Atmospheric Chemistry and Physics Global Isoprene Emissions Estimated Using Megan, Ecmwf Analyses and a Detailed Canopy Environment Model

The global emissions of isoprene are calculated at 0.5 • resolution for each year between 1995 and 2006, based on the MEGAN (Model of Emissions of Gases and Aerosols from Nature) version 2 model (Guenther et al., 2006) and a detailed multi-layer canopy environment model for the calculation of leaf temperature and visible radiation fluxes. The calculation is driven by meteorological fields – air temperature , cloud cover, downward solar irradiance, windspeed, volumetric soil moisture in 4 soil layers – provided by analyses of the European Centre for Medium-Range Weather Forecasts (ECMWF). The estimated annual global isoprene emission ranges between 374 Tg (in 1996) and 449 Tg (in 1998 and 2005), for an average of ca. 410 Tg/year over the whole period, i.e. about 30% less than the standard MEGAN estimate (Guenther et al., 2006). This difference is due, to a large extent, to the impact of the soil moisture stress factor, which is found here to decrease the global emissions by more than 20%. In qualitative agreement with past studies, high annual emissions are found to be generally associated with El Niño events. The emission inventory is evaluated against flux measurement campaigns at Harvard forest (Massachus-sets) and Tapajós in Amazonia, showing that the model can capture quite well the short-term variability of emissions, but that it fails to reproduce the observed seasonal variation at the tropical rainforest site, with largely overestimated wet season fluxes. The comparison of the HCHO vertical columns calculated by a chemistry and transport model (CTM) with HCHO Correspondence to: J.-F. Muller (jfm@aeronomie.be) distributions retrieved from space provides useful insights on tropical isoprene emissions. For example, the relatively low emissions calculated over Western Amazonia (compared to the corresponding estimates in the inventory of Guenther et al., 1995) are validated by the excellent agreement found between the CTM and HCHO data over this region. The pa-rameterized impact of the soil moisture stress on isoprene emissions is found to reduce the model/data bias over Aus-tralia, but it leads to underestimated emissions near the end of the dry season over subtropical Africa.

over the whole period, i.e. about 30% less than the standard MEGAN estimate (Guenther et al., 2006). This difference is due, to a large extent, to the impact of the soil moisture stress factor, which is found here to decrease the global emissions by more than 20%. In qualitative agreement with past studies, high annual emissions are found to be generally associated with El Niño events. The emission inventory is evaluated 15 against flux measurement campaigns at Harvard forest (Massachussets) and Tapajós in Amazonia, showing that the model can capture quite well the short-term variability of emissions, but that it fails to reproduce the observed seasonal variation at the tropical rainforest site, with largely overestimated wet season fluxes. The comparison of the HCHO vertical columns calculated by a chemistry and transport model (CTM) with

