Process based inventory of isoprenoid emissions from European forests : model comparisons , current knowledge and uncertainties

Large uncertainties exist in our knowledge of regional emissions of non-methane biogenic volatile organic compounds (BVOC). We address these uncertainties through a two-pronged approach by compiling a state of the art database of the emissions potentials for 80 European forest species, and by a model assessment and inter-comparison, both at the local and regional scale, under present and projected future climatic conditions. We coupled three contrasting isoprenoid models with the ecophysiological forest model GOTILWA+ to evaluate leaf and ecosystem isoprenoid emissions, build an emissions inventory for European forests, and to consider model behaviour in present climate and under projected future climate change conditions. Hourly, daily and annual isoprene emissions as simulated by the models were evaluated against flux measurements. The validation highlighted a general model capacity to capture gross fluxes but inefficiencies in capturing short term variability. A regional inventory of isoprenoid emissions for European forests was created using each of the three modelling approaches. The models agreed on an average European emissions budget of 1.03 TgC a −1 for isoprene and 0.97 TgC a−1 for monoterpenes for the period 1960–1990, which was dominated by a few species with largest aerial coverage. Species contribution to total emissions depended both on species emission potential and geographical distribution. For projected future climate conditions, however, emissions budgets proved highly model dependent, illustrating the current uncertainty associated with isoprenoid emissions responses to potential future conditions. Correspondence to: T. Keenan (t.keenan@creaf.uab.es) These results suggest that current model estimates of isoprenoid emissions concur well, but future estimates are ighly uncertain. We conclude that development of reliable models is highly urgent, but for the time being, future BVOC emission scenario estimates should consider results from an ensemble of available emission models.


Introduction
Non-methane biogenic volatile organic compounds (BVOC), emitted by most plant species, is a heterogeneous compound class made up of a wide range of reactive volatile hydrocarbons.European forest species emit large amounts of BVOCs, in particular, volatile isoprenoids: isoprene (C 5 H 8 ) and monoterpenes (C 10 H 16 ) (Arneth et al., 2007;Guenther et al., 1995;Simpson et al., 1999).
BVOCs play a significant role in atmospheric chemistry (e.g., Fuentes et al., 2000;Gelencser et al., 2007;Helmig et al., 2006;Kanakidou et al., 2005;Szidat et al., 2006), in particular in the formation of secondary organic aerosols (Kanakidou et al., 2005) and tropospheric ozone at high light T. Keenan et al.: Process based inventory of isoprenoid emissions intensities and temperatures, and in the presence of NO x (Fehsenfeld et al., 1992).Emissions from terrestrial ecosystems also cause a decrease in atmospheric hydroxyl radical (OH) concentrations and thereby result in an increase of the lifetime of methane in the troposphere (CH 4 ) (Poisson et al., 2000;Roelofs and Lelieveld, 2000).The emissions of BVOC have therefore far-reaching implications for air quality (e.g., Bell and Ellis, 2004), both globally and locally.The effects of biogenic emissions on methane lifetime and aerosols can further lead to important feedbacks between emissions and climate change (Kulmala et al., 2004;Peñuelas and Llusià, 2003).
Due to their importance in air chemistry and climatic processes, regional-scale emission inventories are needed to predict regional air quality as well as simulate future climatic conditions (e.g., Collins et al., 2004;Kulmala et al., 2004;Tunved et al., 2006).This requires application of emission models accurately describing the responses of emissions to variation in environmental drivers (Grote and Niinemets, 2008).Temperature and radiation have been found to be the main driving factors for the emission of both isoprene and monoterpenes in broad-leaved species (Dindorf et al., 2006;Hansen and Sharkey, 2001;Monson and Fall, 1989;Loreto and Sharkey, 1990), while only temperature has been suggested to control monoterpene emissions in some conifers (Tingey et al., 1980, but see Staudt et al., 1997).These key findings have driven the development of isoprene and monoterpene emission models from simple empirical models to more process based designs.
Early emission modelling methods took an empirical approach, linking emissions directly with climatic variables (Guenther et al., 1995).These have shown to have high predictive power in some, but not in all situations (e.g., Arneth et al., 2007;Bai et al., 2006;Geron et al., 2002;Harley et al, 2004;Otter et al., 2002Otter et al., , 2003;;Wang et al., 2007).With an improved understanding of the biochemical reaction pathways for the formation of various plant volatiles, more mechanistic models have evolved (Bäck et al., 2005;Martin et al., 2000;Niinemets et al., 1999;Niinemets and Reichstein, 2003;Zimmer et al., 2000).In these process based models, responses of key enzymatic activities to environmental variables set the internal biochemical conditions, which finally determines the emission rates.These advanced models, however, still do not describe several key physiological observations (Monson et al., 2007), such as responses to elevated CO 2 .So far, few attempts have been made to compare the capacity of different emission models to simulate the environmental variability in emissions encountered in the field (Arneth et al., 2007).
Emission models are almost exclusively developed at the leaf level and are then scaled, through spatial and temporal scales, to the canopy, to the stand, and finally to the region level.This requires the coupling of an emission model with a process-based SVAT (soil-vegetation-atmosphere transfer) model.Such models describe leaf scale processes, set the micrometeorological canopy conditions which drive these processes, and describe the biomass and foliage distribution throughout the canopy (Baldocchi et al., 1999;Grote, 2006;Lamb et al., 1993;Lenz et al., 1997;Schaab et al., 2003).Such an approach also requires reliable information of emission potentials of each individual species.
The BVOC emission potential (E S , the maximum emission rate under standard conditions) of terrestrial vegetation is one of the most important variables in modelling BVOC emissions (Arneth et al., 2008b;Grote and Niinemets, 2008).E S strongly varies among species with values near zero to greater than 100 µg g −1 leaf h −1 (Kesselmeier and Staudt, 1999;Wiedinmyer et al., 2004).This high variability also exists between similar taxonomic entities (e.g., Owen et al., 2001) such that the emission potential of species of the same genera may be vastly different (Benjamin et al., 1996;Kesselmeier and Staudt, 1999).For instance, among Quercus species, some species are known to be strong isoprene emitters, some monoterpene emitters, and some species have emissions close to zero (Kesselmeier and Staudt, 1999).While extensive emission potential databases have been collated over recent years (Kesselmeier and Staudt, 1999;Wiedinmyer et al., 2004, http://bai.acd.ucar.edu/Data/BVOC/index. shtml; Nick Hewitt's database: http://www.es.lancs.ac.uk/ cnhgroup/iso-emissions.pdf), these databases include all the estimates made so far without a critical revision.Even for a single species, a large variability exists in the emission factor estimates that is currently not understood.There is a general tendency of modelling studies to use average values of emission factors proposed by emission factor databases (Guenther et al., 2006;Parra et al., 2004;Simpson et al., 1995).However, as non-standardized methods have been used in the estimation of emission factors in the past, using averages of all estimates results in propagation of errors from study to study.Thus, a critical revision of the emission factors is pertinent to standardize the emission factor estimates as much as possible.
In this paper, we first embedded three different isoprenoid emission models into a process based terrestrial biogeochemical model, thus providing a bottom up approach both to quantify the isoprene and monoterpenes emissions being released into the atmosphere under present day and potential future climate change.With these new tools we tested the possible variability due to differences in leaf level models.For this, isoprene emission flux measurements were used to compare diurnal and seasonal emission predictions at two forest sites with contrasting forest structure and species composition, one in the south of France and the other in Michigan, USA.We further revised the emission potentials for 80 dominant European forest species both occurring naturally and in forest plantations.In all cases, original studies were examined and a new consensus estimate was derived for each species.Using these basal emission potentials we derive an emission inventory for both isoprene and monoterpenes emissions from European forests under current climatic conditions using the three different modelling approaches.Finally, European forests were used as an example to analyse and compare large scale model performances under future climatic conditions.

