Isoprene emissions over Europe are calculated here using the MEGAN–MOHYCAN model , based on the widely used MEGAN model for biogenic emissions , coupled with the MOHYCAN multi-layer canopy environment model . Flux=ϵ⋅γ=ϵ⋅CCE⋅γPT⋅LAI⋅γage⋅γSM⋅γCO2

The MEGAN emission model (Eq. ) includes the specification of a standard emission factor ϵ (mg m-2 h-1), representing the biogenic emission under standard conditions as defined in . The distribution of the standard emission factor ϵ (Fig. S1 in the Supplement) is obtained by MEGANv2.1. It is based on species distribution and species-specific emission factors . The MOHYCAN canopy environment model also requires the specification of the plant functional type (PFT). The PFTs are defined by the vegetation map of in 0.1∘×0.1∘ resolution. Seven plant functional types are considered, broadleaf evergreen/deciduous trees, needleleaf evergreen/deciduous trees, shrub, grass, and crops.

The multiplicative factor CCE(=0.52) is adjusted so as γ=1 at standard conditions defined in . The model uses activity factors (γ) to account for the response of the emission to changes in temperature (T), solar radiation (P), leaf age, soil moisture (SM), and the leaf area index (LAI). The activity factor γPT is the weighted average for all leaves of the product of the activity factors for leaf temperature (γT) and photosynthetic photon flux density PPFD (γP). The MOHYCAN model calculates the temperature of both sunlit and shade leaves and the attenuation of light as a function of canopy height, using visible and near-infrared solar radiation values at the top of the canopy, together with air temperature, relative humidity, wind speed, and cloud cover .

The response of the emission flux to leaf temperature is parameterized as γT=Eopt⋅CT2⋅eCT1⋅ACT2-(CT1⋅(1-eCT2⋅A)),A=Tℓ-ToptR⋅Tℓ⋅Topt, where CT1=95×103 J mol-1, CT2=23×104 J mol-1, R is the universal gas constant, Tℓ is the leaf temperature obtained from the MOHYCAN model, Topt is the optimal temperature defined as Topt=313-0.6⋅(T240-297) and Eopt is defined by the average leaf temperature (in K) over the last 24 and 240 h (T24, T240): Eopt=2.034⋅e0.05(T24-297)⋅e0.05(T240-297).

The response to light is expressed as follows: γP=CP⋅α⋅P⋅(1+α2⋅P2)-1/2, with CP=0.0468⋅exp⁡(0.0005⋅(P24-P0))⋅(P240)0.6 and α=0.004-0.0005⋅ln⁡(P240). P is calculated at leaf level, P0 is set to 200 or 50 µg mol m-2 s-1 for sunlit or shaded leaves, respectively, and P24 (P240) is the averages of light intensity over the last 24 (240) h.

The emission response to leaf age is defined as γage=0.05⋅F1+0.6⋅F2+1.125⋅F3+F4, where F1, F2, F3, and F4 represent the fractions of new, growing, mature, and senescent leaves, respectively . The impact of soil moisture stress on isoprene fluxes is highly uncertain, and therefore we assume γSM=1 in this study.

The MEGAN–MOHYCAN model is run at hourly resolution on a 0.1∘×0.1∘ grid. In its current setup, the model requires the following meteorological input data at hourly resolution: downward solar radiation, cloud cover fraction, air temperature above the surface, dew-point temperature (or relative humidity), and wind speed directly above the canopy. Different climatological input data were used depending on the simulation. Table  summarizes all simulations and the corresponding meteorological input. The isoprene emissions for 1979–2014 were obtained by using ERA-Interim ECMWF (European Center for Medium range Weather Forecasts) meteorological fields over the above period.