Introduction
The emissions of biogenic volatile organic compounds (BVOCs) have multiple impacts on the atmospheric composition, including enhanced ozone formation rates in polluted areas, a decreased oxidizing capacity of the global troposphere, and a substantial contribution to tropospheric aerosol abundances in continental regions (Seinfeld and Pan-5 dis, 1998). Among the BVOCs, isoprene is the most largely emitted compound, with global annual emissions on the order of 600 Tg/year (Guenther et al., 2006). Whereas fixed emission inventories have been widely used by global atmospheric chemistry and transport models (CTMs) in the last decade (e.g. Dentener et al., 2006), the importance of meteorology as source of spatiotemporal variability in BVOC emissions has led to 10 the implementation of interactive emission models in CTMs, which make use of the CTM meteorology for estimating the emissions (e.g. Pfister et al., 2007 1 ). It has been also shown that climate change can potentially induce large long-term term changes in global emissions (Sanderson et al., 2003;Guenther et al., 2006) and that meteorological variability, and in particular El Niño events, induce a significant interannual 15 variability of global emissions (Lathière et al., 2006).
Since the first global emission models (Müller, 1992;Guenther et al., 1995), which parameterized the emissions as functions of the instantaneous temperature and radiation levels, the influence of meteorology on the emissions has been seen from measurements to be more complex. Among other factors, the past environmental condi-20 tions (temperature, light) experienced by the leaves, the soil moisture stress, and the age of leaves have well-identified impacts on the emissions, even though their quantitative influence remains uncertain (see Guenther et al., 2006, and references therein). These effects are now parameterized in the MEGAN model (Model of Emissions of Gases and Aerosols from Nature) version 2 (Guenther et al., 2006). This model in-Introduction EGU corporates the results of numerous field and laboratory investigations, and includes a high resolution database for the distribution of plant functional types (PFTs) and of their basal emission factor (i.e. their emission rates in standard conditions), as described in Sect. 2.1. Although the leaf-level radiation fluxes and temperatures are the most important parameters driving the emissions, their parameterizations are generally 5 crude and/or poorly described in past studies of isoprene emissions and their impact on the atmosphere (Guenther et al., 1995;Sanderson et al., 2003;Lathière et al., 2006;Palmer et al., 2006).
where ε is the standard emission factor (mg m −2 h −1 ), i.e. the emission rate at standardized conditions defined in Guenther et al. (2006), and γ, the activity factor, represents the response to deviations from these standard conditions. ρ, which represents the influence of production and losses within the canopy, is taken equal to one in this study. We use the MEGAN EFv2.0 dataset (also used in Guenther et al., 2006), which provides the geographical distribution of both the fractional cover and the standard emission factor of six plant functional types (PFTs): needleleaf evergreen trees, needleleaf deciduous trees, broadleaf trees, shrubs, grass and crops. Here a further distinction between evergreen and deciduous broadleaf trees will be made, since these plant types have different canopy features. The emission flux at any location is therefore a sum of contributions from all PFTs present at this location. The activity factor γ is given by where C CE =0.52 is a factor adjusted so that γ=1 at standard conditions, γ P T is the weighted average (for all leaves) of the product of the activity factors for leaf temperature and PPFD (photosynthetic photon flux density), LAI is the leaf area density 5 (m −2 m −2 ), γ age and γ SM are the leaf age and soil moisture activity factors, respectively. Since leaf temperature and PPFD vary with height due to light attenuation by leaves, the canopy is divided into n layers in the canopy environment model which further EGU distinguishes between sunlit and shade leaves, so that where the index j runs over all layers, ∆LAI j is the partial LAI in layer j , γ j P and γ j T are the PPFD and leaf temperature activity factors at layer k (for either shade or sunlit 5 leaves), and f j sun and f j shade =1−f j sun are the fractional sunlit and shaded area in this layer. The number of layers is taken to eight in this study in order to minimize the numerical error associated to vertical discretization. The leaf area index is evenly distributed between the n layers, i.e. ∆LAI j =LAI/n. Note that γ P T has to be calculated separately for each PFT, because of differences in their canopy characteristics (see Table 1 in the supplement to this article: http://www.atmos-chem-phys-discuss.net/7/ 15373/2007/acpd-7-15373-2007-supplement.pdf).
The light dependence is given by where PPFD is calculated at leaf level (µmol m −2 s −1 ). α and C P depend on the past history of light intensity according to where P 24 and P 240 are the PPFD averages over the past 24 and 240 h, respectively, and P 0 is equal to 200 µmol m −2 s −1 for sunlit leaves and 50 µmol m −2 s −1 for shaded 15378 , where C T 1 =95 000 J mol −1 and C T 2 =230 000 J mol −1 , T l (K) is leaf temperature, R (=8.31 J K −1 mol −1 ) is the universal gas constant, E opt is the maximum normalized emission capacity, and T opt is the temperature at which E opt occurs. These coefficients are estimated as a function of the average leaf temperature over the past 24 h (T 24 ) and 240 h (T 240 ): with T opt = 313 + 0.6 · (T 240 − 297).
The leaf age activity factor γ age is estimated for deciduous canopies as where A new =0.05, A gro =0.6, A mat =1.125, A old =1, and F new , F gro , F mat and F old are the fractions of new, growing, mature and old leaves, respectively. These fractions are parameterized from LAI changes between the current and previous time steps and from the average temperature over the past 15 days, as described in Guenther et al. (2006). Finally, the emission response to soil moisture stress, γ SM , is estimated as

EGU
where f l root is the fraction of roots within the soil layer l , θ l is the volumetric soil water content in this layer (m 3 m −3 ), and θ w is the wilting point. The distribution of roots is estimated following Zeng (2001). Although this distribution is PFT-dependent, the use of a unique profile (26%, 39%, 29% and 6% at the 4 layers of the ECMWF numerical weather prediction model: 0.07 m, 0.21 m, 0.72 m and 1.89 m, respectively) is found to 5 cause negligible errors on the estimation of γ SM in most situations.

