Isoprene chemistry in pristine and polluted Amazon environments:Eulerian and Lagrangian model frameworks and the strong bearing they have on our understanding of surface ozone and predictions of rainforest exposure to this priority pollutant

This study explores our ability to simulate the atmospheric chemistry stemming from isoprene emissions in pristine and polluted regions of the Amazon basin. We confront two atmospheric chemistry models – a global, Eulerian chemistry-climate model (UM-UKCA) and a trajectory-based Lagrangian model (CiTTyCAT) – with recent airborne measurements of atmospheric composition above the Amazon made during the SAMBBA campaign of 2012. The simulations with the two models prove relatively insensitive to the chemical mechanism employed; we explore one based on the Mainz Isoprene Mechanism, and an updated one that includes changes to the chemistry of first generation isoprene nitrates (ISON) and the regeneration of hydroxyl radicals via the formation of hydroperoxy-aldehydes (HPALDS) from hydroperoxy radicals (ISO2). In the Lagrangian model, the impact of increasing the spatial resolution of trace gas emissions employed from 3.75° × 2.5° to 0.1° × 0.1° varies from one flight to another, and from one chemical species to another. What consistently proves highly influential on our simulations, however, is the model framework itself – how the treatment of transport, and consequently mixing, differs between the two models. The lack of explicit mixing in the Lagrangian model yields variability in atmospheric composition more reminiscent of that exhibited by the measurements. In contrast, the combination of explicit (and implicit) mixing in the Eulerian model removes much of this variability but yields better agreement with the measurements overall. We therefore explore a simple treatment of mixing in the Lagrangian model that, drawing on output from the Eulerian model, offers a compromise between the two models. We use this Lagrangian/Eulerian combination, in addition to the separate Eulerian and Lagrangian models, to simulate ozone at a site in the boundary layer downwind of Manaus, Brazil. The Lagrangian/Eulerian combination predicts a value for an AOT40-like accumulated exposure metric of around 1000 ppbv h, compared to just 20 ppbv h with the Eulerian model. The model framework therefore has considerable bearing on our understanding of the frequency at which, and the duration for which, the rainforest is exposed to damaging ground-level ozone concentrations.

borne measurements of atmospheric composition above the Amazon made during the SAMBBA campaign of 2012. The simulations with the two models prove relatively insensitive to the chemical mechanism employed; we explore one based on the Mainz Isoprene Mechanism, and an updated one that includes changes to the chemistry of first generation isoprene nitrates (ISON) and the regeneration of hydroxyl radicals via the formation of hydroperoxy-aldehydes (HPALDS) from hydroperoxy radicals (ISO 2 ). In the Lagrangian model, the impact of increasing the spatial resolution of trace gas emissions employed from 3.75 • × 2.5 • to 0.1 • × 0.1 • varies from one flight to another, and from one chemical species to another. What consistently proves highly influential on our simulations, however, is the model framework itself -how the treatment 15 of transport, and consequently mixing, differs between the two models. The lack of explicit mixing in the Lagrangian model yields variability in atmospheric composition more reminiscent of that exhibited by the measurements. In contrast, the combination of explicit (and implicit) mixing in the Eulerian model removes much of this variability but yields better agreement with the measurements overall. We therefore explore