Leaf level emission algorithms
A recent model comparison study highlighted three isoprenoid emission models as possible candidates for regional or global applications (Arneth et al., 2007).The models take contrasting approaches to modelling emissions, each with different assumptions about the way environmental factors limit the emissions and with different levels of mechanistic detail.Where pertinent, modifications were made in order to ensure consistency between the models (as in Arneth et al., 2007).No direct CO 2 effect on the emissions was applied in the models.The effect of CO 2 on emissions has been previously reported (Rosenstiel et al., 2003), but as yet the mechanistic understanding is poor.

The Guenther et al. model
By far the most widely used models for simulation of natural isoprenoid emissions were developed by Guenther et al. (1991), andGuenther et al. (1993).Their approach was to describe the emission rates by using long-term basal emission factor for isoprene (E I ) and monoterpenes (E M ), and adjusting these basal emission potentials by two empirical factors, one describing the response to light intensity and the other to leaf temperature.The correlation between short term fluctuations, light intensity and leaf temperature is widely studied and much work has gone into validating the Guenther et al. model under different environmental conditions (Monson et al., 1994;Petron et al., 2001).
The emission factors used in the model are emission rates normalized to a leaf temperature (T ) of 30 • C and quantum flux density (Q) of 1000 µmol m −2 s −1 (sometimes for monoterpenes normalized only to temperature) (Guenther, 1991;Guenther et al., 1993Guenther et al., , 1995Guenther et al., , 1997)).For light and temperature dependent isoprenoid emission, the algorithms (based on the equation: E=E S C L C T , where E S is the basal emission potential, C L and C T are the functions of quantum flux density and leaf temperature, respectively.)as outlined in Guenther et al. (2006) were used.
Parameters were determined following the original paramerisations of the Guenther et al. model, including recent algorithms developed by Guenther et al. (2006), which links parameter values to short term (24 h) and long term (10 days) fluctuations in temperatures and light intensity.
For some species, in particular conifers with extensive monoterpene pools with slow turnover compared with the rate of monoterpene synthesis, monoterpene emissions have been related directly to leaf temperature.In such cases, monoterpene emissions were calculated following the Guenther et al. (1993) temperature based monoterpene model, which fluctates emissions based on an empirical temperature response.

The Niinemets et al. model
The Niinemets et al. model for isoprene and monoterpene emissions takes a process-based approach, linking the emission rates to the activity of the synthase enzyme S S to predict the capacity of isoprenoid synthesis pathway and to foliar photosynthetic metabolism via the photosynthetic electron transport rate, J , to predict substrate availability for isoprenoid synthesis (Niinemets et al., 1999(Niinemets et al., , 2002b)).Here, the supply of dimethylallyl pyrophosphate (DMAPP) and NADPH, as affected by the rate of photosynthetic electron transport and the competitive strength of the synthase enzyme for electrons, are considered as the main controlling factors for the rate of isoprenoid synthesis.
Emission rates are calculated through the fraction of total electron flow used for the isoprenoid synthesis, the rate of photosynthetic electron transport, and the cost of isoprenoid synthesis in terms of electrons.Thus, the emissions are linked to the photosynthetic activity of the leaf with the use of only one single leaf dependent parameter, the fractional allocation of electron transport to synthase activity.Martin et al. (2000) developed a process based approach for isoprene emissions using the knowledge of the biochemical pathway of isoprene synthesis, thus providing a highly mechanistic model.

The Martin et al. model
As the rate of isoprene emission will be governed by the rate of the slowest reaction in its biochemical pathway, the emissions can be calculated depending on which partial process is limiting under given environmental conditions.This model formulates the extent different processes in the biochemical pathway of isoprene synthase limit isoprene emission and represents the rate of emission as the minimum of three potentially rate limiting processes: -The supply of carbon to isoprene synthase via pyruvate; -The supply of ATP from phosphorylation; -The temperature dependency of the isoprene synthase reaction rate.
The model applied here is a slight modification (following Arneth et al., 2007) of the original version used by Martin et al. (2000) in order to assure compatibility with the analogous expression in the model of Niinemets et (Lehning et al., 1999;Loreto and Sharkey, 1993;Monson et al., 1992), following the suggestion that Pyruvate is formed from Rubisco carboxylation about 1% of the time (Andrews and Kane, 1991).Implicit in both the Niinemets et al. and the Martin et al. model approaches for the simulation of isoprene emission on the basis of isoprene synthase kinetics (Fall and Wildermuth, 1998) is that DMAPP concentrations must change in response to changes in light availability at any given temperature, i.e. isoprenoid emission rates are substrate-controlled.DMAPP levels have been found to vary within and between days, and varied more in plants which emitted isoprene (Rosenstiel et al., 2002).