The canopy environment model
A canopy environmental model (MOHYCAN, for MOdel for Hydrocarbon emissions by the CANopy) (Wallens, 2004) is used to determine leaf temperature and the radiation fluxes as functions of height inside the canopy. Radiative transfer is based on the 10 framework of Goudriaan and van Laar (1994) (see also Leuning et al., 1995). Solar radiation is attenuated by foliage according to an exponential law, as described in more detail in the supplement (http://www.atmos-chem-phys-discuss.net/7/15373/ 2007/acpd-7-15373-2007-supplement.pdf). Distinct values of the extinction coefficient κ are used for direct and for diffuse light, as well as for visible and NIR (Near Infrared 15 Radiation). The leaves are characterized by diffusion and transmission coefficients.
The direct and diffuse fractions of solar radiation depend on solar zenith angle and cloud optical depth. The latter is estimated from the PPFD at canopy top, based on tabulated irradiances calculated by an atmospheric radiative transfer model (Madronich and Flocke, 1998). Leaf temperature in each canopy layer is determined from the energy balance equation (Goudriaan and van Laar, 1994;Leuning et al., 1995) where Q SW is the absorbed solar (shortwave) irradiation, Q LW is the net longwave radiation emitted/absorbed by the leaf, Q SH is the sensible heat flux, Q LH is the latent heat flux of evaporation, and Q storage is the energy storage change. Q storage is much smaller than the other terms, and can be neglected. EGU energy budget terms Q LW , Q SH and Q LH involves parameterizations of the resistances for the exchange of heat and water vapor, and is described in the supplement. In summary, the input variables of the model are values at canopy top of solar radiation (PPFD+NIR), including their diffuse and direct components, air temperature, relative humidity and windspeed. Air temperature and water vapor pressure are assumed 5 to be constant in the canopy. Attenuation of windspeed by foliage is parameterized, as described in the supplement. Based on these assumptions, the model calculates PPFD and NIR for sunlit and shaded leaves at each level. Leaf temperature is determined iteratively using Eq. (12). The number of required iterations is in general less than 4. 10

Meteorology and LAI dataset
We drive the canopy environment model with ECMWF fields for the downward solar radiation flux, the cloud cover fraction, the soil moisture content in 4 soil layers, and the air temperature, dewpoint temperature, and windspeed directly above the canopy. Reanalysed ERA40 fields are used till 2001, whereas operational analyses are used beyond 15 this date. The data are provided every 6 h on a N80 spectral grid (approximately 1.125 degree in longitude and latitude), and are re-gridded at 0.5×0.5 degree. A sinusoidal fit is applied to air and dewpoint temperature in order to derive hourly values of air temperature and relative humidity. The atmospheric radiative transfer model is used to determine the cloud optical depth from the cloud cover fraction and the solar radiation flux. 20 Hourly values for the diffuse and direct solar radiation fluxes in both clear and cloudy conditions are derived from the assumption of constant cloud cover and cloud optical depth in each 6-h interval. The ratio of PPFD to total solar radiation is taken from the ISCCP D2 dataset (Rossow et al., 1996, http://isccp.giss.nasa.gov/). NIR is assumed to account for the remainder of solar radiation, i.e. UV is neglected (Goudriaan and van 25 Laar, 1994 (Zhang et al., 2004). Monthly climatological LAI values derived from the same dataset are used before this period. As in Guenther et al. (2006), the LAI of vegetated areas is estimated by dividing the MODIS LAI by the 5 vegetated fraction of the grid.

