Influence of isoprene chemical mechanism on modelled changes in tropospheric ozone due to climate and land use over the 21 st century

Isoprene is a precursor to tropospheric ozone, a key pollutant and greenhouse gas. Anthropogenic activity over the coming century is likely to cause large changes in atmospheric CO2 levels, climate and land use, all of which will alter the global vegetation distribution leading to changes in isoprene emissions. Previous studies have used global chemistry–climate models to assess how possible changes in climate and land use could affect isoprene emissions and hence tropospheric ozone. The chemistry of isoprene oxidation, which can alter the concentration of ozone, is highly complex, therefore it must be parameterised in these models. In this work, we compare the effect of four different reduced isoprene chemical mechanisms, all currently used in Earth system models, on tropospheric ozone. Using a box model we compare ozone in these reduced schemes to that in a more explicit scheme (the Master Chemical Mechanism) over a range of NOx and isoprene emissions, through the use of O3 isopleths. We find that there is some variability, especially at high isoprene emissions, caused by differences in isoprene-derived NOx reservoir species. A global model is then used to examine how the different reduced schemes respond to potential future changes in climate, isoprene emissions, anthropogenic emissions and land use change. We find that, particularly in isoprene-rich regions, the response of the schemes varies considerably. The wide-ranging response is due to differences in the model descriptions of the peroxy radical chemistry, particularly their relative rates of reaction towards NO, leading to ozone formation, or HO2, leading to termination. Also important is the yield of isoprene nitrates and peroxyacyl nitrate precursors from isoprene oxidation. Those schemes that produce less of these NOx reservoir species, tend to produce more ozone locally and less away from the source region. We also note changes in other key oxidants such as NO3 and OH (due to the inclusion of additional isoprene-derived HOx recycling pathways). These have implications for secondary organic aerosol formation, as does the inclusion of an epoxide formation pathway in one of the mechanisms. By combining the emissions and O3 data from all of the global model integrations, we are able to construct isopleth plots comparable to those from the box model analysis. We find that the global and box model isopleths show good qualitative agreement, suggesting that comparing chemical mechanisms with a box model in this framework is a useful tool for assessing mechanistic performance in complex global models. We conclude that as the choice of reduced isoprene mechanism may alter both the magnitude and sign of the ozone response, how isoprene chemistry is parameterised in perturbation experiments such as these is a crucially important consideration. More measurements and laboratory studies are needed to validate these reduced mechanisms especially under high-volatile-organiccompound, low-NOx conditions.


Introduction
The emission of volatile organic compounds (VOCs) into the atmosphere in the presence of NO x (the sum of nitric oxide (NO) and nitrogen dioxide (NO 2 )) can lead to the formation of tropospheric ozone (O 3 ), which is a pollutant and greenhouse gas (e.g.Haagen-Smit, 1952).One VOC that contributes significantly to tropospheric O 3 production Published by Copernicus Publications on behalf of the European Geosciences Union.
is the biogenically emitted di-alkene isoprene (2-methyl-1,3butadiene) with annual emissions of ∼ 500 Tg C (Guenther et al., 2006).Isoprene is highly reactive with an atmospheric lifetime on the order of about 1-2 h, and thus has the potential to strongly influence levels of tropospheric O 3 both regionally (e.g.Chameides et al., 1988) and globally (e.g.Wang and Shallcross, 2000).
Isoprene is oxidised in the atmosphere by the hydroxyl radical (OH), O 3 and the nitrate radical (NO 3 ).These reactions initiate a complex cascade of photochemical interactions, which (theoretically) comprise > 10 5 reactions involving > 10 4 species (Aumont et al., 2005).Including all of these reactions in 3-D global modelling studies is too computationally expensive and so isoprene chemistry must be parameterised.Furthermore, our understanding of isoprene oxidation is incomplete; only a small number of these 10 5 reactions are known in depth.Although parameterisation is a necessity, it introduces uncertainties in the chemistry and subsequent calculation of trace gas composition, as multiple species or reactions have to be lumped together.Furthermore, there are several different methodologies for how best to parameterise an explicit chemical mechanism, which has led to the existence of a plethora of different reduced schemes, whose use in models can lead to different results (e.g.Archibald et al., 2010b).Jeffries et al. (1992) laid out a set of basic considerations to make when evaluating a condensed chemical mechanism, which are the points in the process of condensed mechanism development where individual methodologies may diverge.These include the relationship between different lumping groups and explicit species, the method used to select individual lumping groups, e.g. by characteristic reaction times, molecular weight or chemical structure, and the approach to handling chain degradation kinetics for each lumped species.The choices made in developing reduced mechanisms may also have been made with the aim of accurately representing specific timescales (e.g.urban or continental) or species (e.g.O 3 ) (Jeffries et al., 1992).
To date there have been several studies that calculate the effects of future isoprene emission changes caused by potential climate and land use scenarios on surface O 3 (Sanderson et al., 2003;Wiedinmyer et al., 2006;Ganzeveld et al., 2010;Wu et al., 2012;Pacifico et al., 2012), including our recent study (Squire et al., 2014).These 3-D global modelling studies all use (often different) reduced isoprene mechanisms.Very few studies, however, have attempted to quantify the influence of differences in the isoprene scheme on the O 3 response.Previously, von Kuhlmann et al. (2004) compared three different isoprene mechanisms and related parameters such as the deposition of intermediates, the treatment of isoprene nitrates and the emission strength of O 3 precursors, all within a particular global model.Here we explore the behaviour of four reduced schemes, all designed to be used in complex Earth system models (ESMs), in the context of the climate and land use perturbation experiments carried out in Squire et al. (2014).Given the importance of O 3 in the Earth system (Huntingford et al., 2011), our analysis focuses specifically on O 3 and on O 3 precursors.
Even without mechanism reduction, there exist sources of uncertainty in isoprene oxidation that are associated with our fundamental lack of understanding about certain aspects of the chemistry.One such aspect is the degree to which HO x is regenerated from isoprene degradation under low-NO x high-VOC conditions.Several campaigns in such conditions (GABRIEL, Kubistin et al., 2010;INTEX-A, Ren et al., 2008;OP3, Stone et al., 2011;Whalley et al., 2011) reported levels of HO x that were higher than expected, considering the high reactivity of isoprene with OH (k 298 K = 10 −10 cm 3 molecule −1 s −1 ).Proposals have been put forward for missing mechanistic pathways, e.g.peroxy radical isomerisation (Peeters et al., 2009), and epoxide formation (Paulot et al., 2009) which to some extent reconcile these discrepancies (Archibald et al., 2010a;Warwick et al., 2013;Fuchs et al., 2013).It has also been demonstrated that positive biases in the measurement of HO 2 (Fuchs et al., 2011) and OH (Mao et al., 2012) cannot be ruled out in some of those field campaigns listed above.Mao et al. (2012) found that, for a Californian forest environment, taking into account these biases in addition to the proposed mechanistic pathways, gave good agreement between modelled and measured HO x .
The chemistry of isoprene nitrates represents another important source of uncertainty.When hydroxyperoxy radicals from OH-initiated isoprene oxidation (ISO 2 ) react with NO, the major pathway leads to the formation of alkoxy radicals and NO 2 (leading to O 3 formation).However, there is a minor channel that leads to the formation of isoprene nitrates, which act to sequester NO x .There are several uncertainties surrounding the chemistry of isoprene nitrates.First, estimates of the yield of isoprene nitrates from the OH / NO channel range from 4.4 to 15 % (Xie et al., 2013, and references therein).Modelling studies have shown that the assumed yield of isoprene nitrates can have a large impact on tropospheric O 3 (e.g. von Kuhlmann et al., 2004;Wu et al., 2007;Paulot et al., 2012).Second, isoprene nitrates may also be formed from the oxidation of isoprene by NO 3 , which is estimated to account for 30-60 % of isoprene nitrate production (von Kuhlmann et al., 2004;Horowitz et al., 2007;Paulot et al., 2012).The types of isoprene nitrates formed via the NO 3 pathway are distinct from those formed via the OH / NO pathway and details of their atmospheric fates remain relatively obscure (Xie et al., 2013).Third, once formed, isoprene nitrates are readily photooxidised (lifetime ∼ 4 h with respect to OH (OH = 10 6 molecule cm −3 )), leading either to release of NO x , or to second-generation nitrates, retaining the nitrate group.The degree to which NO x is regenerated from isoprene nitrate degradation remains uncertain (Fiore et al., 2012;Xie et al., 2013) and has a significant effect on the O 3 response to isoprene emission changes (Paulot et al., 2012).Fourth, dry deposition of isoprene nitrates, which may represent an important NO x sink in isoprene-rich regions, is also uncertain, with measured deposition velocities ranging from 0.4 cm s −1 (Shepson et al., 1996) to 2.7 cm s −1 (Farmer and Cohen, 2008).Finally, there is evidence for the importance of O 3 -initiated isoprene nitrate degradation (Lockwood et al., 2010) and fast photolysis of isoprene nitrates (Müller et al., 2014).In this study, the isoprene schemes we compare have a range of different parameterisations for isoprene nitrates.
In Sect.2, we describe in detail the four chemical mechanisms used in this study and the methodology for the global perturbation experiments.In Sect.3, we discuss the results of a series of box model simulations with the aim of comparing the four reduced mechanisms to the Master Chemical Mechanism (MCM).This is done for a range of NO x and isoprene concentrations.Global integrations with each mechanism are then conducted to examine the effect of changes in climate, in isoprene emissions with climate, in anthropogenic emissions and in land use.In Sects.4-6, we analyse the results of these global perturbation experiments.