To account for observed solar radiation changes over Europe we performed a second simulation (H2) where the ERA-Interim downward solar radiation fields are adjusted based on homogenized composite time series of ground-based observations from 56 European sites . The sites are grouped in five large European regions (central, northern, eastern, southern, and northwestern Europe, Fig. S2). We calculated the seasonally averaged solar radiation according to ERA-Interim at the locations of the observation sites over 1979–2014 and computed their averages SSR‾ECMWFi,k over each large region i and each season k. The same procedure is applied for the ground-based observations, SSR‾obsi,k. We calculate correction factors fi,k=1+Δ(SSRobsi,k)SSR‾obsi,k-Δ(SSRECMWFi,k)SSR‾ECMWFi,k, where Δ(SSRobsi,k) is the seasonal mean anomaly of solar radiation observed in region i, and Δ(SSRECMWFi,k) is the corresponding anomaly of the ERA-Interim data. The correction factors fi,k are then applied to the solar radiation fields P of Eq. (). The ERA-Interim seasonal surface solar radiation anomalies show a fairly good agreement with the corresponding observed anomalies averaged over five large European regions (central, northern, eastern, southern, and northwestern Europe, Fig. S2) and the calculated correlation coefficient is generally higher than 0.8, except in northwestern Europe (0.75). The ERA-Interim data are found to underestimate the observed decadal trends in all regions and seasons, by a factor of 2–3 in spring and summer. The use of the adjusted observation-based solar radiation fields in the MEGAN–MOHYCAN simulations leads to slightly higher trends in the estimated isoprene fluxes over Europe (see Sect. ), in particular over northwestern Europe.

In order to estimate the impact of climate change, simulations using the regional climate model ALARO were performed. ALARO is the limited-area model version of the ARPEGE-IFS forecast model developed within the ALADIN consortium . These runs were performed following the prescriptions of the international COordinated Regional climate Downscaling EXperiment (CORDEX). Therefore the target domain is the EURO-CORDEX domain (34–70∘ N, 25∘ W–50∘ E, http://www.euro-cordex.net; last access: 15 June 2018) with a horizontal resolution of 12.5 km. As lateral boundary conditions over the European domain, ALARO used the global climate simulations from the CNRM-CM5 model following the guidelines of the fifth Coupled Model Intercomparison Project (CMIP5; ). Validation of ALARO was conducted by comparing observations with model runs forced by realistic boundary conditions from the ERA-Interim reanalysis dataset , and the model was shown to perform in line with other regional climate models (RCMs) of the EURO-CORDEX ensemble over Europe .

With ALARO we assessed the impact of a changing climate following three RCP scenarios, RCP2.6, RCP4.5, and RCP8.5 , which span a range of potential changes in future anthropogenic emissions. The RCP2.6 scenario assumes a peak in radiative forcing at 3.1 W m-2 (490 ppm CO2) by midcentury followed by a decline to 2.6 W m-2 by 2100. In RCP4.5 a moderate increase in radiative forcing to 4.5 W m-2 is assumed until 2050, with a stabilization thereafter (650 ppm CO2). In RCP8.5, emissions continue to rise throughout the 21st century with rising radiative forcing leading to 8.5 W m-2 (1370 ppm CO2) by 2100 . The performed simulations using ALARO meteorology are summarized in Table  for 2070–2099 and the results are compared to the control (CTRL) simulation covering 1976–2005. Additional simulations, accounting for the effects of CO2 inhibition and fertilization, are discussed in Sect. .

Leaf area index is obtained from the MODIS 8-day MOD15A2 (collection 5) composite product generated by using daily Aqua and Terra observations at 1 km2 resolution between 2003 and 2014 . Before 2003, the monthly LAI at every grid cell (x) and month (m) is estimated based on the local temperature of the current and previous months: LAI(x,m)=A(x,m)+B(x,m)⋅(0.65⋅T(x,m)+0.35⋅T(x,m-1)), with A(x,m) and B(x,m) determined from a linear regression between the monthly MODIS LAI data and the ERA-Interim near-surface temperatures between 2003 and 2014. Note that the slope B(x,m) is set to zero when the correlation between LAI and temperature is poor (r<0.3), and in that case the climatological average LAI over 2003–2014 is used. We use the climatological average of the LAI in our standard future (2070–2099) simulations. The increase in LAI associated with CO2 fertilization is accounted for in separate simulations (Table ). Changes in vegetation composition are not considered.