Inventory for year 2003
The monthly averaged isoprene fluxes for January and July 2003 are illustrated on Fig. 1. The global annual isoprene source is estimated to 412 Tg/year in 2003, or about 10 30% less than in the estimations by Guenther et al. (1995) (the GEIA 1995 evaluation) and Guenther et al. (2006). The latter evaluation was based on the MEGAN algorithm and NCEP meteorological data. The datasets used for Leaf Area Index, the distributions of the plant functional types and their associated basal emission factors were identical in this study and in Guenther et al. (2006). Besides the use of NCEP, other 15 differences with the present work included the radiative transfer model, the calculation of leaf temperature, and the wilting point database.
Comparison of our Fig. 1 with the corresponding distributions of Fig. 10 in Guenther et al. (2006) shows a excellent agreement regarding the spatial patterns of the emissions in most regions, with the noticeable exception of Australia and other arid areas. 20 The annual emissions over Northern America calculated in this work are also in excellent agreement with the estimation by Palmer et al. (2006) based on MEGAN and NCEP data, i.e. they are about 10% lower than in the GEIA evaluation (boundaries are as in Fig. 2 in Palmer et al., 2006) when the soil moisture stress effect is neglected in the calculations, in accordance with Palmer et al. (2006). The largest source of tained by using the NCEP/NCAR reanalysis fields for soil moisture (data obtained from www.cdc.noaa.gov/cdc/data.ncep.reanalysis.html) (Kalnay et al., 1996), together with the wilting point used in this reanalysis (=0.1 m 3 m −3 ). In contrast with these results, the use of the wilting point database of Chen and Dudhia (2001) in Guenther et al. (2006) led to a comparatively smaller impact of this activity factor on the emissions 10 (7% globally). Although the high-resolution database of Chen and Dudhia (2001) is probably more realistic than the fixed values used by ECMWF and NCEP/NCAR, it is not appropriate for use in calculations using the soil moisture fields from these analyses, given the importance of the wilting point in the determination of soil moisture in climate models (Maurer et al., 2002;Li and Robock, 2005). As seen on Fig. 2, the 15 emission reduction calculated in this work is largest in subtropical Africa and Australia during the dry season and reaches one order of magnitude in desert areas. Annual North American isoprene emissions are reduced by ca. 10%, mostly due to decreases in the Western U.S.
Other causes might contribute to the lower global emissions estimated in this work, 20 compared to previous estimations. Wallens (2004) estimated that the treatment of light attenuation in the canopy used in the MOHYCAN model leads to lower emissions (10% globally) than the parameterization used in Guenther et al. (1995). As discussed by Guenther et al. (2006), the LAI values from the MODIS dataset are considerably lower than in previous estimations and contribute to lower the global emissions by >20%. 25 The diurnal cycle of temperature, not accounted for in Guenther et al. (1995), contributes to enhance the emissions, but this is compensated by the lower PPFD values from the meteorological analyses, compared with the PPFD fluxes used in Guenther et al. (1995). The use of leaf temperature instead of air temperature in the emission EGU algorithm contributes to increase the global (or the North American) annual emission estimate by 18% according to our calculations. The difference between leaf temperature (average weighted by the emissions) and air temperature is illustrated in Fig. 3.
Leaves are found to be about 1-2 K warmer than their environment over most forest areas, resulting in emission enhancements of ca. 10%. Over savannas and desert ar-5 eas, generally characterized by little cloud cover and high PPFD fluxes, the difference often exceeds 2 K, and leads to emission increases which can exceed 30%. Note that essentially identical results, but of opposite sign, are obtained for the correlation of the emissions with the SOI index. The emissions are positively correlated with ONI over many regions in South America, Africa, Siberia and Alaska, but they are negatively correlated with ONI over the U.S., Australia and many other regions. As a result, global isoprene emissions are not strongly correlated with the ONI (or SOI) in-10 dex. The correlation coefficient between ONI (SOI) and the monthly global emissions is 0.12 (−0.07), i.e. much less than in the studies of Lathière et al. (2006) and Naik et al. (2004). However, isoprene emissions are found to be positively correlated with the ONI delayed by 6 months in almost all regions, as seen on the right panel of Fig. 6. The correlation coefficient between the lagged ONI (SOI) and the monthly global emissions 15 reaches 0.38 (−0.32).

