An estimate of natural volatile organic compound emissions from vegetation since the last glacial maximum

Importance of this paper: Organic chemicals evaporating out of vegetation may a(cid:128)ect various aspects of the earth (cid:213) s environment – including the amount of haze in the air, which a(cid:128)ects the earth (cid:213) s heat balance by reﬂecting sunlight back into space, and also the lifetime of methane gas in the atmosphere. Based on reconstructed vegetation zones for the coldest phase of the Last Glacial Period (about 20,000 years ago) and the warmer present interglacial (the last 11,000 years), we estimated the total amount of the substances evaporating from the world’s vegetation. In the cooler, drier, less forested world of the Last Glacial, the ﬂuxes of these compounds were probably reduced by about one-third to one-half. The resulting reduction in haze might have tended to keep the earth slightly warmer during the Glacial than it would otherwise have been. In addition, Glacial methane levels may have been lowered because of the lack of volatile organics (cid:212) mopping up (cid:213) reactive free radicals that would otherwise break down methane. Abstract The ﬂux of volatile organic chemicals from natural vegetation inﬂuences various atmospheric properties including oxidation state of the troposphere via the hydroxyl radical (OH), photochemical haze production and the concentration of greenhouse gases (CH 4 , H 2 O, CO). Because the Volatile Organic Compound (VOC) ﬂux Increased VOC emissions due to forest expansion and increased tropical temperatures since the LGM may have served as a signiﬁcant driver of climate change over the last 15 ka y through the inﬂuence of VOC oxidation; this can impact tropospheric radiative balance through reductions in the concentration of OH, increasing the concentration of CH 4 .The error limits on past VOC emission estimates are large, given the uncertainties of present-day VOC emission rates, paleoecosystem distribution, tropical paleoclimatic conditions, and physiological assumptions regarding controls over VOC emission. Nevertheless, the potential signiﬁcance of changes in natural VOC emission over the last 20 ka and their inﬂuence on climate are an important unknown that should at least be borne in mind as a limit on the understanding of past atmospheric conditions. Elucidation of the role of VOCs in climate change through paleoclimatic general circulation model simulations may improve understanding of past and future changes in climate. (cid:211) 2000 Elsevier Science Ltd. All rights reserved.

Increased VOC emissions due to forest expansion and increased tropical temperatures since the LGM may have served as a signi®cant driver of climate change over the last 15 ka y through the in¯uence of VOC oxidation; this can impact tropospheric radiative balance through reductions in the concentration of OH, increasing the concentration of CH 4 .
The error limits on past VOC emission estimates are large, given the uncertainties of present-day VOC emission rates, paleoecosystem distribution, tropical paleoclimatic conditions, and physiological assumptions regarding controls over VOC emission. Nevertheless, the potential signi®cance of changes in natural VOC emission over the last 20 ka and their in¯uence on climate are an important unknown that should at least be borne in mind as a limit on the understanding of past atmospheric conditions. Elucidation of the role of VOCs in climate change through paleoclimatic general circulation model simulations may improve understanding of past and future changes in climate. Ó 2000 Elsevier Science Ltd. All rights reserved.

