A global model of natural volatile organic compound emissions

. Numerical assessments of global air quality and potential changes in atmospheric chemical constituents require estimates of the surface fluxes of a variety of trace gas species. We have developed a global model to estimate emissions of volatile organic compounds from natural sources (NVOC). Methane is not considered here and has been reviewed in detail elsewhere. The model has a highly resolved spatial grid (0.5øx 0.5 ølatitude/longitude) and generates hourly average emission estimates. Chemical species are grouped into four categories: isoprene, monoterpenes, other reactive VOC (ORVOC), and other VOC (OVOC). NVOC emissions from oceans are estimated as a function of geophysical variables from a general circulation model and ocean color satellite data. Emissions from plant foliage are estimated from ecosystem specific biomass and emission factors and algorithms describing light and temperature dependence of NVOC emissions. Foliar density estimates are based on climatic variables and satellite data. Temporal variations in the model are driven by monthly estimates of biomass and temperature and hourly light estimates. The annual global VOC flux is estimated to be 1150 Tg C, composed of 44% isoprene, 11% monoterpenes, 22.5% other reactive VOC, and 22.5% other VOC. Large uncertainties exist for each of these estimates and particularly for compounds other than isoprene and monoterpenes. Tropical woodlands (rain forest, seasonal, drought-deciduous, and savanna) contribute about half of all global natural VOC emissions. Croplands, shrublands and other woodlands contribute 10-20% apiece. Isoprene emissions calculated for temperate regions are as much as a factor of 5 higher than previous estimates.

decomposition of organic material, geological hydrocarbon reservoirs, plant foliage and woody material. In addition, there are human influenced natural sources from harvesting or burning plant material. We have estimated emissions of VOC only from oceans and plant foliage. VOC emissions from other sources are very uncertain but probably represent less than a few percent of total global emissions [Zimmerman, 1979;Lamb et al, 1987;Janson, 1992;Eichstaedter et al., 1992]. Further investigation is needed to verify our assumption that these sources are negligible. Section 2.1 describes the methods used to estimate ocean emissions of VOC. The procedures used to estimate foliar emissions are outlined in section 2.2.
We have grouped natural VOC into four categories: isoprene, monoterpenes, other reactive VOC (ORVOC), and other VOC (OVOC). Examples are given for each category in Figure 1. ORVOC are herein defined as compounds with a lifetime, under typical tropospheric conditions, of less than 1 day, while OVOC have lifetimes greater than 1 day. All data sets have been merged into a common grid system with a resolution of 0.5 ø x 0.5 ø latitude/longitude. Hourly emission rates are estimated for one 24-hour period during each month. The daily total emission estimated from the 24-hour data is extrapolated to a monthly emission estimate. The model allows a great deal of flexibility and modularity so that improved data sets and algorithms can easily be incorporated into future versions of the model.

