We have implemented a process-based isoprene emission model in the HadGEM2 Earth-system model with coupled atmospheric chemistry in order to examine the feedback between isoprene emission and climate. Isoprene emissions and their impact on atmospheric chemistry and climate are estimated for preindustrial (1860–1869), present-day (2000–2009), and future (2100–2109) climate conditions. The estimate of 460 TgC/yr for present-day global total isoprene emission is consistent with previous estimates. Preindustrial isoprene emissions are estimated to be 26% higher than present-day. Future isoprene emissions using the RCP8.5 scenario are similar to present-day because increased emissions resulting from climate warming are countered by CO2 inhibition of isoprene emissions. The impact of biogenic isoprene emissions on the global O3 burden and CH4 lifetime is small but locally significant, and the impact of changes in isoprene emissions on atmospheric chemistry depends strongly on the state of climate and chemistry.
 The mechanisms controlling isoprene production at the cellular level are not completely understood but involve both circadian control [Wilkinson et al., 2006; Loivamäki et al., 2007; Hewitt et al., 2011] and photoperiodic up- and down-regulation of isoprene synthase production [Mayrhofer et al., 2005]. These cellular-level controls act to enable isoprene emission in the warmest part of the day and during warm seasons. Emission rates within these intervals are then determined by environmental conditions. The main environmental controls on isoprene emissions are light [e.g.,Monson and Fall, 1989], temperature [e.g., Guenther et al., 1993], and atmospheric carbon dioxide (CO2) concentration [e.g., Monson et al., 2007]. Isoprene emission increases with light and temperature until a temperature optimum of approximately 40°C [Niinemets et al., 1999]. High atmospheric CO2 concentration inhibits isoprene emissions, while low atmospheric CO2 concentration enhances isoprene emissions (see summary of studies in Pacifico et al. ).
 The strong dependence of isoprene emissions on temperature and atmospheric CO2 concentration means that higher temperature in the future should result in an increase in isoprene emissions, while higher atmospheric CO2concentrations should inhibit isoprene emissions. The opposite is likely to happen under preindustrial climate conditions. Changes in vegetation distribution are also likely to impact the global total of isoprene emissions either through anthropogenic land use change or climate-induced changes.
 Several model experiments have been performed to quantify isoprene emissions and their impact on tropospheric O3 concentration and CH4 lifetime under preindustrial, modern, and future climate conditions (see review in Pacifico et al. ), but only Young et al.  have included the effect of CO2inhibition. Here, we simulate isoprene emissions interactively using an Earth-system (ES) general circulation model (GCM) with coupled atmospheric chemistry for preindustrial, present-day, and future climate conditions. We analyze the factors contributing to changes in isoprene emissions at each time and calculate the impact of isoprene emissions on atmospheric chemistry.
 We use HadGEM2 to simulate isoprene emissions for preindustrial (1860–1869), present-day (2000–2009), and future (2100–2109) climate conditions. The causes of differences in isoprene emission are analyzed using additional sensitivity experiments. We then examine the impact of these emissions on atmospheric chemistry and climate. The experiments are summarized inTable 1.
Simulation of 1860–1869, with decadal mean CO2 concentration of 286 ppm
Baseline preindustrial simulation
Simulation of 2000–2009, with decadal mean CO2 concentration of 368 ppm
Baseline present-day simulation
Future (2100–2109) simulation driven by the RCP8.5 scenario, with decadal mean CO2 concentration of 936 ppm
Baseline future simulation
Future (2100–2109) simulation driven by the RCP2.6 scenario, with decadal mean CO2 concentration of 421 ppm
Comparison with the baseline future simulation allows identification of the impact of choice of future scenario on isoprene emission estimate
Present-day decoupled with interactive isoprene
Present-day (2000–2009) simulation where isoprene emissions are modeled interactively; atmospheric chemistry was decoupled from aerosols and radiation
Comparison of these two simulations quantifies the short-term impact of simulating isoprene emissions interactively on atmospheric chemistry and climate
Present-day decoupled with prescribed isoprene and diurnal cycle
Present-day (2000–2009) simulation with prescribed isoprene emissions and superimposed diurnal cycle of isoprene emission linked to the solar zenith angle; atmospheric chemistry was decoupled from aerosols and radiation
Present-day decoupled with prescribed isoprene and no diurnal cycle
Present-day (2000–2009) simulation with prescribed isoprene emissions and no diurnal cycle of isoprene emissions; atmospheric chemistry was decoupled from aerosols and radiation
Comparison with the present-day decoupled with prescribed isoprene simulation quantifies the importance of simulating the diurnal cycle on atmospheric chemistry
Preindustrial with prescribed preindustrial isoprene
Preindustrial (1860–1861) simulation with prescribed preindustrial isoprene emissions