Introduction
Volatile Organic Compounds (VOCs) play an important role in atmospheric chemistry by in¯uencing the observed tropospheric concentrations of ozone and methane. Methane is of particular interest as a signi®cant greenhouse gas which has 10 times the global warming potential of CO 2 on a mass basis (Lashof and Ahuja, 1990); it may also be critical in determining the changes in the earth's radiative balance in both the past and the future. At low VOC concentration, methane is a dominant sink for OH and results in low atmospheric CH 4 concentrations ($ 1 ppm) and a half-life of decades (Prather, 1994). However, at greater concentrations of VOCs that are more reactive towards OH than methane, the concentration and lifetime of CH 4 in the atmosphere might increase (Fehsenfeld, 1992) facilitating warmer global temperatures. From ice cores, methane concentrations are known to have varied by at least 50% during the upper Quaternary climate cycles (Chappellaz et al., 1993), and it is possible that changing VOC emission during that period in¯uenced global temperatures mediated in part by the atmospheric CH 4 concentration. Volatile organic compounds may also decrease regional and global temperatures through the production of atmospheric aerosols that increase atmospheric albedo.
The eect of natural aerosols (terrestrial and oceanic) on the global energy balance in the present-day world is poorly understood, but their cooling eect may well exceed the warming produced by a doubling of the preanthropogenic CO 2 concentration (Charlson et al., 1992;Schwartz, 1996). In spite of the uncertainty regarding the impact of changes in VOC emission on global radiative forcing, it is critical that the process is better understood, in order to predict the drivers of global climate change.
The actual present-day rates of VOC emission from vegetation are still subject to considerable uncertainty, but recently published inventories have been used to make global and continental scale estimates for the present-day world (Geron et al., 1994;Guenther et al., 1995) and projections for future conditions (Constable et al., 1999a). VOCs emitted from vegetation encompass a wide range of compounds originating from the mevalonic acid pathway (including isoprene and monoterpenes) that dier in size, oxygenation and reactivity. The rate at which these compounds are emitted varies according to both the biological properties of the vegetation and the physical environment. Principal biological controls include the ecosystem type, species composition, and canopy structure (Monson et al., 1995). The major physical controls over emission rate are temperature and light intensity.
Simulation of past climates and of past biogeochemical processes is a major challenge for earth system science. The main goal is to better understand why changes in climate have occurred in the past and elucidate the potential driving forces behind future changes. At present, most work in paleoclimatic and biogeochemical modeling is concerned with the global-scalē uctuations in climate which have occurred repeatedly over the last 2.4 million years of the Quaternary period (Crowley and North, 1991). During the last 700,000 years, climate has¯uctuated on a roughly 100,000 year timescale between cool, dry interglacial periods and warmer, more moist interglacials. Although the earth's orbital patterns are known to set the timing of these changes (Crowley and North, 1991), the amplifying factors involved are unclear. Important additional factors in these climate¯uctuations are thought to include changes in ocean plankton ecology, the spatial distribution of terrestrial ecosystems, dust¯uxes from desert surfaces, carbon uptake and release from soil and vegetation, atmospheric water vapor concentration, meth-ane¯uxes from wetlands and variation in¯uxes of VOC from terrestrial vegetation.
At present, global climate models do not predict certain important features of the glacial world, including the degree of aridity and the apparent magnitude of land surface cooling; a limitation that may be due to the omission of key parameters from these models. It is our intention here to explore how global VOC emissions from vegetation may have varied during part of the current glacial±interglacial cycle as a step towards improving understanding of the drivers of paleoclimatological conditions. The goal of this paper is to estimate from a priori principles variation in the¯uxes of three classes of VOCs; isoprene, monoterpenes and other reactive VOCs (ORVOC, reactive VOC with atmospheric lifetimes <1 day) between the extreme of the Last Glacial Maximum (LGM), the early-Holocene ($ 8 ka { thousands of years} BP {before present, in radiocarbon years} and the present Holocene interglacial, while accounting for uncertainty in vegetation distribution and temperature patterns.

Materials and methods
There are two clearly identi®able factors which may have caused large changes in VOC emission between glacial and interglacial conditions. Firstly, alteration in vegetation distribution and composition are well documented for all parts of the world (Adams et al., 1990;Frenzel, 1992;Crowley, 1995). As dierent vegetation classes exhibit a wide range of VOC emission rates (e.g., isoprene varies over 200-fold on a land area basis), shifts in the relative area coverage of the dierent classes likely resulted in an alteration of the potential of vegetation to emit VOCs into the atmosphere (Constable et al., 1991;Turner et al., 1991;Guenther et al., 1993;1999a,b;Martin and Guenther, 1995). Secondly, changes in temperature greatly in¯uence VOC emission (Guenther et al., 1993). Therefore, shifts in annual temperature conditions over a glacial±interglacial cycle could similarly aect VOC emission.
The importance of these two factors in controling VOC emission are essential for accurate emission estimates, however, reconstructions of late Quaternary environments are always problematical with respect to`no analog' conditions (dierent in possibly signi®cant ways from present-day conditions) during the LGM: Perhaps importantly, (i) CO2 concentration was about 80 ppm lower during the LGM than during the Holocene (Alley et al., 1993), but the eects of this change on VOC emission is unknown and (ii) there were dierences in relative abundance of potentially key VOC emitting species that may modify the rate of VOC emission from certain ecosystems. The possible eects of these factors are not explicitly addressed here in this study, but their potential to modify the estimates of VOC emission presented here may be signi®cant and should be borne in mind as a topic for future investigation.