Modifying the leaf-level models to simulate longterm emission responses to environment
All three leaf-level models provide qualitatively similar responses of isoprenoid emissions to changes in temperature and light over short term (Arneth et al., 2007).However, emission rates also adjust to long term modifications in environmental drivers, implying that it is important to consider such effects as well to simulate long-term emission dynamics.
Phenology.Phenology is known to affect isoprenoid emissions and many studies have characterized how emissions respond to phenological events (e.g., Fuentes and Wang, 1999;Kuhn et al., 2004;Monson et al., 1994;Petron et al., 2001;Wilberley et al., 2005).This effect was incorporated into each of the emission models in the same simple manner, by reducing the basal emissions potential which is used in each model by a temperature dependent scalar, K, ranging from 0 to 1.This scalar is calculated following Leinonen (1996), an updated version of the Pelkonen and Hari (1980) approach for calculating the seasonal variations in photosynthetic potential.So, in this way, emissions are scaled in parallel with the photosynthetic potential.In the case of deciduous phenology (Smith et al., 2001) this also leads to a decreasing isoprenoid emission capacity in senescing leaves before the leaves die (Arneth et al., 2007;Geron et al., 2000;Goldstein et al., 1998;Pressley et al., 2005).
Soil water availability effects.Isoprenoid emissions are effected by seasonal water stress, though the exact extent and mechanisms behind the emission responses to water stress are yet unclear (Grote and Niinemets, 2008).A strong reduction of emissions is observed in most cases after strong seasonal water stress (Lavoir et al., 2009;Llusia and Peñuelas, 1998;Pegoraro et al., 2004), though in some cases an increase in emissions has been reported (Pegorara et al., 2007).In the model GOTILWA+, to which the emissions models are coupled, photosynthesis is reduced directly during water stress through a reduction in the rate of electron trans-port, and the maximum carboxylation capacity (Keenan et al., 2009).All three emissions models are indirectly influenced by the soil moisture dependence of stomatal conductance which influences the leaf temperature estimated by GOTILWA+.The Guenther et al. model also includes a direct effect of soil water stress.Emissions are reduced during water stress directly in parallel with photosynthesis, by applying the same function which reduces photosynthetic potential (Grote et al., 2009;Guenther et al., 2006;Keenan et al., 2009).Stress affects the Niinemets et al. model emissions indirectly, as the drought-dependent reductions in the rate of electron transport, J T , lead to reduced substrate availability and thereby to reduced emissions.The Martin et al. model was modified in a similar manner as the Guenther et al. model, but here the water stress function was also applied to the supply of emission precursors.

Species-specific emission potentials
Calculated total emissions are highly dependent on the assigned emission potentials E I and E M that are directly applied in the Guenther et al. model, and determine the isoprenoid synthase activities in the Niinemets et al. and Martin et al. models.The emission potentials have recently been concluded to be one of the most important parameters for modelling regional emissions (Arneth et al., 2008b;Grote and Niinemets, 2008).In most large-scale modelling studies, the values initially specified by Guenther et al. (1995) are applied.Since then, much more information has become available and been integrated into previous estimates (Guenther et al., 2006).However, for most ecosystems and vegetation types measurements of BVOC emissions are still scarce or absent (Grote and Niinemets, 2008;Guenther et al., 2006).Due to the large variation within families, assigning general emission rates to plant functional types is often done by subjective judgment.It often occurs that very different emission potentials are assigned to a certain functional type or species in different modelling studies, leading to very different results (Arneth et al., 2008b for review).We address this problem through a thorough critical literature review to compile the state of the art knowledge of species-specific emission potentials for 80 species included in the GOTILWA+ forest model (Table 1).
In this analysis, all previous known reports were reassessed with regards to measurement methods and derivation of standardized emission potentials.As there is considerable uncertainty in actual light and temperature environment in whole-tree and whole-branch enclosure studies that can seriously affect standardization of emission rates, we preferred estimates conducted with single leaf enclosures.As a variety of temperature responses have been used to convert the emission rates to standardized conditions of 30 • C and 1000 µmol m −2 s −1 , whenever actual response curves were absent, we reconverted the data using the    Guenther et al. (1993) original model.In all cases, units were homogenized (e.g., µg C g −1 hr −1 often used in North American studies vs. µg BVOC g −1 hr −1 often used in European studies).
Errors due to the use of inappropriate values of leaf dry mass per unit area, M A , in converting area based relations to mass basis (for instance converting total leaf area based estimates using projected area-based M A values) were corrected.Three sources of errors were identified in M A values. 1) the reciprocal of M A , SLA (in units cm 2 g −1 ), had been used, and instead of multiplication, area/based values were divided by SLA; 2) M A had been calculated on the basis of total area, but emission measurements were based on projected area (mainly a problem in needle-leaved species), 3) errors in units.These errors were identified on the basis of information provided in the Material and Methods about the methods of area determination and expression of the characteristics, and on the basis of tabulated values of M A and the area-based emission rates.In the case of studies using literature-based M A values, original M A values were looked up from the literature.
Because of the lower sensitivity of analytical techniques in the past, and significant background level of isoprene and monoterpenes in the gas-exchange enclosures either due to isoprenoids in incoming ambient air, or adsorption/desorption processes in the cuvettes, there is considerable uncertainty in detecting minute emission rates with conventional techniques.Due to these uncertainties, emission rates less than 0.1 µg g −1 hr −1 were set to zero in the current compilation.Only the values corresponding to fully-developed leaves in non-stressed conditions were used whenever possible.For several species included in the GOTILWA+ model, reliable emission rates were not available.For these species, emission factors of taxonomically closest species were employed (Benjamin et al., 1996 for discussion).
In addition to the emission factors, for the simulation of monoterpene emissions, it is also important to know whether emissions are only temperature dependent or both light and temperature dependent.In several conifers, the emission may be both light and temperature dependent (Staudt et al., 1997;Shao et al., 2001) and information of the share or these two contrasting emission sources are provided in Table 1.