Ocean Emission of Volatile Organic Compounds
The ocean is supersaturated with VOC with respect to the atmosphere [Frank et al., 1970;Lamontagne et al., 1974].
Because of this imbalance, the ocean is considered a source of these highly reactive compounds to the atmosphere [Broadgate et al., 1994;Plass-Duelmer et al., 1994]. There is experimental evidence that VOC arise from the "photochemical lability" of dissolved organic matter (DOM) in the surface ocean [Ratte et al., 1994], and there appears to be a relationship between the presence of chlorophyll and the lability of DOM in a parcel of surface ocean water [Plass-Duelmer et al., 1993]. Satellitederived ocean color data sets are interpreted as being a proxy for the amount of photochemically labile organic matter, in the form of dissolved organic carbon (DOC), in surface ocean waters containing varying amounts of biological activity. Recent efforts at modeling the air-sea fluxes of trace gases on global grids using general circulation models and Coastal Zone Color Scanner, CZCS, satellite data appear to be promising and these techniques are used here to refine estimates of the global marine source of VOC.
The basic concept of photochemical lability is that higher chlorophyll content of a parcel of water, as sensed by the satellite, is related to a higher production of VOC per unit photon impinging upon the surface ocean. The high spatial variability, or "patchiness," of the chlorophyll in the surface ocean may also explain the high variability of the observations of VOC in the surface ocean. Ratte et al. [1994] found a linear relationship between ethene concentration and DOC per unit photon in seawater. To the extent that photochemical lability covaries with DOC, this finding supports our approach. We compute the global surface ocean concentration of VOC, the global transfer velocity field for VOC, and the global flux field for VOC via a technique similar to that described by Erickson and Eaton [ 1993]. It should be noted that this is a first attempt at creating a global, high-resolution oceanic source term for the emission of VOC to the atmosphere and our estimated errors are at least a factor of 3. We have attempted to constrain the computed surface ocean VOC concentration with the sparse available data and appeal for further measurements.
The air-sea exchange of trace gases is modeled using the standard formula [Liss and Merlivat, 1986]. We calculate the  smaller than Cso for most oceanic regions [Lamontagne et at, 1974;Bonsang et al., 1988;Donahue and Prinn, 1990]. We have selected a transfer velocity formulation based on a stability dependent theory of air-sea gas exchange [Erickson, 1993].
This method of calculating k w gives a global area-weighted transfer velocity for CO 2 of -20 cm h -1, consistent with the 14C inventory estimates [Broecker and Peng, 1974]. Because of a lack of experimental data on the diffusivity of various VOC, we base our estimate of k w for VOC on the diffusivity determined for CO 2. This assumption introduces at least a 50% uncertainty into the calculation.
The theoretical potential of a water parcel to produce VOC is assumed to increase with the amount of fresh, labile organic matter associated with biological activity in the surface ocean as sensed by the satellite. We compute the Crnax term via the simple relationship where D is foliar density (kg dry matter m-2), e is an ecosystem dependent emission factor (gg C m -2 h -1 at a photosynthetically active radiation (PAR) flux of 1000 gmol m -2 s -1 and leaf temperature of 303.15 K), and 7 is a non-dimensional activity adjustment factor that accounts for the influence of PAR and leaf temperature. The methods used to calculate each of these variables are described in this section.
2.2.1. Geographically based input data. Global gridded data sets of ecosystem type, global vegetation indices (GV1), precipitation, temperature, and cloudiness provide the inputs needed to estimate foliar emissions. We have used the data compiled by Olson [1992] to classify ecosystem types. Olson [ 1992] Equations (5a) and (5b) are used to calculate temperature and precipitation limited NPP for each grid cell, and the minimum value is selected as the actual NPP, reflecting the fact that one factor, precipitation or temperature, is the limiting factor in the grid cell. Using the temperature and precipitation data of Leeroans and Cramer [1992], we estimate an annual NPP of 122 Gt (1015 g) dry matter, which is equivalent [Lieth, 1975] Table 1. Temperature and precipitation data from a nearby grid cell were used for the 481 terrestrial grids in the Leemans and Cramer database that are missing temperature and precipitation data. The peak foliar density is the maximum monthly average foliar density that occurs during the year and is estimated as where Dr, is an ecosystem dependent empirical coefficient which we estimated using the data reported by Box [1981]. Table 1 Table 1).

Emission factors. Foliar emissions of individual
VOC range from undetectable to more than 100 gg C g-• h -• for different plant species . A few general rules of foliar VOC emission can be stated (e.g., conifer trees tend to emit monoterpenes), and emissions from a number of plant species have been characterized, but the assignment of emission factors to most ecosystems is limited by a lack of emission rate measurements.
There are two approaches to assigning emission factors to ecosystems. One approach is to quantify the species composition within an ecosystem type, assign an emission rate to each species, and aggregate the resulting emissions from each species. The second approach is to assign an emission rate directly to the ecosystem type and bypass the need for estimates of species composition. The first approach is particularly effective for an area with a low species diversity (e.g., cultivated land), whereas the second approach is best for areas with high species diversity (e.g., tropical forests). Enclosure measurement techniques provide the information needed to define emission rates for individual plant species, while area-averaged flux measurements can provide the information needed to directly assign emission rates to an ecosystem type. Field investigations that use both approaches provide a check on estimates of emission factors.