Reconstruction of past vegetation±climate zones
The uncertainty regarding ®ne scale resolution of shifts in vegetation types is considerable for most paleoecological reconstructions, but broad generalizations can be utilized. We have represented global vegetation as a composite of 29 vegetation classes based on those de®ned by Olson et al. (1983) (Table 1), and previously mapped and published for the late Quaternary period by Faure (1997) or QEN (1995). The distribution of these ecosystem classes for each time period was reconstructed using digitalized hand-drawn maps based on continual checking and input from a large number of Quaternary scientists as a component of the quaternary We have assumed that the widely accepted QEN vegetation/ecosystem maps (Adams and Faure, 1997) represent approximately the dominant ecosystem classes that would have occurred, respectively, during the LGM, the early-Holocene (though with Ôerror limitsÕ to take in most of the range of opinion in the Quaternary community), and the conditions on the globe prior to industrialization and large-scale transformation of natural ecosystems to agricultural ones through anthropogenic activity (present potential vegetation). Furthermore, we have assumed that the analogous ecosystem classes used in this study have broadly similar climatic requirements under both present day and past conditions.
In the QEN (1995) maps, past biome distributions were reconstructed from a combination of indicators: the vegetation conditions indicated by plant fossils (pollen and macrofossils), animals (vertebrates and invertebrates characteristic of particular vegetation or bioclimatic conditions) as well as paleotemperatures of groundwaters from isotopes and noble gas ratios, sediment chemistry and depositional structures indicating a particular bioclimatic regime, etc. (Adams and Faure, 1997). These biome distributions themselves thus implicitly contain climate information (a crucial factor in VOC emission rates), and in this sense are analogous structurally and climatically to the biome distributions in Guenther et al. for present-day world VOC emissions.
For the tropical latitudes, there are various indications (from¯oras, faunas, groundwater composition, oxygen isotopes in eggshells, and from marine indicators from some tropical shelf sea areas) that lowland mean annual temperatures were 5±6°C lower than present throughout the tropics at the LGM. This possibility, perhaps crucial in aecting VOC emission from the tropics at that time because the evaporation process is so temperature dependent  is allowed for in our low-temperature scenario. However, because the scenario of signi®cant tropical temperature lowering at the LGM remains somewhat controversial (due to contradictions with open ocean surface temperature indicators) we have not presented this as our`most likely scenario' in terms of VOC emissions.

Present potential vegetation
For purposes of comparison with the LGM and the early-Holocene periods, the current state of global vegetation is summarized in a map of the present potential vegetation. The present potential vegetation represents that most likely to have occurred prior to anthropogenic modi®cation of ecosystem class (EC) species composi-tion and spatial distribution. The vegetation distribution represents a compilation of the present actual vegetation distribution maps of Olson et al. (1983) and various regional present potential vegetation maps (e.g. Walter, 1971). On an area basis, the present potential vegetation is dominated by tropical rainforest (EC 1) that covers approximately 9.6% of the total vegetated area, this is followed by tropical deserts, tropical scrub, cool temperate forest and savanna (ECs 7, 4, 14 and 9, respectively) ( Table 2).

Early-holocene vegetation
For the early-Holocene ($8000 BP), forest coverage in almost all regions was greater than the present-potential (Frenzel, 1992), with a major forest expansion occurring in Africa (Hamilton and Taylor, 1991) and SE Asia (including China) due to stronger summer monsoons. The areas of the Saharan and central Asian deserts were much smaller due to increased rainfall (Ritchie, 1994) that may have increased the coverage of grassland, and scrub vegetation may have predominated in both these areas at this time (Frenzel, 1992) (Table 2). These changes are incorporated into the maps as increases in the coverage of tropical forests, scrub vegetation and desert (ECs 1±5) in addition to expansions in warm (EC 10) and cool (EC 14) temperate forest, and mid-Taiga (EC 16). The enhancement of coverage by forest vegetation resulted in the decline of coverage for tropical grasslands, tropical deserts, and savanna (EC: 6, 7 and 9, respectively) in addition to reductions in temperate woodland and scrub and desert (EC: 18a, 18b and 22, respectively).

