature (Cordova et al., 2003).
During the LBA-CLAIRE-98 experiment (Andreae et al., 2001) in March 1998, elevated levels of trace gases and biomass burning aerosol were observed at high al-20 titudes (> 9 km) during a flight off the coast of Suriname. Model simulations of CO transport later confirmed the measurements to be the outflow of a deep convective system which had transported biomass burning emissions originating from the northern Amazon (Freitas et al., 2000;Andreae et al., 2001;Gevaerd et al., 2006). During the same experiment, trace gases and CCN spectra were also measured continuously at 25 a ground station in Balbina, near Manaus (Zhou et al., 2002). During the experiment, air masses with origin over undisturbed rainforest and little anthropogenic influence, were sampled at Balbina, yielding O 3 values always less than 20 ppb. Photochemical production of O 3 of up to 15 (Kuhn et al., 2010). Most of the VOC reactivity was provided by isoprene emissions from the surrounding rainforest, and NO x emissions suppressed O 3 production close to urban sources, but stimulated it downwind.
Observations at a pasture site in Rondônia in January-February 1999 during the LBA Wet Season Atmospheric Mesoscale Campaign (WETAMC) showed that down-5 ward convective transport events increased nighttime surface O 3 up to 30 ppb, compared to a background of 3-5 ppb (Betts et al., 2002). During the LBA-EUSTACH experiments, CCN and trace gases (including O 3 , NO x and VOCs) were measured at forest and pasture sites in Rondônia in the wet-to-dry (27 April-29 May 1999) and dryto-wet (12 September-27 October 1999) seasons (Andreae et al., 2002;Rummel et al., 10 2007). The observations showed VOC (isoprene, formaldehyde, acetaldehyde, acetic and formic acid) concentrations 4-5 times higher in the dry than wet-to-dry transition season. The VOC enhancement was a result of both enhanced biogenic emissions and photochemical decomposition due to increased solar radiation. In addition, VOC and O 3 concentrations peaked in the afternoon (around 15:00 LT) in both seasons and at Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | processes include O 3 production from soil NO x emissions and removal via dry deposition to the forest canopy. Aircraft campaigns complete the suite of observations, allowing the examination of convective lofting of surface emissions, with biomass burning emissions of particular importance on the regional scale.
2 Data and methods 5

BARCA aircraft measurements
During the BARCA campaign, in-situ measurements of O 3 were conducted aboard the EMB 110 Bandeirante INPE aircraft using a dual-cell, UV Photometric analyzer (Ozone Analyzer, Model 49i, Thermo Fisher Scientific, United States). During BARCA A, 1 min averages of the original 1 s data were taken. On the other hand, during BARCA B 1 s 10 data were stored, and the detection limit for both campaigns was 1 ppb. The intake for O 3 was forward-facing, located 185 mm from the fuselage on the lower fuselage in front of the propellers to minimize effects of turbulence. The inlet lines consisted of stainless steel tubes with a bend radius of 100 mm and an inner diameter of 11.5 mm. The sample air was not heated or dried before measurement, so reported values are 15 molar mixing rations, nmol mol −1 , abbreviated "ppb", with respect to ambient humid air (Andreae et al., 2012).