We have reviewed 22 field studies of NVOC emission.
These studies were conducted in 12 nations at sites that represent 26 of the Olson [1992] ecosystem types. From the fluxes reported in these studies, we assigned the isoprene and monoterpene emission rate factors, e, shown in Table 1 This relationship appears to be valid for a variety of monoterpene compounds and plant species . There is evidence that monoterpene emission rates from some plants are sensitive to light intensity [Steinbrecher, 1989] but this process has not been described by a numerical model. The light and temperature dependencies of ORVOC and OVOC emission rates are currently unknown.

Canopy radiative transfer model.
We have simulated the variability in solar radiation fluxes across the Earth's surface and within a vegetation canopy using simple models that account for a majority of observed variation. The astronomical routines described by Iqbal [1983] are used to compute hourly solar elevation angles and above canopy direct and diffuse PAR. The incoming solar radiation is modified for the effects of monthly average cloud cover [Hostlag and Van Ulden, 1983].
The effects of canopy shading are determined with the canopy radiative transfer model described by Norman [1982].  Emissions have been summed for each of the ecosystem types in the Olson database and are shown in Table 1

southeast Asia, and tropical woods of South America and
Africa. January maximum monoterpene and ORVOC emissions are predicted for tropical woods in South America and Africa. As expected, emissions of all NVOC compounds are relatively higher in woodland areas. This result is due to higher foliar densities and base emission rate factors. In addition, higher emissions are associated with areas of higher temperatures. Differences in the spatial distributions predicted for isoprene, monoterpenes, and other VOC are primarily due to differences in the emission factors assigned to various ecosystems (Table 1).

Emission Rate Comparisons
Estimates of global NVOC emissions are compiled in Table  4

1991; Turner et al., 1991; Mueller, 1992] used higher resolution climate and land use databases (monthly to seasonal and 1-to 5-degree grid cells). Our estimated isoprene emission rate of 503 Tg C yrl is slightly higher than the highest previous estimates, which range from 175 to 450 Tg C yr -1 The estimated monoterpene emission rate of 127 Tg C yr -1 falls just below the lower end of previously reported values (143 to 480
Tg C yr-1). Several previous global isoprene and monoterpene emission rate estimates were based entirely or primarily on the field measurements reported by Zimmerman [1979]. The higher isoprene and lower monoterpene emission rates reported by the present study are due primarily to differences in base emission factors that are based on the results of the 21 studies listed in Table 1. The isoprene emission factors for temperate regions in our model are as much as a factor of 5 higher than previous estimates. Emission rate factors for tropical areas are higher than those used in previous efforts, but this is offset by our use of a canopy light extinction model. The global totals are similar because tropical regions dominate total emissions. Table 5. Table   Table 3

. Global Area-Weighted Mean Ocean VOC Flux Equation Values
January  1) but assumed that some Northern Australia landscapes, e.g., croplands and barren desert, had lower rates. Table 5 indicates agreement between regional and global model estimates when similar emission factors are used, e.g., for monoterpene emissions from Sweden and isoprene emissions from northern Australia. Our isoprene emission estimates are higher than previous estimates for the United States and for European countries. Our isoprene estimates for the United States are 50% higher than the estimate of Zimmerman [1979]    Estimates are in 109 g C yr -1. RVOC includes isoprene, monoterpenes and other VOC with a lifetime of less than 1 day (ORVOC).