Comparison with campaign measurements
The inventory is tested against campaigns measurements at mid-latitudes (Harvard forest) and in Amazonia (Tapajós). EGU on a 30 m tower extending 7 m above the canopy. The forest is composed of red oak (a strong isoprene emitter) and other species. Needleleaf evergreen and broadleaf deciduous trees represent 35% and 64% of the site area, respectively (Goldstein et al., 1998), in good agreement with the PFT distribution used in MEGAN (63% and 67% of broadleaf trees at the two nearest gridcells). 5 Our model calculations are compared with the measurements in Fig. 7 and Fig. 8. Although the observed diurnal cycle is relatively well reproduced by the model, an underestimation is noted (35% on average), which probably reflects an underestimation of the standard emission factors in the model. The underestimation is highest around noon (40%), and lowest at high solar zenith angles.
When corrected for the 35% bias, the model results reproduce remarkably well the seasonal as well as the day-to-day variations of isoprene fluxes between June and mid-September (Fig. 8), with a correlation coefficient of about 0.90. Before day 160 and after day 260, however, the model largely overestimates the fluxes. The leaf age factor γ age calculated according to Eq. (10) lowers the emissions in spring and fall (as 15 compared to summertime), but this reduction appears to be much too weak, or the response of the emissions to LAI variations and the past weather conditions might be possibly underestimated.

Tapajós
Isoprene fluxes from a primary tropical rainforest in Brazil were measured during three 20 separate field campaigns: April 2001 during the wet season, July 2000 at the end of the wet season, and October-November 2003 during the dry season. The technique used to collect these datasets was the eddy covariance-fast isoprene system (EC-FIS) technique (Guenther and Hills, 1998). All the measurements were conducted at the Floresta Nacional do Tapajós site (2 • 51 ′ S, 54 • 58 ′ W) in the state of Pará run by 25 S. Wofsy's group from Harvard University. This long-term CO 2 flux tower was sponsored by the Large-scale Biosphere-atmosphere experiment in Amazonia (LBA). The July 2000 dataset has been previously reported (Rinne et al., 2002). passing inlet air through a heated plantium catalyst. Due to importation difficulties, no isoprene standard was available on-site in 2003, but the FIS was calibrated both before departure and upon return to the laboratory in the United States. Calibrations were performed by dilluting a high-concentration gas standard in 2001. Standard eddy covariance methodology was used to compute half-hour fluxes, but no corrections (e.g., 15 the Webb correction) were applied to the data except for a 2-D wind rotation to ensure a zero vertical velocity. The teflon tube introduced a 5-6 s delay between the datasets which was determined by examing the lag correlation for the half-hour periods. The daily averaged emission fluxes are shown on Fig. 9. The model results agree well with the dry season measurements (red diamonds) when the standard emission factor is reduced by a factor 1.7. The model succeds in reproducing the steep decrease (factor of 3) in the emission rates in the course of the measurement period, between day 300 and day 308. This decrease is due to rapid changes in meteorological conditions during that period. The modelled emissions during the wet season (February-July) are almost a factor of 2 lower than during the dry season, due to lower LAI (Huete et al., 2006), lower PPFD fluxes and lower temperatures during that time period. Although this seasonality is much more pronounced than in the inventory of Guenther et al. (1995) (with only 15% difference between April and September emissions at that site), the flux measurements at Tapajós indicate a even much stronger EGU seasonality of isoprene fluxes. This result reinforces conclusions already drawn by e.g. Kuhn et al. (2004), based on isoprene emission capacity measurements at another Amazonian site, and Trostdorf et al. (2004), based on ambient isoprene measurements at Tapajós in 2001. For example, the measured fluxes in April 2001 are almost an order of magnitude lower than the dry season fluxes. In other terms, the standard emission factor should be a factor of 2-5 lower during the wet season, compared to the dry season. This probably cannot be explained by soil moisture effects, since the soil moisture stress factor (γ SM ) is found to be always equal to one at this location, although it cannot be excluded that this parameterization is inappropriate for tropical rain forests. Trostdorf et al. (2004) have proposed to introduce a precipitation-based activity factor for isoprene emissions in order to better match the observations: where P 3 is the average precipitation rate during the past 3 months, P 3,max is the maximum value of this average. Using precipitation rates from the ECMWF/ERA40 dataset, this factor is found to reduce wet season fluxes by a factor of 1.5, compared to the dry season fluxes, and is therefore not sufficient to reconcile the model with observations. Alternative models relating the emissions not only to environmental parameters, but 5 also to physiological parameters like stomatal conductance, assimilation and intercellular CO 2 concentration are more likely to help improving the prediction of isoprene emissions in tropical rainforests (Simon et al., 2005).