Comparison of these two experiments quantifies the impact of isoprene emission on preindustrial atmospheric chemistry and climate
Preindustrial with prescribed present-day isoprene
Preindustrial (1860–1861) simulation with prescribed present-day isoprene emissions
Present-day with prescribed preindustrial isoprene
Present-day (2000–2001) simulation with prescribed preindustrial isoprene emissions
Comparison of these two experiments quantifies the impact of isoprene emission on present-day atmospheric chemistry and climate
Present-day with prescribed present-day isoprene
Present-day (2000–2001) simulation with prescribed present-day isoprene emissions
Future no CO2 inhibition
Future (2100–2109) simulation driven by the RCP8.5 scenario with the CO2 inhibition factor in the isoprene emission scheme switched off
Comparison with the baseline RCP8.5 future simulation quantifies the effect of CO2 inhibition on future isoprene emissions
2.1. Model Description
 HadGEM2 (Hadley Centre Global Environment Model 2 [Collins et al., 2011; Martin et al., 2011]) is a fully coupled Earth-system model. It is built around the HadGEM2 atmosphere-ocean GCM and includes a number of ES components, which can be coupled individually to HadGEM2. ES components that are part of the HadGEM2 Earth-system modeling framework include the ocean biosphere model diat-HadOCC (Diatom-Hadley Centre Ocean Carbon Cycle, a development of the HadOCC model ofPalmer and Totterdell ), the Top-down Representation of Interactive Foliage and Flora Including Dynamics (TRIFFID) dynamic global vegetation model [Cox, 2001], the land-surface and carbon cycle model MOSES2 (Met Office Surface Exchange Scheme [Cox et al. 1998, 1999; Essery et al. 2003]), the interactive BVOC (iBVOC) emission model, and the United Kingdom Chemistry and Aerosol (UKCA) model [Morgenstern et al., 2009]. In its default configuration (N96L38), the model has a horizontal resolution of 1.875 × 1.25° (∼200 × 140 km) and 38 vertical levels extending up to 39 km altitude. HadGEM2 runs at a 30 min time step with the exception of global radiation, which is updated every 3 h and provided between those time steps via interpolation.
 The TRIFFID vegetation module of HadGEM2 simulates the dynamics of five plant functional types (PFTs): broadleaf trees, needleleaf trees, shrubs, and C3 and C4 grass (i.e., grasses using the C3 and C4 photosynthetic pathway, respectively). Changes in the extent of croplands over time are not simulated but are prescribed from land use maps prepared for the Coupled Model Intercomparison Project 5 (CMIP5; K. E. Taylor, R. J. Stouffer, and G. A., Meehl, A summary of the CMIP5 experiment design, 2009, http://www-pcmdi.llnl.gov; www.iiasa.ac.at/web-apps/tnt/RcpDb). Here we use the historic (1850–2000 [Hurtt et al., 2009]) and the Representative Concentration Pathways (RCP8.5; 2000–2100 [Riahi et al., 2007]) data sets, as described in Jones et al. . A further four surface types (urban, inland water, bare soil, and ice) are used in the land-surface scheme for the calculation of water and energyexchanges between the land and the atmosphere. Each model grid box can include varying proportions of several vegetation and/or surface types.
 The isoprene emission scheme is part of the iBVOC model coupled to the leaf physiology scheme in MOSES2. The BVOC emission scheme sits on top of the leaf physiology scheme. As a consequence, the BVOC emission scheme is sensitive to changes in photosynthetic activity, including changes induced by water stress, and vegetation competition. Thus, both leaf physiology and BVOC emissions are a consequence of light, temperature, precipitation, soil water content, plant nitrogen content, and CO2levels. In the BVOC emission scheme, the rate of isoprene production is controlled by the production of electrons during photosynthesis (PSII: Photosystem II) and the estimated energy and redox-equivalents requirements to reduce isoprene from the initial steps of carbon assimilation. The inhibitory effect of increasing CO2 concentration is taken into account using an empirical relationship derived from Arneth et al.  that is dependent on leaf internal CO2concentration and implicitly includes the short-term response of isoprene emission to drought stress [Arneth et al., 2007]. In MOSES2, the ratio of leaf internal CO2 concentration and atmospheric CO2 concentration can be considered constant at first approximation [Cox, 2001]. The scheme uses PFT-specific base emission rates for isoprene (or isoprene emission factors), specifically 35μgC/gdw/h (where gdw is gram dry weight) for broadleaf trees, 12 μgC/gdw/h for needleleaf trees, 16 μgC/gdw/h for C3 grass, 8 μgC/gdw/h for C4 grass, and 20 μgC/gdw/h for shrubs [Pacifico et al., 2011]. The off-line version of this isoprene emission scheme reproduces the diurnal, day-to-day, and seasonal variability of above-canopy isoprene fluxes as measured at several flux sites and also the spatial patterns of isoprene emission over Amazonia and Asia as recorded by satellite-derived data [Pacifico et al., 2011]. The model overestimates isoprene emissions compared to flux-tower observations by a factor of 1.7 to 3.9, reflecting the use of highly uncertain generic emission factors for different types of vegetation as well as not including isoprene loss through the canopy.