We account for the direct effect of atmospheric CO2 concentration changes on isoprene emissions through the activity factor γCO2 in Eq. (). This factor is applied to the historical simulation (H3) and to the ALARO simulations, as shown in Table . Two different parameterizations were tested, (WI) and (PH). The empirical parameterization by is given by Eq. (), γCO2=Ismax/(1+(Ci/C⋆)h), where Ismax=1.344, Ci is the leaf internal CO2 concentration at non-water-stressed conditions, which is equal to 70 % of the atmospheric CO2 concentration, C⋆=585 ppm, and h=1.4614. The γCO2 is equal to 1 at the atmospheric CO2 concentration of 402.6 ppm. This parameterization was determined empirically based on growth experiments with two aspen tree species (Populus deltoides and P. tremuloides) grown at four different CO2 concentrations (400, 600, 800, 1200 ppm), and was used to determine the impact of CO2 inhibition in the future atmosphere .

The parameterization of is obtained by an empirical nonlinear least-squares regression, based on a combination of laboratory and field observations obtained from 10 different studies on various plant species including tropical and temperate tree species as well as herbaceous plant species γCO2=a/(1+a⋅b⋅C), where C is the atmospheric CO2 concentration, and a=8.9406 and b=0.0024 ppm-1 are fitting parameters; γCO2 is equal to 1 at the CO2 concentration of 370 ppm.

For CO2 concentrations higher than 380 ppm the PH parameterization induces a relatively stronger inhibition (1 to 0.3) compared to the WI parameterization (1 to 0.4) (Fig. ). The parameterizations result in similar γCO2 values at concentrations corresponding to the historical simulations and to the RCP2.6 scenario, but differ by around 20 % for the RCP4.5 and RCP8.5 scenarios. In both schemes the inhibition factor behaves linearly at very high CO2 levels. Here we use the more recent PH parameterization in the historical H3 simulation (Table ). Both parameterizations are tested in the case of ALARO simulations, thus providing a range of the CO2 inhibition effect in the projected emission estimates.

Lastly, we estimated the effect of CO2 fertilization on the projected emissions through the expected enhancement in leaf biomass densities and LAI based on a recent study . Using long-term (1982–2009) satellite LAI records and ecosystem models, obtained a widespread increase in LAI over the majority of vegetated areas on the global scale and attributed the major part of the observed greening trends to CO2 fertilization. This is crudely parameterized here as a linear LAI increase of 15 % per 100 ppm of CO2 concentration (Table ). Dynamical vegetation models, e.g., ORCHIDEE , would be required in order to provide a more mechanistic simulation of the LAI variations and of the distribution and structure of the natural vegetation, but this lies beyond the scope of the present study. Note, however, that dynamical vegetation models have identified weaknesses related to the use of a limited number of static plant functional types, and to the poor representation of species competition .

In comparison to other bottom-up isoprene inventories, the MEGAN–MOHYCAN estimated emissions are generally lower. Averaged over 1980–2009 in the same EURO-CORDEX domain, our estimates amount to 7.3 Tg yr-1, and are by 22 % lower than in the MEGAN-MACC inventory (9.4 Tg yr-1, ), and about 3 times lower than in the GUESS-ES model (20.1 Tg yr-1, ). Similarly, satellite-based isoprene emission estimates, obtained using observations of formaldehyde, a high-yield isoprene oxidation product, indicate slightly higher isoprene emissions with respect to our estimates. For instance, an inversion study constrained by OMI (Ozone Monitoring Instrument) formaldehyde observations over a decade (2005–2014) suggested top-down isoprene emissions amounting to 8.4 Tg yr-1, i.e., 20 % higher than in the a priori MEGAN–MOHYCAN inventory . In the same line, an independent study using OMI formaldehyde observations from 2005 inferred an average increase of isoprene emissions by 11 % over Europe and emission decreases of 20–40 % in southern Europe with regards to their a priori MEGAN estimate .