Molnar, 1990] are based on the Lamb et al. [1987] analysis of the field measurements reported by Zimmerman [1979]. A review of recent measurements and a revision of the Zimmerman
[1979] data by Guenther et al. [1994] suggests that the previous isoprene emission rate factors may be as much as a factor of 5 too low for U.S. woodlands.
Although there is a considerable range in estimates of NVOC emission, there is general agreement that the global emission rate is at least 400 Tg C yr 'l. Our NVOC flux estimate of 1150 Tg C yr -1 is more than a factor of 7 greater than estimated global anthropogenic VOC emissions [Mueller, 1992] and is more than a factor of 2 greater than estimated annual methane emissions [Taylor et at., 1991 and temperature, but which include humidity, CO 2 concentration, stomatal conductance, leaf development, time of day, season, and environmental stresses. The relative importance of these factors has been investigated for monoterpene and isoprene emission in a few plant species but is not well understood The importance of many of these factors is not known at all for many NVOC. Some factors, e.g., the relationship between CO 2 concentration and NVOC emission, may have a negligible impact on current estimates of natural VOC emissions but may be important for estimating future emission scenarios. The sensitivity of photochemical models to natural VOC emission rate estimates, reviewed in section 6, indicates a need for further reductions in natural VOC emission rate uncertainties. In this section, factors contributing to current uncertainties are discussed, and future research priorities are outlined.

Ocean VOC Emissions
The surface ocean VOC concentrations used in our model are within the range of the existing observations, but these data are sparse. Since fluxes are always calculated, either with modeled or observed Cso values, the observed fluxes have considerable uncertainty. Improved estimates of VOC emissions from oceans may be limited by our understanding of the sources of these compounds. VOC in oceanic waters are thought to be produced by phytoplankton and abiotically by the oxidation of planktonically derived polyunsaturated lipids. Wilson et al. [1970] found in laboratory experiments that dissolved organic matter produced by diatoms, a common marine phytoplankton in temperate neritic waters, could produce ethene and propene if illuminated. They also found that if living cells were present, production was enhanced, and ethane and propane were also found, though in lesser amounts than the alkenes. Clearly, more observations of VOC in the surface ocean are needed to adequately constrain the computed values presented here.

Canopy Deposition
Tropospheric chemistry and transport models require estimates of surface emission and deposition fluxes. Most terrestrial surfaces are covered by a vegetation canopy, and flux estimates should represent the flux across the boundary between the top of the canopy and the bottom of the atmospheric boundary layer. Since many of our emission factors are based on leaf and branch enclosure measurements, emissions into the atmospheric boundary layer will be overestimated if chemical and deposition losses within the canopy are significant and unaccounted for. Flux estimates based on above-canopy concentration gradients agree reasonably well with enclosure measurements, indicating that most VOC escape into the atmosphere above the canopy [Lamb et aL, 1986]. The amount of VOC deposited on ground and canopy surfaces is not well known but could be significant. Reactions with OH could produce compounds that are quickly deposited on canopy surfaces. This probably has a small impact on daytime fluxes, since the time scale for turbulent diffusion in the canopy, <100 s [LeClerc and Shaw, 1988], is much less than the typical tropospheric lifetime, >1000 s [Atkinson, 1990], of most VOC. Under stable nighttime conditions, the time scale for diffusion in the canopy could be much larger. Field estimates of withincanopy deposition factors are needed so that this loss can be accounted for in VOC surface emission models.

Ecological Modeling
Uncertainties in foliar density and species composition estimates are a significant component of the overall uncertainties in emissions from land surfaces. VOC emission estimates for the United States and Europe have been improved by using the detailed foliar density and species composition estimates described by Geron et al. [1994] and D. Simpson et al. [1995]. Detailed vegetation inventories for many regions are currently unavailable. Existing global ecosystem distribution estimates include static databases [e.g., Olson, 1992] and estimates based on dynamic models [e.g., Bergengren and Thompson, 1994].
The estimates of Olson [ 1992] are based on a great deal of effort to validate individual grids and appear to provide the best means for estimating 1990 emissions. The incorporation of an accurate dynamic model of ecosystem distributions would be a significant improvement to our global VOC modeling procedures, because it provides a capability for investigating potential changes in emissions due to climate, succession, disturbance, and land use related changes in ecosystem distributions.
Foliar VOC emission rate modeling efforts could also be improved by coupling the existing model with process-based models of global carbon and nitrogen dynamics. Ecosystem models of carbon and nitrogen fluxes and pool sizes have recently been extended to global scales by Melillo et al. [1993]. Some of these monthly variables (e.g., gross primary productivity, carbon, and nitrogen pool sizes) may be closely related to VOC fluxes although these relationships have not yet been quantified.  and Fall, 1993a]. If these measurements are confirmed, the global emission of this compound alone could be more than 100 Tg C. In addition, a wide variety of alkane, alkene, aromatic, sesquiterpene, alcohol, aidehyde, ester, ketone, and ether compounds are emitted from plants in at least trace amounts [Zimmerman, 1979;Winer et al., 1992].