Introduction
Though not all plants emit isoprene, around 500 Tg(C) isoprene are emitted from vegetation globally each year (Guenther et al., 2006;Arneth et al., 2008). This exceeds the 5 annual carbon flux attributable to anthropogenic non-methane volatile organic compounds (NMVOCs) by a factor of ten; see, e.g., WMO (1995). Isoprene is also highly reactive, with a lifetime of the order of minutes with respect to oxidation by the hydroxyl radical (OH), compared to around 10 years in the case of methane (CH 4 ). Biogenic isoprene emissions thus have the potential to profoundly affect tropospheric OH.
Moreover, since OH is the main chemical species responsible for removing gaseous pollutants, including some potent greenhouse gases (e.g. CH 4 ), isoprene emissions have a bearing on both air quality and climate. These influences are furthered through the impact that isoprene emissions have on tropospheric ozone (O 3 ), which is both a respiratory aggravant and a greenhouse gas (see, respectively, WHO, 2000; IPCC, 15 2002). High O 3 concentrations at ground level also cause visible leaf-injury to plants, reducing the rates at which they photosynthesise and increasing their requirements of resources to detoxify and repair (e.g. Ainsworth et al., 2012); their detrimental impact on crop yields, for example, is well documented (e.g. Avnery et al., 2011). There is thus a feedback between the health of the Amazon rainforest and the composition of 20 air above it, with potentially significant implications for the tropical carbon cycle (Sitch et al., 2007;Pacifico et al., 2015); it is interesting that the chemistry in the pristine rainforest environment ("NO x limited"; see later) tends to yield low O 3 concentrations -in favour of the forest's health. However, despite these important effects, our ability to capture the influence of isoprene emissions on tropospheric OH and O 3 has been 25 in question since Lelieveld et al. (2008) and Butler et al. (2008) reported observations of unexpectedly high OH concentrations in the boundary layer over the Amazon basin iment with a Learjet campaign of 2005 (GABRIEL; see Atmospheric Chemistry and Physics Special Issue 88). High isoprene emissions were expected to sustain high isoprene concentrations whilst suppressing OH concentrations (due to the rapid reaction between the two). It was the unexpectedly high OH concentrations that led Lelieveld et al. (2008) to speculate that the chemistry models were missing a mechanism by 15 which some of the OH initially consumed in isoprene oxidation was "recycled". Meanwhile, Butler et al. (2008) explored the role that the physical separation, or "segregation", of air masses containing isoprene emissions could play in resolving the apparent paradox, as have Pugh et al. (2011) since; see later. Here, we confront two atmospheric chemistry models, and modelling frameworks (Eulerian and Lagrangian), lated using parameterisations based on the Model of Emissions of Gases and Aerosols from Nature (MEGAN; Guenther et al., 2006), with vegetation simulated offline using the Sheffield Dynamic Global Vegetation Model (SDGVM; Beerling et al., 1997;Beerling and Woodward, 2001) as described by Lathière et al. (2010). Before exploring the impact of changes in isoprene emissions, Squire et al. (2014) demonstrated that 15 UM-UKCA showed some skill at reproducing recent observations of tropospheric O 3 when employing present day emissions: they compared their simulated profiles of O 3 with sonde profiles from the Southern Hemisphere ADditional OZonesondes network (SHADOZ; Thompson et al., 2003). Notably, however, this network did not offer measurements of tropospheric O 3 above the Amazon rainforest -globally, responsible for 20 almost half of all biogenic NMVOC emissions (Guenther et al., 1995) and the greatest source of isoprene (see, e.g., Fig. 2 of Squire et al., 2014). Squire et al. (2015) then explored the sensitivity their projections of future tropospheric O 3 showed to the chemical mechanism they employed. However, they did not explore the impact of this mechanism on their ability to reproduce present day observations. 25 Here, we test the ability of (i) a nudged version of UM-UKCA and (ii) a Lagrangian model, the Cambridge Tropospheric Trajectory model of Chemistry And Transport (CiT-TyCAT; Pugh et al., 2012), to simulate SAMBBA measurements above the Amazon. In each model, we carry out (otherwise identical) integrations employing two of the four Introduction  Squire et al. (2015): the UM-UKCA Chemistry of the Troposphere (CheT), in which isoprene oxidation follows the Mainz Isoprene Mechanism (MIM; Pöschl et al., 2000); and an updated version of this mechanism (CheT2) that incorporates the recent developments in our understanding of this chemistry compiled for the UK Met Office by Jenkin (2012). CheT, being based on the MIM, contains 5 similar chemistry to the models with which Lelieveld et al. (2008) and Butler et al. (2008) were unable to simulate the simultaneously high isoprene-and OH concentrations observed during the GABRIEL campaign. Meanwhile, the updates in CheT2 include an efficient route by which OH initially consumed in isoprene oxidation may be regenerated at low ambient concentrations of nitrogen oxides (NO x = NO + NO 2 ): the formation of 10 hydroperoxy-aldehydes from hydroperoxy radicals and their subsequent rapid release of OH (Peeters et al., 2009;Crounse et al., 2011). Questions remain regarding the effect that the "segregation" of air masses containing isoprene emissions has on the chemistry ensuing therein. In large-eddy simulations of a convective boundary layer, Krol et al. (2000) found that it could reduce the 15 effective rate of reaction between OH and a generalised hydrocarbon by as much as 30 %, relative to that simulated in a box model. They obtained the largest reductions when the hydrocarbon was emitted non-uniformly and assumed to react rapidly with OH. Recall that the emissions of isoprene are not expected to be uniform, since not all plants emit isoprene, and isoprene is highly reactive towards OH. It was in this 20 context that Butler et al. (2008) explored the role that segregation could play in reconciling the simultaneously high isoprene-and OH concentrations observed during the GABRIEL campaign. They found that a 50 % reduction in the "effective rate constant" for this reaction was required, implying a high degree of segregation. Though Butler et al. (2008) recognised that the measurements were not of sufficient spatial-and 25 temporal resolution to confirm such segregation, they believed that the high degree of variability observed in isoprene concentrations rendered this not implausible. Pugh et al. (2010) subsequently explored a 50 % reduction in the rate of this reaction, in an effort to reconcile measurements from the Oxidant and Particle Photochemical Pro-Introduction   with simulations using CiTTyCAT. Later, explicitly exploring segregation using observed variability in isoprene, Pugh et al. (2011) found that segregation alone could not fully account for the difference between modelled and measured OH. CiTTy-CAT is ill-suited to exploring the effects of segregation in a convective boundary layer, 5 since the transport of each air parcel is described by a trajectory that at best captures regional-to continental scale convection (see next section). However, it does allow us to simulate, independently, the chemistry taking place in air parcels arriving at points spaced only a short time apart on a flight track, in other words assuming no mixing occurs between them; we have the option of then adding a simple treatment of mixing 10 such as relaxation towards background composition. More sophisticated treatments of mixing between ensembles of trajectories are possible using a variant of the CiTTyCAT model (see Pugh et al., 2012;Cain et al., 2012), but we begin here by focussing on the simulation of independent air parcels. This makes for an interesting comparison with simulations using UM-UKCA, which implicitly mixes air on the scale of the model grid 15 (3.75 • × 2.5 • in "climate mode"). CiTTyCAT should yield greater variability in simulated isoprene concentrations: Lagrangian box models, run in "domain-filling" mode, capture much of the strain-induced stirring of the atmosphere, with no mixing, whilst Eulerian models capture the large-scale stirring but inevitably introduce mixing as they average concentrations across grid boxes (see, e.g., Dragani et al., 2002;Methven et al., 2003). 20 The mixing of air within the models, or lack thereof, has further effects on the chemistry simulated. Most pertinent to this study is the effect of mixing on the concentration of NO x , specifically the NO x : VOC (CH 4 + NMVOCs) ratio; see, for example, the results of the Empirical Kinetics Modelling Approach of Dodge (1977) and Sillman and He (2002). At very low NO x : VOC ratios, we expect isoprene to remove O 3 (through direct 25 reaction with it); at moderate-to-high NO x : VOC ratios, OH-initiated isoprene oxidation should produce O 3 with increasing efficiency as the NO x : VOC ratio increases; until, at very high NO x : VOC ratios encountered only in the most polluted urban and industrial environments, this O 3 production is partially offset by O 3 titration. The effects, therefore, still intercepted air influenced by biomass burning upwind, but considerably less than those flights specifically aiming to probe this influence. Chemically, we focus on measured concentrations of five species central to gas-phase tropospheric chemistry: O 3 , NO, NO 2 , isoprene (C 5 H 8 ) and carbon monoxide (CO). We have no measurements of OH concentration with which to compare our simulated OH concentrations. However, 10 the comparison of measured and modelled O 3 concentrations is nonetheless valuable in the context of isoprene oxidation and its impact on OH, since: OH is derived from O 3 (by O 3 photolysis and reaction of the resulting excited O( 1 D) oxygen atoms with water vapour); and, if a model were to reproduce measurements of OH concentration but not O 3 concentration, it would be simulating OH well but for the wrong reason(s). The thick black lines in Fig. 1 illustrate the flight tracks in longitude and latitude of flights B735, B744, B745, B749 and B750. (The variation in altitude during each flight is illustrated in subsequent figures comparing measured and modelled trace gas concentrations.) Superimposed on each of these panels in Fig. 1 of air parcels backwards in time (i.e. with increasing distance from the fight tracks), which offers the Lagrangian model an opportunity to capture diversity in initial composition that may be lost or less well resolved in the Eulerian model; and the general consistency of colour along any one trajectory bound for a flight track, indicating that many air parcels have spent much of the 7 days prior to arrival on the flight track at 5 approximately the same pressure. This implies that air parcels encountered in lowaltitude portions of the flights have often spent much of the last 7 days at low altitudes and, thus, exposed to trace gas emissions (and dry deposition) in the boundary layer; air parcels encountered in high-altitude portions of the flights, on the other hand, have often spent much of the last 7 days well above the boundary layer and hence exempt 10 from these influences. The trajectories bound for the site downwind of Manaus, meanwhile, appear to descend over the course of 7 days from origins in the mid troposphere to the site in the boundary layer (900 hPa) and are thereby predominantly exposed to trace gas emissions (and dry deposition) in the latter half of their journeys. 15 O 3 was measured at 0.1 Hz by a Thermo Environmental UV absorption photometer model 49C, traceable to the UK National Physical Laboratory primary ozone standard with an uncertainty of 2 %, and a precision of 1 ppb.