Last glacial maximum vegetation
Maps of global vegetation coverage for the period around the LGM have been produced by several different sources over the past few years. Crowley (1995) performed a series of calculations for LGM carbon storage based on a 220-site pollen database. Peng (1994) produced a set of Northern Hemisphere/global estimates of carbon storage based on the vegetation maps of Grichuk (1992) and the climate reconstruction maps of Frenzel (1992). Published maps also exist of regions or sub-regions of the world for the LGM and Holocene (Frenzel, 1992;Clapperton, 1993). The maps utilized here represent an approximate`consensus' picture developed by the QEN (Adams, 1995;Adams and Faure, 1997).
Evidence used to support the QEN ecosystem reconstructions include (1)¯ora and fauna fossil data and (2) sedimentological indicators such as buried soil types, sedimentary structures and sediment chemistry. Plant fossils are the preferred direct indicators of past vegetation, but can produce misleading results for local vegetation if concentrated in sediments by selective transport or preservation. Furthermore, plant fossil data is relatively sparse for the period around the LGM due to aridity. The QEN ecosystem reconstructions emphasize 14C-dated evidence, whether palynological, sedimentological or zoological, however, the error bars on the age of each site often span a few thousand years. Nevertheless, it is reasonable to suggest that there was an overlapping period when cool and arid conditions occurred across many dierent regions of the globe, although not necessarily the LGM sensu stricto. This would produce a global vegetation distribution much like that which has been reconstructed in the QEN maps for the LGM. The climatic changes associated with the LGM reduced the coverage of tropical vegetation (43% of the total vegetated area), whereas polar/boreal vegetation expanded (34% of the vegetated area) ( Table 2).
As there are still signi®cant uncertainties concerning the actual global vegetation cover over the last 25,000 years, the ecosystem distribution during the LGM was bounded by creating three variations in forest ecosystem distribution that covers the range of forest cover estimates. Our baseline forest cover distribution scenario is thus derived from the QEN ecosystem maps (Adams and Faure, 1997) and is designated the medium forest area (MFA) distribution ( Table 2). The general consensus among the Quaternary community, re¯ecting in the MFA, is that there was a major (>50%) loss of forest cover from the tropics and expansions of desert zones throughout the globe relative to the present potential vegetation distribution. The QEN (Adams and Faure, 1997) maps for the LGM show large reductions in forest vegetation, especially in the higher latitudes where the forests were mostly replaced by ice sheets, polar desert and by the arid, cold-climate steppe tundra vegetation (Adams and Faure, 1997). In the QEN maps and the sources upon which they are based, temperate forest vegetation seems almost to have disappeared from Table 2 Reconstructed global areas of ecosystem types, from Adams and Faure (1997). Bracketed values are for the low forest and high forest scenarios, respectively. Values are in Mkm 2

Ecosystem code
LGM 8000 ya 5000 ya Present pot. Europe (Frenzel, 1968(Frenzel, , 1992Turner and Hannon, 1988;Bennett et al., 1991) and boreal forest disappeared from most of Eurasia (Velichko and Spasskaya, 1991;Velichko and Isavea, 1992), leaving only isolated pockets scattered through mountainous areas. In China and Japan temperate forest vegetation was replaced mainly by open woodlands, forest steppe and grasslands (Tong and Shao, 1991;Frenzel, 1992;Ooi, 1992;Wang and Sun, 1994;Liew et al., 1995). Lower latitudes also showed major desert expansions in Australia (Bowler, 1976), north Africa (Schulz, 1986), northwest India (Singh et al., 1974) and parts of South America, and retreat of forests and woodlands in Africa (van Neer, 1984;L ezine and Cassanova, 1989;Hamilton and Taylor, 1991), Australia (Thom et al., 1994), central and South America (Clapperton, 1993;Leyden et al., 1993;Suguio et al., 1993;van der Hammen and Absy, 1994) and southeast Asia Thomas and Thorp 1991). However, signi®cant uncertainties do remain for the forest distribution during the LGM, and other scenarios are presented to take account of the full range of existing and possible interpretations in the literature. The second forest distribution scenario is the`high forest area' (HFA) scenario that is more closely in line with the mapping eorts of Peng (1994) and Crowley (1995); in which the LGM has greater coverage of forest and other moist climate vegetation such as woodland and moist steppe than the MFA (Table 2). Data supporting the HFA scenario is principally derived from use of a relatively limited amount of pollen data that require moist depositional conditions for preservation and therefore tend to concentrate at more moist localities favoring interpretation as increased forest distribution. The third forest distribution scenario is a`low forest area' (LFA) scenario suggesting vegetation distributions in which forest, woodland and moist steppe were restricted in distribution (Frenzel, 1992;Clapperton, 1993) as suggested primarily by sedimentological and geomorphological data in conjunction with limited use of plant fossil evidence (Table 2).