Methods
In this section, we outline the experiments conducted to ascertain the effect of using different reduced isoprene chemical mechanisms in the context of global climate, emissions and land use change experiments (Sect.2.2.2).In Sect.2.1, details of the reactions and species that make up the reduced mechanisms are given.

Isoprene chemical mechanisms
The species included in each mechanism are given in Table 1, whilst a comparison of the reactions is given in Table 2.The different isoprene mechanisms were each embedded in an otherwise identical tropospheric chemistry mechanism simulating the chemistry of methane, ethane, propane, HO x and NO x , following O' Connor et al. (2014).
The first simulation used the UM-UKCA Chemistry of the Troposphere (CheT) mechanism, as was done for all integrations in Squire et al. (2014).The CheT isoprene mechanism consists of 16 species and 44 reactions (see Tables 1  and 2), and is based on the Mainz Isoprene Mechanism (MIM) (Pöschl et al., 2000).MIM was developed from a reduction of the Master Chemical Mechanism (version 2) (Jenkin et al., 1997), by lumping species based on their structure (e.g.all hydroxyperoxy radicals were lumped as ISO 2 , and methacrolein and methyl vinyl ketone as methacrolein, MACR).The overall CheT mechanism also forms the base case against which all the schemes studied, are compared.
Since the creation of MIM, there have been a number of developments in our understanding of isoprene chemistry concerning issues such as those discussed in the Introduction.In a report compiled for the UK Met Office (Jenkin, 2012), these new developments were incorporated into the current CheT framework.The resulting updated mechanism (which will be referred to as CheT2, see Tables 1 and 2) is the most complex mechanism used in this study, consisting of 24 species and 59 reactions, and is traceable to the MCM version 3.2 (MCMv3.2).
The following is a summary of the changes introduced into the CheT mechanism to create CheT2.First, changes to the chemistry of first-generation isoprene nitrates (ISON) were made.In CheT, NO x is regenerated from ISON by photolysis or conversion to second-generation nitrates (NALD) followed by reaction with OH.In CheT2, the overall yield of NO x from ISON was increased, in line with recent measurements (Perring et al., 2009), by increasing the ISON photolysis rate and adding an ISON + OH → NO 2 reaction channel.O 3 initiated degradation of ISON was also added based on the evidence of Lockwood et al. (2010) -it should be noted, however, that Lee et al. (2014) found that Lockwood et al. (2010) substantially overestimated the rate of this reaction.Secondly, CheT2 includes the formation of hydroperoxy-aldehydes (HPALDs) from ISO 2 and subsequent rapid release of OH (Peeters et al., 2009).This leads to more HO x regeneration in low-NO x high-isoprene conditions, bringing modelled and measured values closer together (e.g.Archibald et al., 2010a).The formation of isoprene epoxydiols (IEPOX) from the oxidation of isoprene hydroxy-hydroperoxides (ISOOH), a potential source of secondary organic aerosols (Paulot et al., 2009), was also included in CheT2.Finally, the yield of peroxymethacrylic nitric anhydride (MPAN) from isoprene oxidation was revised down from its CheT value (Jenkin, 2012).
The Air Quality in the Unified Model (AQUM) scheme, which was developed to deliver regional air quality forecasts and conduct air quality studies to inform emission control policies (Savage et al., 2013), was also investigated.The mechanism has a more anthropogenic VOC focus and a less detailed isoprene scheme compared with CheT (17 species, 23 reactions).Two important simplifications in the isoprene scheme are that (1) isoprene nitrates are not formed from the OH initiated pathway via the reaction of ISO 2 with NO, and (2) there is no production of MPAN.
The last and most simple isoprene scheme investigated was the super-fast chemistry scheme developed at the Lawrence Livermore National Laboratory (LLSF) (Cameron-Smith et al., 2009) for use in the Community Earth System Model (CESM -http://www.cesm.ucar.edu/models/cesm1.0/).The LLSF isoprene scheme only considers the reactions of isoprene with OH and O 3 , and was parameterised based on the net effect of a more complex isoprene mechanism (Cameron-Smith et al., 2009).Aside from not including isoprene chemistry at all, it is about as simple an approximation of isoprene chemistry as is currently used in ESMs, but is still a significant improvement over neglecting isoprene chemistry altogether (Cameron-Smith et al., 2009).The scheme was developed for use in very long Table 1.List of chemical species included in each of the isoprene mechanisms.Note that names of some of the species in AQUM were changed from the names given in Savage et al. (2013) to be consistent with the other schemes.These are as follows: "HOIPO2" -ISO 2 , "MVK" -MACR, "MVKOOH" -MACROOH, "HOMVKO2" -MACRO 2 .global 3-D integrations, where reducing computational cost is paramount.