Satellite and ground-based O 3 and meteorological data
In addition to the in-situ O 3 data, the model results were compared with OMI/MLS monthly mean tropospheric ozone mixing ratios and total column ozone (http://acd-20 ext.gsfc.nasa.gov/Data_services/cloud_slice/#pub) (Ziemke et al., 2006). Tropospheric values were estimated by subtracting the stratospheric contribution from total column measurements. A cloud-slicing method was used to detect O 3 inside optically thick clouds. This method was able to detect elevated O 3 levels of around 50 ppb in the upper parts of convective clouds over South America and Africa, comparable to background cloud-free levels in the tropics (Ziemke et al., 2009). The model total tropospheric O 3 column and mean tropospheric O 3 mixing ratio were calculated by summing O 3 mixing ratios, weighted by the model level air density, from the first model level to the level below the tropopause. The tropopause level was determined by the World Meteorological Organization (WMO) definition of a temperature lapse rate less than 2 K km −1 (Logan, 5 1999). The models were also compared with soundings measuring O 3 , temperature, and relative humidity conducted at sites in Paramaribo, Surinam (5. shadoz/) (Thompson et al., 2003a(Thompson et al., , b, 2007 Mean daily cycles of fluxes of sensible and latent heat and radiation were obtained from flux tower measurements for the wet (February-March 1999, January-March 2000 and dry (July-September 1999

Model description and simulation setup
Simulations of BARCA A and B were conducted with the Chemistry Coupled Aerosol-Tracer Transport model to the Brazilian developments on the Regional Atmospheric Modeling System (CCATT-BRAMS;   Soil moisture was initialized with the TRMM-based soil moisture operational product (GPNR) developed by Gevaerd and Freitas (2006). The PBL parameterization in CCATT-BRAMS is based on Mellor and Yamada (1982), while in WRF-Chem the Mellor-Yamada-Janjic (MYJ; Janjiae, 1994) scheme was used. In CCATT-BRAMS, shallow and deep convection are parameterized based on the 10 mass-flux approach of Grell and Dévényi (2002). CCATT-BRAMS also uses the Turbulent Kinetic Energy (TKE) from the Planetary Boundary Layer (PBL) scheme to determine if convection will be triggered within a grid cell. In WRF-Chem the Grell 3-D (G3) scheme was used, which includes shallow convection and subsidence spreading of convective outflow into neighboring grid cells. The Noah land surface model 15 (Koren et al., 1999) was used in WRF-Chem and the Land Ecosystem-Atmosphere Feedback model v.2 (LEAF-2; Walko et al., 2000) was utilized in CCATT-BRAMS. Land use was provided by the United States Geological Survey (USGS) global 1 km vegetation dataset, updated with a land cover map for the Brazilian Legal Amazon Region for use in meteorological models (PROVEG) (Sestini et al., 2003). PROVEG is based   Longo et al., 2009). Anthropogenic emissions were estimated from the RETRO, GOCART and EDGAR v4.0 global databases updated with South American inventories (Alonso et al., 2010). Biogenic emissions were provided by a monthly climatology for the year 2000 produced with the Model of Emissions of Gases and Aerosols from Nature (MEGAN; Guenther 5 et al., 2006). In WRF-Chem, the same Gaussian diurnal cycle with peak at 15:00 UTC (11:00 LT) is applied to both anthropogenic and biogenic emissions, while in CCATT-BRAMS the diurnal cycle of biogenic emissions follows the solar radiation cycle. In both models, the biomass burning daily cycle peaks at 18:00 UTC (15:00 LT). In both CCATT-BRAMS and WRF-Chem, the Regional Atmospheric Chemistry Mechanism (RACM) 10 was used (Stockwell et al., 1997). In WRF-Chem, the Goddard Chemistry Aerosol Radiation and Transport (GOCART; Chin et al., 2002) aerosol scheme was used with aerosol direct radiative effects. CCATT-BRAMS has a smoke aerosol scheme with intensive optical properties (extinction efficiency, single scattering albedo and asymmetry parameter) calculated in an offline Mie code based on observations of climatological 15 size distribution and complex refractive index from AERONET sites in the southern Amazon (Rosario et al., 2011(Rosario et al., , 2013. CCATT-BRAMS includes scavenging of soluble species in the convective scheme following Berge (1993), as described in Freitas et al. (2005), where the wet removal rates are a function of the precipitation rate, liquid water content and precipitable water. In the cloud microphysics scheme the wet depo-20 sition follows Barth et al. (2001), whereby low solubility species partition into the liquid phase according to Henry's Law and high solubility species by diffusion-limited mass transfer. In WRF-Chem, at the convective-parameterizing scale, a constant fraction of gas and aerosol species in convective updrafts are removed (complete removal for sulfur dioxide -SO 2 , sulfate -H 2 SO 4 , ammonium -NH 3 , nitric acid -HNO 3 and sea salt; 25 no removal for hydrophobic organic (OC) and black carbon (BC) and dimethyl sulfide (DMS); and 50 % removal for all other aerosol species). On the other hand, no wet scavenging is included for cloud water and precipitation resolved by the microphysics ACPD 14,2014 Ozone production and transport over the Amazon Basin scheme, because this option is not currently available in WRF-Chem for the RACM chemical mechanism. The CCATT-BRAMS simulations employ a lightning NO x parameterization based on convective cloud top height (Stockwell et al., 1999). In WRF-Chem, lightning production of NO x was not included, because these parameterizations have not yet been evalu-5 ated for the Amazon region. Uncertainties remain about the scavenging efficiencies of soluble species by deep convective storms. Simulations of an idealized thunderstorm by several cloud-resolving models yielded varying results for CH 2 O, H 2 O 2 and HNO 3 in convective outflow due to differing microphysics and assumptions about retention of chemical species during cloud drop freezing (Barth et al., 2007). In the tropics, 10 over continents, lightning production is comparable to other sources of NO x , including biomass burning and soil release, and it is the primary source over oceans (Bond et al., 2002). Since lightning NO x production peaks in the upper troposphere, it could be an important contributor to ozone production. The impact of wet deposition and lightning NO x production on O 3 distributions will be more closely examined in future 15 modeling studies of tropospheric chemistry in the Amazon.