In the following sections, the isoprene emissions estimated by the H3 simulation (Table ) are compared directly to isoprene flux measurements in Europe. Section  presents a comparison of modeled isoprene emissions with campaign-averaged isoprene fluxes measured at seven different locations. Section  investigates the ability of the model to reproduce the temporal variations as observed in Vielsalm (Belgium) and in Stordalen (Sweden).

Figure  shows the monthly averaged midday fluxes estimated in the H3 simulation at the model grid cells corresponding to the location of nine field campaigns , using either the MEGAN emission factors or using local emission factors (see further below).

Differences between field measurements and modeled data were expected, since the local vegetation around the measurement site differs from the heterogeneous vegetation mix of the model grid cell (in addition, the effect of the footprint on the flux measurements is also not taken into account by the model). The PFT fractional areas of the local vegetation are compared to the model PFT fractions of the corresponding grid cell in Fig. S5. Many field campaigns were conducted in forests whereas the corresponding model grid cells consist for a large part (15 to 91 %) of low isoprene-emitting PFTs such as crops, grass, and bare soil. At these sites (ECHO, Lochristi, Haute Provence, and Bosco Fontana), this discrepancy explains the large underestimation of model estimates using MEGAN emission factors. At Castelporziano, however, the relatively open local landscape is not well represented by the 0.1∘×0.1∘ vegetation map, which suggests a substantial fraction of needleleaf forest, partly explaining the emission overestimation at this location.

In order to correct for this effect, we re-calculated the model isoprene fluxes using local emission factors. These emission factors are based on the local PFT fractions (Fig. S5) combined with the standard emission factors (SEFs) given for the different PFTs in : 10 mg m-2 h-1 for the broadleaf deciduous sites (ECHO, Lochristi, Haute Provence, Bosco Fontana), 5.3 mg m-2 h-1 at Vielsalm, 1.8 mg m-2 h-1 at Castelporziano, and 1.6 mg m-2 h-1 in Stordalen. Overall, the use of local emission factors significantly improves the model performance and reduces the average bias for all sites from -70 to +5 % (Fig. ).

Note, however, that local emission factor estimates based on SEFs defined for broad PFTs are still crude approximations for the local SEFs. For instance, the SEF at the ECHO site is likely too high since it is dominated by non-isoprene emitters such as Fagus sylvatica and Betula pendula . Similarly, the vegetation at Castelporziano is a mixture of low-isoprene-emitting species like Quercus ilex and Arbutus unedo (0.1 µg gDW-1 h-1, where DW denotes dry weight of leaf biomass) and non-isoprene emitters such as Erica multiflora, Rosmarinus officinalis, and Phillyrea angustifolia, and therefore the SEF calculated assuming a large fraction of strongly emitting shrubs is likely too high. For Vielsalm, a local SEF of 2.88 mg m-2 h-1 is used, adjusted to minimize the average bias between the model and the observations in 2010 (see next section).

The model overestimation at the poplar plantation in Lochristi (Fig. ) is unexpected, given that Populus sp. is a strong isoprene emitter . However, the plantation was coppiced 6 months before the measurements, and new shoots started to sprout only in May 2012 , possibly explaining the difference between the modeled and the measured isoprene fluxes at that site (Fig. S6).