Box model experiments
A box model comparison study was performed with the different isoprene schemes to establish any inherent differences in the schemes that do not arise from the complexity present in a global 3-D model.This also allows us to compare the reduced schemes with a more complex scheme, the MCMv3.2(Jenkin et al., 1997;Archibald et al., 2010b), which is too complex to put into a global 3-D chemistry-climate model.The detailed nature of the MCM lends itself to being a benchmark mechanism against which the others can be compared (e.g.Archibald et al., 2010b).However, the MCM still contains approximations; e.g.many of the rate constants are inferred from other reactions using structure reactivity relationships (SARs, e.g.Kwok and Atkinson, 1995;McGillen et al., 2011), and only four of the six ISO 2 isomers are included.Furthermore, many of the recent discoveries such as the isomerisation chemistry of ISO 2 and methacrolein are not included.
For our box model comparison, the Kinetic PreProcessor (KPP) solver (Sandu and Sander, 2006) was used, with a model time step of 20 min.The model was set up so that different emissions of NO x and isoprene were input, allowing us to study how the mechanisms compared over a wide range of NO x -to-isoprene ratios.NO x emissions between 0.001 and 0.5 mg N m −2 h −1 and isoprene emissions between 0.0001 and 6 mg C m −2 h −1 were used, with emission rates  being constant for the duration of a given model run.Atmospheric pressure (1 × 10 5 Pa) and temperature (298 K) were kept constant, and the amount of light varied through the day as in a grid cell at 14 • latitude on Julian day 172 (solar declination angle = 23.44 • ).To ensure that differences in the oxidation chemistry were not due to differences in photolysis between the mechanisms, the MCM photolysis parameterisation was used in all cases.Details of how photolysis coefficients are calculated using this parameterisation are given in Jenkin et al. (1997).The model was initialised with 30 ppb O 3 , 1820 ppb CH 4 , 102 ppb CO, and run with a fixed amount of H 2 O (50 %).The box model does not include any advection or deposition processes, and, as such, O 3 values are likely to be higher than those measured in the field or calculated in UM-UKCA.Other consequences of including emissions but not removal pathways are that steady state will never be reached and longlived reservoir species will accumulate.For example, OH could be modified by accumulation of H 2 O 2 via OH+H 2 O 2 .To minimise such effects on oxidant fields, a relatively short run length of 3 days for the simulations was chosen.In all simulations, the maximum O 3 value on the third day was used to provide a consistent point of comparison between mechanisms.The results of the box model comparison are given in Sect.3.

Global perturbation experiments
To investigate the influence of variations in the isoprene mechanism on potential changes in tropospheric O 3 over the 21st century, a global chemistry-climate model (the UK Met Office Unified Model coupled to the UK Chemistry and Aerosol model, UM-UKCA) was used, as specified in Squire et al. (2014).For each mechanism investigated, a present-day (2000) integration was conducted, following the model setup described for the BASE run in Squire et al. (2014).Then, for each mechanism four future (2095) integrations were conducted to investigate (1) CC, climate change only, (2) IC, isoprene emission change with climate, (3) AC, anthropogenic emission change, and (4) LC, land use change, with each integration set up as described in Squire et al. (2014).The effect of mechanistic changes on the O 3 response to including the CO 2 inhibition of isoprene emissions was not investigated in this study.It should also be noted that with the change in land use, we assume no change in NO x emissions.
For CC, all parameters, including isoprene emissions, remained as in the present day BASE run except sea surface temperatures, sea ice concentrations and greenhouse gas concentrations.In IC, isoprene emissions were allowed to vary with a scenario of future climate change.This led to globally higher isoprene emissions (545 Tg C yr −1 ) than in the BASE run (467 Tg C yr −1 ), largely due to the effect of extended CO 2 fertilisation of the biosphere under the elevated CO 2 levels.AC was characterised by stringent emission cuts across much of the Northern Hemispheric developed regions, leading to lower NO x levels there.For LC, we used a scenario of future cropland expansion that is dominated by the replacement of tropical broadleaf trees with crops (see Squire et al. (2014) for details).As the crops emit less isoprene than broadleaf trees (Guenther et al., 2006;Lathiere et al., 2010), this causes a decrease in isoprene emissions (190 Tg C yr −1 globally).

Mechanism intercomparison with a box model
In this section we discuss the results of the simulations we performed to assess the performance of the isoprene mecha- nisms we investigated in UKCA using a simple box model.The simulations we performed were aimed at understanding how mechanistic assumptions affected the simulation of O 3 by comparing the results of the reduced mechanisms to the MCM.
As discussed in Sect. 1, the inherent assumptions regarding mechanism reduction (Jeffries et al., 1992), and specifically the omission of unique isoprene oxidation products, may lead to large discrepancies for simulated O 3 between the mechanisms.Here we hypothesise that assumptions about the sequestration of HO x (via organic hydroperoxides ROOH) and NO x (via organic nitrates RONO 2 ) are a key cause of inter-mechanism variability and lead to changes in the abundance and hence lifetime of NO x .
To test this hypothesis, for each mechanism, box model runs were performed at a series of different NO x and isoprene emission rates, so that an O 3 isopleth plot could be constructed, similar to those found in Dodge (1977) and Sillman and He (2002).In general, all mechanisms produce the same overall behaviour of increasing O 3 with increasing isoprene and NO x emissions.These general features can be seen in Fig. 1 1 is shown to be related to the overall behaviour of the NO x lifetime (Fig. 2), defined using Equation 1 from Browne and Cohen (2012).Fig. 2 shows that the NO x lifetime varies by over 2 orders of magnitude over the range of NO x and isoprene emissions the simulations covered.A general feature of Fig. 2 is that the NO x lifetime shows a degree of anticorrelation with the O 3 mixing ratios simulated.Regions of the O 3 isopleth where net O 3 destruction occurs coincide with regions where the NO x lifetime is greatest.The key cause of this is a reduction in the abundance of OH.
Our main focus is on the comparison of the reduced mechanisms with the MCM for the simulated O 3 mixing ratios and NO x lifetimes.Figures 3 and 4 show the differences in the O 3 and NO x lifetimes simulated using the reduced mechanisms relative to the MCM.These differences provide us with useful information about variations in chemical oxidation between the schemes, which may be used to help diagnose the impact of isoprene mechanism differences in the more complex context of the global model experiments (Sect.4).

LLSF
LLSF is the most simplistic scheme with no unique species used to represent the chemistry of isoprene, beyond those used to describe the oxidation of methane.Thus, there is only a small amount of NO x locked up in methyl nitrate formation (reflecting the small yield of methyl nitrate production) and HO x is efficiently sequestered into organic hydroperoxides (reflecting the slow rates of reactivity of methyl hydroperoxide).Figure 3d shows that using LLSF results in the highest biases in O 3 compared to the MCM of all the mechanisms considered.Under low-NO x high-isoprene emissions, LLSF is biased high by up to ∼ 80 %.Because the reaction between MeO 2 and HO 2 is slower than for the major isoprene-derived peroxy radicals in the MCM, mixing ratios of peroxy radicals are found to be higher in this region, and the enhanced rate of RO 2 + NO leads to high O 3 .This is consistent with the differences in the NO x lifetime (Fig. 4d) which show that under these conditions, the LLSF mechanism simulates an overall shorter NO x lifetime.This shorter lifetime is reflected across most of the VOC : NO x emission space with the exception of high-NO x emissions, where there is better agreement with the MCM (biases ∼ 10 %).A band of low-bias in O 3 (Fig. 3d) is correlated with a band of high-bias in NO x lifetime (Fig. 4d) under a range of low isoprene emissions and moderately high NO x emissions.Both of these features are consistent with the parameterised instantaneous loss of two OH for the isoprene + OH reaction in LLSF.With LLSF, the low bias at low isoprene emissions and high bias at high isoprene emissions largely cancel each other out, leading to a small overall MB (−2.6 %) for O 3 with respect to the MCM (Fig. 3d).Despite a small overall bias, the majority of the regions where isoprene chemistry is known to be important for the O 3 budget are the regions where this mechanism shows greatest bias.

AQUM
As discussed in Sec.2.1, AQUM was developed for representing the oxidation of isoprene in the context of simulating regional air quality.As such AQUM does not account for the formation of organic nitrates, other than from the reaction of isoprene with NO 3 .Under low-NO x high-isoprene emissions, AQUM is biased high by ∼ 20 % (Fig. 3c).The figure shows that there is a band of large negative MB (−25 %) in O 3 , compared to the MCM, under NO x emissions of 0.01-0.1 mg N m −2 h −1 .There is very little correlation between the biases in O 3 and biases in the NO x lifetime (Fig. 4c) under these emissions.In regions of high-isoprene high-NO x emissions AQUM simulates much greater values of NO x lifetime than the MCM, consistent with a reduction in the abundance of NO x (Browne and Cohen, 2012) and a reduction in OH.Whilst the reduction in the abundance of NO x , as evidenced by the increase in NO x lifetime, is consistent with a reduction in the potential for producing O 3 , it cannot be as- cribed to sequestration of NO x in organic nitrates and the attribution of the origin of this behaviour lies beyond the scope of this investigation.