 Atmospheric chemistry in HadGEM2 is simulated by the UKCA model. In this study, we use UKCA in the Extended Tropospheric Chemistry (UKCA-ExtTC) version. The isoprene oxidation mechanism used in this study is the Mainz Isoprene Mechanism as described inPöschl et al. . This is the same isoprene mechanism as the one used in Telford et al. , but it does not include any OH recycling within the canopy or the boundary layer, although recent work suggests it may be important [Lelieveld et al. 2008; Hofzumahaus et al. 2009]. UKCA-ExtTC simulates the spatial distribution and evolution in time of 89 chemical species, 63 of which are model tracers. UKCA-ExtTC includes emissions from anthropogenic, biogenic, soil, and wildfire sources for 17 species: nitrogen oxides (NOX = NO + NO2), CH4, carbon monoxide (CO), hydrogen (H2), methanol, formaldehyde, acetaldehyde and higher aldehydes, acetone, methyl ethyl ketone, ethane (C2H6), propane (C3H8), butanes and higher alkanes, ethylene (C2H4), propane (C3H6), isoprene, (mono-)terpenes, and a lumped species representing aromatics (toluene + xylene) from anthropogenic sources. Anthropogenic emissions are prescribed from CMIP5 inventories, while wetland methane emissions are simulated with data fromGedney et al. . Emissions of biogenic species (isoprene, terpenes, methanol, acetone) are computed by iBVOC and provided to UKCA at every time step. The isoprene emission scheme is that of Pacifico et al. , as described above. Terpenes, methanol, and acetone emissions are simulated with the model described in Guenther et al. . Anthropogenic and wildfire emissions are prescribed from monthly mean emission data sets prepared for CMIP5: we use the historic data set [Lamarque et al., 2010] for our preindustrial simulations and the RCP8.5 [Riahi et al., 2007] emission scenario for future simulations. Soil-biogenic NOx emissions are prescribed using the monthly distributions provided by Global Emissions Inventory Activity (http://www.geiacenter.org/inventories/present.html), which are based on the global empirical model of soil-biogenic NOx emissions ofYienger and Levy . NOX emissions from global lightning activity are parameterized based on the convective cloud top height following Price and Rind [1992, 1994]and are thus sensitive to the model climate. UKCA also includes a dry deposition scheme based on the resistance in-series approach as outlined inWesely . Physical removal of soluble species is parameterized as a first-order loss process based on convective and stratiform rainfall rates.
2.2. Description of the Baseline Simulations
 We made decade-long simulations of preindustrial (1860–1869), present-day (2000–2009), and future (2100–2109) conditions, using HadGEM2 in its atmosphere-only configuration, i.e., with all (implemented) couplings between atmosphere and land surface (including carbon cycle) active but without coupling to the ocean and ocean biosphere. HadGEM2 was initialized with equilibrium concentrations of the major chemical components (O3, CO, H2, total reactive nitrogen (NOy), BVOCs) taken from the CMIP5 simulation (see description of the simulations in Jones et al. ). Methane concentrations were prescribed for each of the time periods as specified by CMIP5, with values of 805, 1750, and 3750 ppb for the preindustrial, present-day, and RCP8.5 future climate simulation, respectively. The decade-mean CO2atmospheric concentration was 286, 368, and 936 ppm for the preindustrial, present-day, and RCP8.5 future climate simulation, respectively.
 Monthly means of sea surface temperature and sea ice cover were prescribed for each of the three simulation periods using climatologies derived from the appropriate decade of the Hadley Centre CMIP5 1850–2100 transient climate run [Jones et al., 2011]. The vegetation distribution for each of our simulations was prescribed using the simulated vegetation averaged for the same decade from this transient climate run, on which we superimposed crop area as given in the CMIP5 historic and future land use maps [Hurtt et al., 2009; Riahi et al., 2007]. The preindustrial and present-day simulations were driven by historical scenarios [Lamarque et al., 2010] of anthropogenic and wildfire emissions. Future emission data sets for anthropogenic and wildfire sources were taken from the RCP 8.5 scenario [Riahi et al., 2007]; these are standard input data sets specified for CMIP5 as used for the HadGEM2 CMIP5 simulations [Jones et al., 2011].
 The model was run for 11 years, and the last 10 years were used for subsequent analysis. This set of experiments allows us to simulate isoprene emissions and their impact on atmospheric chemistry and thus to quantify the sensitivity of isoprene emissions to past, present, and future environmental conditions.
 To assess the impact of isoprene emissions on atmospheric chemical composition, we have analyzed the change in tropospheric O3 burden and CH4 tropospheric lifetime. The tropospheric domain in this context extends from the ground to the chemical tropopause defined as the 150 ppbv O3 contour level. The tropospheric CH4 lifetime has been derived using the chemical definition [Seinfeld and Pandis, 1998] with respect to reaction with the hydroxyl radical (OH) only and does not consider other CH4 sinks (e.g., dry deposition at the surface, stratospheric exchange). Methane chemical lifetime has been applied consistently in all cases.
2.3. Description of the Additional Experiments
 We performed several additional experiments to study (a) the impact of different future scenarios on isoprene emissions and (b) the importance of modeling isoprene emissions interactively as opposed to using prescribed emissions and to quantify (c) the impact of isoprene emission on atmospheric chemistry and climate and (d) the effect of CO2 inhibition on future isoprene emissions.
 To study the impact of different future scenarios on isoprene emission estimate, we performed an additional future (2100–2109) simulation driven by the RCP2.6 scenario (Table 2) [Moss et al., 2010]. We used RCP8.5 and RCP2.6 because they represent, respectively, the “high” and the “low” future scenario in terms of global warming.
