Response of isoprene emission to ambient CO2 changes and implications for global budgets

We explore the potential role of atmospheric carbon dioxide (CO2) on isoprene emissions using a global coupled land–atmosphere model [Community Atmospheric Model–Community Land Model (CAM–CLM)] for recent (year 2000, 365 ppm CO2) and future (year 2100, 717 ppm CO2) conditions. We incorporate an empirical model of observed isoprene emissions response to both ambient CO2 concentrations in the long‐term growth environment and short‐term changes in intercellular CO2 concentrations into the MEGAN biogenic emission model embedded within the CLM. Accounting for CO2 inhibition has little impact on predictions of present‐day global isoprene emission (increase from 508 to 523 Tg C yr−1). However, the large increases in future isoprene emissions typically predicted in models, which are due to a projected warmer climate, are entirely offset by including the CO2 effects. Projected global isoprene emissions in 2100 drop from 696 to 479 Tg C yr−1 when this effect is included, maintaining future isoprene sources at levels similar to present day. The isoprene emission response to CO2 is dominated by the long‐term growth environment effect, with modulations of 10% or less due to the variability in intercellular CO2 concentration. As a result, perturbations to isoprene emissions associated with changes in ambient CO2 are largely aseasonal, with little diurnal variability. Future isoprene emissions increase by more than a factor of two in 2100 (to 1242 Tg C yr−1) when projected changes in vegetation distribution and leaf area density are included. Changing land cover and the role of nutrient limitation on CO2 fertilization therefore remain the largest source of uncertainty in isoprene emission prediction. Although future projections suggest a compensatory balance between the effects of temperature and CO2 on isoprene emission, the enhancement of isoprene emission due to lower ambient CO2 concentrations did not compensate for the effect of cooler temperatures over the last 400 thousand years of the geologic record (including the Last Glacial Maximum).


Introduction
Isoprene (C 5 H 8 , 1-methyl-1,3-butadiene) makes up the largest fraction of nonmethane isatile organic compounds emitted into the atmosphere, with an estimated global source of 440-660 Tg C yr À1 (Guenther et al., 2006). This highly reactive compound plays a key role in tropospheric chemistry and climate, as a precursor to both ozone (Wang & Shallcross, 2000) and secondary organic aerosol formation (Kroll et al., 2006), and as a control over the lifetime of tropospheric methane (Kaplan et al., 2006). The dominant isoprene source (490%) is emission from vegetation, and these emissions are highly sensitive to temperature (Monson et al., 1992). Thus, not only does isoprene emission influence climate, but also climate influences isoprene emission, thereby representing an important climate-driven feed-back between photochemical processes in the atmosphere and biogeochemical processes in the terrestrial biosphere. Understanding the isoprene emission response to the meteorological and phenological environment is vital to predicting the evolution of tropospheric composition and climate forcing. Here, we investigate the implications of the isoprene response to changes in atmospheric carbon dioxide (CO 2 ) concentrations on present day and future global isoprene budgets.
Isoprene emission rates from plants were first measured by Sanadze (1959), and placed into the context of global atmospheric processes by Rasmussen & Went (1965) and have since been characterized for a range of ecosystems via enclosure studies (as summarized by Wiedinmyer et al., 2004). Several theories have been offered to explain the role of isoprene biosynthesis in plant processes, including thermal protection (Sharkey & Singsaas, 1995), protection against ozone damage (Loreto et al., 2001) and a 'safety valve mechanism' to maintain metabolic homeostasis (Rosenstiel et al., 2004). Although individual factors have been shown to dictate the behavior of individual species, for example (Behnke et al., 2007), the ultimate reasons(s) for isoprene production in all plants remains unresolved. However, several controlling environmental factors have been identified. Early work recognized the temperature and light sensitivity of isoprene emission and these meteorological drivers were the basis of the first empirical emission models (Tingey et al., 1981;Guenther et al., 1991Guenther et al., , 1993Lamb et al., 1993). Isoprene emissions respond exponentially to short-term (minute-to-minute) increases in temperature (up to a threshold in the range of 40-50 1C) (Monson & Fall, 1989;Singsaas et al., 1999) as well as to longer term (weekly-to-seasonal) increases in temperature Sharkey et al., 1999;Petron et al., 2001) which contributes to the large seasonal and interannual variability in emission rate at temperate latitudes (Abbot et al., 2003), and implies emission increases in the face of global warming (Liao et al., 2006). The observed (Pegoraro et al., 2004) inhibition of isoprene emission in drought conditions and in young or aged leaves is also included in the recent Model of Emissions of Gases and Aerosols from Nature (MEGAN) (Guenther et al., 2006). Additional factors that may modulate isoprene emissions, such as nutrient availability, physical stress and ozone exposure Alessio et al., 2004;Velikova et al., 2005), are not included in current emission algorithms, largely due to insufficient data (Guenther et al., 2006).