CheT
CheT, which is based on the MIM (Pöschl et al., 2000) was developed for global-scale modelling studies of atmospheric chemistry.As with AQUM, it accounts for the formation of isoprene nitrates produced via NO 3 chemistry but also includes the generation of isoprene nitrates from the OH initiated oxidation (the primary route for isoprene oxidation).Fig. 3a shows the bias in O 3 simulated using CheT compared with the MCM.The bias in O 3 is generally within ±20 % of the MCM over all the emissions of isoprene and NO x we investigated.Under high-NO x and high-isoprene emissions, the bias is higher, with CheT simulating up to 40 % less O 3 than the MCM.Under these conditions CheT also simulates an increase of the NO x lifetime, consistent with a reduction in the abundance of NO x (Fig. 4a).Overall the mean bias (MB) is −5.7 %, indicating a weak negative bias compared to the MCM.This result is consistent with the work of Archibald et al. (2010b) who showed that the CheT scheme (UKCA in their runs) simulated lower levels of O 3 than the MCMv3.1.

CheT2
The last reduced mechanism that was tested in the box model was CheT2.CheT2 is based on CheT but includes recent updates to the oxidation of isoprene including the formation of IEPOX (Paulot et al., 2009) and modifications to the treatment of isoprene nitrates, allowing regeneration of NO x and formation of multigeneration nitrates.Figure 3b shows that the low bias in O 3 simulated at high O 3 precursor emissions is much less pronounced in CheT2 than in CheT, being within ±20 % of the MCM.This is consistent with the lower rate of MPAN formation in CheT2 compared to CheT, meaning that more NO x is available for O 3 formation.This is also observed in the comparison of the NO x lifetime.Figure 4b highlights that there is overall a shorter NO x lifetime (reduction of 50 %) across the range of isoprene and NO x emissions simulated.Under high-isoprene low-NO x conditions, the effect of the addition of the ISO 2 HO x regeneration pathways present in CheT2 (the Peeters mechanism, Peeters et al., 2009) is seen.These were not included in the MCMv3.2, and they serve to enhance HO x , and decrease NO x , similar to LLSF.This finding is consistent with the enhanced O 3 seen in Archibald et al. (2011) when CheT (UKCA in their work) was run with inclusion of the Peeters mechanism.Overall the MB of CheT2 with respect to the MCM is lower (−1 %) than for CheT (−5.7 %).

Summary of box model results
These box modelling results show that biases in simulated O 3 are related to the treatments of organic nitrate and organic hydroperoxide chemistry, and to the simulated abundance of NO x (and its lifetime).The box modelling allows us to examine the effect of the complexity of the chemical processes included and we have shown that we are able to begin to understand the effects of mechanistic differences.
It is clear that conclusions drawn solely on the basis of e.g.NO x lifetime are necessarily incomplete, e.g. in the case of the AQUM biases, and that a detailed analysis of individual reaction fluxes is required to quantitatively understand the origin of biases over the range of emissions used in global model simulations.However, we emphasise that we have not attempted to perform a systematic study on the isoprene oxidation mechanism.A more rigorous approach would be to examine the individual effects of omissions of mechanistic pathways through one-at-a-time sensitivity tests using one base mechanism, for example the MCM.This is far beyond the scope of our study, which aims to qualify the role of uncertainty in the chemistry vs. the uncertainty in forcings (e.g.emissions) as drivers for the uncertainty of global simulations of O 3 .Rather, we aim to link the results from the box model simulations with the global model results (Sect.4) to provide insight into the behaviour of the isoprene-NO x -O 3 system.In the following section we will discuss the results of simulations with a global chemistry-climate model where each of the reduced mechanisms is used to assess the impact of using different chemical mechanisms on simulations of climate, land use and emission changes.