Other
Further field measurements of ORVOC and OVOC in a variety of ecosystems are needed to establish these flux rates. Earlier reports of emissions of oxygenated VOC from plants are summarized by Fehsenfeld et al. [1992]. The biochemical and physiological mechanisms controlling the emission of these compounds is almost completely unknown at present.

Isoprene and Monoterpenes
Uncertainties in isoprene and monoterpene emission factors and the influence of light, temperature, and humidity are discussed in sections 5.5.1 and 5.5.2. Factors which may play an important role in regulating isoprene and monoterpene emissions but are not addressed in our current modeling procedures are discussed in the rest of this section. 5.5.1. Ecosystem-average emission factors. The major difference between the annual isoprene and monoterpene emission rates estimated by this study and the results of previous efforts are the emission factors assigned to various ecosystem types. Many of the emission factors compiled in Table 1 [1992], although isoprene emission occurs via stomatal pores in leaves, the emission rate is generally a function of isoprene synthesis rate and not stomatal conductance. Modeling of isoprene emission from leaves can be simplified by ignoring stomatal conductance except that this is an important parameter in determining leaf temperature. One area of uncertainty is the light dependence of monoterpene emission. Monoterpene emissions are light dependent for some plants [Steinbrecher, 1989] but not for others [Guenther et al., 1991]. Lerdau, [1994] found that first-year needles of douglas fir (Pseudotsuga menziesii) have light-dependent monoterpene emission rates, whereas older needles do not. It may be that light has an effect only when plants are actively synthesizing monoterpenes. [Dement et al., 1975]. Correlations between monoterpene emission and foliar moisture have also been reported [Lamb et al., 1985]. 5.5.3 Plant development and growth environment. Plant developmental processes influence isoprene and monoterpene emissions in a variety of ways. Yokouchi and Ambe [1984] report that monoterpene emission rates vary with season but the mechanism responsible was not identified. A number of studies have found that young leaves tend to have much lower or no isoprene emissions [Guenther et al, 1991;Grinspoon et al., 1991]. Kuzma and Fall [1993] have shown that this is due to a lack of isoprene synthase activity, and that increased isoprene emission in older leaves is associated with increased levels of this enzyme. Growth environment also plays an important role in determining how quickly new foliage will begin to emit isoprene at significant levels. Monson et al. [1994] found that aspen leaves began to emit isoprene after cumulative daily maximum temperatures above 0øC reach approximately 400 degree-days. $harkey and    Fall, 1993b]. These types of events could significantly alter VOC emission rates on regional scales and should be further investigated. 5.5.4. Nutrient, water, and injury status. Lerdau et al. [ 1994] have shown that monoterpene emissions from ponderosa pine are correlated with needle monoterpene concentrations. Since needle monoterpene concentrations are influenced by nitrogen availability, this may be an important control over monoterpene emissions. Steinbrecher [1989] observed significant differences in the pattern of emitted monoterpenes and the needle monoterpene concentrations for Norway spruce, suggesting that this relationship may not be straightforward. Harley et al. [ 1994] observed increased isoprene emissions with increased nitrogen availability. Sharkey and Loreto [1993] found dramatically increased isoprene emissions from plants subjected to water stress. This could have a significant impact on regions undergoing drought. Ayers and Gillett [ 1988] found that isoprene emission was much higher during the wet season in tropical Australia. This may have been due to the increased biomass during this period rather than the increased plant water status. Long-term water stress leads to increased monoterpene emission from cypress trees [Yani et al., 1993]. [Lewinsohn et al., 1991] and decreases in isoprene emissions from certain vines . The effects of wounding on monoterpene emission rates have not been quantified, but it is well known that physical leaf disturbance and wounds in monoterpene emitters lead to a large short-term increase in emissions [Zimmerman, 1979]. Effects on isoprene emission after wounding are related to transmissible wound signals ].