Global temperature conditions
The temperature conditions experienced by vegetation are as important as its structure and composition in determining VOC emission (see below). We assume here that each EC had broadly similar climate requirements to its closest analogs in the present-day world, even if the detailed mosaic of community composition was dierent. In this study, the temperature conditions were assumed to be roughly constant for present-day and past biomes, with the biomes having shifted in the past to position themselves along temperature gradients. Thus for most biomes, the emission factors for each biome used in our main source reference on global land VOC emissions  automatically include temperature eects.
There are some examples in which temperature relationships of species and biomes during the LGM could have diered from present conditions (e.g. an extinct warm-temperate form of Picea along the Gulf coast of North America during the LGM), however, in most areas both vegetation and other independent temperature indicators suggest roughly similar biome temperature requirements in the past.
Only for the tropics (below) is this distinction important; the tropical lowlands represent the extreme upper end of the global temperature scale, and if global temperature everywhere cooled substantially, the tropics themselves would have been existing under lower mean temperatures (other bioclimatic zones could shift latitudinally, but there would be nowhere else for tropical zones to shift to if they all cooled substantially). Thus, one potentially important dierence between the present conditions and those of the LGM was the lower tropical lowland temperatures (Broecker, 1985) that could (i) eliminate high VOC emitting vegetation assemblages, by in eect pushing the climatic tropics equatorwards, and even with the tropics reducing the abundance of characteristic`tropical' forms and replacing them with lower montane and warm temperate forms and (ii) reduce the VOC emission from remaining vegetation. A 5°C lowering of mean annual temperature in most areas of the lowland tropics now seems likely on the basis of¯oristic, isotopic and other indicators from the tropical landmasses, and also from certain oceanic indicators (Broecker, 1985;Crowley, 1995). A low tropical temperature scenario (LT) was superimposed upon the LGM MFA distribution to assess the potential eects of lower tropical temperatures on total annual VOC emission. The Low Temperature (LT) scenario reduced the mean annual temperature for tropical ecosystem classes (EC: 1±9) by 5°C. The mean annual temperatures of tropical montane forest (EC 12) was not reduced by 5°C as the hypothesized reduction was con®ned to lowland locations (it is assumed that montane forest would move downslope, keeping a steady emission rate). Furthermore, tropical montane forest contributes only a small fraction to VOC emission both from tropical ecosystems as a whole and to the total annual global VOC emission. The reductions in temperature were achieved by subtracting 5°C from the mean annual temperature value used in calculating emissions rates for total areas which contained tropical ecosystem classes 1± 9.