NO and NO 2 measurements
NO x was measured from the aircraft using a high sensitivity NO x chemiluminescence 20 system built by Air Quality Design, Inc. The instrument has a dual channel architecture for independent quantification of NO and NO 2 , with each channel having a sample flow of 1 L min −1 . NO is measured in one channel by an established chemiluminescence technique (Lee et al., 2009) Pollack et al. (2010) evaluated the relative high NO 2 affinity for conversion of NO 2 to NO using 395 nm blue light LEDs. They highlighted the low probability of other species within the gaseous chemical matrices such as nitrous acid (HONO), being affected by the 395 nm light, so in turn reducing possible non NO 2 species interfering with the measurement. NO x was then quantified by 5 ozonation of the subsequent total NO present in the reaction vessel after conversion with NO 2 derived from the difference between NO x and NO mixing ratios. The instrument was calibrated by adding a small flow (5 sccm) of known NO concentration (5 ppmv -Air Liquide) into the ambient sample flow, resulting in around 10 ppbv of NO. The conversion efficiency of the NO 2 converter was measured during each calibration by gas phase titration of the NO to NO 2 by addition of O 3 . In flight calibrations were always carried out above the boundary layer, thus ensuring low and stable background levels of NO x . Typically calibrations are carried out at the beginning and end of a flight, with sensitivities and conversion efficiency interpolated between the two and applied to all data. Detection limits for the 10 s averaged data were ∼ 10 pptv for NO and 15 pptv for NO 2 with approximate total errors at 1 ppbv being 10 and 15 % for NO and NO 2 respectively.

CO measurements
CO was measured at 1 Hz by an Aerolaser VUV fluorescence analyser model AL5002 (Gerbig et al., 1999). The instrument was calibrated in-flight using an air standard, 20 traceable to the World Meteorological Organisation CO scale X2004, with a 2 % uncertainty, and 3 ppb precision.