Scaling from the leaf to the landscape
The three emission models simulate isoprenoid emissions from any given leaf in dependence on climatic conditions.To scale from the leaf to the landscape, the emission models were coupled to the biogeochemical forest model, GOTILWA+.This model described leaf structural and chem-ical characteristics, and thus foliage physiological potentials.This model also described the microclimatic conditions and forest structure necessary to scale from the leaf to the canopy, and further to the region.Using forest inventories and regional databases of climate on a 10 ×10 scale, simulations were run for each of the forest stands in EU15+2 (pre-enlargement Europe, Norway and Switzerland).

Scaling from the leaf to the stand: the GOTILWA+ model
Each emission model considered was coupled separately to the photosynthetic submodel of the GOTILWA+ terrestrial biogeochemical model (Gracia et al., 1999;Keenan et al., 2008; http://www.creaf.uab.es/GOTILWA+).GOTILWA+ is a process-based forest model that has been developed to simulate carbon and water fluxes from forest ecosystems and to explore how the functioning of forests is influenced by climate, tree stand structure, management techniques and soil properties.Carbon and water fluxes of forests are simulated for different environmental conditions, for different tree species, and under changing environmental conditions that result from either climatic modifications or from alterations in management regime.
The GOTILWA+ model includes a two-leaf canopy photosynthetic model (Wang and Leuning, 1998;Dai et al., 2004).The photosynthesis sub-model treats the C 3 photosynthetic pathway.The canopy is divided into sunlit and shaded leaves, with the amount of intercepted diffuse and direct radiation depending on the time of the day, season, and the area of leaf exposed to the sun (Campbell, 1986(Campbell, , 1990)).Foliage net assimilation rates are calculated using the Farquhar et al. (1980) photosynthesis model, with dependencies on intercepted quantum flux density, species-specific photosynthetic capacities, leaf temperature, and leaf intercellular CO 2 concentration (C i ).Stomatal conductance is calculated using the Leuning et al. model (Leuning et al., 1995) that is the advancement of the Ball et al. model (Ball et al., 1987).Water stress affects the photosynthesis-conductance coupling by directly reducing the photosynthetic potential through a nonlinear relation to soil water content (Keenan et al., 2009).Canopy isoprene and monoterpene emission rates were calculated on an hourly basis as the sum of sunlit and shaded leaf fractions using their specific leaf temperature and incident radiation values.

Scaling from the stand to the region
To supply the input data required by the model, an extensive database has been built within the framework of the European ATEAM (Advanced Terrestrial Ecosystem Analysis and Modelling) and ALARM (Assessing Large-scale Risks for biodiversity with tested Methods) projects, connecting diverse information sources at a European level and adapting them to fit the same spatial resolution of 10 ×10 .The database contains data related to forest species, forest area cover, forest structure, forest function (photosynthesis, respiration rates), soil hydrology, organic matter decomposition rates and management strategies (Schröter et al., 2005).Area explicit estimates of forest cover were made available which specify the eighty dominant forest species in Europe.Simulations were run with GOTILWA+ for each 10 ×10 scale forested pixel in Europe (EU15+2).For parameterisation of the forest structural components in GOTILWA+, three forest functional types (temperate deciduous, temperate broadleaf evergreen, and temperate needle-leaf evergreen) were considered.

Scaling through time
Simulations were run with each emission model coupled to the GOTILWA+ model for a two hundred year period from 1900 to 2100.From 1900 to 2000, a reconstructed climatic data time series based on the CRU05 (1901CRU05 ( -2000) ) monthly dataset (New et al., 1999) was used, with global atmospheric concentrations of CO 2 from 1901 to 2000 obtained from the Carbon Cycle Model Linkage Project (McGuire et al., 2001).The IPCC socioeconomic analysis (IPCC, 2001(IPCC, , 2007) ) resolved several possible standard scenarios for CO 2 emissions (A1, A2, B2, B1).Different global circulation models use these scenarios to generate future climatic projections.In this work we have applied the climatic projection for period 2001-2100 generated by the HadCM3 global circulation model using the A2 scenario as an indicator for the effect of possible future climate change on the BVOC emissions.This scenario uses an estimated increase in atmospheric CO 2 to 709 ppm by 2080.Using this scenario, HadCM3 predicts an increase in temperature of 2.8 • C by 2080 for the area included in this study (in comparison to the average temperatures for the 1960-1990 period), and regional changes in precipitation.Species distributions for the past 100 years is assumed equivalent to current species distribution, and no explicit changes were applied for simulations over the 21st century.

Stand scale validation
Recent developments in direct methodologies such as eddycovariance, disjunct eddy covariance, relaxed eddy accumulation and disjunct eddy accumulation techniques provide high quality quantitative measurements of isoprenoid emissions (Businger andOncley, 1990, Hills andGuenther, 1998;Rinne et al., 2000;Karl et al., 2002;Ciccioli et al., 2003, Spirig et al., 2005).So far, few such measurements are available over forest ecosystems.Here, one relatively short-term series (26-days) of diurnal time-courses of isoprene emissions with a half-hour time-resolution (Arneth et al., 2007) and one long-term series of seasonal time-courses of isoprene fluxes with daily resolution (Pressley et al., 2005(Pressley et al., , 2006) ) are used to validate the implementation of the isoprene models at the canopy level.Three simulations were run at each site, one for each model coupled to the GOTILWA+ model, using stand and species parameters gathered from the literature (Arneth et al., 2007;Curtis et al., 2001;Goldstein et al., 1998;Pressely et al., 2006;Waring et al., 1998).Emissions factors were used from a previous modelling study at these sites (Arneth et al., 2007).Meteorological data was gathered using eddy-covariance measurements techniques at each site.