For model results evaluation, the mean vertical O 3 profiles for observations, CCATT-BRAMS and WRF-Chem were calculated for the regions to the west, north, south, east, and around Manaus. Horizontal flight legs were excluded from analysis to eliminate the influence of plumes in the boundary layer. To calculate the mean simulated profiles, the 20 four grid points closest in latitude and longitude to each observation were determined at the two model hours that bracket the observations. At each of these grid points and hours, vertical profiles were extracted from the model output and then linearly interpolated to the observed GPS height. The four points from each time were averaged, weighting by the inverse distance to the observed longitude and latitude. Finally, 25 the prior and posterior hour values were averaged with appropriate weights. Thus, 16 model points were used with spatial and temporal weights to obtain each model value for comparison to observations. The observed and model time series were then separated into five regions to the west, north, east, and south of Manaus, and in the region of ACPD 14,2014 Ozone production and transport over the Amazon Basin 2-3 km, indicating the influence of small-scale fire plumes. This variability of O 3 in the PBL presents a challenge to the regional models, since the effects of small-scale processes such as plume rise and convection are parameterized and averaged across the grid cell. 20 In addition to surface emissions and chemical sources and sinks of O 3 , several meteorological processes are key drivers of tropospheric O 3 distributions, including solar radiation, tracer transport and removal. Thus, first we evaluate the ability of the models to represent these processes and their seasonalities. In the dry-to-wet transition season, a band of increased precipitation extends in TRMM 3B43 observations from the northwest Amazon to southeast Brazil, and precipitation is also intense in the ITCZ at 10 • N (Fig. 4). In the wet-to-dry transition season, increased precipitation extends from the western Amazon to the northeast coast of Brazil (Fig. 5). Incident solar radiation is higher in the dry than wet season for both 5 sites (Figs. 6-8). At the forest and pasture sites, peak sensible heat flux is higher in the dry-to-wet than wet-to-dry transition seasons, and higher at forest than pasture sites for both seasons, while latent heat flux is higher in the wet-to-dry than dry-to-wet transition seasons for both sites, and higher at the pasture site for both seasons (Figs. 7 and 8).

ACPD
In radiosoundings, a decrease in dew point temperature is observed in upper levels 10 (300-400 hPa) from 0 to 12 or 18Z, likely due to precipitation, more pronounced in the wet-to-dry transition season (Figs. 9 and 10).
Mean precipitation during the dry-to-wet (November 2008) and wet-to-dry (May 2009) transition seasons was calculated for the TRMM 3B43 data and the CCATT-BRAMS and WRF-Chem models for three regions: the Amazon ( Table 2. The mean precipitation on the 35 km resolution domain for the two months is shown in Figs. 3 and 4, respectively, as well as the delineations of the subregion boxes. The signs of NE-SE differences are correctly modeled by both models during both seasons, i.e., the NE is drier than the SE 20 during November and vice-versa during May. For the Amazon, CCATT-BRAMS slightly underestimates the precipitation rates in both seasons, but the rate in WRF-Chem is about twice that of TRMM 3B43. The models were also evaluated against TRMM 3B42 3 hourly precipitation rates at the 78 surface station locations in the Amazon (Table 3). Both models had a positive bias in both seasons, but WRF-Chem had a higher bias 25 and RMSE than CCATT-BRAMS.
In the dry-to-wet transition season, for both CCATT-BRAMS and WRF-Chem, the mean daily cycle of surface incident shortwave radiation calculated for the Manaus AERONET site for October-November 2008, falls within one standard deviation of Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the mean AERONET observations (Fig. 6), but is closer to the upper limit, possibly due to underestimated cloudiness or AOD in the models. For the forest and pasture sites, both models represent the daily cycles of incident shortwave and ingoing and outgoing longwave radiation, although incident shortwave is slightly overestimated (by 50-100 W m −2 ) at peak (Fig. 7). During the wet-to-dry transition season, both models 5 overestimate peak incident shortwave radiation by about 100 W m −2 (Fig. 8), suggesting that they underestimate cloudiness.