C 5 H 8 measurements
Isoprene was measured using an on-board proton transfer reaction mass spectrometer containing a quadrupole detector (PTR-MS, Ionicon Analytik GmbH, Innsbruck, Aus-Introduction bons and oxygenated hydrocarbons with a typical cycle time of 3-5 s. Isoprene mixing ratios were determined using a dynamically-diluted calibrated gas standard (∼ 500 ppb in nitrogen, uncertainty ±5 %, Apel-Reimer, Boulder CO). For the purposes of this comparison we have applied a 15-point smoothing function to the high frequency data to give an approximately 1 min moving averaged mixing ratio. The mean limit of detection 5 for isoprene under these conditions was 110 ppt. The overall measurement uncertainty is estimated to be ±15 %. Full instrumental, operational and calibration details are described in Murphy et al. (2010). Additionally, whole air samples (WAS) were collected on flights B735 and B749, and subsequently analysed to measure isoprene mixing ratios. The WAS system, described in greater detail by Lidster et al. (2014), comprises sixty four canisters with fused silica deactivated inner surfaces, each with three litre internal volume. Individual canisters were filled at operator-determined times using a double-headed metal bellows pump (all stainless steel components) to a final pressure of up to 40 psi and shipped back to the UK for analysis within one month of collection. Analysis was performed using 15 a dual channel gas chromatograph with flame ionisation detectors, described in detail by Hopkins et al. (2011), which was calibrated using a certified standard supplied by the National Physical Laboratory (Ozone precursors mix, cylinder number D641613). Detection limits were in the single parts per trillion range with typical calculated uncertainties of between 3 and 20 %. For full details, the reader is referred to Squire et al. (2014). Here, we simply note that the model was run in "climate mode" -at a relatively low spatial resolution, N48 L60 (3.75 • longitude × 2.5 • latitude; 60 hybrid height levels stretching from the surface to around 84 km) -and employed the standard tropospheric chemistry mechanism, 5 CheT. This is the setup that Squire et al. (2015) subsequently used in their "BASE CheT" experiment. Their "BASE CheT2" experimental setup was identical except for employing the updated CheT2 chemistry. Here, we carry out two integrations with UM-

CiTTyCAT (Lagrangian model)
CiTTyCAT r4.2.1 (Pugh et al., 2012) is a Lagrangian model of atmospheric chemistry and transport, stemming from the Cambridge Tropospheric Trajectory model of Chemistry And Transport (Wild et al., 1996). This is not the first time that CiTTyCAT has been used to simulate atmospheric chemistry and composition over a tropical rainfor-5 est: Pugh et al. (2010), as briefly referred to in the introduction, tested the performance of the model in two-box mode (two boxes, to account for the nocturnal collapse of the boundary layer and development of a residual layer above it), confronting it with measurements made during the OP3 campaign at Danum Valley, Malaysian Borneo. We use the model in single trajectory mode (moving a single model box along one tra-10 jectory at a time) many times over as we loop over all back-trajectories bound for (a) the arrival points spaced one minute apart on the five SAMBBA flights, and (b) the receptor site downwind of Manaus at 6 hourly intervals throughout September 2012. The single trajectory mode has been used extensively in previous studies of long range transport (see, e.g., Wild et al., 1996;Evans et al., 2000;Real et al., 2007Real et al., , 2008. Note 15 that the treatment of transport constitutes the main difference between CiTTyCAT and UM-UKCA: transport in the Lagrangian framework is described by discrete trajectories (series of times and locations) calculated offline, as opposed to fluxes between adjacent model boxes in a fixed 3-D Eulerian grid. The back-trajectories, illustrated in Fig. 1, are calculated using ROTRAJ (Methven,20 1997) in conjunction with ECMWF ERA Interim analyses, as previously outlined by Pugh et al. (2012). The analysed wind fields, available at 6 hourly intervals (00:00 UT, 06:00 UT, 12:00 UT and 18:00 UT) are interpolated linearly in space and time. The location of each trajectory is then calculated by integrating the interpolated wind velocities with respect to time according to the fourth order Runge-Kutta method (Methven, 1997 To ensure the transport in the two models is broadly consistent, we use the same analyses to calculate the trajectories as we use to nudge UM-UKCA (see above). However, two key differences remain. Firstly, the trajectory calculations exploit the full-resolution of the analysed winds (roughly 0.7 • × 0.7 • ) whilst UM-UKCA is nudged towards these winds following degradation to the resolution of its Eulerian grid 5 (3.75 • × 2.5 • in "climate mode"). The transport in CiTTyCAT is therefore more finely resolved and should yield greater structure in the composition of air it simulates along each flight track, and downwind of Manaus, particularly when combined with high resolution trace gas emissions. The transport in CiTTyCAT, however, only includes convection as captured by the analyses (i.e. large-scale convection) whilst UM-UKCA explicitly 10 adds updrafts and downdrafts associated with convection on smaller scales, following Gregory and Rowntree (1990) and Gregory and Allen (1991). CiTTyCAT therefore lacks a certain amount of vertical mixing. Some mixing within the boundary layer is included implicitly, as the addition of emissions (conversion from mass fluxes to enhancements in concentration) depends on a length scale associated with the height of the boundary 15 layer, but no ventilation of the boundary layer or exchange with the free troposphere is included. We focus first on the simulation of independent air parcels -with no vertical (or horizontal) mixing -to explore the influence of contrasting air parcel histories on the chemistry ensuing therein. However, we subsequently explore the sensitivity of some of our results to a simple treatment of diffusive vertical mixing. 20 The treatment of diffusive vertical mixing, described by Pugh et al. (2012), comprises relaxation towards background composition at rates specified by free-troposphere and boundary-layer diffusion coefficients, κ FT and κ BL . These yield relaxation timescales of  15,2015 Isoprene chemistry in pristine and polluted Amazon environments J. G. Levine et al. Mix3) that span the ranges suggested by Pugh et al. (2012); these are given in Table 1 together with the τ FT and τ BL they yield.  Table 2. We simulate the measurements made on the five SAMBBA flights (B735, B744, B745, B749 and B750) subject to setups 1-4, exploring the impact of different chemical mechanisms (CheT and CheT2; described in more detail in the next section) and different resolutions of trace gas emissions ("UKCA res" and "High res"; described in Sect. 2.4). With setups 5 and 6, we then explore the sensitivity of 15 our simulations of flight B735 to the inclusion of diffusive vertical mixing as described above. We do so in order to assess to what extent differences between simulations with CiTTyCAT and UM-UKCA can be thus reduced; setups 5 and 6 therefore employ UKCA res emissions. Finally, we model the composition of the atmosphere downwind of Manaus throughout September 2012 subject to setups 3-8. Our aim here is to capture the 20 episodic influence of anthropogenic emissions from Manaus, and we therefore chiefly adopt setups employing High res emissions (3, 4, 7 and 8). We continue, however, to explore the sensitivity of our results to emission resolution, hence including setups 5 and 6, in addition to both the choice of chemical mechanism and the inclusion/exclusion of vertical mixing. 25 To reduce the differences between the two models, aside from those intrinsic to their different frameworks, we ensure that precisely the same chemical mechanisms and chemical reaction rate coefficients are used in CiTTyCAT as in UM-UKCA (Squire et al., 2014(Squire et al., , 2015. Likewise, we employ the same dry deposition velocities and Henry ACPD 15,2015 Isoprene chemistry in pristine and polluted Amazon environments J. G. Levine et al. coefficients in the two models (Squire et al., 2014(Squire et al., , 2015, and we use 3-D fields of precipitation, output from UM-UKCA every 20 min timestep, to drive the wet deposition in CiTTyCAT. This is in addition to initialising the composition of air parcels in CiTTy-CAT with the concentrations of species simulated in UM-UKCA (subject to the same chemical mechanism) as described at the end of the last section.