Several studies have shown that elevated atmospheric CO 2 concentration inhibits isoprene production, as summarized by Arneth et al. (2007b). This suggests a potential self-regulation of isoprene emission in plants, where in a warmer climate, CO 2 -rich atmosphere emis-sions of isoprene remain relatively unperturbed. Indeed, Arneth et al. (2007a) used a process-based isoprene emission model to show that predicted isoprene emission enhancements induced by temperature and vegetation increases in 2100 were offset by inhibition due to ambient CO 2 levels. The direct processbased response of isoprene emission to ambient CO 2 concentrations in this model followed the same behavior predicted by Possell et al. (2005) who developed a parameterized response based on a number of plant studies. In a recent study by Wilkinson et al. (2008) (hereafter referred to as W08), the first attempt was made to separate and model (1) the long-term response of plants to CO 2 concentrations in the growth environment and (2) the instantaneous response of isoprene emissions to changes in intercellular CO 2 concentration (C i ) when atmospheric CO 2 concentrations are varied in a gas-exchange cuvette. The latter is a function of the balance between leaf CO 2 assimilation rate and leaf stomatal resistance, and can be affected by drought, leaf temperature and light intensity. On the basis of observed responses to both the short-term and the longterm CO 2 levels, W08 developed separate empirical scaling factors for basal emission rates of the form used by Guenther et al. (2006). Here, we go beyond previous work by using these scaling factors to test the growth and instantaneous CO 2 effects on simulated global isoprene emissions.
Reduction in isoprene emission potential from exposure to elevated CO 2 may be counter-acted to some degree by the fertilization effect of CO 2 which enhances photosynthesis, water use efficiency and hence vegetation productivity (Drake et al., 1997;Kö rner, 2000). This suggests that in a CO 2 -rich atmosphere, net primary productivity (NPP) may increase, but the direct inhibition due to elevated CO 2 concentration will also increase. Furthermore, changes in global land cover may shift the isoprene emission regions northwards with the expansion of the boreal forests (Lathière et al., 2005), placing isoprene-emitting forests into a cooler climate.
Predicted changes in isoprene emissions are predicated on an accurate model description of isoprene production, vegetation distribution and future climatic conditions, and are therefore highly uncertain. And yet, the increase in isoprene emission in a warmer climate is among the most important drivers of change in atmospheric chemical composition and oxidative capacity projected by models (Brasseur et al., 2006;Liao et al., 2006). Past projections of change in isoprene emission, and therefore atmospheric photochemistry in general, have been made using models that emphasize the effects of future climate warming on the highly temperature-sensitive biochemical processes underlying isoprene biosynthesis and/or change in NPP due to increases in atmospheric CO 2 concentration (Constable et al., 1999;Tao & Jain, 2005). Using such models, rising isoprene production in plants is predicted to propel surface ozone levels upwards by 10-30 ppb in the next 100 years, with associated exceedance of air quality standards in some regions (Sanderson et al., 2003). The methane lifetime is also projected to increase in a warmer climate as a result of increased competition with isoprene for oxidative radicals (Shindell et al., 2007), although there are uncertainties in our understanding of the impact of isoprene on atmospheric oxidation capacity (Lelieveld et al., 2008). Enhanced isoprene emission is also projected to increase the burden of biogenic secondary organic aerosols by 20% by the year 2100 (Heald et al., 2008). Monson et al. (2007) have criticized this approach as ignoring the potential direct effects of increases in atmospheric CO 2 on isoprene. In this study, we use a global coupled landatmosphere model, along with recently derived models at the leaf level that describe long-and short-term CO 2 effects, to explore the degree to which the inhibition of isoprene emissions under elevated CO 2 concentrations opposes the large increases in isoprene emission predicted for future climate warming scenarios and in the presence of increased global NPP.