Site 1
We used measurements from two field campaigns at two Mediterranean Quercus pubescens stands in southern France (43 • 39 N, 6 • E) conducted during 2000 and 2001.In these campaigns, isoprene fluxes were measured with the eddy covariance technique for approximately two weeks in the summer of each year.The fast isoprene sensor (Hills and Zimmerman, 1990) was employed in these measurements, and half hour values were stored.These data have been used to examine the diurnal course of emissions predicted by each model.Average diurnal courses were constructed by taking the emission value for each hour of each day.

Evaluating isoprene model emissions: diurnal time-courses
The data from the French site were used to test the capacity of the models to reproduce the diurnal time-courses of isoprene emissions.Each model compared well against the average diurnal time-course (Fig. 1).Temperature and available light are the two main drivers for each of the models.The Guenther et al. and Martin et al. models showed similar responses to both temperature and radiation in 2001 (Fig. 2).Each showed a non-linear response to temperature with emissions exponentialy increasing with increasing temperature.The Niinemets et al. model was less sensitive to higher temperatures than the other two models (Fig. 2).Each of the models showed a similar response to global radiation (Fig. 2), though the slope of the response differed slightly between models a = 0.015, 0.016, and 0.014 for the Guenther et al., Martin et al., and Niinemets et al. models respectively.

Evaluating isoprene model emissions: seasonal time-courses
The effect of phenological events during spring and autumn was well captured in 2000 and 2001 (Fig. 3).In 2002, the spring emissions were overestimated, though the higher proportion of missing data during this period complicated the comparison.The autumn reduction in isoprene emission was correctly predicted by the models for each year.Day-today variability and summer maxima were reproduced with a lesser accuracy than the long-term phenological effects.In particular, the models strongly underestimated emissions in 2000.

Current emissions from European forests
Total simulated isoprene emission from pre-enlargement European Union with Norway and Switzerland (EU15+2) was 1.03 TgC a −1 for the period between 1960 and 1990.A broad range of canopy emissions was observed for the species considered (Fig. 4), with order of magnitude differences per unit ground area.However, when considered together with species aerial coverage and climatic distribution, the strongest emitters were not necessarily the largest contributors to the European emission budget.The Europe-wide isoprene emission was dominated by a few species.Quercus robur was by far the highest contributor, with 32% of total Europe-wide emission.Quercus pubescens and Quercus petraea contributed 18% and 16%.Eucalyptus species, Quercus frainetto, and Quercus pyrenaica each contributed around 6% to total emissions.Six species contributed between 1 and 2% (Quercus faginea, Quercus rubra, Populus spp., Robinia pseudacacia, Picea abies), with the remaining species contributing less than 1% of the total European emission budget.Monoterpene emissions showed a similar pattern, with a very broad range of species emissions per unit stand area, Atmos.Chem.Phys., 9, 4053-4076, 2009 www.atmos-chem-phys.net/9/4053/2009/(Arneth et al., 2007), compared with the simulation results using GOTILWA+ with the three isoprene emission models (Guenther et al., Niinemets et al., and Martin et al. models).The average diurnal time-course from the measurements is also compared with the simulation results.but with only a few species dominating the total budget of emissions from European forests when both emission potentials and geographical distribution were taken into account.
Total simulated emissions amounted to 0.93 TgC a −1 .Of this total, 24% was emitted by forests dominated by Quercus ilex, 21.5% by Pinus sylvestris, 16% by Fagus sylvatica, 11% by Picea abies and 7% by Quercus suber.Seven further forest species (Fig. 5) contributed between 1 and 2% to total emissions, with the remaining forest species contributing less than 1%.
Emissions showed a strong regional pattern, being influenced by the distribution of the species across Europe, and regional weather patterns.For isoprene emission, France showed the highest emissions for the period 1960-1990 (Fig. 6), followed by the north-western area of the Iberian Peninsula.The Mediterranean zone, although it is subject to higher temperatures, and radiation, showed lower emissions than central mainland Europe due to lower plant coverage and leaf area index as well as due to suppression of emissions during extensive summer drought periods (as reported by e.g.Grote et al., 2009).(Pressley et al., 2005) and fluxes simulated by GOTILWA+ with the three emission models compared (Guenther et al., Martin et al., and Niinemets et al. models) over three years (left panels) and the regressions of measurements vs. simulations (right panels).a represents the slope of the presented linear regressions, and RMSE is the root mean squared error.Simulated monoterpene emissions were lower than those of isoprene, with lower interspecific emission rate variability.The emissions of monoterpenes also showed a more uniform spatial distribution.The highest emissions were predicted for the southern Iberian Peninsula, and areas of the Mediterranean zone (Fig. 7).

Model comparisons for the period 1960-1990
The period 1960-1990 is often used as a baseline for comparing regional models in the past and the future (Schröter et al., 2005).For isoprene, slight differences were observed between the emission model predictions for this period, with the Guenther et al. and Martin et al. models giving 8.8% and 6.1% higher emissions than the Niinemets et al. model.For monoterpenes, the Guenther et al. model gave slightly higher (4.6%) emissions than the Niinemets et al. model for the simulated period, perhaps due to a slightly stronger temperature response (Arneth et al., 2007).