CheT and CheT2 chemical mechanisms
The standard tropospheric chemistry mechanism, CheT, includes 56 chemical tracers and 165 photochemical reactions, of which 16 tracers and 44 reactions comprise the MIM (Pöschl et al., 2000). It is the result of a systematic reduction of version 2 of the Master Chemical Mechanism (MCM; Jenkin et al., 1997), in which species are lumped together based on their structure, for example all hydroxyperoxy radicals as "ISO 2 ". CheT2 differs only with respect to isoprene oxidation, with 24 tracers and 59 reactions in place of the previous 16 and 44 respectively, and is traceable to MCM version 3.2 (MCMv3.2). The differences, reflecting the updates compiled by Jenkin (2012)  3. The inclusion of the formation of isoprene epoxydiols (IEPOX) from the oxidation of isoprene hydroxyl-hydroperoxides (ISOOH); Paulot et al. (2009) identified these as a potential source of secondary organic aerosols.

4.
A reduction in the yield of peroxymethacrylic nitric anhydride (MPAN) from isoprene oxidation relative to that adopted in CheT; see Jenkin (2012) for details.

5
In this study, however, we are less concerned with the differences between the two mechanisms, which have already been explored at length (see, e.g., Archibald et al., 2010a, b;Squire et al., 2015), than we are with their relative abilities to reproduce observations of atmospheric composition above the Amazon rainforest -and the latter subject to different model frameworks (Eulerian and Lagrangian) and trace-gas emis-10 sions.

Trace gas emissions
The trace gas emissions are comprised of: anthropogenic emissions taken from EDGAR version 4.2 (http://edgar.jrc.ec.europa.eu); and biogenic emissions calculated with the Organising Carbon and Hydrology In Dynamic Ecosystems land surface model 15 (ORCHIDEE), with the exception of NO 2 emissions from soils that are taken from the Global Emissions Inventory Activity (GEIA; Yienger and Levy, 1995). The annual total emission of each species, globally, is given in seasonality to be relatively low in the tropics. The NMVOC emissions come lumped together as a single carbon flux. We derive emissions of ethane (C 2 H 6 ), propane (C 3 H 8 ), formaldehyde (HCHO), acetone (CH 3 C(O)CH 3 ) and acetaldehyde (CH 3 CHO) from this using the IPCC (2002)'s speciation of industrial-and biomass burning emissions; see their Table 4.7(b). This speciation is crude: we assume, for example, that "ketones" are 5 entirely comprised of CH 3 C(O)CH 3 , and "other aldehydes" solely CH 3 CHO. However, our priority is to start from anthropogenic emissions of sufficient spatial resolution to resolve the city of Manaus, and EDGAR 4.2 is unique in providing emissions of this resolution, globally. The CiTTyCAT integrations employing "High res" emissions exploit their full 0.1 • ×0.1 • resolution. For use in UM-UKCA, and the CiTTyCAT integrations employing "UKCA res" emissions, the emissions are degraded to 3.75 • × 2.5 • ; see Table 2 and accompanying text. As stated above, biogenic emissions of C 5 H 8 , HCHO, CH 3 C(O)CH 3 and CH 3 CHO are calculated with ORCHIDEE. This includes parameterisations based on Guenther et al. (1995) and Lathière et al. (2006), modified according to Guenther et al. (2012) 15 and more recent findings to take into account the progress of our knowledge in this field (see Messina et al., 2015). ORCHIDEE is forced with 2012 National Centers for Environmental Prediction meteorological analyses (NCEP v5.3) from the Climatic Research Unit of the US National Centre for Atmospheric Research. These daily (24 h average) emissions are used at full spatial resolution (0.5 • × 0.5 • ) in the "High res" integrations 20 with CiTTyCAT but, just as for the anthropogenic emissions, degraded to 3.75 • × 2.5 • for use in UM-UKCA and the CiTTyCAT integrations employing "UKCA res" emissions. We apply a diurnal cycle -the same in both models, based on the division of each 24 h period into 20 min intervals -to the emissions of C 5 H 8 but not HCHO, CH 3 C(O)CH 3 or CH 3 CHO. The NO 2 emissions from soils are taken from GEIA dataset, soilNOXmn1.1a 25 (Yienger and Levy, 1995). In view of the uncertainty in these, they are used in both models, in all integrations, at a resolution of 3.75 • × 2.5 • . Figure 2 illustrates the total "UKCA res" emissions of each species (anthropogenic + biogenic) on the 1 January and 1 July 2012. Recall, only the species including a bio-  Fig. 3 compares the monthly mean total "UKCA res" and "High res" emissions of each species encountered in September 2012, over the same domain as the flight tracks and back-trajectories are illustrated in Fig. 1; note that the influence of the low resolution NO 2 emissions from soils are still visible in the "High res" NO 2 emissions here.