Table 2. Global Annual Total Isoprene Emission, Global Annual Total GPP, Global Annual Mean CO2Atmospheric Concentration, and Global Annual Mean Air Temperature at 1.5 m for the HadGEM2 Preindustrial, Present-Day, and Future Climate Simulations
Future (RCP 8.5) 2100–2109
Future (RCP 2.6) 2100–2109
Isoprene emissions (TgC/yr)
Air temperature (°C)
 To evaluate the short-term impact (3 hourly) of simulating isoprene emissions interactively, we compared a present-day experiment where isoprene emissions are modeled interactively to two experiments with prescribed isoprene emissions. The model setup was the same as in the present-day base simulation described above, except that atmospheric chemistry was decoupled from aerosols and radiation in all three experiments to ensure that they all have the same meteorology. In the two noninteractive experiments, monthly mean climatologies of isoprene emissions were prescribed to represent the seasonal cycle of isoprene emission. The climatologies were constructed from monthly mean isoprene emission fluxes from the present-day interactive-emission simulation averaged over the entire decade. In one of these experiments, there is no diurnal cycle of isoprene emissions, while in the other a simple diurnal cycle linked to the solar zenith angle was superimposed to avoid simulating isoprene emission during the night. These last two simulations were performed to test the impact of using a mean value of daily isoprene emission as opposed to imposing a diurnal cycle.
 To evaluate the impact of isoprene emissions on atmospheric chemistry, we performed two preindustrial simulations and two present-day simulations, with prescribed preindustrial isoprene emissions for one of each pair and prescribed present-day isoprene emissions for the other. Isoprene emissions were prescribed as monthly mean climatologies, constructed as described above. Atmospheric chemistry was decoupled from aerosols and radiation to ensure the same meteorology in the preindustrial pair of experiments and in the present-day pair of experiments.
 To evaluate the impact of the CO2 inhibition effect on isoprene emission, we performed a RCP8.5 future simulation where the model setup was the same as the baseline future simulation (section 2.2), but with the CO2 inhibition factor in the isoprene emission scheme switched off.
3.1. Sensitivity of Isoprene Emissions to Past, Present, and Future Climate
 HadGEM2 simulates global annual isoprene emissions under preindustrial, present-day, and 2100 RCP8.5 climate conditions of 579, 460, and 456 TgC/yr, respectively (Table 2). This is a decrease of 20% in simulated isoprene emissions between the preindustrial and the present day and a further decrease of 1% between the present day and the 2100. The global annual total gross primary productivity (GPP) increases by 23% between the pre-industrial and present-day simulations and by 64% between the present-day and 2100 runs (Table 2). The simulated GPP in the future run using the RCP2.6 scenario is 162 PgC/yr, which is 32% lower than the GPP simulated using the RCP8.5 scenario. However, simulated isoprene emissions for 2100–2109 using the RCP2.6 scenario are only 1% higher (461 TgC/yr) than those simulated using the RCP8.5 scenario, which is not a significant difference as it is less than the inter-annual variability.
 The superimposed anthropogenic land use change is the major driver of change in vegetation cover on top of simulated natural vegetation dynamics. The fractional cover of shrubs, broadleaf trees, and needleleaf trees decreases by 4.9%, 4.1%, and 2.9%, respectively, between the preindustrial and present-day simulations driven largely by anthropogenic land use change. The fractional cover of grasses is increased between the two simulations by 8.5% and 2.4% for C3 and C4 grasses, respectively (Table 3 and Figure 1), reflecting an increase in crop and pasture area. The fractional cover of needleleaf trees is decreased 0.3% between the present-day and RCP8.5 future simulations, while the fractional cover of C3 grass, shrubs, and broadleaf trees increases by 6.6%, 1.7%, and 0.4%, respectively. The difference in C4grass fractional cover between the present-day and future simulations is not significant, as it is less than the inter-annual variability (Table 3 and Figure 1).
Table 3. PFT-Specific Isoprene Emission Factors (IEFs) Used in the HadGEM2 Simulationsa
Percentage of land covered by each PFT and percentage of isoprene emitted by each PFT for the HadGEM2 preindustrial, present-day, and future climate simulations.
Preindustrial land fraction (%)
Present-day land fraction (%)
Future (RCP 8.5) land fraction (%)
Preindustrial isoprene emission (%)
Present-day isoprene emission (%)
Future (RCP 8.5) isoprene emission (%)
 The changes in vegetation cover between the three experiments lead to differences in the partitioning of isoprene emissions by PFT. Isoprene emissions from shrubs, needleleaf trees, and broadleaf trees decrease between the preindustrial to present-day runs by 3%, 4%, and 1%, while isoprene emissions from C3 and C4 grasses are increased by 6% and 1% (Table 3). Isoprene emissions of needleleaf trees decrease by 3% between the present-day and RCP8.5 future simulations; emissions from C3 grass, shrubs, and broadleaf trees increase by 2%, 1%, and 1%, respectively, while emissions from C4 grasses are unchanged (Table 3).
 Tropical areas are the main source of isoprene for all the time periods considered (Figure 2). The area of rain forest in Amazonia, characterized by high isoprene emissions, varies through time resulting in higher emissions in the preindustrial and lower emissions in the future (2100–2109 RCP8.5 scenario). Similarly, isoprene emissions over Southeast Asia and Indonesia are higher in the preindustrial and lower in the future simulation. There is little change in emissions from tropical Africa and extratropical regions such as North America, Europe, and Australia.