Projected future European emissions
Although the choice of the model had little effect on the estimated budget of isoprenoids during the period 1960-1990, Values represent average estimates from the three emissions models (Guenther et al., Niinemets et al., and Martin et al.).Black bars denote the percent contribution of each species to the total European isoprene emissions budget for this period (for contributions of greater than 1%).Error bars represent the spatial standard deviation from the mean for the model ensemble.Species emission potentials are according to Table 1.
projected future emission estimates were greatly affected by the choice of the emission model.In the case of isoprene, the models showed up to a two-fold difference in predicted emission rates by 2100 (Fig. 8).Differences were not fully evident until about 2050, when large disparities began to appear between the models.All models agreed on a general strong increasing trend in both isoprene and monoterpene emissions, though the dynamics of this trend differed between models.For isoprene emission, the Martin et al. model showed the strongest response with average per pixel emission of 4400 mg C m −2 a −1 from European forests.This gave a total European budget of 2.36 TgC a −1 for the period 2080-2100, i.e. more than double the average emission predicted for the period 1960-1990 with the same model.The Guenther et al. model gave slightly higher emissions than the other models for the early 21st century, but did not respond as strongly as the Martin et al. model   Estimated average annual monoterpene emissions (mg C m −2 a −1 ) from European forest species for the period 1960-1990 (grey bars).Values represent average estimates from the two emissions models (Guenther et al., and Niinemets et al.).Black bars denote the percent contribution to total European monoterpene emissions budget for this period (for contributions of greater than 1%).Error bars represent the spatial standard deviation from the mean for the model ensemble.Species emission potentials follow Table 1.
as strongly as the other models to future climate change, giving a total of 1.58 TgC a −1 .On average, 98% increase in isoprene emission was predicted for the period 2080-2100 relative to the emission in the period 1960-1990 with the three emission models.
Of the two monoterpene emission models, the Guenther et al. model consistently predicted higher emissions than the Niinemets et al. model throughout the 21st century, and responded much more strongly to climate change, perhaps reflecting its higher temperature sensitivity as suggested by Arneth et al. (2007).For the period between 2080 and 2100, the Guenther et al. model predicted a total European monoterpene budget of 1.27 TgC a −1 , which represents a 31% increase in monoterpene emissions with respect to the emissions for the period   emission from European forests of 1.07 TgC a −1 , which represented an 11% increase with respect to the period from 1960 to 1990.As an average of the two models, monoterpene emissions from European forests were predicted to increase by 21% for the period 2080-2100 relative to the period 1960-1990.