Model description
We use here the coupled Community Atmospheric Model (CAM3) and Community Land Model (CLM3.5) of the global NCAR Community Climate System Model (CCSM3) . The CLM simulation is driven by the CCSM-simulated climate. The CAM is run in prognostic mode for present-day (2000) and future (2100) conditions. Simulations are performed with a 30 min time step at a 21 Â 2.51 horizontal resolution with 26 vertical levels from the surface to the lower stratosphere ($ 4 Pa).
The CLM simulates the biogeophysical processes associated with land-atmosphere exchange (Dickinson et al., 2006). Vegetation is described by 16 plant functional types (PFTs). Thornton & Zimmermann (2007) provide details on the canopy scheme and how simulated NPP for the CLM model shows generally good agreement with global observations across a range of vegetation types. We use the diagnostic CLM mode, with fixed land surface parameters and with the same spatial and temporal resolution of CAM3. The latest version of CLM (v3.5) surface datasets include leaf area index (LAI) based on MODIS v4 and PFT distributions for present day from a combination of MODIS, AVHRR and crop data as described by Lawrence & Chase (2007). For the set of future simulations which include the effects of land cover change and CO 2 fertilization, we use surface data from the dynamic vegetation CLM simulation of Alo & Wang (2008) (see 'Future projections (2100) with fixed vegetation' for further details). CLM photosynthesis is based on the dePury & Farquhar (1997) sun-shade model, which is a big leaf model that dynamically treats sunlit and shaded photosynthesis and irradiance separately and thus achieves good agreement with more complex multilayer canopy flux models. Photosynthesis rates (A) are dependent on vegetation surface temperature, CO 2 concentrations at the source of carboxylation (after accounting for estimated internal cellular resistance), soil moisture and irradiance. Intercellular CO 2 concentrations within the leaf (C i ) are controlled by leaf boundary layer resistance (r b ) and stomatal resistance (r s ). The Ball-Berry model (Ball et al., 1987) is used in CLM as follows: where C s is the concentration of CO 2 at leaf surface, C a is the atmospheric concentration of CO 2 and P atm is the atmospheric pressure. Leaf boundary resistance (r b ) is calculated as a function of wind speed after fixing the leaf-to-air turbulent transfer coefficient at 0.01 m s À1/2 and the characteristic leaf dimension at 0.04 m (following the procedures described in Oleson et al. (2004). Stomatal resistance is solved iteratively based on rates of photosynthesis, and C i is then resolved following Eqn (1) (Oleson et al., 2004). Future climate conditions are based upon the IPCC SRES A1B scenario (IPCC, 2001), with atmospheric CO 2 concentrations fixed at 717 ppm for the year 2100. Present-day (which we take as the year 2000) CO 2 concentrations are fixed at 365 ppm. Sea surface temperatures are specified from previous NCAR CCSM climate change experiments using the SRES A1B emissions (Meehl et al., 2006). The transient climate sensitivity of the CCSM3 fully coupled model is 2.47 1C (Kiehl et al., 2006). A full analysis of the CCSM3 future climate simulation is not the objective of this work, see Meehl et al. (2006) for further details.
All analyzed simulations are initialized following a 1-year spin-up simulation. Future snapshot simulations are performed for 1 year and are repeated for 10 years to assess the magnitude of interannual climate variability in the vicinity of the snapshot. We perform three sets of simulations here: (1) present-day vegetation and climate (2000), (2) future climate with fixed vegetation and (3) future climate with projected 2100 vegetation. We use the fixed vegetation simulations as our standard present-day and future simulations.