Comparing Eulerian and Lagrangian model frameworks
We  5,13.5,14.5 and 15.5 UT;08:30,09:30,10:30 and 11:30 LT). This is consistent with the air parcels during those periods generally not encountering the boundary layer (altitudes of less than 1000-2000 m) during the previous 7 days; see Fig. 1 and accompanying text. Since these air parcels are subject to neither emissions nor dry deposition, the small differences between the green and dashed-blue 5 lines are due to intervening chemistry and/or wet deposition. The [O 3 ] simulated with CiTTyCAT also shows much more structure in these regions than that simulated with UM-UKCA. The initialisation is responsible for this additional structure -the result of divergent air parcel histories (see Sect. 2.1.1 and, again, Fig. 1) -note that it is largely retained in CiTTyCAT over the course of 7 days but almost entirely lost in UM-UKCA. 10 Compared to the measurements, CiTTyCAT shows too much structure (perhaps due to the lack of explicit mixing in the model) whist UM-UKCA shows too little (presumably due to mixing on the scale of its 3. The measurements of [C 5 H 8 ] reflect its short lifetime with respect to oxidation by OH, reaching up to around 10 ppbv low down in the boundary layer (close to its sources) but swiftly decreasing with increasing altitude; see Fig. 6. Both models capture the rapid decrease in [C 5 H 8 ] with altitude but both tend to overestimate [C 5 H 8 ] at low altitudes with respect to the measurements, most particularly CiTTyCAT. It is possible 5 that the C 5 H 8 emissions we employ are too high but, in terms of their global annual total, they lie towards the lower end of literature estimates: 354 Tg(C) year −1 compared to 300-600 Tg(C) year −1 (e.g., Guenther et al., 2006;Arneth et al., 2008). Moreover, year-round measurements of C (CheT), so the more marked overestimation in CiTTyCAT is at least partly due to some other reason. Again, this difference between the two models extends across all five flights (see Figs. S1-S4) and we suggest that it is consistent with a lack of vertical mixing in CiTTyCAT: air parcels bound for low altitude portions of the flight are exposed to C 5 H 8 emissions 25 in the boundary layer but subject to no ventilation of the boundary layer and, hence, exchange with free tropospheric air of lower, if not zero, [C 5 H 8 ]. In Sect. 3.4, we will explore the impact of introducing a simple treatment of diffusive vertical mixing in CiT-TyCAT. In intervening sections, however, we first explore the effects of a change of Introduction

Comparing chemical mechanisms (CheT and CheT2)
We find that the chemical mechanism employed has a negligible effect on the  Perring et al. (2009) and the addition of O 3 -initiated ISON degradation (Lockwood et al., 2010)  CiTTyCAT are also greater for the latter two flights, and [O 3 ] demonstrates sensitivity to the chemical mechanism employed during some portions of flight B750 too. The difference in [C 5 H 8 ] simulated with CiTTyCAT and UM-UKCA, however, remains large, and we remain interested in the impact of vertical mixing. 25 Switching from "UKCA res" to "High res" emissions also has negligible effect on the [O 3 ] we simulate for flight B735 in CiTTyCAT, everywhere but the beginning and end of the flight; see Fig. 9 High res emissions (irrespective of the chemical mechanism employed) is presumably attributable to the latter's greater ability to resolve elevated emissions of O 3 precursors associated with the respective cities/airports (Porto Velho and Manaus). As with the change of chemical mechanism explored in the last section, the increase in emission resolution has only a modest effect of the [NO], [NO 2 ] and [CO] we simulate, and is 5 limited to the low altitude portions of the flight -in which the corresponding air parcels have spent much of the last 7 days in the boundary layer exposed to emissions. Whilst the High res emissions yield lower [C 5 H 8 ] (again irrespective of the chemical mechanism employed), CiTTyCAT continues to greatly overestimate [C 5 H 8 ] relative to the measurements and the difference in [C 5 H 8 ] simulated with the two models is reduced 10 but not removed; for clarity, the [C 5 H 8 ] simulated with UM-UKCA is not included in Fig. 9 (please refer back to Fig. 8).