 To quantify the impact of vegetation change alone on isoprene emissions, we scaled present-day isoprene emissions with the preindustrial and 2100 PFT distribution map. This causes a 19% (87 TgC) increase in global annual total isoprene emissions for preindustrial compared to present-day (Table 4) due to an increase in isoprene emissions from broadleaf trees (especially over southern Amazonia) and from shrubs (especially over central Africa) and a general decrease of emissions from C3 grass (Figure 3). Scaling present-day isoprene emissions with the RCP8.5 2100 PFT distribution map causes an 8% (37 TgC) decrease in isoprene emissions compared to present-day (Table 4). These future changes in vegetation cover do not have a major impact on isoprene emissions from any of the PFTs: there is a small increase of emissions from C3 grass over central Africa, a small decrease of emissions from broadleaf trees over tropical areas, and a general decrease of emissions from needleleaf trees and shrubs (Figure 3).
Table 4. Quantitative Attribution of Causes of Differences Between Present-Day and Preindustrial and Between Present-Day and Future Isoprene Emissions
Change to Present-Day Isoprene Emissions
Implied Change in TgC
Preindustrial PFT distribution
Preindustrial CO2 atmospheric concentration
Future PFT distribution
Future CO2 atmospheric concentration
 Lower CO2 atmospheric concentration during the preindustrial period directly contributes to higher isoprene emission than would be expected under preindustrial climate conditions. To quantify the direct effect of changes in CO2atmospheric concentration between preindustrial and present-day, we scaled present-day isoprene emissions with the preindustrial CO2 factor [Pacifico et al., 2011]. For simplicity, the CO2 factor was specified according to the ratio of atmospheric CO2 concentration rather than the ratio of leaf internal CO2 concentration. The impact of using the preindustrial CO2factor for present-day isoprene emissions is a 29% (133 TgC) increase in global total annual emissions compared to the present-day simulation (Table 4). The remaining 22% (101 TgC) reduction in global total isoprene emissions in the preindustrial compared to present-day simulations is caused by the impact of climate on isoprene emissions (Table 4). The higher atmospheric CO2 concentration (936 ppm) in the RCP8.5 scenario leads to a decrease in isoprene emission. The impact of using the future CO2factor over present-day isoprene is a 61% (283 TgC) decrease in isoprene emissions compared to the present-day simulation (Table 4). Thus, the remaining 69% (316 TgC) increase in global total isoprene emissions in the future compared to present-day simulation is due to climate (Table 4).
3.2. Impact of Modeling Isoprene Emissions Interactively
 The diurnal cycle (3 hourly resolution) of isoprene concentrations over North America, South America, Europe, Southeast Asia, and Africa from the simulation with prescribed isoprene emissions and superimposed diurnal cycle and that with interactive isoprene emissions are highly correlated (correlation coefficients >0.9) for every region. However, the diurnal cycle of isoprene concentrations from the simulation with prescribed isoprene emissions and no diurnal cycle is anticorrelated with the simulation with interactive isoprene emissions: it unrealistically shows higher isoprene concentrations at nighttime than during the day (Figure 4).
3.3. The Impact of Isoprene Emission on Atmospheric Chemistry and Climate
 The tropospheric ozone burden increases by 40% (99 Tg) between the pre-industrial and present-day simulations, and by 15% (54 Tg) between the present-day and future RCP8.5 simulations (Table 5). The increase in O3 burden is mainly due to the increased atmospheric concentration of most O3 precursors, such as NOX, CO, CH4, and nonmethane hydrocarbons, in response to human activities [Denman et al., 2007].
Table 5. Estimates of O3 Burden, CH4Lifetime, SOA From Terpenes, SOA From Aromatic Compounds, SOA From Isoprene, and Total SOA for the HadGEM2 Preindustrial, Present-Day, and Future Climate Simulations
O3 Burden (Tg)
CH4 Lifetime (years)
SOA From Terpenes (Tg)
SOA From Aromatic Compounds (Tg)
SOA From Isoprene (Tg)
Total SOA (Tg)
Future (RCP 8.5)
 Methane lifetime decreases by 18% (2.49 years) between the preindustrial and present-day simulations and by 2% (2.4 months) between the present-day and future simulations (Table 5). The decrease in methane lifetime in the present-day and future simulations is due to the increase in the oxidizing power of the atmosphere, as a consequence of human activities. The presence of NOX, CO, and various hydrocarbons in the troposphere affects the concentration of the OH radical and therefore the oxidizing power of the atmosphere. Also in most regions of the atmosphere, the photolysis of ozone is a source of OH too [Denman et al., 2007].
 The impact of changes in isoprene emissions between preindustrial and present-day has different impacts on atmospheric chemistry depending on the background state. Under preindustrial conditions, a 20% decrease in isoprene emissions leads to a 0.7 Tg (0.2%) increase in the O3 burden and an ∼9 month decrease in the CH4 lifetime. The decrease in CH4 lifetime causes an 80 ppb decrease in CH4 atmospheric concentration and a decrease in CH4 radiative forcing of 44 mW/m2.