In the dry-to-wet transition season (Fig. 7), the peak latent heat flux at 13:00 LT is higher at the forest site than at the pasture site (460 W m −2 vs. 268 W m −2 ) whereas the sensible heat flux shows an opposite difference (180 vs. 215 W m −2 ), due to lower 10 evapotranspiration and higher surface temperatures in the pasture. As a result, the observed Bowen ratio (sensible/latent heat flux) is lower at the forest site than the pasture site (0.23-0.38 vs. 0.8). However, in WRF-Chem, the Bowen ratio at 13:00 LT shows a smaller contrast between the forest and pasture sites (0.40 vs. 0.51), due to underestimated sensible heat flux at the pasture site. In the wet-to-dry transition 15 season (Fig. 8), as for the dry-to-wet transition, peak latent heat flux at 13:00 LT is higher at the forest site than at the pasture site (401 W m −2 vs. 324 W m −2 ). However, the sensible heat flux is also higher at the pasture site (119 W m −2 vs. 168 W m −2 ) and Bowen ratio is lower at both forest and pasture sites for this season ( Figs. 9 and 10. For BARCA A, while the temperature profile is well represented by the models, the dew point temperature in CCATT-BRAMS is approxi-25 mately 10 K too high above 500 hPa and 5 K too low below 500 hPa. The wind speed is overestimated by both models above 500 hPa and underestimated below 500 hPa. For BARCA B, dew point temperature is about 5 K too high in CCATT-BRAMS above 500 hPa. Wind speed is about 2 m s −1 too low above 600 hPa in both models. The models were evaluated against data from 26 METAR (airports) and 52 Synop (INPE) surface meteorological stations, whose locations are depicted in Fig. 11. Values of Root Mean Squared Error (RMSE) and bias for various meteorological parameters for CCATT-BRAMS and WRF-Chem simulations for BARCA A (October-November 2008) and BARCA B (April-May 2009) are shown in Table 3. Both mod-5 els overestimate precipitation on average, with a RMSE of 2.4-3.0 mm h −1 and bias of 0.3-3.5 mm h −1 for CCATT-BRAMS, and RMSE of 4.5-7.1 mm h −1 and bias of 3.5-5.8 mm h −1 for WRF-Chem. Dew point temperature is underestimated by 1-2 K and surface pressure is underestimated by 1-2 hPa. Wind speed is underestimated by CCATT-BRAMS and overestimated by WRF-Chem by 0.1-0.6 m s −1 . Temperature is 10 underestimated in all cases by 0.1-2.4 K except by CCATT-BRAMS during BARCA A, which overestimated temperature by about 1 K. Now we summarize the key findings of the model-data meteorological comparison and their implications for the chemistry simulations. The models capture the seasonal spatial distribution of precipitation over northern South America, although the mean 15 precipitation rates are slightly lower (CCATT-BRAMS) and substantially higher (WRF-Chem) than the TRMM retrievals in the Amazon region. This may indicate errors in the strength and vertical distribution of convective transport and the amount of convective wet removal. Peak shortwave radiation is slightly overestimated by both models, which may be related to low cloudiness (convection is triggering too early) or AOD 20 (too much aerosol wet removal). This will increase O 3 production from photolysis, as well as increase surface temperature and evaporation. Although biogenic emissions are not coupled with meteorology in these simulations, this may increase biogenic emissions in future studies that include online biogenic emissions. WRF-Chem predicts a nearly constant Bowen ratio at forest and pasture sites. This indicates that the 25 Noah land surface model is not properly representing the impact of conversion of forest to pasture and the resulting increase in sensible heat flux. The models generally show good agreement with soundings, but excess moisture in CCATT-BRAMS above 500 hPa may stimulate excess O 3 production. Despite these limitations, the models are able to capture the meteorological contrast between the dry-to-wet and wet-to-dry transition seasons.