Discussion
To our knowledge, this is the first time different BVOC emission modelling approaches have been run in parallel on a regional scale.The results serve to both reduce and highlight uncertainty in the modelling of current emissions of isoprene and monoterpenes from European forest species through a two-pronged approach: the compilation and assessment of species specific emission potentials, addressing the broad variability of values published in the literature, and the comparison of three distinct methodologies for the prediction of regional emissions.So far, all regional emission inventories in Europe have been based on uncritical use of species emission potentials collected in databases that have not been updated since late 90s (e.g., Nick Hewitt's database: http://www.es.lancs.ac.uk/ cnhgroup/iso-emissions.pdf) (Parra et al., 2004;Projections, 2007;Simpson et al., 1999;Solmon et al., 2004).However, new information has become available on the emission characteristics of some important species such as important Mediterranean species Quercus suber that has been traditionally considered "non-emitting" species in emission Fig. 7.Estimated annual monoterpene emissions (gC m −2 a −1 ) from European forest canopies over the period 1960-1990, using average estimates from the two monoterpene emission models (Guenther et al. and Niinemets et al.) coupled to GOTILWA+.models.Recent data show that this species strongly emits monoterpenes (Pio et al., 2005;Staudt et al., 2004).Analogously, Fagus sylvatica, a dominant component in European temperate deciduous forests has previously been reported to be a minute monoterpene emitter (König et al., 1995;Steinbrecher et al., 1993), while recent data demonstrated that this species is moderate to high monoterpene emitter (Dindorf et al., 2006;Luchetta, 1999;Moukhtar et al., 2005).While for some species, reliable information of emission potentials is still not available, the aerial coverage of these species is generally small and only minor improvement of large-scale emission estimates is expected.
Model comparisons with tower flux measurements at midlatitude forest sites show that the models do not differ greatly in their ability to reproduce the short-term variations in iso-prene emission, accurately capturing the diurnal time-course of isoprene emission driven by modifications in light and temperature.The Guenther et al. model, being solely based on light and temperature, was not effective at capturing the mid-day decline in the emission in drought conditions, leading to a general overestimation of the total emission at the Mediterranean site.The Guenther et al. model is based on two driving variables, and assumes a fixed shape of the response of isoprene emission to these variables.Yet, other factors have been known to affect the emission rates as well (e.g., Affek and Yakir, 2002;Loreto et al., 2001;Loreto and Velikova, 2001;Peñuelas and Llusia, 2001;Sharkey and Singsaas, 1995;Sharkey and Yeh, 2001).Given its simplicity, its performance under present-day conditions is surprisingly comparable with the other two more plastic models., 2001, 2007).The displayed error bars (in grey) represent the standard error from the mean due to spatial variation.The Mean Trend is a quadratic regression (y=y 0 +ax+bx 2 ) of the displayed data.
The Niinemets et al. model performed better for simulation of average diurnal emission time-course, but it was unresponsive to the high temperatures experienced in the last few days at the French site.The response of the Niinemets et al. model to temperature was reported in a previous study (Arneth et al., 2007, Fig. 1), though not discussed in detail.There, it was also shown to be slightly less responsive to temperature than the other two models.The mid-day decline of the Niinemets et al. model is related to its stronger dependence on light through the electron transport activity (Arneth et al., 2007).As numerous studies investigating diurnal variations in chlorophyll fluorescence in drought-stressed plants demonstrate, electron transport activity is also downregulated around mid-day, in line with the rates of photosynthesis.In addition, the model of Niinemets et al. based on overall energy level predicts that the mid-day decline is associated in downregulation of DMADP pool size (that is linked to the rate of electron transport).Recent support to this hypothesis comes from the study of Magel et al. (2006).The Martin et al. model, which takes a more detailed approach to describe the limiting precursors of emissions, better reproduced the diurnal time-series, accurately capturing both the day-to-day variability and the average diurnal time-course.Long-term (daily to seasonal) emissions were moderately well reproduced by the models.The low correlation with the data highlights the fact that a complete understanding of emissions drivers and controls is missing from the models, and we are indeed a long way from fully realistic BVOC emission estimates (Arneth et al., 2008b;Grote and Niinemets, 2008;Monson et al., 2007).
The analysis of the models diurnal responses to light and temperature (Fig. 2) suggests that the temperature response is responsible for most of the model differences.This became more apparent at higher temperatures, and explains the increasing differences between model predictions as we move to a projected future warmer climate.Resolving this temperature response will be one of the challanges for future VOC modelling studies.More long term measurements of emissions data from forest canopies in regions which experience high summer temperatures would be of great benefit.
Although there are many strong isoprene and monoterpene emitting tree species in Europe, the regional distribution of these species leads to only a few of these strong emitters being important for the estimation of total European emission budget.These species were not necessarily the strongest emitters, but tended to be species with large coverage in regions with high temperatures and radiation.Total European forest isoprene emissions for the late 20th century were strongly dominated by three Quercus species, making up over two thirds of the total isoprene emission.It is therefore of considerable importance to focus research efforts on accurately quantifying the emission potentials from these species, and their emission responses to environmental drivers.Monoterpene emissions were similarly dominated by a few species, with five species contributing 80% of the total emission budget.Having more accurate data on these species could greatly improve the reliability of estimates of present day emission budgets.This dominance of few species over the total European emissions budget means that the use of an improved basal emissions database did not lead to a large difference in the total budget when compared to previous studies.We expect, however, that local budgets should be improved by the use of a more up to date basal emissions database.As we have confirmed the total budget using more rigorous methods, we can also have more confidence in current model predictions.
The resulting inventory of isoprene emission is similar to the value presented by Simpson et al. (1999) (1.4 TgC a −1 ; Simpson et al., Table 18), and Arneth et al. (2008a, 1.2 TgC a −1 ) for the same area.Since isoprene emission estimates are highly uncertain, due to their linear dependency on the leaf emission potentials assigned to a species or vegetation type, and due to strong dependence on the kind and quality of the land cover information used (Guenther et al., 2006), the similarity of the calculated annual totals is remarkable.Recently it has been suggested that such agreement between model simulations is due to compensation efforts applied to move model estimates closer to the hypothetical "real" regional or global emission value (Arneth et al., 2008b).We made no such effort, and would suggest that the similarity of different modelled isoprene estimates derives from the fact that regional isoprene emissions are dominated by a few highly emitting and well documented species.
Such similarities do not apply for monoterpene emission estimates, with a broad range of global emissions reported in the literature (Adams et al., 2001;Guenther et al., 1995;Kaplan et al., 2006;Lathiere et al., 2006;Levis et al., 1999;Naik et al., 2004;Tao and Jain, 2005;Valdes et al., 2005).Our estimate of total monoterpene emissions from the European forests of 0.97 TgC a −1 is lower than the emission estimate published in Tao and Jain (2005; 1.73 TgC a −1 ).However, their simulation is based on a larger part of Europe than ours, and their parameterization also differs from ours by considering six plant functional types only, rather than species-specific parameterization.To our knowledge, no study has simulated the emission of monoterpenes from European forests on a European scale.
Phenology is becomming increasingly recognised to have a strong affect on regional emissions (e.g., Simon et al., 2006;Tarvainen et al., 2007).Here we have applied a simple function to the basal emission potential in order to simulate a common phenological response in each of the models.Such functions have recently been highlighted as one of the biggest uncertainties in regional modelling (Keenan et al., 2009).The extrapolation of phenological functions to the future is also subject to great uncertainty.Most current phenological models predict a lenghtening of the growing season in Europe due to projected future climatic change.This would acount for some of the future increase in emissions reported in this study, but would not lead to between model differences in estimated emissions.
Although both the isoprene and monoterpene emission models performed comparably under current climatic conditions, large differences were observed in the emission estimates for realistic future climatic change scenarios.The differences observed here between model responses for the late 21st century result from the small differences in temperature, radiation and atmospheric CO 2 concentration responses reported in the isoprene model (Arneth et al., 2007, Fig. 1, 2 for review).The isoprene model of Martin et al. is more sensitive to temperature, followed in sensitivity by the Guenther et al. model (Fig. 2).The long-term simulations under gradually increasing temperature and atmospheric CO 2 (Fig. 8) highlight the implications of such differences in sensitivities between models.Our study demonstrates that the choice of the model used can greatly alter the final result.The nonconcurrence of the emission models in simulations of future scenarios calls into question the validity of numerous conclu-sions regarding future emissions, and the resulting effects on atmospheric chemistry, made so far on the basis of only one emission model.
A recent study hypothesised that we are overconfident about our ability to accurately model BVOC emissions from terrestrial vegetation, and, according to the terminology of Le Quere (2006), we are in the "illusion" phase of model development (Arneth et al., 2008b).When we consider the estimates of present day emissions that were obtained by applying the same methodology and with the models coupled to the same terrestrial vegetation model, the comparability of model performance does not seem to support such a conclusion.However, when moving to projected future climatic conditions, it becomes clear that we are far from reaching a clear understanding of the processes governing emission rates (Monson et al., 2007), and their potential responses to future climate change.
Current model development efforts are focused on the improvement of simple empirical algorithms (Guenther et al., 2006) and the development of more process-based emission models (Arneth et al., 2006;Back et al., 2005;Grote et al., 2006;Niinemets et al., 2002a;Martin et al., 2000;Niinemets et al., 1999;Zimmer et al., 2000).Each approach lends itself more easily to different applications, such as the easy implication of simple empirical models in atmospheric chemistry models (e.g., Guenther et al., 2006), or the detailed study allowed for by the more intricate models (e.g., Grote et al., 2006).Despite the big differences in model structures, none of the models included in this study outperformed the others.In fact, no existing approach has been shown to perform consistently better (Arneth et al., 2007).This is due both to a lack of detailed model inter-comparisons, and a lack of good quality data with which to test the models.Further efforts in both fields, together with the development of new modelling approaches and synergies (Grote and Niinemets, 2008;Monson et al., 2007) will be needed to advance our ability to reliably simulate emissions.Meanwhile, we urge that results from different model approaches should be considered in any simulation project dealing with terrestrial emissions of BVOCs, particularly if considering future climate change scenarios.