 However, under present-day climate conditions, the same 20% decrease in isoprene emissions leads to a 2 Tg (O3) decrease in the O3 burden and an ∼3 month decrease in the CH4 lifetime. This suggests that 2% of the simulated decrease in O3burden between the preindustrial to present-day simulations is caused by biogenic isoprene emissions. The decrease in CH4 lifetime leads to a 59 ppb decrease in CH4 atmospheric concentration and decreases the CH4 radiative forcing by 22 mW/m2.
 Regionally, the impact of the 20% decrease in global total isoprene emissions on the spatial patterns of tropospheric O3 under preindustrial climate conditions is to reduce tropospheric O3 over America, South Africa, and most of Europe and Asia (Figure 5). The same change in isoprene emissions under present-day climate inverts the O3 trend (now positive) over the eastern United States and most of Europe and Asia (Figure 5). Our model does not include oceanic isoprene emissions, hence the impact over these areas is minimal.
 The impact of changes in isoprene emissions on atmospheric chemistry depends strongly on the state of climate and chemistry; in particular, a decrease in isoprene emissions leads to an increase in O3 burden under preindustrial climate conditions and a decrease in O3burden under present-day climate conditions because of the different level of NOX in the atmosphere. Under preindustrial conditions, NOXemissions are very low as they are essentially only due to wildfires, soil-biogenic NOx, and lightning, with the latter being smaller at the preindustrial compared to present-day because of cooler climate conditions and less convection. In these conditions, isoprene reactions with OH and O3 compete with each other and the net results is a destruction of O3. At the present-day NOXemissions are much higher. In our simulations, the increase is mainly due to anthropogenic activity (fossil fuel burning) but also a prescribed increase in wildfires and a simulated increase in lightning because of warmer climate and more convection. Under present-day conditions, isoprene reactions with OH dominate the isoprene sinks and isoprene becomes an effective O3 precursor, apart from remote/pristine areas with low NOX where isoprene destroys O3. The impact of isoprene on O3burden is bigger in magnitude at present-day (2 Tg decrease in O3 burden) compared to preindustrial (0.7 Tg increase in O3burden). This is because in the present-day atmosphere, the NOX/VOC ratio is closer to the optimum where O3 formation is most efficient, so that the NOX/VOC ratio is more sensitive to changes in VOCs and isoprene. The preindustrial atmosphere is much less saturated in NOX and so the NOX/VOC ratio is much further away from the optimum; hence, the atmosphere is less sensitive to changes in VOCs and isoprene in terms of O3 production efficiency.
 Isoprene has its primary sink through reactions with OH, competing with CH4. Under both preindustrial and present-day climate conditions, lower isoprene emissions increase the OH sink for atmospheric CH4 and cause a decrease in CH4 lifetime. Decrease in CH4lifetime is 9 months at preindustrial and 3 months at present-day; this difference in magnitude is due to the different state of atmospheric chemistry. In the preindustrial atmosphere, the total amount of reactive carbon (CH4+ CO + VOC) is dominated by natural sources of reactive carbon. Under present-day conditions anthropogenic emissions of CO and VOC are much higher [Denman et al., 2007]. Hence, the preindustrial atmosphere is much more sensitive to changes in isoprene because isoprene accounts for a substantially higher portion of the reactive carbon emission flux than under present-day conditions.
 The impact of the 20% decrease in global total isoprene emissions under both preindustrial and present-day climate conditions leads to a decrease in SOA burden from isoprene of 0.1 Tg (20% decrease).
3.4. Study of the Impact of the CO2 Inhibition Effect on Future Isoprene Emissions
 When the CO2 inhibition effect on isoprene emissions is not taken into account, simulated isoprene emission for 2100 is 859 TgC/yr under the RCP8.5 scenario. Including the CO2 inhibition effect on isoprene emissions causes a 47% decrease in these emission estimates. The effect of including the CO2 inhibition effect on isoprene emissions results in 4 Tg increase in O3 burden and a 10 month decrease in CH4 lifetime. The impact of including the CO2 inhibition effect leads to high variability in tropospheric O3 concentrations, with more tropospheric O3 over tropical forested areas, and over some areas of western China and western Siberia, and less tropospheric O3 over Europe, the Sahara desert, and the Arabian peninsula (Figure 6). CO2 inhibition of isoprene emissions causes a decrease in isoprene and, consequently, a decrease in O3 in high NOX areas, where isoprene is an O3 precursor, and an increase in low NOX areas, where isoprene is responsible for O3 destruction.
4. Discussion and Conclusions
 Our estimate of 460 TgC/yr for present-day global total isoprene emission is consistent with previous estimates, which range between 400 and 600 TgC/yr [Arneth et al., 2008]. This estimate is slightly lower than the value obtained with the off-line version of our model (535 TgC/yr) [seePacifico et al., 2011], but the off-line estimate was made with a different model setup (different vegetation distribution and climate) and for the interval 1990–1999 rather than 2000–2009 as in the present case. A large factor of uncertainty in the estimate of isoprene emissions is given by basal isoprene emission factors; these factors do not change with time, so their uncertainty will affect absolute emissions but not the relative changes we simulate.