Observed and simulated chemistry
Lower levels of O 3 in the rainy season are largely associated with increased presence of convective clouds and precipitation. Decreased surface temperatures and incident 5 solar radiation due to cloudiness suppress emissions of biogenic VOCs such as isoprene. In addition, higher surface humidity and precipitation decrease the occurrence of fires that emit NO x and VOCs. O 3 precursors are further decreased by wet removal within the storms. On the other hand, during the dry-to-wet transition season, increased solar radiation, latent heat and temperature stimulate the production of OH and other 10 HO x radicals that can stimulate net O 3 production. For BARCA, the simulated mean monthly emission rates for two O 3 precursors, NO x (anthropogenic and biomass burning) and isoprene (biogenic) are shown in Fig. 12. In November 2008, elevated NO x emission rates of up to 5 × 10 −5 kg m −2 d −1 are seen from an area of intense biomass burning in the northeast Amazon, as well as from more 15 scattered fires in the southeast Amazon. These are both regions that were overflown by the aircraft (Fig. 12). In May 2009, the Amazon region is largely free of fire. Because biogenic NO emissions (e.g., from soil) were not included in the MEGAN climatology used in this study, background NO emissions (in absence of fire) are likely too low. Typical model anthropogenic NO x emissions values over the Amazon, primarily from 20 biofuel sources, were 0.008-13 µg N m −2 h −1 N. These NO x emissions included in the models were less than one third of the mean values of 44 ± 14 µg N m −2 h −1 NO measured by Kaplan et al. (1988) during ABLE-2A. This is considered a threshold value for NO x -driven O 3 production to be the dominant O 3 source in the PBL. The model emissions were also much lower than the mean emissions from forest of 35.8 µg N m

15
These values are also seen along the northern Andes and southeast Brazil, due to recirculation of O 3 -rich air. In the wet-to-dry transition season, O 3 deposition is at a minimum in the western Amazon, with values of 0.5-1 nmol m −2 s −1 for CCATT-BRAMS and 2 nmol m −2 s −1 for WRF-Chem. For both models, deposition velocities are higher over the rainforest than in the savanna to the east or south of the Amazon Basin, and 20 higher in the wet-to-dry transition than in the dry-to-wet transition. These patterns are also seen in the tower observations in Table 4. O 3 surface fluxes and dry deposition velocities predicted by the models were compared with observations from several field campaigns (Table 4). These include during the dry (May 1999) and wet (September-October 1999) seasons at Reserva Biológica 25 Jarú (RBJ, forest) and Fazenda Nossa Senhora (FNS, pasture) from LBA-EUSTACH (Rummel et al., 2009;Kirkman et al., 2002) and during the wet season at Reserva Ducke (RD, forest tower near Manaus,2.95 • S, 59.95 • W) from ABLE 2B (April-May 1987) (Fan et al., 1990) and at FNS from LBA-TRMM (January-February 1999) (Sigler ACPD 14, 14005-14070, 2014 Ozone production and transport over the Amazon Basin  , 2002). For the observations, the means of the hourly (WRF-Chem) and 3 hourly (CCATT-BRAMS) O 3 dry deposition fluxes (nmol m −2 s −1 ) and the medians of the daytime (11:00-21:00 UTC) hourly mean deposition velocities (cm s −1 ) are shown. The values were extracted from the grid points closest to the tower locations. In the observations, O 3 fluxes are larger in the dry season, due to higher O 3 mixing ratios. However, 5 deposition velocities are higher in the wet season, and O 3 deposition to the Amazon forest constitutes a globally significant O 3 sink (Rummel et al., 2009). Both models capture these patterns, but the models underestimate the deposition velocities by 15-75 %, which may be partially responsible for the low O 3 fluxes at the Jarú forest site in both seasons and the pasture site in the dry season. the overall distribution (inflow in NE Brazil in November 2008, lower values over the Amazon Basin, elevated at 30 • S) but values are underestimated relative to OMI/MLS. However, the BARCA observations are generally lower than the models in the boundary layer, indicating that the satellites appear here to be dominated by the middle troposphere and long-range transport. An example of observed and simulated O 3 during 5 the flight legs between Manaus and Belém in BARCA A and B is shown in Fig. 17. While the models capture the pattern of increasing O 3 values with height, the models underestimate elevated O 3 values from 2.5-4.5 km, and overestimate low values near the surface (1-1.5 km). The models also do not reproduce the variability in the high values, likely due to the aircraft intersection of biomass burning plumes. This is expected 10 given the coarse horizontal grid resolution. Thus, mean profiles are analyzed in order to study differences among the regions and seasons and to assess the models' abilities to capture the impacts of such small-scale processes on regional O 3 distributions.