MEGAN v2 isoprene emission scheme
Isoprene emissions in CLM follow the MEGAN v2.0 with detailed canopy light and temperature algorithms (Guenther et al., 2006). Basal emission factors (e j ) at standard conditions of light, temperature and leaf area are specified for each PFT (j), for each grid box to account for species-wide divergence in emission capacities. Total canopy-level fluxes (F, in units of mg C m À2 h À1 ) are calculated by summing the emissions across all vegetation types with fractional area coverage (w j ) in the grid box and modulating the basal emission rate with an emission activity factor (g): where r is the canopy loss and production factor, set here to unity, as recommended for isoprene by Guenther et al. (2006). The activity factor accounts for emissions response to phenological and meteorological conditions and includes scaling factors for light (g P ), temperature (g T ), leaf age (g age ), soil moisture (g SM ) and LAI: The activity factors in Eqn (3) are calculated based on the instantaneous temperature, radiation, soil moisture and LAI at each time step in the CLM, as well as the average temperature and radiation conditions over the last 24 h and 10 days. The radiation response is applied separately for the sunlit and shaded leaves in the forest canopy environment. The canopy environment constant (C CE ), a factor used to set emission activity to unity at standard conditions, is set to 0.40 for the CLM model at the standard conditions specified by Guenther et al. (2006).

Activity factor for the CO 2 response
Recent studies have shown a significant inhibition of isoprene emission rate in the presence of elevated atmospheric CO 2 concentration for several species plants and for both short-term exposure (affecting C i ) and long-term exposure (due to increase in C a ) as summarized by Arneth et al. (2007b). Recently, W08 described a series of experiments using four plant species Populos tremuloides (aspen), Populos deltoides (cottonwood), Liquidambar styraciflua (sweetgum) and Eucalyptus globulus (eucalyptus), in which, once again, a consistent inhibition of isoprene emission rate was shown in the presence of elevated C i over the short term and C a over the longer term. In that study, a model was developed to describe both the short-and longterm response to CO 2 separately based on the observed response of the aspen plants, for which a greater number of growth environment experiments were performed. We use this model to describe the generalized CO 2 response for all isoprene-emitting vegetation in the global analysis reported here. We note that this is a simplification and requires the assumption that all isoprene-emitting species function in the same way with regard to CO 2 sensitivity. The parameterization may be expanded as further plant species are tested for their CO 2 sensitivity; it is particularly important that tropical plants are investigated as they dominate most global inventories of the total isoprene flux. In support of our simplification, however, we note that both Possell et al. (2005) and W08 find that a number of different herbaceous and woody species respond consistently to changes in the atmospheric CO 2 concentration, and thus the parameterized response observed by W08 may indeed provide an adequate description of midlatitude or even global vegetation.
W08 find that isoprene emission rates decrease nonlinearly with instantaneous changes in intercellular CO 2 concentration. They suggest that this response can likely be linked to changes in metabolite pools and enzyme activity within the leaf. The activity factor (g Ci ) is modeled as a sigmoidal response curve: where I s,max is the estimated asymptote at which further decreases in intercellular CO 2 have a negligible effect on isoprene emission, C* is a scaling coefficient and h is an exponential scalar. The sensitivity of isoprene emission rate to C i decreases with long-term exposure to elevated atmospheric CO 2 ; thus, isoprene emission response curves are fit individually to the responses reported in W08 for plants grown at 400, 600, 800 and 1200 ppmv atmospheric CO 2 (W08, Fig. 5). The parameters for Eqn (4) were obtained from these response curves and are provided in Table 1. Enhanced CO 2 concentrations in the long-term growth environment of the plant were also found to induce a negative response in isoprene synthesis, likely the result of changes in gene expression in the plant (W08). This response exhibited the same sigmoidal functionality as the short-term response: The parameters for Eqn (5) are given in Table 2. Note that W08 defined these parameters as a function of C i , which they assumed was equivalent to 0.7 of the observed atmospheric CO 2 concentration (C a ) and we include this conversion in Eqn (5). W08 show how this parameterization agrees well with the response estimated by Possell et al. (2005) over the range of atmospheric CO 2 concentrations considered (but diverges at C a o400 ppm), as shown here in Fig. 1. The parameterization of Possell et al. (2005) gives isoprene emission activity factors of 1.19 and 0.60 for atmospheric CO 2 concentrations of 365 ppm and 717 ppm, respectively, implying a 49% decrease in isoprene emission efficiency from 2000 to 2100. Conversely, the long-term growth environment parameterization of W08 implies a 28% drop in efficiency (activity factors of 1.04 and 0.75 for 365 ppm and 717 ppm, respectively), and if C i is assumed to be 0.7 of C a globally, then the additional shortterm effect yields a total 37% decrease in isoprene emission efficiency.
