The effect of climate change on vegetation, isoprene emissions and surface ozone levels was investigated using a global three-dimensional general circulation model coupled to a dynamic vegetation and chemistry models. Integrations for the 1990s and 2090s were performed. Isoprene emissions increased from 549 Tg yr−1 (1990s) to 736 Tg yr−1 (2090s) as a result of climate change with fixed vegetation. However, the isoprene emissions only increase to 697 Tg yr−1 in the 2090s when vegetation changes were included. Surface ozone levels rose by 20–30 ppbv in some locations with fixed vegetation owing to increases in precursor gases, but only by 10–20 ppbv if the vegetation changes were included. Ozone levels in parts of China, Korea and the eastern USA are predicted to exceed the World Health Organisation health limit of 60 ppbv, however, when changes in vegetation are accounted for, a much smaller area exceeds this limit.
 Plants emit a wide range of hydrocarbons and other compounds [Kesselmeier and Staudt, 1999], of which one of the most important is isoprene. This compound is highly reactive, and if sufficient levels of nitrogen oxides (NOx) are present, will produce ozone as it degrades. If the levels of NOx are low, isoprene will react directly with ozone and reduce its levels. The isoprene emission flux is estimated to lie between 200 and 570 Tg yr−1 (Kesselmeier and Staudt, , and references therein). Wang and Shallcross  studied the effect of isoprene on surface ozone levels for the present day using a coupled land surface-chemistry-transport model. They found that if isoprene emissions were included, ozone levels were 4 ppbv larger over the oceans, and 8–12 ppbv greater over mid-latitude land areas than the same simulation with no isoprene emissions.
 Isoprene also has a potentially very important role in future biosphere-climate-chemistry feedbacks [Shallcross and Monks, 2000]. The climate of the Earth at the end of the 21st century is predicted to be significantly warmer than the current climate, and so isoprene emissions are likely to increase [Turner et al., 1991]. Isoprene plays an important role in controlling levels of the OH radical due to their rapid reaction. Reaction with the OH radical is the major sink for methane. Consequently, increased emissions of isoprene may change the growth rate of atmospheric methane [Wang and Shallcross, 2000].
 Depending on the behaviour of the various plant species, isoprene emissions by 2100 could be several times larger than their current values [Shallcross and Monks, 2000]. However, these estimates do not take into account the shifts in the vegetation cover itself in response to climate change. Cox et al.  performed a climate change experiment including an interactive vegetation model and found strong feedbacks between the climate and the terrestrial biosphere. One significant impact was the drying of the Amazonian climate, resulting in significant loss of forest in that region. This change will have a strong impact on isoprene emissions, as these forests are the primary emitters of isoprene. In this paper, we examine the effect of vegetation changes as a response to climate change on isoprene emissions and surface ozone levels. Human-induced forest clearence and fires will clearly reduce the area of forest too, but are not considered in the present study.
2. Coupled Model Integrations
 The modelling was done in two stages. The first model used here is a version of the Hadley Centre Climate Model HadCM3 [Gordon et al., 2000]. Land surface processes are modelled with the MOSES2 scheme [Essery et al., 2001]. It simulates the physiological processes of photosynthesis, respiration and transpiration for each of 5 Plant Functional Types (PFTs: broadleaf tree, needleleaf tree, C3 grass, C4 grass and shrub) on the basis of the near-surface climate and atmospheric CO2 concentration. In this study atmospheric CO2 levels are prescribed according to the IS92a scenario [Pepper et al., 1992].
 The distribution of the PFTs is modelled with the TRIFFID Dynamic Global Vegetation Model [Cox, 2001], which takes the net carbon fluxes for each PFT as input and provides the fractional cover and Leaf Area Index (LAI) of each PFT as output. Competition between different PFTs is included. MOSES2 and TRIFFID therefore allow biogeophysical feedbacks between the terrestrial biosphere and the atmosphere. The HadCM3 model coupled to MOSES2 and TRIFFID was integrated continuously from 1860 to 2100. The experiment is similar to that of Cox et al.  but prescribes the level of atmospheric CO2 according to the IS92a scenario. Hence there is no direct feedback between climate and CO2, but the vegetation can change dynamically in response to changes in climate. It should be noted that the TRIFFID model does not currently include human-induced forest clearance or fires, and so will tend to overestimate forest areas.