 Preindustrial isoprene emissions are significantly larger than present-day emissions in our simulations. Although the climate warming between the preindustrial and present-day simulations tends to increase isoprene emissions, this is more than offset by the decreased emissions as a consequence of CO2 inhibition and, though to a lesser extent, land use and vegetation changes.
 Global isoprene emissions by 2100 under the RCP8.5 scenario are similar to simulated present-day emissions because of the cancellation of climate-induced increases in emissions by decreases due to the CO2 inhibition effect. Using the alternative RCP2.6 future climate scenario does not have a significant impact on future isoprene emission estimates. This is perhaps surprising given the approximately 3°C difference in global mean temperature between the two simulations and very different atmospheric CO2 by 2100 (Table 2). This demonstrates that the cancellation of temperature and CO2 effects holds true for a wide range of future scenarios. The main reason behind this cancellation is due to the fact that the CO2 inhibition effect scales as 1/CO2 [Pacifico et al., 2011], while isoprene emissions scale as exp(T) [see Pacifico et al., 2011]. But the temperature change driven by CO2 radiative forcing scales with log(CO2). Hence, the combined total cancels out.
 Changes in vegetation cover between the present-day and future simulations have only a small impact on isoprene emissions (8% decrease of isoprene emissions; seeTable 4), much smaller than the impact of such changes between the preindustrial and present-day simulations (19% increase of isoprene emissions; seeTable 4). However, the simulated change in vegetation, and particularly in the cover of high emitters (broadleaf trees and shrubs), is also much smaller between the present-day and future simulations than between the preindustrial and present-day simulations. A similar impact of future vegetation distribution on isoprene emission (7% decrease in isoprene emissions from 1990s to 2090s) was simulated inSanderson et al. , albeit with a different isoprene emission scheme and a different model setup. However, future land use varies between scenarios and is hard to project with confidence [Hurtt et al., 2009; Thomson et al., 2010; Tilman et al., 2011]. Different land use and land use change could have significant impacts on biogenic isoprene emissions [e.g., Ashworth et al., 2012].
 Although isoprene emissions depend on GPP, the two show an opposite pattern of behavior between the set of simulations: isoprene emissions decrease while GPP increases between the preindustrial and present-day and between the present-day and future simulations. The lower GPP in the preindustrial simulation is a result of cooler climate and lower CO2 atmospheric concentrations. The high isoprene emissions in the preindustrial simulations are a consequence of the greater extent of highly emitting PFTs (broadleaf trees and shrubs) and the CO2enhancement effect on isoprene emissions. GPP is higher in the future than in the present-day simulation because of warmer temperatures and the CO2 fertilization effect. However, the CO2inhibition effect causes a decrease in isoprene emissions in the future simulation. This is compounded by a reduction in the extent of high-emitting PFTs like broadleaf trees and shrubs.
 Our simulations show that CO2inhibition/enhancement is more important than climate-induced changes in vegetation distribution in determining isoprene emissions under historic and future climate conditions. Other modeling studies [Heald et al., 2009; Young et al., 2009] have also shown that CO2 inhibition offsets the large increases in future isoprene emissions predicted in earlier studies [see, e.g., Sanderson et al., 2003; Lathière et al., 2005; Wiedinmyer et al., 2006]. The inclusion of CO2 inhibition/enhancement in current models is based on observational studies (see review in Pacifico et al. ). However, there are still only a limited number of studies of isoprene inhibition/enhancement, mostly focusing on leaf-level measurements on temperate species, and thus some uncertainty in the exact form of the response curve [seeYoung et al., 2009] and its applicability outside temperate biomes. Given the apparent importance of CO2inhibition/enhancement of isoprene emissions, as shown by our simulations, more and longer-term measurements on a wider range of species would be useful in order to refine this function for further modeling work.
 Including the CO2 inhibition effect leads to an overall decrease in the O3 burden in our future simulation. However, there is considerable spatial heterogeneity in the changes in tropospheric O3, with increases over tropical forested areas, China and Southeast Asia (Figure 6). Ozone formation in low-NOXenvironments is very sensitive to the type and lifetime of organic nitrates formed in the isoprene oxidation processes. Ozone increase in low-NOX environments has already been observed by Young et al.  and attributed to (a) a decrease of direct destruction of O3 through ozonolysis of isoprene and (b) reduced sequestration of NOX by isoprene oxidation products (nitrates and PeroxyAcyl nitrates (PAN)), which increased NOX levels and O3 production. Due to the increase in the ozone, more OH is formed and the chemical lifetime of isoprene nitrate and PAN decreases, releasing even more NOX to the atmosphere that would otherwise be deposited to the ground.
 Simulating isoprene with prescribed isoprene emissions without a superimposed diurnal cycle does not provide realistic results. Simulating isoprene emissions interactively rather than with prescribed emissions and a superimposed diurnal cycle does not produce a big impact on atmospheric chemistry on a global and regional level, but it is the most realistic way to estimate isoprene emissions for past and future climate conditions. Furthermore, interactive simulations could be used to investigate how stresses on vegetation, such as ozone damage of plant leaves [Sitch et al., 2007], might affect isoprene emissions.