The activity factors for the CO 2 response are implemented in the CLM MEGAN2 emission scheme described in 'MEGAN v2 isoprene emission scheme' as additional activity factors in Eqn (3). The parameters for Eqn (4) are obtained by linearly interpolating the values in Table 1 corresponding to intervals of long-term growth CO 2 concentration to the specified atmospheric CO 2 concentration. Within CLM, the short-term activity factor (g Ci ) is then calculated following Eqn (4) based on simulated instantaneous C i at each location and timestep. The activity factor for long-term growth environment (g Ca ) is calculated from Eqn (5) given a fixed atmospheric CO 2 concentration and is globally constant in these simulations. In the sections that follow, the short-term and long-term activities factors are treated as one single activity factor (g C ¼ g Ci g Ca ) for all figures.

Global isoprene budgets
Present-day (2000) Figure 2 shows the seasonal distribution of simulated C i and the CO 2 activity factors, which generally range between 1.01 and 1.05 for present-day conditions, over different latitude bands. The activity factor associated with long-term exposure to CO 2 concentrations of 365 ppm derived from Eqn (5) is 1.04. This value is greater than unity because CO 2 concentrations are less than the standard conditions of 400 ppm. The activity factor for short-term CO 2 exposure is within 5% of unity. The range of simulated activity factors (for both long-and short-term effects together) in 2000 is plotted on Fig. 1, and agrees well with both W08 and Possell et al. (2005) fits. The net result of both activity factors is a slight increase (3%) in global isoprene emissions from 508 Tg C yr À1 with standard MEGAN2 to 523 Tg C yr À1 when CO 2 activity factors are included (Table 3). This increase is significant when compared with the simulated variability in global total isoprene emissions due to interannual variability in climate (standard deviation of AE 8 Tg C yr À1 in global total emissions over the 10-year simulation). Relative emission increases are fairly uniform globally, with the largest absolute increases in Australia, sub Saharan Africa and South America (Fig.  3). Figure 4 shows that the CO 2 activity factor is largely aseasonal, compared with the other meteorological and phenological activity factors of Eqn (3).

Future projections (2100) with fixed vegetation
Atmospheric CO 2 concentrations in 2100 (717 ppm) result in a 0.75 isoprene emission activity factor (or 25% reduction) associated with long-term growth environment following Eqn (5) (Fig. 2). The short-term response yields activity factors of 0.86-1.03 depending on local C i (see further discussion below), which when combined with the long-term effect, results in a significant net reduction in simulated isoprene emissions compared with the standard MEGAN2 [as described by Eqn (3)]. Global total emissions drop 31% from 696 to 479 Tg C yr À1 (Table 3, Fig. 3). Relative emission reductions are generally globally uniform. Figure 3, therefore, shows that the largest absolute changes in isoprene emission when analyzed with the CO 2 activity factor included in the model are found in the largest isoprene emission regions, in Australia, sub-Saharan Africa and the Amazon region of South America. Figure 5 shows the seasonal simulated CO 2 activity factors, C i (normalized by C a ) and the factors which control these concentrations for the year 2100. Controlling factors included the rate of photosynthesis and stomatal resistance. For example, increased photosynthesis rates (e.g. throughout Russia in summer) have the effect of consuming atmospheric CO 2 and therefore  they also cause decreases in C i within the leaf [as predicted by Eqn (1)]. Similarly, increased stomatal resistance impedes the diffusion of CO 2 from the atmosphere into the leaf, also forcing a reduction in C i . The seasonal variability in C i due to the interacting effects of photosynthesis and stomatal resistance are apparent over the boreal forests of the northern high latitudes, where isoprene sources are less abundant compared with temperate and tropical latitudes. We note that although the average ratio of C i to C a is close to 0.7 as assumed by W08, there is significant regional variability (Fig. 5). However, in general, the activity factor associated with the CO 2 effect varies little, particularly when compared with the seasonal response to light and temperature (Fig. 4). The nearly constant response of isoprene emission rate to CO 2 is a consequence of the dominance of the long-term growth CO 2 response over the short-term C i -driven response. As seen in Fig. 1, the global variability in simulated isoprene emission response to C i is nominal on a monthly mean basis.  Despite the large range of simulated C i around the world shown in Fig. 5, the resulting sensitivity of isoprene emissions generally varies by less than 10% in 2100. In addition, the mean response to C i is generally a 10% decrease in isoprene emission or less. In fact, this simulation suggests that the discrepancy between the two model fits (Possell et al., 2005 vs. W08) may be resolved by accounting for the short-term C i -driven response. The fits of Possell et al. (2005) and W08 differ by $ 10% in 2100, including the short-term response reduces the W08 prediction by close to this amount, bringing the results into closer agreement. Although the seasonal role of C i variability is small, the diurnal response may be important at local scales. To test this, we examine the diurnal profile of C i and the associated CO 2 activity factor in two important isoprene source regions: North and South America. Figure 6 shows both the mean and range of activity factors associated with C i (normalized by C a ) in summer (June-August) of 2100. Again, we illustrate here that a large range in C i translates to modest differences in activity factor. Above all, this figure shows that C i does not vary sufficiently on a diurnal basis to significantly modify isoprene emission, at least at the local landscape scale. We compare the diurnal variability in activity factor associated with changes in CO 2 concentration with that associated with light to make this distinction.