At Bosco Fontana, where a mixture of strong emitters (Quercus robur and Quercus rubra) and low emitters (Quercus cerris and Carpinus betulus) is present, a good agreement between modeled and measured flux is obtained, suggesting that the SEF of 10 mg m-2 h-1 is representative for this landscape. At the site in Haute Provence, dominated by a strong isoprene emitter (Quercus pubescens), an excellent agreement is obtained for the field campaign in June 2012 , whereas the model is somewhat too low in August 2010 .

The model potential to capture temporal flux variations is evaluated against flux measurements at the Vielsalm site located in a temperate mixed forest in the Belgian Ardennes (50.30∘ N, 5.99∘ E). The site consists of a mixture of evergreen needleleaf trees (mainly Pseudotsuga menziesii, Picea abies, and Abies alba) and deciduous broadleaf tree species (mainly the non-isoprene emitter Fagus sylvatica). Those tree species are generally weak isoprene emitters, explaining the low local SEF of 2.88 mg m-2 h-1. The main isoprene emitters are likely green needleleaf trees, especially the Abies alba .

The flux measurements used were obtained by disjunct eddy covariance by mass scanning technique during two field campaigns at the Vielsalm site: July–October 2009 , and May–September 2010 . The isoprene measurements were performed with an hs-PTR-MS (proton transfer reaction mass spectrometer, Ionicon, Innsbruck, Austria). Ambient air was continuously sampled at the top of a tower at a height of 52 m a.g.l. The instrument performs one measurement of isoprene fluxes every 2 s, and half-hourly averages are used for comparison with the model.

Figure  displays the evolution of the daily averaged measured and modeled fluxes (panels a and b) as well as their monthly averaged diurnal cycles (panel c). The model averages are calculated with the same temporal sampling as the observations. Both the day-to-day and the diurnal variability are well represented by the model for this site, as reflected by the high correlation coefficients of 0.92 for 2009 and 0.91 for 2010. Whereas the overall bias is small for both field campaigns (-8.3 % for 2009 and -0.8 % for 2010), the modeled seasonal pattern differs from the observed fluxes. The model is biased highly in May (+33 %) and June (+10 %), but it is biased low in September (-18 %) and October (-63 %). A possible explanation for this discrepancy might be that the leaf age factor described in Eq. (), i.e., the emission from new and growing leaves, might be overestimated, whereas the emission from senescent leaves might be underestimated. It should be noted that the activity factors γP and γT have their own uncertainties which might also impact the modeled seasonal variation.

A second model validation is performed for a sub-arctic wetland ecosystem at Stordalen in northern Sweden (68.33∘ N, 19∘ E, 351 m a.s.l.), 200 km north of the Arctic Circle . The region is characterized by a short but intensive growing season (from mid-May to mid-September) and is influenced by discontinuous permafrost conditions affecting surface hydrology and, thus, the growth conditions of the vegetation. The vegetation in the vicinity of the measurement tower was dominated by species such as Eriophorum ssp., Carex ssp. and Sphagnum ssp., all known to be low isoprene emitters .

Isoprene was measured using a hs-PTR-MS, which was combined with a sonic anemometer to estimate ecosystem-scale fluxes using disjunct eddy covariance. Measurements were taken at a height of 2.95 m a.g.l. (vegetation height ca. 50 cm) and fluxes from May to September 2006 reported at a temporal resolution of 30 min . For isoprene fluxes, the mean estimated error (2σ) was found to be 0.03 mg m-2 h-1.

The daily averaged observed and modeled fluxes as well as the diurnal cycles of fluxes are shown in Fig. . The model is biased low by ca. 40 % on average over the campaign, possibly suggesting an underestimation of the SEF used in the calculation (1.6 mg m-2 h-1) for arctic C3 grass . However, the model is able to capture the day-to-day variability (correlation coefficient of 0.84) in spite of the low fluxes at that site, frequently of the order of (or even lower than) the estimated error on the fluxes. The low bias of the model might be partly due to a low bias in the LAI values from MODIS used in the model, equal to ca. 0.88 at that site, to be compared with locally measured LAI reaching up to 3.5 at the most dense spots of the wetland sedges. In addition, the MEGAN algorithm might not be optimal for this subarctic vegetation type. As proposed by , vegetation in this area is especially well adapted to survive under conditions of short active seasons. The subarctic sedges start photosynthesizing in early spring under still cool temperatures, possibly resulting in isoprene emission induction occurring sooner than in other extratropical ecosystems. This hypothesis is supported by the stronger negative bias in June (-68 %) compared to July and August (ca. -35 %).