Evaluation against formaldehyde data from a satellite
Isoprene being a major precursor of formaldehyde in the atmosphere, the vertical col-10 umn distributions of this compound obtained from satellite instruments provide the opportunity to test and possibly improve the emission inventories. The GEOS-CHEM tropospheric chemical/transport model (CTM) has been used in several studies by the 15388

EGU
Harvard group to provide improved estimates of isoprene emissions based on HCHO columns retrieved from the GOME (Global Ozone Monitoring Experiment) instrument, in particular over the United States (Palmer et al., 2003;Abbott et al., 2003;Palmer et al., 2006), over China and Southeast Asia (Fu et al., 2007), and on the global scale (Shim et al., 2005). In regions where isoprene is the dominant precursor of formaldehyde, like the Eastern U.S. during summertime, the estimated uncertainty on these emissions is ∼30% (Palmer et al., 2006), and is mainly related to uncertainties in the isoprene chemical mechanism. In tropical regions, the derivation of emissions from GOME data is made more difficult. This is to a large extent caused by the strong contribution of biomass burning to the observed HCHO signal, difficult to separate from the 10 biogenic VOC contribution, due to its large uncertainty and spatiotemporal variability.
In the global inverse modeling study of Shim et al. (2005), for example, the biomass burning source of non-methane organic compounds was increased by a factor of 2-4 in the optimization, which however failed to provide a satisfactory match between the modelled and observed HCHO distributions over Africa. 15 We use here formaldehyde columns retrieved from GOME at IASB-BIRA (De Smedt et al., 2007a). They differ from previous HCHO retrievals (e.g. Chance et al., 2000;Wittrock et al., 2000) by the choice of the wavelength interval used for DOAS (Differential Optical Absorption Spectroscopy) fitting, taken to be 328.5-346 nm. This choice improves the slant columns and decreases the fitting residuals in tropical regions, 20 compared with retrievals obtained with the usual fitting window (337. 5-359 nm). Slant columns are converted to vertical columns from detailed radiative transfer calculations and vertical profile shapes of formaldehyde concentrations taken from an updated version of the IMAGES model (Müller and Stavrakou, 2005). A more detailed description of the retrieval methodology is provided in De Smedt et al. (2007a, b 2 ). 25 The meteorological fields in IMAGES are obtained from ECMWF analyses for the 2 De Smedt, I., Van Roozendael, M., Müller, J.-F., Stavrakou, T., Eskes, H., and Van der A., R.: Ten years of tropospheric formaldehyde retrieval from GOME and SCIAMACHY, in preparation, 2007b EGU winds, convective fluxes, temperature, and water vapour. The chemical mechanism for isoprene degradation is adapted from the MIM mechanism (Pöschl et al., 2000), with a HCHO yield at high NO x about 20% higher than the corresonding GEOS-Chem yield, which was found to be consistent with aircraft observations over the United States (Millet et al., 2006). The biomass burning emissions are based on the GFED v1 inventory for burnt biomass (van der Werf et al., 2003) with emission factors of Andreae and Merlet (2001). The modelled HCHO columns between 1997 and 2001 are compared with the GOME retrievals on Fig. 10. The blue and red lines correspond to simulations using either GEIA or MEGAN, respectively. In all regions except Southern Africa, the MEGAN-based inventory brings the seasonal variation of the modelled columns closer to the observations. Over Northern Australia, the MEGAN emissions appear to be overestimated, although the excellent agreement regarding the seasonal variation might indicate a systematic bias in the model and/or the data, since biogenic emissions have a strong seasonality in this region (Fig. 1). The overestimation of HCHO columns is 15 worsened when the soil moisture stress activity factor is not considered in the determination of the emissions (dashed red line in Fig. 10). Over Northern Africa, the strongly reduced wet season (May-November) emissions from MEGAN compared to GEIA appear to be supported by the HCHO comparison. The wintertime discrepancies for this region are probably related to biomass burning, but the model appears to provide a 20 better match with the data at the end of the dry season (February-April) when the soil moisture activity factor is taken equal to one. Over Southern Africa, the use of MEGAN emissions leads to a general underestimation of HCHO columns by the model, except at the peak of the dry season, when fires are the dominant source of reactive hydrocarbons. Over Western Amazonia, where biomass burning emissions are generally low, 25 the lower isoprene emissions of the MEGAN-based inventory lead to a spectacular reduction of the model/data discrepancies, an improvement found at most locations in South America. At the model grid cells closest to the Tapajós forest site in the Pará province of Brazil, the model matches very well a the HCHO data, except in August-