Comparing trace gas emissions of different resolutions
A different story emerges for the other four flights; see Figs. S9-S12. [C 5 H 8 ], simulated subject to both CheT and CheT2 chemistries, decreases markedly: practically to zero on flights B744 and B745 (Figs. S9 and S10), now actually underestimating the 15 measurements; and, though still overestimating, much closer in line with the measurements on flights B749 and B750 (Figs. S11 and S12). Furthermore, the [O 3 ] simulated, again subject to both CheT and CheT2 chemistries, changes considerably. Notably, at mid-to-high altitudes, CiTTyCAT's overestimation of the measurements is much reduced on flights B744, B749 and B750 (Figs. S9, S10 and S12) and largely removed 20 on flight B745 (Fig. S10). So, in this one regard -the insensitivity our simulations show to the resolution of trace gas emissions employed -flight B735 appears to be an exception. Remember, however, that we explore the impact of emission resolution in the Lagrangian environment alone, where we can exploit a resolution of 0. We observe this behaviour across all five case study flights, and it is this central finding, which appears to result from the choice of model framework alone, on which we mean to focus from here on and speculate is the result of differences in vertical mixing.

Exploring sensitivity to vertical mixing in CiTTyCAT
To explore the impact of introducing a simple treatment of diffusive vertical mixing in 5 the Lagrangian framework, we return to flight B735, employing "UKCA res" emissions and CheT chemistry. We explore three formulations of mixing, or relaxation, as outlined in Sect. 2.2.2 and , the inclusion of this relaxation, subject to all three formulations (Mix1-3), brings the concentrations simulated with CiTTyCAT, originally with no mixing (dotted green lines), much closer in line with those simulated with UM-UKCA (dashed blue lines). It would therefore appear that vertical mixing (or the lack thereof) has po-20 tential to explain some of the differences observed between the two models/model frameworks.
Of course, the close agreement between UM-UKCA and CiTTyCAT, on including mixing in the latter, is only to be expected for [O 3 ], since we relax the [O 3 ] simulated with CiTTyCAT towards monthly mean values simulated with UM-UKCA (see Sect. 2.2.2). 25 We do likewise for other species of intermediate lifetimes: CO, C 2 H 6 , C 3 H 8 and PAN. Meanwhile, the concentrations of all short-lived species, including C 5 H 8 , are relaxed towards zero concentrations -characteristic of free tropospheric air. The most rapid 24278 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | relaxation, Mix1, yields the best agreement between modelled and measured [C 5 H 8 ]. Indeed, the agreement is excellent. However, Mix1 yields the worst agreement between modelled and measured [O 3 ] in the low altitude portions of the flight -in these regions, considerably worse than the simulation without mixing, consistently overestimating the measurements by 15-20 ppbv in absolute terms, and close to 100 % in relative terms.

5
The slower relaxations, Mix 2 and Mix 3, yield somewhat higher [C 5 H 8 ] -greater than that measured but a significant improvement over that simulated with CiTTyCAT without mixing, and better in two out of three portions of the flight than that simulated with UM-UKCA. Mix2 and Mix3, meanwhile, yield better agreement between modelled and measured [O 3 ] in the low altitude parts of the flight. Furthermore, Mix3 starts to retain  Table 1), we judge Mix3 to be the most appropriate overall.

Modelling atmospheric chemistry downwind of Manaus
We now move from what has predominantly been an exploration in the spatial do-20 main -in other words, comparing modelled and measured trace gas concentrations on flight tracks -to an exploration of the temporal domain.  (Ainsworth et al., 2012). 15 Below, we discuss an AOT40-like exposure metric based on a 30 day time series, and refer to the metric as "AOT40" to remind readers of the non-standard accumulation period we use. Our study is not designed to calculate annual exposure metrics, but simply to highlight the sensitivity of exposure metrics to the modelling method used (where measurements are not available). In view of the sensitivity that our simulations 20 with CiTTyCAT show to the inclusion of a simple treatment of diffusive vertical mixing (relaxation towards background composition; see previous section), we simulate the [O 3 ] downwind of Manaus with and without relaxation formulations, Mix1-3. We start, as before, by using CheT chemistry but employ High res emissions (0.1 in an attempt to resolve the episodic influence of anthropogenic emissions from the city 25 100 km upwind. We compare the results of these integrations with the [O 3 ] simulated with UM-UKCA in "climate mode" (employing anthropogenic emissions at 3.75 • × 2.5 • ) in the top left of Fig. 11; the corresponding "box and whisker" plots of the absolute min- pending on the speed of relaxation imposed, this structure is suppressed to a greater or lesser extent, and the simulations with CiTTyCAT (solid red, blue and green lines) can generate more or less variability in the time series than UM-UKCA. Mixing formulation, Mix3, judged in the last section to yield best agreement between modelled and measured [O 3 ] over a range of altitudes, including specifically low altitudes, yields a dis-10 tribution of [O 3 ] that has a higher median value than UM-UKCA (34.0 cf. 31.7 ppbv), a higher 75th percentile (38.3 cf. 34.9 ppbv), and a higher absolute maximum ( ozone metrics, such as POD Y (see above), we would expect a less marked difference in the latter; it would depend, however, on how peak [O 3 ] values correlate with turbulence and plant physiology (e.g. stomatal opening). The broader message is that all three metrics investigated here change on moving from a global chemistry-climate model run in "climate mode" to a combination of 5 this model and a Lagrangian model capable of exploiting very high resolution anthropogenic emissions and retaining compositional structure. The combination of models may yield a not-dissimilar background exposure, but one punctuated episodically with acute, high [O 3 ] events. It is interesting that this does not appear to be the result of the Lagrangian model's ability to exploit very high resolution anthropogenic emissions, 10 despite the site being just 100 km downwind of Manaus; recall, this insensitivity was observed in our simulations of flight B735, bound for Manaus. Instead, it appears to be the result of its ability to retain heterogeneity in the origins of air parcels, and their histories over the previous seven days. 15 We have confronted two atmospheric chemistry models -a global Eulerian chemistryclimate model, UM-UKCA O'Connor et al., 2014), and a trajectorybased Lagrangian model, CiTTyCAT (Pugh et al., 2012) -with airborne measurements of atmospheric composition above the Amazon rainforest (O 3 , NO, NO 2 , C 5 H 8 and CO) from the 2012 SAMBBA campaign (see Darbyshire and Johnson, 2013). To our 20 surprise, the simulations with the two models proved relatively insensitive to the chemical mechanism employed (CheT or CheT2; see Sects. 3.2 and 3.5). Explored only in the Lagrangian environment, the sensitivity our simulations showed to the spatial resolution of trace gas emissions (0. was the result of the different treatments of transport in the two model frameworks, and their consequences for mixing (see Sect. 3.1). The lack of explicit mixing in the Lagrangian model yielded structure in its simulations of atmospheric composition, particularly [O 3 ], that was more reminiscent of that largely (but not always) exhibited by the measurements, though perhaps a little too pronounced. Meanwhile, extensive mixing 5 in the Eulerian model removed much of this structure but yielded better overall agreement with the measurements in terms of magnitude (see Sect. 3.1). We found a simple treatment of mixing -relaxing the composition of air parcels in the Lagrangian model towards monthly mean data from the Eulerian model -to offer a compromise between the two, combining some of the benefits of both. Applied to the simulation of boundary 10 layer [O 3 ], we showed that the choice of model framework affects our understanding of both the frequency at which the rainforest is exposed to damaging [O 3 ] and the duration for which it is so exposed, as quantified for example in terms of a version of the accumulation-over-threshold metric, AOT40 (see Sect. 3.5).