A comparison between the control ALARO (CTRL, 1976–2005, Table ) and the historical ERA-Interim surface temperature and solar radiation fields is presented and discussed in the Supplement (Fig. S7). The use of the ALARO control fields results in lower mean isoprene fluxes by 37 % over the domain (Table ), caused by a negative bias of the ALARO surface temperature fields compared to the ECMWF reanalysis. The CTRL fields are, however, not used here for emission estimation, but as a reference with which the projected isoprene emissions (2070–2099) will be compared. Surface temperature, precipitation, and surface shortwave radiation for the different RCP scenarios are compared to the CTRL fields in Fig. S8.

The absolute difference between the projected (2070–2099) and the control (1976–2005) mean temperature, solar radiation, and precipitation over the European domain, as simulated with the ALARO model for the climate scenarios (Table ), are displayed in Fig. . An average temperature increase of 0.9, 2.2, and 4 ∘C is found for RCP2.6, RCP4.5, and RCP8.5, respectively, with respect to the control simulation. The change in temperature presents a similar geographic distribution for the three scenarios, with the strongest temperature increases predicted over European Russia and Scandinavia. The simulated pattern as well as the range of temperature changes are consistent with results from other EURO-CORDEX model simulations and projections from the Coupled Model Intercomparison Project (CMIP5; ). The intercomparison shows that the largest model disagreements in summer occur in France and in the Balkans, suggesting a higher uncertainty for temperature projections in these regions.

The mean downward solar radiation is decreased over the domain, by up to -4 W m-2 for the RCP8.5 simulation compared to the control simulation. This average decrease is due to the combination of higher radiation in southern European countries and France (up to +8 W m-2) and decreases elsewhere (up to -10 W m-2). The amplitude of the expected changes in solar radiation and the simulated pattern are in line with results from the EURO-CORDEX ensemble . Note, however, that the different climate simulations in the EURO-CORDEX ensemble show large discrepancies over France, central Europe, and the coastal areas of Italy, Greece, and Turkey, underlining a higher uncertainty in projections of solar radiation in these regions .

Finally, the model predictions suggest a drier Mediterranean and wetter northern and eastern Europe (Fig. ). This pattern agrees reasonably well with previous studies and with the EURO-CORDEX ensemble . The latter suggests a robust increase in precipitation in central and northern Europe (up to 25 %), as well as a drop in precipitation in southern Europe (by up to 25 %). Note that according to the EURO-CORDEX ensemble, future precipitation projections show strong variability across different simulations at the 45∘ N latitude band, including southern France, northern Italy, and central Romania .

The impact of climate change on annual isoprene emissions according to the different RCP scenarios, upon neglecting the CO2 inhibition effect, is shown in the first column of Fig. . Whereas the RCP2.6 simulation suggests very weak changes in isoprene emissions (lower than 20 %), RCP4.5 and RCP8.5 indicate local emission increases reaching 40 and 110 %, respectively. In all simulations the strongest increase is found in southern Europe, European Russia, and Finland. This pattern, consistent with independent simulations , reflects the patterns of changes in temperature and solar radiation. The higher isoprene emissions in northeastern Europe are mainly a result of the strongly increased temperatures, and are somewhat counteracted by the decreasing solar radiation. In southwestern Europe the higher emissions are due to the combined effect of moderate temperature increases and cloud cover decreases.