The mean vertical O 3 profiles for observations, CCATT-BRAMS and WRF-Chem for the regions to the west, north, south, east and around Manaus are shown for BARCA 15 A and B in Figs. 18 and 20, respectively, and NO profiles corresponding to the aircraft tracks are depicted in Figs. 19 and 21, respectively. Mean profiles from longitudinal surveys over Amazônia of O 3 during ABLE-2A  and ABLE-2B (Harriss et al., 1990) and NO during ABLE-2A (Torres and Buchan, 1988)  Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | region, where they are lower (5-15 ppb). These results suggest an increasing influence of fire emissions to the east and south of Manaus, but that O 3 in clean regions has not changed much. Both models generally overestimate O 3 from 1-2 km and underestimate O 3 from 3-4.5 km. As seen in the CO results shown in Andreae et al. (2012), the model pro-5 files have steeper slopes than the observations, except in the polluted south, possibly due to excessive vertical mixing of precursors. In addition, the models may be missing sources of O 3 and/or precursors at 3-4.5 km in the model inflow boundary conditions. In general the models overestimate O 3 in the PBL compared to aircraft measurements, but underestimate the total column values relative to the OMI/MLS satellite product. 10 This suggests that the total column values in Amazonia are dominated by global pollution from Africa, rather than local O 3 production from biomass burning. A typical OMI averaging kernel (cloud-free ocean conditions) shows maximum sensitivity from 594-416 hPa (Zhang et al., 2010). Therefore, OMI may not be detecting O 3 in the PBL from local sources, but rather primarily seeing global pollution from Africa. 15 The excess O 3 in the PBL in the models could be due to either a low deposition sink, as O 3 dry deposition velocities in the models are about half of observed values, or excessive model sensitivity to NO x emissions, or both. Two additional simulations were conducted with WRF-Chem to evaluate the model sensitivity to these processes: (1) doubling the calculated deposition velocity for O 3 only (2DEPVEL) and (2) halving 20 the NO x surface emission rates (0.5ENOx). The O 3 profiles corresponding to BARCA flights for these two simulations are also included in Figs. 18 and 20. The corresponding NO profiles from all model simulations as well as a mean profile over Amazônia from ABLE-2A are depicted in Figs. 19 and 21. The 0.5ENOx simulation reduces O 3 more than 2DEPVEL throughout the entire profile. In the dry-to-wet transition, 2DE-25 PVEL reduces O 3 in the lower PBL by about 25 %, while 0.5ENOx decreases O 3 by around 40 %, and in the wet-to-dry-transition the reductions are about 10 and 30 %, respectively. In general the 0.5ENOx O 3 profiles are lower than observed in the first 500 m above the surface, but they provide the best representation of the data for the north and west regions in the dry-to-wet transition. They also provide a similarly good fit as 2DEPVEL for the east, Manaus and south regions, while in the wet-to-dry transition 0.5ENOx is closer to the observed value from 0-500 m in all regions except the north. During BARCA A, NO in all WRF-Chem simulations in the north, west, and Manaus regions is 10-15 ppt from 0-500 m above the surface, increasing to a maxi-5 mum of 20-50 ppt at 2 km a.g.l., and is generally lower than the ABLE-2A observations in the PBL. In the east and south, where biomass burning influence was seen, NO in 0-500 m a.g.l. increased from 20-50 ppt in the base simulation to 35-60 ppt in 2DE-PVEL due to decreased O 3 and conversion of NO to NO 2 , and was generally within one standard deviation of the ABLE-2A measurements in the PBL. In BARCA B, NO 10 simulated by WRF-Chem is very low, 5-10 ppt in the entire profile, except for the west region, where a mean NO of 30 ppt is seen from 0-500 m a.g.l. This is again due to very low O 3 , and for the Manaus region, where anthropogenic NO x sources may have contributed to NO values of 20 ppt. These results suggest that adjustment of dry deposition parameterizations are needed to increase O 3 deposition velocities by about 15 a factor of two in agreement with ground observations. Future research will compare simulated NO x fields with observations from more recent field campaigns, as the results of these simulations also suggest that O 3 in WRF-Chem is very sensitive to NO x emissions. Above the boundary layer, from 3-4 km a.g.l., chemical inflow at the eastern bound-20 ary of South America may contribute to O 3 elevated above background. In order to evaluate the model representation of this inflow, vertical profiles from SHADOZ soundings on the northeast coast of South America during the BARCA A and B periods were compared with