Present-day mechanism intercomparison with a global model
Using the four reduced schemes, global simulations of the present-day atmosphere were conducted.Figure 5a shows near-surface O 3 for the present day using the CheT isoprene scheme.Figure 5b-d illustrates the change in this O 3 caused by the use of different isoprene chemical schemes.All schemes simulate a present-day tropospheric O 3 burden that is within 1 standard deviation of the model ensemble mean from the ACCENT study (344 ± 39 Tg) (Stevenson et al., 2006).As may be expected from a comparison of isoprene chemical mechanisms, the largest differences between the schemes occur where isoprene emissions are highest (tropical regions and the southeast USA).In these regions (mean isoprene emissions > 0.1 mg C m −2 h −1 ), the mean surface O 3 for CheT is 41 ppb, whilst for AQUM and LLSF the values are higher -46 ppb (+11 %) and 50 ppb (+18 %) respectively.In some places (e.g.Amazonia, Central Africa), this equates to surface O 3 that is at least 10 ppb higher than with CheT.By comparison, surface O 3 in CheT2 is very similar to that of CheT, even in the high isoprene emitting regions.The regions of high isoprene emissions, where the largest differences between the mechanisms are calculated, are generally situated away from areas of intense anthropogenic activity.As a result, these areas tend to have low NO x emissions.To understand the changes occurring in this low-NO x high-isoprene regime, Table 3 gives the mean O x budget fluxes for near-surface (below 720 m) grid cells with monthly mean NO x emissions less than 0.03 mg N m −2 h −1 and monthly mean isoprene emissions greater than 0.1 mg C m −2 h −1 (roughly matching the bottom right-hand quarter of Fig. 1).Here we define O x as O 3 P + O 1 D + O 3 + 2 × NO 3 + NO 2 + 3× dinitrogen pentoxide (N 2 O 5 ) + nitric acid (HNO 3 ) + peroxynitric acid (HNO 4 ) + PAN + peroxypropionyl nitrate (PPAN) + MPAN. Figure 6 shows geographical loca-Table 3. Near-surface (below 720 m) mean O x budget fluxes (mol gc −1 s −1 ) for regions with high isoprene emissions (greater than 0.1 mg C m −2 h −1 ) and low NO x emissions (less than 0.03 mg N m −2 h −1 ).Values from the BASE run are given.Also given are the differences caused by climate change (CC) and land use change (LC).See Fig. 6 3. Units range from 0 to 12, indicating the number of months per year that each grid cell was included in the calculation.Using emissions from the BASE run, only those months when mean isoprene emissions were greater than 0.1 mg C m −2 h −1 and mean NO x emissions were less than 0.03 mg N m −2 h −1 were included.Blue indicates that, based on this criterion, the grid cell was not included in the calculation at all.
tions of those grid cells included in this emissions regime, and also indicates how many months per year each grid cell was included.
From the budget terms in Table 3 for the BASE integrations, total mean O x production varies across the schemes from 74 mol gc −1 s −1 (here gc is grid cell) (CheT and CheT2) to 200 mol gc −1 s −1 (LLSF).The majority of this variance is due to differences in the peroxy-radical + NO reactions (HO 2 + NO, MeO 2 + NO and other peroxy-radicals (RO 2 ) + NO).In CheT, CheT2 and AQUM, RO 2 is primarily ISO 2 , MACRO 2 (see Table 1 for definitions) and the peroxy acetyl radical (MeCO 3 ).
In all schemes, the oxidation of isoprene by OH is a source of peroxy radicals.In CheT, CheT2 and AQUM, the initial isoprene + OH reaction leads exclusively to the production of ISO 2 , whilst for LLSF MeO 2 is produced instead.Both MeO 2 and ISO 2 may react with NO producing O x (propagation), or with other peroxy radicals producing peroxides (termination).k ISO 2 +HO 2 (similar in all schemes that include ISO 2 ) is three times higher than k MeO 2 +HO 2 (identical in all schemes), and k MeO 2 +NO (identical in all schemes) is two times higher than k ISO 2 +NO (similar in all schemes that include ISO 2 ).This suggests that the scheme that produces the largest fraction of MeO 2 from isoprene oxidation (LLSF) should also show the highest total RO 2 + NO flux and consequently highest O 3 levels, exactly as calculated (see Fig. 5d).In CheT, CheT2 and AQUM, once ISO 2 is formed, it may be further oxidised to produce second-generation peroxy radicals such as MACRO 2 .In AQUM, reactions of ISO 2 and MACRO 2 with NO lead to greater production of O 3 , as evident from the higher mean RO 2 + NO flux (Table 3): 51 (AQUM), 31 (CheT), 29 mol gc −1 s −1 (CheT2).The reason for this is the inclusion in CheT and CheT2 of competing peroxy radical + NO reaction channels that do not lead to O 3 formation.AQUM does not include the isoprene nitrate formation pathway from ISO 2 , which accounts for 4.4 and 10 % of the total ISO 2 + NO flux in CheT and CheT2 respectively (Jenkin, 2012).Additionally, AQUM does not include MPAN formation from MACRO 2 + NO, which contributes to a lower mean MACRO 2 + NO → NO 2 flux in CheT and CheT2 compared to AQUM: 7.0 (CheT), 6.7 (CheT2) and 12 mol gc −1 s −1 (AQUM).
Figure 7 shows total peroxyacyl nitrates ( PAN = PAN + MPAN + PPAN) near the surface in (a) CheT and (bd) the difference between CheT and the other schemes.Figure 7d shows that, compared to CheT, there is much less PAN in LLSF (the PAN tropospheric burden in LLSF is 1.49 Tg compared to 3.57 Tg in CheT).This follows since in LLSF, no PAN precursor radicals (MeCO 3 nor MACRO 2 ) are produced from isoprene oxidation.As PANs are a source of O x to remote regions, the low PAN in LLSF is likely the cause of the low O 3 compared to CheT over the remote tropical oceans (a mean reduction of 10 % between ±20 • lat., Fig. 5d).Another consequence of reduced PAN formation is that more NO x stays close to the isoprene source region.This contributes to the higher total RO 2 +NO flux, and hence higher O 3 , in these regions in LLSF. Figure 7b indicates that PAN in CheT2 is marginally lower than in CheT (the PAN tropospheric burden is about 6 % lower).The cause of this is that the MPAN production rate in CheT2 is set to be 10 % of that in CheT.The CheT2 rate is the value we would recommend, as it has been adjusted to take account of the fact that in UKCA, the species MACRO 2 represents a set of peroxy radicals, not just the MPAN precursor methacrolyl peroxy radical (Jenkin, 2012).
Figure 7c shows that PAN in AQUM is again marginally lower than in CheT (tropospheric PAN burden is 7 % lower), this time due to the total absence of MPAN formation.However, the difference is small owing to the fact that in AQUM PAN production is faster, a result of higher production of the PAN precursor radical MeCO 3 from isoprene oxidation.The mean mixing ratio of MeCO 3 is 29 % higher than the average of that in CheT and CheT2.Possible additional sources of MeCO 3 in AQUM are the higher yield of methyglyoxal (MGLY), which rapidly reacts to form MeCO 3 (CheT and CheT2 = 16 Tg MGLY yr −1 , AQUM = 40 Tg MGLY yr −1 ).The higher yield of the MeCO 3 peroxy radical also accounts for a fraction of the higher RO 2 + NO flux, and hence higher O 3 in AQUM.
Extending the comparison to the wider troposphere, Table 4 gives the summed total O x budget fluxes for the different schemes up to the tropopause.Fig. 8 shows the zonal mean ozone for the entire troposphere.The tropopause is shown by the black line.It is immediately apparent that the differences in O 3 at the surface are not representative of the net effect on O 3 over the entire troposphere.Whilst the O 3 burdens of CheT and CheT2 are very similar (379 and 380 Tg, respectively), AQUM has a lower burden (374 Tg) and LLSF lower still (360 Tg).This is consistent with the zonal difference plots (Fig. 8b-d), which show that away from the surface, both AQUM and LLSF give lower O 3 than CheT, most notably in the tropical tropopause region.Although the highest total tropospheric net chemical O 3 production rate is calculated for LLSF, (499 Tg yr −1 ), overall the O 3 burden is lower due to the higher rate of dry deposition (1180 Tg yr −1 ) compared to CheT (1155 Tg yr −1 ) and CheT2 (1154 Tg yr −1 ) (see Fig. 9).The rate of dry deposition in AQUM is also high (1191 Tg yr −1 ) (Fig. 9c).In UM-UKCA dry deposition only occurs at the surface and is highest over forested regions.As AQUM and LLSF both produce higher O 3 near the surface and notably over forested regions (high isoprene emitting regions), dry deposition is likely to be higher.This is indeed the case as illustrated by Fig. 9, which shows much higher O 3 dry deposition fluxes over forested regions (e.g.Amazonia, central Africa) in AQUM and LLSF compared to CheT.CheT and CheT2 have higher rates of PAN formation, leading to more O 3 production away from forested regions and the surface in general, thus resulting in lower O 3 dry deposition and higher overall tropospheric O 3 burdens.
Although tropospheric O 3 varies little between CheT and CheT2 (Figs. 5 and 8), there are larger changes in other key oxidants, notably OH.Due to the inclusion of additional HO x regeneration pathways within the isoprene oxidation mechanism of CheT2 (namely the Peeters mechanism, Peeters et al., 2009), one would expect CheT2 to show higher levels of OH over high isoprene emitting regions.Figure 10 shows that OH in CheT2 is indeed higher than in CheT over the main isoprene emitting regions, with maximum increases of approximately 50 %.Warwick et al. (2013) also calculated that including the Peeters mechanism in UM-UKCA gave higher OH, improving agreement between modelled and measured values.
Levels of the main night-time oxidant, NO 3 , are higher in CheT2, AQUM and LLSF than in CheT (not shown).By percentage, the largest increases are calculated in the main isoprene emitting regions (Tropics).Here CheT2 shows increases in NO 3 compared to CheT of around 30 %, whereas AQUM and LLSF show much greater increases in NO 3 -up to 7 times more.This is a consequence of the lower formation rates of nitrate reservoirs in these schemes -thus NO 3 formed from the base non-isoprene chemistry, by reactions such as N 2 O 5 photolysis, is removed more slowly.These differences in NO 3 levels have implications for the simulated rate of oxidation at night.As key oxidants, differences in both OH and NO 3 are important for secondary organic aerosol (SOA) formation, which requires the formation of oxidised organic products.
Another mechanistic difference between CheT2 and CheT that has the potential to affect SOA production is the inclusion of epoxide formation in CheT2, based on the work of Paulot et al. (2009).In the tropics high levels of epoxides (50-70 ppt) reach an altitude of nearly 5 km, and similar mixing ratios are present even in the lower Tropical Tropopause Layer (TTL) (10-13 km).Isoprene-derived epoxides are known to be precursors of organic aerosol formation (Paulot et al., 2009;Surratt et al., 2010), and as such, the presence of epoxides at high tropical altitudes could have important implications for cloud formation (e.g.Froyd et al., 2010).