 The simulated overall impact of biogenic isoprene emission on global O3 burden and CH4 lifetime is small. This is in line with previous studies reporting that between 91% and 92% of the simulated 21st century higher concentrations in O3 is related to direct effects of anthropogenic emissions, with the remainder of the increase attributable to secondary effects of climate change combined with biogenic precursor emissions [Denman et al., 2007]. The contribution of biogenic isoprene emissions to the methane radiative forcing at present-day with respect to preindustrial is about one order of magnitude smaller than that caused by anthropogenic change in methane concentration [Denman et al., 2007]. Decreased isoprene emissions leads to an increase in O3 burden under preindustrial climate conditions and a decrease in O3burden under present-day climate conditions because of the different level of nitrogen oxides in the atmosphere. Under preindustrial conditions, NOXemissions are very low, while present-day NOX emissions are much higher. In our simulations, the increase is mainly due to anthropogenic activity (fossil fuel burning) but also a prescribed increase in wildfires (from the CMIP5 scenario [Lamarque et al., 2010]) and a simulated increase in lightning because of a warmer climate and more convection. Recent work [Marlon et al., 2008; Wang et al., 2010] indicates that the assumed increase in wildfires and hence wildfire emissions during the 20th century is unrealistic, since observations show a pronounced downturn in fire during this interval and biomass-burning levels today are lower than during the preindustrial period. Use of a more realistic scenario for biomass burning would produce changes in our estimates of OH and CH4, but the differences would be small because the change in NOXbetween preindustrial and present-day is largely conditioned by the fossil-fuel burning.
 Soil-biogenic NOxemissions are prescribed in our simulations and do not change between preindustrial, present-day, and future climate conditions. Soil-biogenic NOx emissions are important in remote and rural areas [Ganzeveld et al., 2002] where isoprene emissions can be high. Since NOx emissions are controlled by temperature [Arneth et al., 2010], we would expect a reduction in soil NOx in preindustrial times and increased NOx under future conditions (i.e., changes in the same sense as those from biomass burning and anthropogenic activity). However, NOx emissions are also influenced by vegetation type, and changes in vegetation cover could compensate, to some extent, for temperature-induced changes. Further uncertainty is introduced because of the potential impact of the postindustrial use of nitrogen fertilizer on soil NOx emissions [Lathière et al., 2006]. Given these uncertainties, we have not included a time-varying estimate of soil NOx emissions in our experimental setup. However, as is the case for biomass burning, we expect changes in soil NOx to have only a small impact on OH and CH4budgets between preindustrial and present-day because the change in NOXis overwhelmingly driven by fossil-fuel burning.
 Our model overestimates isoprene emissions compared to flux-tower measurements [Pacifico et al., 2011]. There are large differences between sites in the magnitude of the overestimation and too few observations to be able to derive a robust estimate of the implied downscaling factor that would be required to correct this bias. This bias is one of the reasons that we focus on the differences between the preindustrial, present-day, and future simulations. However, this overestimation of isoprene emission probably means that we also overestimate the impact of isoprene on atmospheric chemistry; more realistic simulation of isoprene emissions would reduce this impact still further.
Hewitt et al.  included a parameterization of circadian control on isoprene emissions in the Model of Emissions of Gases and Aerosol from Nature (MEGAN) model of Guenther et al. and showed that this reduced isoprene emissions and improved the agreement with observations of isoprene emissions and ground-level ozone. However, it is unclear that circadian control of isoprene synthase production has any impact on plant- or ecosystem-level emissions independent of other environmental controls, and the results from MEGAN are not conclusive because there are many other molecular-level controls on isoprene emission that are not taken into account in this (as in other) models.
 Our estimate of SOA burden of 0.5 Tg for present-day (Table 5) is in agreement with previous studies [e.g., Carlton et al., 2009; Lin et al., 2012]. We have shown that the decrease in isoprene emissions between the preindustrial and present-day caused a decrease in SOA burden of approximately 0.2 Tg. A recent plant chamber experiment with species typical of boreal forest has shown evidence that isoprene can significantly inhibit new particle formation, i.e., aerosol nucleation rates [Kiendler-Scharr et al., 2009]. This potential effect is not taken into account in the current modeling scheme and needs to be investigated further. In particular, this result has been debated in a model simulation showing that an increase in isoprene emission in pristine and moderately polluted environments may not suppress aerosol formation, [Taraborrelli et al., 2012].
 There are clearly a large number of uncertainties associated with predictions of changes in isoprene emissions and on the relative magnitude of emission-chemistry-climate feedback: additional field studies, both of CO2inhibition/enhancement and of isoprene-inhibition of SOA formation, would be helpful to reduce these uncertainties. Nevertheless, coupled modeling provides an important tool for the exploration of this complex system and shows that simplistic projections of both the likely trajectory of isoprene emissions and their impact are likely to be unrealistic because of the complexity of competing influences on both emissions and atmospheric chemistry. This study has shown the importance of considering these processes in a coupled system.
 This paper is a contribution to the GREENCYCLES (MRTN-CT-2004512464) (F.P., S.P.H.). This work has been partly funded by EUCAARI (European Integrated project on Aerosol Cloud Climate and Air Quality interactions) 036833-2. F.P., G.A.F., C.D.J., and W.J.C. were supported by the Joint DECC/Defra Met Office Hadley Centre Climate Programme (GA01101).