Printer-friendly Version
Interactive Discussion EGU November 1997 when forest fires were most intense. This good agreement contradicts the analysis of the surface flux measurements discussed in Sect. 4.2, which suggested a large overestimation of isoprene fluxes at this location, in particular during the wet season. Possible explanations include the spatial variability of the emissions, and a poor representativity of the Tapajós site; the oxidation of other biogenic organic com- 5 pounds not accounted for in the model; and the possible existence of large biases in the budget of oxidants, most importantly OH, as indicated by recent findings from field campaigns in the Amazonian rainforest (Kubistin et al., 2007;Kuhn et al., 2007;Karl et al., 2007). 10 We have presented a global isoprene emission inventory covering the period 1995-2006, based on the MEGAN model. The general features of the emission distribution for the year 2003 are very consistent with the corresponding distribution calculated by Guenther et al. (2006), a logical result since the emission algorithm, but also the distributions used for LAI and the standard emission factors are adopted from this work.

Conclusions
However, the global annual emission calculated for 2003 is about 30% lower than in Guenther et al. (2006), to a great extent because of a stronger emission limitation due to drought calculated in our work in arid areas like Australia, subtropical Africa and the Western United States. Besides the direct impact of soil water stress on the emissions (through the γ SM activity factor of Eq. 11), drought also influences the emissions 20 through the stomatal resistances and the leaf temperatures. We calculate that the use of leaf (instead of air) temperature in the emission algorithm increases the global annual emission by almost 20%. Neglecting the soil moisture effect on the stomatal resistance calculation would not imply a large change, because the low relative humidities generally associated with drought conditions already lead to a large resistance 25 increase.
The interannual variability of isoprene emissions is found to be higher than in a pre-Introduction EGU vious study, with up to 20% difference between the global annual emissions of different years. This larger influence of meteorology on the emissions might be due to the ECMWF meteorological analyses adopted in our calculations and also to the dependence on past temperatures and radiation levels of the emissions in MEGAN. The highest annual global emissions are estimated for years following an El Niño event 5 (e.g. 1998 and 2005). More precisely, the emissions are positively correlated with the Oceanic Niño Index lagged by 6 months (correlation coefficient of 0.38). The influence of El Niño is significant in both the Tropics and the higher latitudes.
Comparisons with tower flux measurements at a mid-latitude forest site and in the Amazonian rain forest show the ability of the model to reproduce the short-term vari-10 ations in isoprene emissions. Long-term variations are not so well reproduced, as illustrated by the strong overestimation of the modelled fluxes during the wet season (in April and July) at Tapajós. The average model/data biases at Harvard forest during the summer (underestimation by factor 1.35) and at Tapajós in the dry season (overestimation by factor 1.7) might be indications that the standard emission rates used 15 in MEGAN are inappropriate at these locations; however, the representativity of these sites for larger-scale flux estimations might be limited (e.g. Karl et al., 2007). Further measurements are obviously needed to better ascertain the spatiotemporal variability of the emissions, especially over tropical rainforests. Satellite measurements of formaldehyde, a major isoprene degradation by-product, might prove to be very useful 20 for better constraining the emissions and their variability, as illustrated by comparisons of GOME vertical columns with global models over the United States (Palmer et al., 2006), over Southeast Asia (Fu et al., 2007), or over other regions like Africa, South America and Australia (Fig. 10). Further work will be essential in order to improve the CTMs, e.g. regarding the chemical mechanism in low-NO x conditions, the emissions 25 and chemistry of other biogenic NMVOCs, and the emissions and chemistry of compounds released by vegetation fires, which also contribute to the total HCHO signal observed from the satellites. Synergies should be also developed for a better integration of surface (or aircraft) campaign measurements in conjunction with analyses using and monoterpene emissions from Amazonian tree species with physiological and environmental parameters using a neural network approach, Plant Cell Environ., 28, 287-301, 2005. 15388 15396 Introduction