Interplay between drought and C i likely exists in certain environments as the maximum rate of carboxylation is adversely affected by drought. We hypothesize that decreases in soil moisture could thus scale back isoprene emissions both directly and indirectly (as shown previously by Pegoraro et al., 2004). However, as we have shown, differences in C i do not appreciably alter isoprene emission rates in 2100 using the model of W08. For example, although there are strong seasonal trends in soil moisture in Amazonia which are associated with reduced photosynthesis in the dry season, the resulting differences in C i (of $ 100 ppm) modify isoprene emissions by only a few percent. Standard MEGAN2 algorithms would predict a 37% increase in isoprene emission from 2000 to 2100, largely a result of rising temperatures (Fig. 7). Figure 4 shows that in this simulation with fixed vegetation, the only activity factor of Eqn (3) predicted to change significantly under future conditions is that related to temperature. In 'Activity factor for the CO 2 response', we noted that if C i is assumed to be 70% of C a isoprene emission efficiency due to both the long-term and short-term CO 2 effects decreases by 37% from 2000 to 2100. Thus, the isoprene response to projected increases in temperature and CO 2 would offset each other exactly. However, average C i in the model is higher than 70% of C a and therefore the model predicts an 8% decrease in emissions by 2100 (compared with 2000) when the activity factor accounting for CO 2 concentration is included. Projected regional trends match the global average picture, with large increases in isoprene emission rates from 2000 to 2100 due to climate warming, negated by including the inhibition of isoprene emission by CO 2 .

Future projections (2100) with predicted vegetation
Dynamic vegetation models generally predict the future degradation of vegetation in Amazonia and the regrowth of the high-latitude boreal forests in concert with a lengthened growing season and associated increases in LAI over vegetated regions (Joos et al., 2001;Gerber et al., 2004;Lathière et al., 2005). Reductions in rainfall may lead to water stress and regional dieback (Niyogi & Xue, 2006). Guenther et al. (2006) find that isoprene emissions may decrease by up to 30% when future vegetation distributions are used to drive MEGAN, although this projection is based on a specific land-use change scenario that assumes a very large increase in global cropland area. They also show projections of LAI which more than double in some regions by 2100; this is primarily due to projected increases in NPP. The linear dependence of isoprene emission rate on LAI (which is modulated by a nonlinear decrease due to the light-dependent activity factor g P particularly at high LAI), described in Eqn (3), highlights the critical importance of understanding the effects of terrestrial CO 2 fertilization for future prediction of isoprene emission.