Future perturbation experiments
In the previous section, we compared the different isoprene mechanisms under present-day conditions.In this section we examine how the mechanisms compare in the context of the future climate change (Sect.5.1) and future isoprene emission change (Sect.5.2) perturbation experiments described in Sect.2.2.2.

Climate change
Figure 11a shows the change in near-surface O 3 caused by our climate change scenario (CC) using the CheT scheme, as in Squire et al. (2014).Figure 11b-d shows the effect of CC using the CheT2, AQUM and LLSF isoprene schemes respectively.The general pattern of near-surface O 3 changes is similar in all schemes.There are reductions over the oceans due to increased water vapour and subsequent loss of O 3 via increased O 1 D + H 2 O.Over land where O 3 production dominates (e.g.polluted Northern Hemispheric regions), near-surface O 3 increases as the flux through O x producing reactions usually increases with temperature.In regions with high isoprene emissions such as the Tropics, O 3 also tends to increase, due to changes in PAN.PAN decomposition exhibits a strong temperature dependence, such that under the higher temperatures of climate change PAN decomposes faster, thus more NO x will be present near the isoprene emission source.As a result, the mean RO 2 (= HO 2 + MeO 2 + RO 2 ) + NO flux increases in these re- gions (see CC entries in Table 3) (AQUM = +15, CheT = +10, CheT2 = +9.3mol gc −1 s −1 ), leading to higher O 3 near the isoprene emission source.Unlike with the other schemes, O 3 decreases in highisoprene low-NO x regions in simulations using LLSF.This is because LLSF produces very little PAN compared to the other schemes (see Sect. 4), so no increase in NO x with temperature is calculated as would occur with increased PAN decomposition.The fact that O 3 actually decreases is due to the negative temperature dependence of k C 5 H 8 +OH .The flux through this reaction under climate change decreases by ∼ 20 % in all schemes, leading to associated increases in OH and isoprene export.Lower isoprene oxidation rates lead to a lower rate of peroxy radical production, and thus the HO 2 + NO and MeO 2 + NO reaction fluxes decrease in LLSF (−1.8 and −3.2 mol gc −1 s −1 , respectively, Table 3).In schemes other than LLSF, this effect is masked by the large increase in NO x from increased PAN decomposition.Despite large changes in tropospheric net chemical production due to climate change, the tropospheric O 3 burdens in the CC experiment remain unaltered (Table 3).
We also explored how ozone changes with our future anthropogenic emission scenario (AC, not shown).This scenario is characterised by large reductions in NO x emissions over the USA, Europe and Japan.The O 3 response was remarkably similar for all the different isoprene mechanisms, presumably because the largest changes in anthropogenic emissions occur away from regions of high isoprene emissions.We conclude that the O 3 -NO x response in these regions is controlled largely by the simple NO x -HO x chemistry which is the same in all chemistry schemes.If, instead, the scenario had included large NO x changes in the Tropics where isoprene emissions are high, it is likely that the schemes would respond differently.It has previously been shown that changes in tropical NO x associated with increased anthropogenic activity can lead to large changes in O 3 , e.g.Paulot et al. (2012) found that tropical O 3 increased by 30 % when NO x emissions everywhere were set to those of the USA in terms of GDP per capita.Conducting a similar experiment with different isoprene chemical mechanisms would be a worthwhile extension to our work.

Isoprene emission change
In this section, we examine the results of the two isoprene emission change experiments: IC (the change in isoprene emissions with climate) and LC (the change in isoprene emissions with land use).
Figures 12 and 13 show the changes in surface O 3 that occur for each of the different isoprene chemical mechanisms in the IC and LC experiments respectively.In both cases, the isoprene mechanism sensitivity is more pronounced than for the CC experiment (Sect.5.1), which may be anticipated given that the perturbations in IC and LC specifically involve isoprene.On the scale of the whole troposphere, the O 3 burden is enhanced in IC and reduced in LC for all schemes (Table 4).This is expected as in IC, ultimately, there is more O 3 precursor and in LC there is less.
In the next three sections, we analyse the O 3 trends in Figs. 12 and 13 using the corresponding O x budget terms in Tables.3 and 5.This is done for each distinct O x production regime; Sect.5.2.2 -NO x -limited regions where isoprene emissions increase, Sect.5.2.3 -NO x -limited regions where isoprene emissions decrease, and Sect.5.2.4 -VOC-limited regions where isoprene emissions increase.In the next section (Sect.5.2.1), we discuss how each of these regimes is defined.

Defining distinct O x production regimes
In the IC experiment, a mean global increase in isoprene emissions (+78 Tg C yr −1 ) is calculated.Within the high isoprene emitting regions, there are three distinct regimes of change, which we will denote as IC regions 1, 2 and 3 (ICr1,  ICr2 and ICr3).Each regime is defined on a per-month-pergrid-cell basis as follows: 1. ICr1 is months when isoprene emissions in a grid-cell increase by more than 0.05 Tg and the environment is NO x -limited.
2. ICr2 is months when isoprene emissions in a grid-cell decrease by more than 0.05 Tg and the environment is NO x -limited.
3. ICr3 is months when isoprene emissions in a grid-cell increase by more than 0.005 Tg and the environment is VOC-limited.
The isoprene emission change criteria is an order of magnitude smaller for ICr3 than for ICr1 or ICr2, owing to the greater sensitivity of increasing isoprene emissions in a VOC-limited environment compared to a NO x -limited environment.Here we define VOC-limited as where the ratio of L N (loss of radicals from reactions with NO and NO 2 ) to Q (the sum of all radical sinks) is more than 0.5 (Kleinman et al., 1997;Wiedinmyer et al., 2006).NO x -limited is defined as where L N /Q is less than 0.5 (Kleinman et al., 1997;Wiedinmyer et al., 2006).To ensure that each regime includes the same grid cells in CheT, CheT2, AQUM and LLSF, L N /Q values from CheT were used in all cases.The geographical location of those grid cells included in each regime are shown Fig. 14, indicating also how many months per year each grid cell was included.Table 5 gives mean O x budget fluxes for these three regimes, which will be discussed in Sects.5.2.2-5.2.4.In LC, the pattern of change in all high isoprene emitting regions is the same as that of ICr2: reductions in isoprene emissions (−190 Tg C yr −1 globally) in a NO x -limited environment.As such, LC and ICr2 will be discussed together.Note that those high isoprene-emitting regions that were VOC-limited in IC (ICr3, e.g.southeastern USA) are NO xlimited in LC owing to the inclusion of an anthropogenic emission scenario of large Northern Hemispheric NO x emission reductions.The mean O x budget terms for LC are given in Table 3.These were calculated using the same grid cells as the other budgets in this table (see Fig. 6).  5. Units range from 0 to 12, indicating the number of months per year that each grid cell was included in the calculation.For each region (ICr1, ICr2 and ICr3) different criteria were used to select which months a grid cell should be included, as follows: (a) ICr1 is months when isoprene emissions increase by more than 0.05 Tg and the environment is NO xlimited.(b) ICr2 is months when isoprene emissions decrease by more than 0.05 Tg and the environment is NO x -limited.(c) ICr3 is months when isoprene emissions increase by more than 0.005 Tg and the environment is VOC-limited.See text for how NO x -limited and VOC-limited are defined.Blue indicates that, according to the above criteria, the grid cell was not included in the calculation at all.