When considering the effect of CO2 fertilization, we obtained a significant enhancement of the emissions, by +15 % (RCP2.6), +52 % (RCP4.5), and +141 % (RCP8.5), compared to the control simulation, and an increase by +8 % (RCP2.6), +15 % (RCP4.5), and +32 % (RCP8.5) compared to the simulation accounting only for climate effects (Fig. , Table ). The combined effect of climate change and CO2 inhibition is also shown in Fig. . Since both are of similar magnitude, but of opposite sign, considering both effects leads to isoprene fluxes similar to the control emissions. The strength of the CO2 inhibition, however, is different for the two parameterization schemes tested here . In comparison to the control simulation, total projected isoprene fluxes are 11 % lower and 26 % higher in the RCP8.5 scenario following or , respectively. For the other RCP scenarios, the simulated changes in isoprene emission range between -7 and 17 %. Note that the spatial pattern of the emission change is not influenced by introducing the CO2 inhibition effect since CO2 is uniformly distributed. When incorporating all the above effects, the end-of-century modeled isoprene fluxes are found to range either between 0 % (RCP2.6) and +17 % (RCP8.5) (using ) or between 11 and 65 % (using , not shown), with respect to the control fluxes. Note, however, that recent studies suggest that the CO2 inhibition of isoprene is reduced at high temperatures and therefore it may not have a large influence in the warmer Europe predicted in future climate scenarios .

Precipitation plays only a minor role in most regions, although the drier future summers simulated for Mediterranean regions should lead to enhanced soil moisture stress, which is believed to inhibit isoprene emission , and therefore tend to decrease the fluxes. As the present study neglects the effect of soil moisture on isoprene fluxes, the present and future fluxes are likely to be somewhat overestimated, in particular over southern Europe. In this region the increasing temperatures and the decreasing precipitation trends ( and Fig. ) should result in enhanced soil moisture stress, possibly causing a decline of isoprene fluxes over time. However, the influence of soil moisture stress on isoprene fluxes is still highly uncertain; for example, the MEGAN parameterization implemented with soil moisture fields from ECMWF reanalyses has been found to overestimate this effect over arid and semiarid regions .

Our simulations predict isoprene emission changes falling within the range of previous studies, i.e., between +90 % and -55 % on the global scale, and between +85 % and -30 % over Europe (Fig. ). The large dispersion of the different estimates of Fig.  is, to a large extent, explained by the diversity of model setups, namely the climate scenario, the study period, and most importantly, the choice of driving parameters which are allowed to vary (i.e., the climate fields, the CO2 activity factor, and/or the vegetation distribution). The increase in isoprene emission as a result of climate change of +70 % globally and of +85 % over Europe are very close to the predicted emission change in our study when only climate changes are considered. Weaker emission changes are induced when incorporating the CO2 inhibition effect, being between -10 % and +25 % compared to present-day emissions, in good consistency with the emission changes simulated in the present study.

Considering future changes in vegetation induces an additional decrease or increase in isoprene emissions depending on the simulation setup. The use of a dynamical vegetation model generally leads to higher isoprene flux estimates due to the increasing biomass as result of rising temperatures, radiation, and CO2 fertilization . Overall, most studies using a dynamical vegetation model agree on a relatively strong flux increase in the wide range of 27 % to 360 % . Human-induced land use changes generally cause less drastic emission changes . Significant cropland expansion is likely to result in lower isoprene fluxes globally, at most 41 % lower than present-day emissions . A recent study reported that, globally, human-induced land cover change is expected to have a more significant impact than natural vegetation changes, leading to a relative decrease of future isoprene emissions up to 33 % . Note, however, that afforestation is expected to be the dominant land use change over Europe, and therefore the combination of natural and human-induced vegetation changes could induce a significant increase in isoprene emission of up to 40 % . The application of land use change scenarios (e.g., those of the ALARM project, ) to projected isoprene emission estimates with MEGAN–MOHYCAN will be carried out in future work.