To investigate the relative sensitivity of isoprene emission to vegetation distribution and density, we use land surface parameters, including global PFT distribution and LAI, projected with a dynamic global vegetation model (DGVM). Our goal here is not to capture the range in potential vegetation response which can vary significantly with climatic predictions (Alo & Wang, 2008) and among models (Scholze et al., 2006), but rather to use a single realization of future vegetation to compare the relative effects of changes in vegetation distribution and productivity vs. the direct effect of CO 2 inhibition on the projected global isoprene emission rate. Alo & Wang (2008) used the CLM DGVM to investigate the response of the terrestrial ecosystem to changes in climate projected by eight general circulation models. We use their results for 2100 under the A1B scenario driven by the climate predictions of the CCSM, consistent with the model and future conditions employed here. Although the global vegetation expansion projected by the CLM DGVM driven by the CCSM climate is consistent with the increases seen in seven of the eight models, regional responses in vegetation may differ. In particular, this model does not project the Fig. 6 Diurnal trend in summer (June-August) 2100 mean (line) and range (shaded region) of intercellular CO 2 concentrations (normalized by atmospheric CO 2 , blue) and CO 2 activity factor (black) for North America and South America. Also shown is the mean diurnal profile of isoprene emission activity factor associated with light (red).  present-day (2000, dotted) and future with fixed vegetation (2100, solid) simulation with (green) and without (black) CO 2 activity factor. extensive deforestation of the Amazon projected in some previous studies (Lathière et al., 2005) and seen in one model considered by Alo & Wang (2008). The CCSM model predicts wetter conditions in 2100 than any of the other models considered, and thus water limitations do not instigate large-scale vegetation dieback. We note here that these simulations account only for natural changes in vegetation and do not include the effects of urbanization and cropland expansion. Global mean LAI over vegetated surfaces more than triples from 1.2 m 2 m À2 in the MODIS-based present-day conditions of Lawrence & Chase (2007) to 3.8 m 2 m À2 in the 2100 simulation of Alo & Wang (2008). Figure 8 shows that foliar expansion due to CO 2 fertilization and enhanced NPP is projected throughout the world and is not limited to specific regions. Vegetation cover is projected to increase in general, with a northward expansion as projected in previous studies (Joos et al., 2001;Gerber et al., 2004;Lathière et al., 2005), particularly for broadleaf trees (shown in Fig. 8), the highest emitters of isoprene among PFTs. For further discussion of the potential changes in the terrestrial ecosystem, we refer the reader to Alo & Wang (2008).
When the effects of dynamic vegetation and enhanced NPP from CO 2 fertilization are included, global isoprene production is projected to more than double by 2100 (1240 Tg C yr À1 ), compared with present-day levels (note that emissions are even higher when CO 2 inhibition is not accounted for, Table 3). This effect is significantly larger than the inhibition of isoprene emission predicted from the results presented in W08. However, recent work has shown that the biosphere's capacity to absorb CO 2 has been overestimated by up to 74% in models which do not account for nutrient limitation , and thus future LAI increases may be significantly more modest. Indeed, if LAI increases by only a quarter of the predictions of Alo & Wang (2008), isoprene increases associated with this effect would be comparable with decreases due to the direct effect of CO 2 inhibition. Arneth et al. (2007a) found that CO 2 inhibition compensated fully for both temperature and vegetation changes projected for 2100 in the LPJ-GUESS model. This substantially different level of compensation likely arises from the use of very different climate drivers, where the study of Arneth et al. (2007a) was driven by the one model (HadCM) highlighted by Alo & Wang (2008) to project large decreases in natural vegetation, in stark contrast to the CCSM climate used here.
Implications for historical isoprene emissions Figure 9 shows how isoprene emission activity factors associated with temperature and CO 2 levels evolved  (Thornton & Zimmermann, 2007) and as predicted by the Community Atmospheric Model (CLM) dynamic vegetation model for 2100 (A1B, right) (Alo & Wang, 2008). Color scales are saturated at respective values. throughout the recent geological past. Atmospheric CO 2 concentrations over the Vostok ice core record are lower than present day, and thus isoprene emission is expected to be enhanced by CO 2 , not inhibited as seen for future conditions. However, over this time period, atmospheric CO 2 concentrations rarely deviate from concentrations of 150-250 ppm and thus the CO 2 enhancement of isoprene production remains modest and fairly constant. Therefore, although the balance between temperature and CO 2 activity factors is critical to future predictions of isoprene (our results for 2100 are shown as circles on Fig. 9), large temperature fluctuations in the geological past remain the primary control on isoprene emissions over the last 400 thousand years. Adams et al. (2001) and Kaplan et al. (2006) suggest that the drop in methane concentrations observed in ice core samples during the Last Glacial Maximum (LGM, $ 20 000 years BP) could be explained by an enhanced oxidative sink (hydroxyl) resulting from lower isoprene emissions in a cooler climate with reduced vegetation. However, the enhancement of isoprene emission associated with depressed CO 2 concentrations in the past would act against this, perhaps nullifying any effect on OH. Indeed, Arneth et al. (2007a) predict a strong CO 2 enhancement at the LGM following the parameterization of Possell et al. (2005). However, as seen in Fig. 1, the W08 parameterization shows a much more modest enhancement of isoprene at low CO 2 concentrations which, when compared with the effect of temperature at the LGM (Fig. 9), would imply a net drop in isoprene emission, consistent with the historical record of methane. The divergence of these parameterizations plainly highlights the need for further study of the direct effect of CO 2 on isoprene emission at low concentrations.