NO
x -limited regions where isoprene emissions increase (ICr1) In ICr1 (where isoprene emissions increase in a NO x -limited environment), both total chemical O x production and total chemical O x loss increase in all schemes, owing to greater O 3 precursor emissions.Changes in O x loss are similar in all schemes, being driven largely by an increase in isoprene ozonolysis (in the range +21 to +27 mol gc −1 s −1 across the schemes).On the other hand, total O x production varies considerably between schemes, from ∼ +1 mol gc −1 s −1 in CheT and CheT2, to +90 mol gc −1 s −1 in LLSF.The overall result is a decrease in net O x production for CheT and CheT2 (each −16 mol gc −1 s −1 ), close to no net change in AQUM, and a net increase in LLSF (+50 mol gc −1 s −1 ).As explained in Sect.4, the primary peroxy radical produced from isoprene oxidation in LLSF is MeO 2 , whilst in the other schemes it is ISO 2 and MACRO 2 .MeO 2 has a higher propensity for reaction with NO than ISO 2 or MACRO 2 , thus an increase in isoprene emissions (as in ICr1) will increase the total RO 2 + NO flux by a greater amount in LLSF than in the other schemes.Note that the MPAN and isoprene nitrate formation pathways that compete directly with O x production from isoprene-derived peroxy radicals in CheT and CheT2 are not included in AQUM.Accordingly, increasing    isoprene emissions in AQUM leads to a larger increase in the RO 2 + NO flux than in CheT or CheT2.

NO x -limited regions where isoprene emissions decrease (LC and ICr2)
In LC and ICr2 (where isoprene emissions are reduced in a NO x -limited environment), the opposite trend is calculated compared to ICr1.Both O x loss and production decrease due to lower levels of O 3 precursor emissions.As with ICr1, the change in O x loss is similar in all schemes, being driven by reductions in isoprene ozonolysis (on average −11 (LC) and +1.4 mol gc −1 s −1 (ICr2, CheT2)) or close to no change (−0.4 mol gc −1 s −1 (ICr2, CheT)).

VOC-limited regions where isoprene emissions increase (ICr3)
For ICr3, where isoprene emissions increase in a VOClimited environment, all schemes show the same trend of increased near-surface O 3 .In such an environment, the effect of adding isoprene favours O 3 production to a far greater extent than O 3 loss, owing to the availability of NO x .The result is that in all schemes, even CheT and CheT2 that have a lower overall propensity for O x production, net O x production increases (+25 (CheT), +26 (CheT2), +46 (AQUM), +47 mol gc −1 s −1 (LLSF), Table 5).

A comparison of O 3 sensitivity to precursor emissions in global and box models
Figure 15 shows O 3 isopleths as a function of NO x and isoprene emission, similar to that in Fig. 1, but for the reduced schemes and, in this case, using O 3 mixing ratio data from the global UM-UKCA simulations.Data from all of the experiments discussed in Sect. 5 were included in Fig. 15 to maximise the NO x -isoprene emission space that was covered (the exact same emission values were earlier used to produce Figs. 1 and 3).The arrows in Fig. 15 indicate the mean emission of NO x and isoprene in the Amazon region, before and after land use change (i.e.those emissions used in the AC and LC integrations).For CheT and CheT2, the gradient of the contours is such that O 3 increases with the isoprene emission change, but for the other two schemes, O 3 decreases.This is consistent with the picture presented in Fig. 13.Comparing Fig. 15 to Figs. 1 and 3, the principal features of the isopleths derived from the global model are captured well by the box model simulations.This includes resolution of the differences between the schemes, such as the higher O 3 in LLSF and AQUM at high isoprene emissions compared to CheT and CheT2.In fact, compared to the global model, the box model simulations tend to accentuate the chemical differences between the schemes.With the isopleths derived from the global model simulations, the effects of advection and deposition somewhat buffer the chemical differences, leading to a narrower range of O 3 mixing ratios and more similar isopleths.Consequently, the O 3 levels reached in the global simulations (less than 70 ppb) are generally lower than in the box model (approaching 170 ppb).The qualitative agreement between the isopleths derived from the box model and global model suggests that the method of constructing such isopleths with the far less computationally expensive box models is a convenient way to quickly and accurately assess differences between chemical mechanisms.The fact that Figs. 1, 3 and 15 show good agreement, gives us confidence that use of the box model to compare to a near-explicit mechanism (the MCM) with reduced schemes is relevant for global model experiments.

Conclusions
Here we have examined the effect on tropospheric O 3 of using different reduced isoprene chemical mechanisms currently used in ESMs, to simulate the impact of climate change (CC), isoprene emission changes with climate (IC), anthropogenic emission changes (AC), and land use change (LC).Between the CheT and CheT2 schemes, there is no significant difference in near-surface O 3 , though OH is higher in isoprene-emitting regions in CheT2 due to the inclusion of additional HO x regeneration pathways from isoprene oxidation.For the BASE run, in the major isoprene emitting regions AQUM and LLSF give O 3 levels that are at least 10 ppb higher than with the other schemes.This is due to differences in the speciation of peroxy radicals produced by the schemes.LLSF produces a large yield of MeO 2 that rapidly reacts with NO to form O 3 .The other schemes produce ISO 2 , which has a higher rate of radical termination than MeO 2 , thus leading to less O 3 formation.AQUM produces more O 3 than CheT and CheT2 because the scheme makes less ISON and no MPAN -both important NO x sinks near the isoprene emission source.
Turning to the future perturbation experiments, in CC, the O 3 climate change sensitivity is similar in all schemes, though LLSF responds differently over the Amazon, due to the fact that no PANs are produced.In the anthropogenic emission change experiment (AC) (characterised by large NO x emission reductions in the Northern Hemisphere), all mechanisms respond in a similar way.This suggests that the O 3 -NO x response is driven largely by the simple NO x -HO x -alkane chemistry, which is the same for all schemes.
With the isoprene emission change experiments (IC and LC), there are changes in both isoprene ozonolysis (O x loss) and the RO 2 + NO flux (O x production).For the land use change experiment (LC), isoprene emissions decrease leading to a reduction in both processes.The ozonolysis changes are the same in all schemes, but the RO 2 + NO reductions differ widely between schemes.For LLSF, reductions are largest owing to the high yield of MeO 2 , which favours reaction with NO compared to higher isoprene-derived peroxy radicals (ISO 2 and MACRO 2 ).These are produced by the other schemes, leading overall to a smaller reduction in RO 2 + NO.In LLSF and AQUM the reduction in RO 2 + NO is sufficient to cause a net decrease in near surface O 3 in response to land use.This is not the case for CheT and CheT2, however, due to the formation of MPAN and ad-ditional ISON.For IC (increase in emissions), the opposite trends are calculated, though AQUM is in closer agreement with CheT and CheT2.This is most likely due to smaller net isoprene emission changes in IC compared to LC.In IC (where isoprene emissions increase under VOC-limited conditions (e.g.southeast USA)), all schemes show a net increase in near surface O 3 owing to an increase in isoprene emissions that favours O x production under such conditions.
Using the CheT scheme, Squire et al. (2014) found that the calculated increases in O 3 due to cropland expansion (LC) were too small to cause a significant increase in O 3induced vegetation damage.As the three additional mechanisms examined here simulate similar or negative changes in O 3 with cropland expansion, this conclusion would not change with the use of these schemes, and further calculations (not shown) demonstrate this to be the case.
Using the emissions and O 3 data from all of the global model experiments, we were able to construct O 3 isopleths in terms of NO x and isoprene emissions.These isopleths explain the global model response to isoprene emission changes.Using these same O 3 precursor emissions, we constructed O 3 isopleths using a box model.We find there is good qualitative agreement between those isopleths derived from the global model and those from the box model.This suggests that comparing chemical mechanisms within a box model is a computationally efficient and useful tool for assessing mechanistic performance in complex global models.Furthermore, the agreement between box and global models gives us confidence that the comparison between a box model and a near-explicit mechanism is relevant to the global model experiments.The findings reported here should help to guide mechanistic development strategies.For example, we find that the LLSF scheme produces much higher O 3 near isoprene source regions than the other three schemes.This was the only scheme where only simple peroxy radicals were produced, and, crucially there was no PAN production from isoprene chemistry.Adding simple parameterisation of PAN formation would improve the distribution of O 3 .Here, we have shown that the magnitude and even the sign of the O 3 response is affected by the choice of reduced isoprene mechanism.