CCATT-BRAMS and WRF-Chem (Fig. 22) . A previous analysis of ozone soundings and aircraft measurements at Natal suggested that increases in tropospheric ozone in the Southern Hemisphere springtime may be due to stratospheric intrusion (Logan, 1985). This is consistent with the November 2008 profiles at Natal; the models may not be capturing the intrusion of stratospheric air masses seen in the observations, indicated by 15 upper tropospheric (> 500 hPa) layers with elevated O 3 and very low relative humidity (< 20 %). On the other hand, at Paramaribo on November 6 and November 25, 2008 and at Paramaribo on 4 May 2009, when air masses at all levels were of Northern Hemisphere origin, the models reproduced the nearly constant with altitude O 3 values of 30-40 ppb. 20 In summary, chemistry simulations of the BARCA periods with CCATT-BRAMS and WRF-Chem overestimated O 3 in the PBL by 5-10 ppb in the wet-to-dry transition (BARCA B), with background levels observed (10-20 ppb) in all regions. In the dryto-wet transition (BARCA A), the models generally reproduced elevated O 3 levels in the northeast and southeast Amazon where biomass burning emissions of precursors 25 led to significant enhancements of ambient O 3 . The models overestimate O 3 in the PBL by 5-10 ppb, whereas from 2-4 km the modeled values are generally lower than observations. These discrepancies of models with observations may result from an overly-mixed (constant with altitude) profile due to overly active vertical mixing from the PBL scheme from 1-2 km or too much downward convective transport of O 3 from 2 km to the surface, as observed by Betts et al. (2002). In the lower boundary layer, the surface sink of O 3 (dry deposition) may be too low, or overestimation of NO x sources may produce too much O 3 . Additional simulations with WRF-Chem showed that O 3 in the lower boundary layer was about twice as sensitive to increases in O 3 deposition 5 velocity as reductions in NO x emissions, but both simulations achieved better agreement with observations. Although NO emissions over the forest were less than half of observed values, likely due to the lack of inclusion of soil emissions, sufficient O 3 production occurred to match or exceed aircraft observations, suggesting that the model chemistry is overly NO x -sensitive.

Conclusions
The BARCA campaign offered the first regional aircraft survey of O 3 in the Amazon Basin in both the dry-to-wet and wet-to-dry transition seasons. In both seasons, extremely low background O 3 values (< 20 ppb) were observed to the west and north of Manaus, and in the wet-to-dry transition low O 3 was also measured to the east and 15 south and in the region around Manaus. These background values are the lowest observed on Earth, due to a combination of isolation from anthropogenic and biomass burning NO x sources and O 3 deposition to the forest canopy, and the ecosystem and atmospheric chemistry is adjusted to these very low values. According to the models, the chemistry in the Amazon is very sensitive to NO x emissions from soils, so that even 20 a small overestimate of NO x emissions generates too much O 3 in the PBL. However, it is likely that the model chemistry is incorrect in the PBL, because the models have about the right amount of NO x but far too much O 3 in the PBL. Therefore, we conclude that the current model chemistry produces much more O 3 per unit NO x than the atmosphere at very low NO x , but may be about right in polluted conditions. Further Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | and the NO x emissions. In addition, simulated O 3 was lower than both the OMI/MLS total tropospheric O 3 and the BARCA observations in mid-levels, indicating that the O 3 retrieved by satellites is dominated by the middle troposphere and long-range transport and does not represent well boundary layer O 3 values. As the regional population grows in the Amazon basin, leading to increases in both urban and fire NO x sources, 5 this is indeed a big concern because PBL O 3 is lower in clean areas than the models predict, so that the change to polluted conditions is larger, and that the chemistry to define the path to higher NO x conditions is poorly represented. Future modeling studies can include more complete organic chemistry and biogenic emissions, including NO emissions from soil, as well as improved representation of lightning NO x production, 10 dry deposition, convective transport and wet scavenging processes, to address this NO x sensitivity. Additionally, future field campaigns in the Amazon that include aircraft observations of nitrogen oxides and hydrocarbons and ground-based measurements of NO flux from the forest canopy may allow better constraints on the Amazonian O 3 budget.