Conclusions
We use a global coupled land-atmosphere model here to show that the inhibition of isoprene emission with increasing CO 2 concentrations that has been observed in leaf-level studies is a key control on future projections of global isoprene emission rate. In particular, we find that the CO 2 inhibition predicted in 2100 under the A1B IPCC SRES scenario may completely offset the large temperature-driven increase in isoprene emission predicted by standard models. This suggests that future isoprene production may to a large degree be buffered by competing influences. In contrast to current model prediction, Lelieveld et al. (2008) have proposed that tropical forests are able to maintain a strong atmospheric oxidation capacity even while emitting copious Fig. 9 The geological record of temperature anomaly (black) and atmospheric CO 2 concentrations (grey, thick) from the Vostok ice core (top) and the associated isoprene emission activity factors (below). Activity factors are normalized to present-day values. The CO 2 activity factor includes both long-and short-term effects. The short-term CO 2 activity factor over the geological past depends on parameters for plants grown at 400 ppm in Table 1, as values for lower CO 2 growth environments are not available. Predicted activity factors for 2100 are shown as dots. The Last Glacial Maximum (LGM) is noted. Vostok data source: Petit et al. (1999). quantities of reactive isoprene. This remarkable capability for sustaining the atmosphere's cleansing ability may be enhanced at millennial scales by the opposing response of isoprene to temperature and CO 2 . We note, however, that this simple picture does not account for changing vegetation. Although uncertainty remains as to how effectively CO 2 fertilization might enhance global NPP and how this may be limited by nutrient availability, it is certainly likely that fertilization under higher atmospheric CO 2 concentrations will contribute to foliar expansion. Rising atmospheric CO 2 concentrations are also likely to result in significant changes in species composition (Mohan et al., 2007) that will lead to substantially increased isoprene emissions in at least some landscapes. To what degree either of these potential forcings would enhance future isoprene emissions remains unclear.
We find here that the response of isoprene to CO 2 concentrations is dominated by the long-term growth effect. The results of W08 suggest that the shorter-term influences driven by dynamics in C i modulate isoprene emission by less than 10% under CO 2 concentrations in present day and in a 2100 scenario. Furthermore, the aseasonality of the activity factor associated with C i throughout most of the world and the consistent diurnal profile imply that this 10% modulation does not introduce important temporal variability in isoprene production. Although the long-term growth parameterization of W08 is easily incorporated in any chemical transport model (CTM), few CTMs are interfaced with an active land model for prognostic estimation of C i in plants. Our results suggest that neglect of the short-term response of isoprene emission to C i would imply at most a 10% overestimate in the prediction of isoprene emissions. This falls within existing uncertainties on isoprene emission estimates. Alternatively, given the modest variability in CO 2 activity factor associated with variable C i , fixing the C i to a globally constant fraction of C a (such as 0.7 as previously suggested) in CTMs would adequately describe the short-term response.
The dampening of future predicted isoprene emission increases via CO 2 inhibition points towards a more constant chemical composition of the troposphere than previous projections suggest. The feedback of a warmer climate on isoprene emission from vegetation and the associated anticipated enhancements in ozone, organic aerosol and methane may be significantly muted in a high CO 2 environment.
The W08 parameterization implies that the enhancement of isoprene emission due to low ambient CO 2 concentrations during the LGM ( $ 20 000 years ago) is dwarfed by the effect of the cooler climate. This supports previous conclusions that reductions in glacial isoprene emission may explain some of the glacial-interglacial changes in methane concentrations observed in ice cores.