Summary and discussion
We noted in the introduction that Lagrangian models, run in "domain-filling" mode, 15 capture much of the strain-induced stirring of the atmosphere, with no mixing, whilst Eulerian models capture the large-scale stirring but inevitably introduce mixing as they average concentrations across grid boxes (see, e.g., Dragani et al., 2002;Methven et al., 2003). The Lagrangian model thus, predictably, proved capable of retaining much more structure than the Eulerian model: heterogeneity in the chemical composition of 20 air parcels bound for points spaced only a few minutes apart along a flight track, or arriving at the same location at regular intervals (every 6 h) over an extended period of time. We found that this additional structure largely reflected the model's ability to resolve the different origins of air parcels (7 days previously) and, only to a lesser extent, differences in: the routes by which they were subsequently transported; the emissions 25 and deposition to which they were exposed en route; and hence the chemistry that took place within them. The composition of each air parcel was simulated independently in the Lagrangian model (i.e. assuming no mixing took place between them). There was not only no mixing in the horizontal, but no mixing in the vertical; the Lagrangian model Introduction The simple approach to diffusive vertical mixing that we later introduced into the Lagrangian model has been used in previous studies; see Pugh et al. (2012)  Lagrangian model contributed to its overestimation of [O 3 ] in the free troposphere, it likely also contributed to its simulation of low [O 3 ] in the boundary layer; the lack of ventilation of the boundary layer, and exchange with free tropospheric air aloft, would have neglected mixing with, and enrichment by, higher [O 3 ] air above. This raises the question whether something else might be amiss in the model(s), leading to an over-5 estimation of boundary layer [O 3 ] in this environment. One possibility is that the dry deposition of O 3 to the rainforest is underestimated. Hardacre et al. (2015) recently highlighted the crude treatment of dry deposition to tropical forests in current global chemistry-climate models. In both the Lagrangian model and the Eulerian model, we have employed a 1 m dry deposition velocity (V d ) of 0.5 cm s −1 to the forest. This compares favourably with the measurements of Rummel et al. (2007) in the Amazon, which yielded a "mean midday maximum" O 3 V d of 0.5 cm s −1 in the dry season. However, such measurements in the Amazon are sparse and the V d of O 3 could vary from one region to another: more O 3 V d measurements are called for.
We used the Lagrangian model, the Eulerian model, and the combination of the 15 two models, to simulate [O 3 ] at a site in the boundary layer, approximately 100 km downwind of Manaus, over a period of a month; see Sect. 3.5. The choice of site was arbitrary, but chosen in anticipation of demonstrating the need to employ very high resolution anthropogenic emissions (capable of resolving a city of the order of 10 km × 10 km) to correctly capture the chemistry and composition downwind. To our 20 surprise, our simulations again proved relatively insensitive to the resolution of emissions (and chemical mechanism) employed, consistent with our earlier findings when simulating the SAMBBA measurements. However, again, they demonstrated a high degree of sensitivity to the model framework; see top left and top right of Fig. 11 The key message here is that both the frequency at which the rainforest is exposed to damaging [O 3 ], and the duration for which it is so exposed, change with 5 the model framework. The choice of model framework therefore has a strong bearing on predictions of the exposure of tropical forests to ground-level ozone, and hence our understanding of the health of the rainforest, the tropical carbon cycle, and how these might change under future climate-and trace gas emission scenarios. It is not clear from the current time series study whether it is possible to derive a transfer function by 10 which to modify chemistry-climate simulations of ozone over tropical rainforest, but our results clearly motivate a further study to seek such a function.