 In the second stage, surface temperatures and sea ice extent from the coupled HadCM3-MOSES2-TRIFFID integrations were used to drive an atmosphere-only version of the climate model (HadAM3) which was coupled to the Met Office Lagrangian tropospheric chemistry model STOCHEM. This model and its performance have been discussed elsewhere [Johnson et al., 2002, and references therein]. In STOCHEM, anthropogenic emissions of precursor hydrocarbons and nitrogen oxides were in accordance with the Special Report on Emission Scenarios (SRES) A2 scenario [IPCC, 2000].
 Emission of isoprene was calculated using the well-established algorithms given by Guenther et al. . The isoprene emission is calculated from a standard flux for each class of plant species under specified conditions, which is modified by two dimensionless coefficients that are non-linear functions of photosynthetically active radiation and temperature respectively. Appropriate coefficients for these algorithms for each PFT were based on tabulated values for many different ecosystems given by Guenther et al. .
 The isoprene emission fluxes are highly uncertain, and fluxes from only a few plant species out of the large number that exist have been studied. Rosenstiel et al.  studied isoprene production from a cottonwood plantation with current and elevated levels of CO2 and found that at elevated levels of CO2, the isoprene emission was significantly reduced. It is also possible that plants may adapt to a warmer climate and consequently emit less isoprene than that calculated from the algorithms of Guenther et al. . There is insufficient data available at present to include these long-term effects. The response of different plant species to higher temperatures and elevated CO2 may also be different. In the present study, it is assumed that the emission factors for each PFT are constant.
 For the second set of integrations, the coupled chemistry-climate model was integrated for two years, and results from the second year were used in the analysis. Three chemistry-climate integrations were done and are summarised in Table 1. The first simulation is for the present day and acts as a control. Simulation 2 is for the 2090s, and includes the changed vegetation cover from the effects of climate. Simulation 3 is also for the 2090s but uses the vegetation distribution for the 1990s. A comparison of the results from simulations 2 and 3 will demonstrate the effect of the vegetation changes on isoprene emissions and surface ozone levels in the 2090s. These two simulations are identical in every other way. The global mean surface temperature differs by 4.7 K between simulation 1 and simulations 2 and 3 owing to the particular years used; however, the decadal mean surface temperature difference between the 1990s and 2090s in these runs is smaller, at 3 K.
Table 1. Summary of Coupled Chemistry-Climate Model Integrations
Isop emiss/Tg yr−1
3. Results and Discussion
 Significant changes in the vegetation cover in some regions are observed from the coupled atmosphere-ocean-land surface model integrations. As an example, the fraction of each model grid box covered by broadleaf forest in the tropics is shown in Figure 1, as this species is the primary isoprene emitter in the present work. However, emissions of isoprene by the other PFTs are also important. In Figure 1, the simulated distribution of the broadleaved forests in the tropics for the 1990s is shown in the top panel. The areas of trees predicted by the model are in good agreement with those from vegetation databases. However,the area of trees in the Amazon region is too large as forest clearance and fires are not considered. In the lower panel of Figure 1, the same map for the 2090s is shown. The dieback of forest in the Amazon region is clear. Interestingly, the model predicts little change in Africa or Indonesia.
 Needleleaved trees are found to increase in area in Alaska and Canada, Scandinavia and parts of northern Siberia, owing to warmer temperatures and elevated levels of carbon dioxide. However, a decrease is predicted to occur in parts of southern Africa, and the eastern USA. Little change in the areas of C3 and C4 is predicted to occur, except for a growth in the area of C4 grass in central Amazonia as the forest dies away. The areas of shrub in the tundra regions increase in the warmer climate.