Estimating voc emission from vegetation and paleovegetation
VOC output from vegetation depends on such biological factors as leaf area index (LAI, leaf area per unit ground area), speci®c VOC emission rate of foliage, and canopy temperatures. The estimates of Guenther et al. (1995) for present-day base VOC emission rates of ecosystem classes can be tentatively applied to the past vegetation, assuming (i) minimal alteration of climatic and structural relationships between past ecosystem classes and those occurring in the present-day world and (ii) limited changes in the types of VOCs emitted by vegetation and the environmental control over their biochemical production and emission. The emission factors assigned to vegetation types in this study are derived from the base VOC emission factors of Guenther et al. (1995) that were developed from a literature survey of current VOC emission studies and are considerably dierent from those used in most previous studies (Table 3).
Addressing the in¯uences of temperature (see above) are essential for VOC emission simulations of both monoterpenes and isoprene, whereas accurate illumination conditions are critical for isoprene alone (Guenther et al., 1993. Algorithms for estimating VOC¯ux from vegetation at the foliar level under standard environmental conditions, utilizing empirical emission coef-®cients to account for the eects of illumination and temperature, have achieved considerable predictive success (Guenther et al., 1993). Additional work  scaled leaf level VOC emission to the canopy on a monthly time frame incorporating information on monthly NPP and leaf area index (LAI) from satellite-derived normalized dierence vegetation index (NDVI). Additional modi®cations incorporated the in¯uence of speci®c leaf weight (g m-2) and canopy structure that de®ned either sun or shade leaves to improve emission estimates. The Guenther et al. (1995) methods were applied to the QEN ECs (Table 1) to calculate global¯uxes of isoprene; monoterpenes; and ORVOCs during the LGM, early-Holocene and present potential vegetation conditions.
Estimating VOC emission from vegetation and paleovegetation. VOC output from vegetation depends on such biological factors as leaf area index (LAI, leaf area per unit ground area), speci®c VOC emission rate of foliage and canopy temperatures. The estimates of Guenther et al. (1995) for present-day base VOC emission rates of ecosystem classes can be tentatively applied to the past vegetation, assuming (i) minimal alteration of climatic and structural relationships between past ecosystem classes and those occurring in the present-day world and (ii) limited changes in the types of VOCs emitted by vegetation and the environmental control over their biochemical production and emission. Here we have assigned VOC¯ux values on a land surface area basis to the QEN classi®cation based on those estimated for current ecosystems by Guenther et al. (1995) ( Table 3).
Addressing the in¯uence of temperature (see above) is essential for VOC emission simulations of both monoterpenes and isoprene, whereas accurate illumination conditions are critical for isoprene alone (Guenther et al., 1993. Algorithms for estimating VOC¯ux from vegetation at the foliar level are scaled from standard environmental conditions, utilizing empirically determined modi®ers account for the eects of illumination and temperature (Guenther et al., 1993). The emission of monoterpenes is calculated as an exponential function from emission rate measured under standard environmental conditions modi®ed by an empirical coecient de®ning the temperature dependence of monoterpene emission (Guenther et al., 1993). Isoprene emission is calculated by scaling emissions measured under standard conditions using empirically derived factors accounting for the eects of temperature and illumination. The illumination factor describes an approximately linear increase in isoprene emission up to a saturation point, whereas the temperature factor approximates an inverted parabola simulating the change in enzymatic activity with temperature (Guenther et al., Table 3 Per-unit area emission rates for VOCs (Tg Mha y) (after Guenther et al., 1995)    scaled leaf level VOC emission to the canopy on a monthly time frame incorporating information on monthly NPP and leaf area index (LAI) from satellite-derived normalized dierence vegetation index (NDVI). Additional modi®cations incorporated the in¯uence of speci®c leaf weight (g m À2 ) and canopy structure that de®ned either sun or shade leaves to improve emission estimates. The Guenther et al. (1995) methods were applied to the QEN ECs (Table 1) to calculate global¯uxes of isoprene; monoterpenes; and ORVOCs during the LGM, early-Holocene and present potential vegetation conditions.