Emissions
Tropospheric photochemical model results are sensitive to VOC emission estimates and indicate that current uncertainties in NVOC estimates should be reduced. In addition, global chemical and transport models provide a means of testing VOC flux estimates by comparing modeled and observed concentrations of chemical species, such as carbon monoxide, that are dependent on VOC fluxes.
Concentrations of ozone and its precursors over North America were investigated by Jacob et al. [ 1993], using a threedimensional, continental-scale photochemical model. They found that a doubling of isoprene emissions resulted in less than a 4-ppb increase in mean 0 3 concentration anywhere in their model domain. This is to be expected for any region where 03 production is NO x limited. Shutting off isoprene emissions completely in the model resulted in 5 to 15 ppb decreases of 03 over most of the eastern United States. With a relatively low isoprene emission rate, less than half of the amount estimated by our global model, Jacob et al. [1993] found that isoprene contributed 27% of the total source of CO over North America. Hough and Johnson [1991] examined the budgets of photochemical oxidants on a global scale, using a zonally averaged two-dimensional chemistry and transport model. They found that a 20% reduction in annual isoprene emissions, from 450 to 360 Tg, resulted in changes in global average concentrations of +3.6% for OH, 0% for 0 3 , -3.0% for peroxyacetyl nitrate (PAN), -2.2% for H 202, and -9.3% for CH3COO2H. A 20% reduction in annual monoterpene emissions, from 550 to 440 Tg, resulted in changes in global average concentrations of +0.5% for OH, 0% for 03 , -0.6% for PAN, -0.4% for H20 2, and -0.4% for CH3COO2H. The strong seasonality and distinct spatial distribution of natural VOC should result in detectable changes in the atmospheric CO signal attributable to this source. A combination of global atmospheric modeling and satellite data could provide a means for validating VOC emission estimates. Accurate estimates of CO emissions from global biomass burning would also be needed for this analysis.

Conclusions
The NVOC emission rate estimates described in this paper are our current best estimates for use in 3-D global computer models. Estimated isoprene emissions in temperate regions are considerably higher than previous estimates, but the global totals are similar, since they are dominated by emissions in the tropics. Isoprene and monoterpene emissions are estimated to contribute 57% and 14%, respectively, of the total reactive VOC flux and are primarily emitted from woodlands. About half the total global VOC flux is estimated to be from compounds other than isoprene and monoterpenes. Ocean emission estimates are considerably lower than most previous estimates but can still play an important role in the remote marine boundary layer.
The model described here has been used to generate an inventory of estimates for 1990. These data are available in digital format from the IGAC-GEIA archive. In addition to the IGAC-GEIA inventory, the emission model components have been incorporated into regional and global 3-D chemistry and transport models and are being used to investigate the interactions between global change and trace gas biogeochemistry. One of the most critical aspects of creating these highresolution, global estimates of trace gas fluxes is to emphasize the errors and need for an enlarged observational database to check these model results.
Uncertainties associated with isoprene and monoterpene emissions in some temperate regions are at least a factor of 3. Fluxes of isoprene and monoterpenes in tropical regions and fluxes of other VOC in all regions have even higher uncertainties. Field measurements of regional VOC fluxes from surfaces and vegetation types where few or no data exist will provide some sorely needed constraints on calculations such as this. A better understanding of the processes controlling NVOC emission will also lead to improvements in existing model algorithms.