3.1. Changes in Isoprene Emissions
 A comparison of simulations 1 and 3, which have the same vegetation distribution, show that in the future climate the isoprene emissions have increased by 187 Tg yr−1 owing to the increases in temperature. However, in simulation 2, where the vegetation change has been included, the isoprene emissions have increased by a smaller amount, 148 Tg yr−1. Therefore, vegetation changes have reduced the isoprene emission in the 2090s by 39 Tg yr−1. This reduction is mostly due to the decreased area of forest in the Amazon region. However, the model also predicts slightly smaller isoprene emissions over parts of central and southern Africa, northern China and the eastern USA when the vegetation changes are included. Conversely, isoprene emissions from southern China are predicted to increase when the vegetation changes are considered.
3.2. Changes in Surface Ozone Levels
 Anthropogenic emissions of hydrocarbons and NOx have increased considerably in the SRES A2 scenario between the 1990s and the 2090s. These extra emissions result in greater production of ozone at the emission sources, and also around these areas as the primary pollutants are transported away. The warmer temperatures in the 2100 climate mean that the chemical reactions proceed at a faster rate.
 The difference in the monthly mean surface ozone levels for July between the two future simulations (2 and 3) and the present day (simulation 1) are shown in Figures 2 and 3. At this time of year, temperatures and radiation intensities in the northern hemisphere will be near their maximum values, and hence ozone production will be at or close to its greatest. The ozone increase between 1990 and 2090 for July, with fixed vegetation can be seen in Figure 2. Generally, ozone levels over the continents have increased by 10–20 ppbv, except for northern Canada and northern Siberia, where ozone levels have fallen by about 5 ppbv. Larger increases in ozone (30 ppbv) have occurred over the eastern USA, central Europe, and parts of India. However, over northern China, Korea and Japan ozone levels are between 30 and 50 ppbv larger.
 The effect of the vegetation change on the surface ozone levels can be seen in Figure 3. Ozone levels over the Amazon are 5 ppbv smaller than in Figure 2. This change must be due to reduced emission of isoprene. Over the eastern USA ozone levels are up to 20 ppbv smaller, whereas they are 10–15 ppbv larger over central USA and Canada. Over southern China ozone levels are up to 25 ppbv larger, but 20–30 ppbv smaller over northern China and Korea.
 The changes in ozone levels are closely linked to changes in isoprene emissions. Where the isoprene emissions have fallen as a result of climate-induced vegetation changes (northern China, Korea, Amazonia, eastern USA), the ozone levels have also decreased. Where the isoprene emission has increased (southern China), ozone levels have also increased. The effect is much less marked over the Amazon region and Africa because the levels of nitrogen oxides (NOx) are smaller.
 Calculation of future isoprene emissions without considering vegetation changes could lead to an overestimation of 39 Tg yr−1. Consequently, predicted ozone levels at some continental locations are 5–30 ppbv larger. It must be stressed that these results were obtained by using only one future emission scenario. Combinations of different emission scenarios and vegetation models will doubtless produce different isoprene emissions and hence ozone levels. The isoprene emission factors for each vegetation type are highly uncertain, and the use of different factors would clearly change the results. It has also been assumed that these factors remain constant, but some evidence shows that they could change [Rosenstiel et al., 2003]. However, our results do indicate the importance of changes in vegetation coverage when calculating future levels of ozone. The increase in ozone levels of 5–30 ppbv in the present results when vegetation changes are ignored is significant. The ozone levels in July over the eastern USA and China are 60–80 ppbv, which exceeds the World Health Organisation (WHO) recommendation of 60 ppbv as an upper limit. Above this level significant damage to lung tissue and exacerbation of asthma is evident [World Health Organisation, 1994]. If the vegetation changes are included, the ozone levels in the same areas are significantly lower, and a much smaller area exceeds the WHO limit.
 We would like to thank the UK Department for Environment, Food and Rural Affairs for support through contracts EPG 1/3/164 (Air and Environmental Quality Division) and PECD 7/12/37 (Global Atmosphere Division).