Present potential vegetation and VOC emission
The present potential vegetation of the globe ( Fig.  1(a)) includes 26 of the 29 QEN ecosystem classes, not present in the distribution maps are both forms of steppe-tundra (EC 20a, b) and forest tundra (EC 26). Both steppe-tundra and forest tundra occur in localized areas in the present-day world, but their extent was considered too small to map. Of the remaining ecosystem classes, 41% of the total vegetated area is covered, in order, by tropical rainforest, tropical desert, tropical scrub, cool temperate forest and savanna (EC: 1, 7, 5, 14 and 9). The combination of present potential vegetation and climatic conditions estimates a global total annual VOC emission of 912.9 Tg y À1 (61% isoprene; 13% monoterpenes; 24% ORVOC) (Figs. 2±4). Present potential VOC emissions are dominated by tropical rainforest (EC 1) representing 41% of the total, however, signi®cant contributions are provided by savanna (EC 9), cool temperate forest (EC 14), and monsoon forest (EC 2) that account for 11.4%, 5.7% and 5.3% of the annual total VOC emission, respectively. High rates of VOC emission (>15 Tg Mha À1 y À1 ) are geographically centered within 10°latitude around the equator in Africa, but extends as far as 15°around the equator in the Americas ( Fig. 1(a)). In Asia, high VOC emission rates are limited in their southern extension, but approach 30°N latitude in northern India and around the Bay of Bengal ( Fig. 1(a)). Regions of low VOC emission (<2 Tg Mha À1 y À1 ) are apparent throughout north Africa, the Arabian peninsula and central Asia, whereas more limited low VOC emission regions are con®ned to deserts in North and South America, southern Africa, and Australia ( Fig. 1(a)).

Vegetation distribution and VOC emission during the early-Holocene and Mid-Holocene
The total vegetation cover of the earth during the early-Holocene ($8 and $5 ka BP) (diered relatively little from the vegetation cover of the present-potential. However, there were signi®cant expansions in the areal coverage of tropical rainforests, monsoon forests and tropical grasslands ( Table 2) that are all signi®cant emitters of VOCs (Table 3). Additional expansions occurred in savanna and a range of ecosystem classes covering more minor areas ( Table 2). The expansion of the above ecosystem classes occurred at the expense of others that were reduced considerably in areal coverage including tropical scrub, tropical desert and lowland tundra, respectively. Our calculation suggest that the alteration in ecosystem distribution and a shift towards the relative balance of ecosystems that emit VOCs at high rates to those at low rates resulted in a 20% increase in total annual VOC emissions over the present potential vegetation to 1095 Tg y À1 (Figs. 2±4). The eects of the warmer and more moist environmental conditions on VOC emissions of the globe are the result of large reductions in ecosystem classes emitting VOCs at low rates (<2 Tg Mha À1 y À1 ) ( Fig. 1(c) and (d)). During the early-Holocene, desert regions are dramatically reduced in area, while regional deserts in North America and Australia remain those in South America are eliminated, whereas the vast deserts extending from North Africa to Central Asia become con®ned to Central Asia alone ( Fig. 1(c) and (d)). In contrast to the marked reduction in desert ecosystem distribution, those occurring in ecosystem classes emitting VOCs at the greatest rates (>15 Tg Mha À1 y À1 ) are only slightly expanded relative to present potential conditions (Fig. 1(c)).