Conclusions
The coupling of the three different model approaches (The Guenther et al., Niinemets et al. and Martin et al. models) to an ecophysiological forest model provides a unique opportunity to explore the time-dependent changes in modelled biogenic emissions due to differences in model structure and model responses to changes in climatic and physiological processes.The modelled emissions from present day European forests were shown to be independent of the emission model used, with estimates of 1.03 TgC a −1 for www.atmos-chem-phys.net/9/4053/2009/Atmos.Chem.Phys., 9, 4053-4076, 2009 isoprene emission and 0.93 TgC a −1 for monoterpenes, giving a consistent emission inventory for BVOCs from European forests.
Coincidence of model estimates of emissions for current and past climatic conditions in Europe suggest that the present day inventories of BVOC emissions provide realistic estimates.However, model-dependent differences in simulated estimates of future emissions of both isoprene and monoterpenes highlight the fact that we are in the early stages of the path towards a full understanding of the processes governing BVOC emissions.This has important implications for any study seeking to model future BVOC emissions.Many studies involving modelled future BVOC emissions (e.g., potential offsetting of emissions by rising CO 2 concentrations, relative effects of changes in land use on quantitative emission estimates, effects on future emissions on tropospheric O 3 concentrations and air quality) may need to be revised to take into account the inherent variability introduced by the choice of the emission model used.

Fig. 1 .
Fig. 1.Hourly measured canopy isoprene emissions (mg C m −2 h −1 ), from the French site for two two-week periods in 2000 and 2001(Arneth et al., 2007), compared with the simulation results using GOTILWA+ with the three isoprene emission models(Guenther et al.,  Niinemets et al., and Martin et al. models).The average diurnal time-course from the measurements is also compared with the simulation results.

Fig. 2 .
Fig. 2. Temperature ( • C, left) and radiation (W m −2 , right) responses of hourly simulated canopy isoprene emissions (mg C m −2 h −1 ), from the French site for the two-week period in 2001, for each of the three isoprene emission models(Guenther et al., Martin et al., and Niinemets  et al. models).Regressions against temperature are polynomial regressions given by the equation y=y 0 +ax+bx 2 , whilst regressions against radiation are linear regressions of the form y=y 0 +ax, with a root mean squared error of the regressions of -Temperature: 2.7, 3.5, 1.8 and Radiation: 2.5, 2.8, 2.3 for the Guenther et al.,Martin et al., and Niinemets et al. models, respectively.

Fig. 3 .
Fig. 3. Comparison of daily integrated canopy isoprene emission fluxes from the UMBS forest (Pressley et al., 2005) and fluxes simulated by GOTILWA+ with the three emission models compared (Guenther et al., Martin et al., and Niinemets et al. models) over three years (left panels) and the regressions of measurements vs. simulations (right panels).a represents the slope of the presented linear regressions, and RMSE is the root mean squared error.

Fig. 4 .
Fig. 4. Estimated average annual isoprene emissions (log scale) from European forest species for the period 1960-1990 (grey bars).Values represent average estimates from the three emissions models(Guenther et al., Niinemets et al., and Martin et al.).Black bars denote the percent contribution of each species to the total European isoprene emissions budget for this period (for contributions of greater than 1%).Error bars represent the spatial standard deviation from the mean for the model ensemble.Species emission potentials are according to Table1.
under conditions of more severe climate change (3720 mg C m −2 a −1 for 2080-2100).Total European isoprene emission simulated with the Guenther et al. model for the period 2080-2100 was 2 TgC a −1 .The Niinemets et al. model predicted the lowest emission of the three models throughout the century, and did not respond

Fig
Fig. 5.Estimated average annual monoterpene emissions (mg C m −2 a −1 ) from European forest species for the period 1960-1990 (grey bars).Values represent average estimates from the two emissions models(Guenther et al., and Niinemets et al.).Black bars denote the percent contribution to total European monoterpene emissions budget for this period (for contributions of greater than 1%).Error bars represent the spatial standard deviation from the mean for the model ensemble.Species emission potentials follow Table1.

Fig. 6 .
Fig.6.Estimated annual isoprene emissions (gC m −2 a −1 ) from European forests over the period 1960-1990, using average estimates from the three isoprene emissions models coupled to GOTILWA+.

Fig. 8 .
Fig. 8. Average per m 2 modelled isoprene and monoterpenes emissions from European forest canopies from 1900 to 2100.Climate from the CRU (New et al., 1999) was used for the period 1900 to 2000.Results from 2001 to 2100 correspond to climate from HadCM3 global circulation model using climate change scenario A2 (IPCC, 2001, 2007).The displayed error bars (in grey) represent the standard error from the mean due to spatial variation.The Mean Trend is a quadratic regression (y=y 0 +ax+bx 2 ) of the displayed data.

Table 1 .
Compilation of isoprene (E I ) and monoterpene (E M ) emission factors and projected leaf dry mass per unit area (M A ) in 80 dominant species of European forests.
The Guenther et al. model failed to reproduce the midday decline in emissions, and thus, overestimated the emissions at this site.The Martin et al. model also overestimated the midday isoprene emission rate, but to a smaller extent.The Niinemets et al. model accurately reproduced the early morning and late evening emissions, and also gave accurate estimates of the average diurnal midday emissions.When looking at the complete time series, no noticeable differences among model predictions can be discerned, except for the last five days of the measurement campaign in 2001.During this period, a large increase in the emissions was observed, which was accurately reproduced by both theGuenther et al. and the Martin et al. model.The  Niinemets et al. model, however, provedunresponsive during this period.This increase in the emissions was correlated with an increase in temperature during the last four days.Over the entire time series, all models performed well with an r 2 correlation of 0.71, 0.74, and 0.72 for theGuenther et  al., Niinemets et al., and Martin et al. models respectively.
The actual data exhibited a larger variability than the models in each year, in particular in 2000 and 2001.Both theGuenther et al. and the Niinemets et al. modelshowed nonlinear responses in their ability to simulate emissions, accurately capturing low level emissions (∼<20 mg C m −2 d −1 ), but underestimating higher emissions.The slope of the linear regression of both against the measurements were comparable and reflect this underestimation (Fig.3).The Martin et al. model showed a more linear response, with a slight tendency to underestimate emissions in all cases.Although the slope of the response of the Martin et al. model was better than that of the other two models, the root mean squared error of the Martin et al. model results was 46% higher than that of the Guenther et al. model and 30% higher than that of theNiinemets et al. model.