Figure 1 .Figure 2 .
Figure 1.O 3 (ppb) isopleth plot as a function of NO x and isoprene emissions for the Master Chemical Mechanism (MCMv3.2).This was created from a series of box model runs.

Figure 3 .
Figure3.O 3 (percentage difference from the MCM, Fig.1) isopleth plot as a function of NO x and isoprene emissions for different isoprene chemical mechanisms.Also quoted for each plot is the mean bias (MB) from the MCM.

Figure 5 .
Figure 5. Difference in the present-day (2000, BASE) 5-year mean near-surface (< 720 m) O 3 (ppb) between CheT isoprene chemistry and other isoprene chemical mechanisms.The stippling indicates where the difference is significant at the 5 % level (greater than ±2.5 × the standard error).

Figure 6 .
Figure 6.Grid cells included in the calculation of the mean O x budget fluxes reported in Table3.Units range from 0 to 12, indicating the number of months per year that each grid cell was included in the calculation.Using emissions from the BASE run, only those months when mean isoprene emissions were greater than 0.1 mg C m −2 h −1 and mean NO x emissions were less than 0.03 mg N m −2 h −1 were included.Blue indicates that, based on this criterion, the grid cell was not included in the calculation at all.

Figure 7 .
Figure 7. Difference in the present-day (2000, BASE) 5-year mean near-surface (< 720 m) PAN (ppb) between CheT isoprene chemistry and other isoprene chemical mechanisms.The stippling indicates where the difference is significant at the 5 % level (greater than ±2.5 × the standard error).

Figure 8 .
Figure 8. Difference in the present day (2000, BASE) five year mean zonal O 3 (ppb) between CheT isoprene chemistry and other isoprene chemical mechanisms.The stippling indicates where the difference is significant at the 5 % level (greater than ±2.5 × the standard error).

Figure 9 .
Figure 9. Difference in the present-day (2000, BASE) 5-year mean O 3 dry deposition rate (mol gc −1 s −1 ) between CheT isoprene chemistry and other isoprene chemical mechanisms.The stippling indicates where the difference is significant at the 5 % level (greater than ±2.5 × the standard error).

Figure 10 .
Figure 10.BASE 5-year mean near-surface (below 720 m) OH (10 6 molecule cm −3 ) in (a) CheT and (b) the difference between CheT and CheT2.The stippling indicates where the difference is significant at the 5 % level (greater than ±2.5 × the standard error).

Figure 11 .
Figure 11.Changes in 5-year mean near-surface (< 720 m) O 3 (ppb) (2095-2000) caused by climate change (CC) for different isoprene chemical mechanisms.The stippling indicates where the difference is significant at the 5 % level (greater than ±2.5 × the standard error).

Figure 12 .
Figure 12.Changes in 5-year mean near-surface (< 720 m) O 3 (ppb) (2095-2000) caused by the change in isoprene emissions with climate (IC) for different isoprene chemical mechanisms.The stippling indicates where the difference is significant at the 5 % level (greater than ±2.5 × the standard error).

Figure 13 .
Figure 13.Changes in 5-year mean near-surface (< 720 m) O 3 (ppb) (2095-2000) caused by land use change (LC) for different isoprene chemical mechanisms.The stippling indicates where the difference is significant at the 5 % level (greater than ±2.5 × the standard error).

Figure 14 .
Figure 14.Grid cells included in the calculation of the mean O x budget fluxes reported in Table5.Units range from 0 to 12, indicating the number of months per year that each grid cell was included in the calculation.For each region (ICr1, ICr2 and ICr3) different criteria were used to select which months a grid cell should be included, as follows: (a) ICr1 is months when isoprene emissions increase by more than 0.05 Tg and the environment is NO xlimited.(b) ICr2 is months when isoprene emissions decrease by more than 0.05 Tg and the environment is NO x -limited.(c) ICr3 is months when isoprene emissions increase by more than 0.005 Tg and the environment is VOC-limited.See text for how NO x -limited and VOC-limited are defined.Blue indicates that, according to the above criteria, the grid cell was not included in the calculation at all.

Figure 15 .
Figure 15.Monthly mean surface O 3 (ppb) as a function of monthly mean NO x and isoprene emissions from UM-UKCA.Data from all of the UM-UKCA experiments in this study are used to generate this plot.The NO x and isoprene emission rates used are identical to those used in Figs. 1 and 3.

Table 2 .
Isoprene mechanism for CheT and differences between the CheT mechanism and the CheT2, AQUM and LLSF mechanisms.If a reaction is blank then it is exactly the same as in CheT, such that only the differences are shown.All rate constants (k) are in units of 10 −14 cm 3 molecule −1 s −1 .Those species not defined in Table1or in the main text are defined here: HCOOH -formic acid, H 2 O 2hydrogen peroxide, HCHO -formaldehyde.
for the MCM.When both NO x and isoprene are low, O 3 stays around the initial concentration (30 ppb).As emission rates of both O 3 precursors increase, O 3 increases reaching a maximum of 140-160 ppb at the highest emission rates used (0.5 mg N m −2 h −1 of NO x and 6 mg C m −2 h −1 of isoprene -top right-hand corner).When isoprene emissions are low and NO x emissions are high (top left-hand corner) net O 3 destruction occurs that is consistent with high nitric acid formation (O x loss via NO 2 +OH).When isoprene emissions are high and NO x emissions are low, as in a tropical rainforest, (lower right-hand corner) net O 3 destruction occurs as is consistent with high levels isoprene ozonolysis and loss of O 3 via reactions with HO x .Considering that the box model in some cases does not reach equilibrium, the precise numbers reported here (e.g.140-160 ppb) are not of significance to the real world where removal processes (i.e.deposition of intermediate compounds and O 3 and NO x ) exist.The overall behaviour of O 3 in Fig.
for which grid cells were used to calculate the values in this table.RO 2 = HO 2 + MeO 2 + RO 2 .

Table 4 .
Whole tropospheric O x budget (Tg yr −1 ) and O 3 burden (Tg) in the BASE run for different mechanisms, and changes due to climate change (CC), isoprene emissions with climate change (IC), anthropogenic emissions (AC), and land use (LC).

Table 5 .
Changes in the near-surface (below 720 m) mean O x budget fluxes from the IC experiment (mol gc −1 s −1 ).The budget is split into three regimes; ICr1 -NO x -limited regions with large increases in isoprene emissions, ICr2 -NO x -limited regions with large decreases in isoprene emissions, and ICr3 -VOC-limited regions with large increases in isoprene emissions.See Fig.14for which grid cells were included in each regime, and see text for precise definitions of the regimes.
Finally, for CheT and CheT2, the increase in O 3 caused by the reduction in isoprene ozonolysis outweighs reductions in O 3 caused by reductions in RO 2 + NO, leading overall to increases in net O x production (each +6.7 (LC), For AQUM, the lower rate of formation of NO x reservoir species compared to CheT or CheT2 leads to a greater reduction in RO 2 + NO, overall leading to a moderate reduction in net O x production (−4.0 (LC), −53 mol gc −1 s −1 (ICr2)).