Vegetation distribution and VOC emission during the last glacial maximum
The cooler and drier glacial conditions of the LGM produced signi®cant changes in the relative areas covered by dierent ecosystem classes ( Fig. 1(b)), relative to the present potential conditions. Conditions during the LGM favored the development of open herbaceous steppe-tundra across mid and northern latitudes; this vegetation type that does not occur over any signi®cant area in the present potential vegetation. LGM cold and aridity virtually eliminated coastal giant conifer, Mediterranean forest, south taiga and bog/swamp (Table 2).
While the general pattern of increasing aridity during the LGM is supported by data across many continents, reconstruction of vegetational distribution and emissions on a ®ner scale is confounded by the fact that detailed information on changes in species composition over time are generally only available for the forest regions of North America and Europe. The most problematic case is the`steppe-tundra' ecosystem (vegetation classes 20a, 20b) which covered much of the northern latitudes during the LGM (9.4% of the global land area), but is absent from the present potential vegetation maps; the nearest analogs occur on isolated south facing slopes in north-eastern Siberian mountains, and even here there are important dierences in plant assemblages from the LGM Ôsteppe-tundraÕ. A range of¯oristic, zoological and sedimentological indicators suggest that the LGM steppe-tundra ecosystem was a fairly sparse vegetation (at least 50% open ground) resembling semi-desert or dry steppe in most areas, existing under fairly brief cool dry summers and very cold winters (Velichko and Spasskaya, 1991;Spasskaya, 1992;Kolpakov, 1995;Adams and Faure, 1997). By analogy with present-day ecosystems existing under relatively similar conditions, we have assumed a`dry tundra' set of VOC emission characteristics for the steppe-tundra. Even if one assumes a Ôdry steppeÕ or Ôcold semi-desertÕ set of emissions characteristics for the steppe-tundraÕ, this would not profoundly aect the global VOC emission.
Relative to the VOC emission from the present potential vegetation, emission from vegetation of the LGM MFA is 29% less at 643 Tg y À1 (Figs. 2±4). Although the percentage contribution from ORVOC emission to the total is similar to present potential vegetation, there is a shift towards slightly lower isoprene emission (59% of the total) and slightly higher monoterpene emission (16% of the total) (Figs. 2±4). These shifts re¯ect the vegetational distribution towards lower coverage of the high isoprene emitting tropical forests and greater coverage of temperate and polar/boreal ecosystems that emit more monoterpenes. An examination of the change in VOC emission of speci®c ecosystem classes reveals large shifts in the proportion of the total VOC emission originating from each ecosystem class (Figs. 2±4). During the LGM, low VOC emitting vegetation expanded to cover substantial fractions of both North and South America; similarly vegetation of North Africa, the majority of Europe and Asia are all converted to low VOC emitting forms of vegetation ( Fig. 1(b)). As the low VOC emitting ecosystem classes expanded during the LGM there was a contraction in the area covered by ecosystem classes emitting VOCs at high rates. During the LGM tropical forests became con®ned to a region east of the Andes Mountains in South America, isolated populations in central Africa and tropical southeast Asia ( Fig.  1(b)). The Low Forest Scenario assumes further large reductions in forest and other moist climate vegetation types, resulting in a further reduced VOC emission relative to the present (Figs. 2±4 and Tables 4±7).

Eects of reduced tropical temperatures on VOC emission during the last glacial maximum
The incorporation of the possible LGM lowering of tropical lowland temperatures into the estimation of total annual VOC emissions results in a signi®cant alteration in the total annual VOC emissions and the relative proportions of isoprene, monoterpenes and ORVOCs.
The 5°C reduction of tropical temperatures reduced emission of all VOC classes from tropical vegetation by approximately 50% resulting in an annual VOC emission of 241 Tg y À1 , as compared to 486 Tg y À1 for all tropical ecosystem classes in the LGM MFA scenario. The eects of the lower VOC emission from tropical vegetation lowered total annual global VOC emission during the LGM MFA by 38% to 395 Tg y À1 (Figs. 2±4). The trend observed between the LGM and present potential vegetation in terms of the relative proportions of VOCs originating from dierent VOC classes is repeated when the simulation temperatures for tropical lowland forests are reduced (i.e., the fraction attributed to isoprene) declines (56% of total), whereas the fraction attributable to monoterpenes increases (17% of total); the fraction attributable to ORVOCs increases slightly (Figs. 2±4).

Discussion
Our a priori study of possible changes in VOC emission suggests that the change between LGM and Holocene conditions could well have been large. Comparison of the emission estimates for the present potential vegetation simulations serves to validate the general approach. The 561 Tg y À1 annual isoprene emission estimates from this study are considerably greater than estimated by Turner et al. (1991),  Zimmerman (1979Zimmerman ( ) or M uller (1992 at 285, 350 and 250 Tg y À1 , respectively (though these other estimates were for the present-actual world without agricultural eects on forest cover). But they are of similar magnitude to the $450 Tg y À1 estimated by both Dignon and Logan (1990) and Rasmussen and Khalil (1988); and the 503 Tg y À1 estimated by Guenther et al. (1995). Similarly, the published range of annual monoterpene emission varies considerably, but the 125 Tg y À1 simulated here are very close to those of M uller (1992) and Guenther et al. (1995), but approximately one-fourth those estimated by Zimmerman (1979). With respect to the total annual VOC emission, those estimated here are 21% lower than the 1150 Tg y À1 estimated by Guenther et al. (1995), but almost double the 491 Tg y À1 estimated by M uller.
The range of published estimated values for both total VOCs and speci®c VOC components can be traced to a range of methodological dierences between the models including varying estimates of ecosystem class coverage and diering incorporation of temperature and/or illumination in¯uences on VOC emission. However, the greatest cause of the variation lies in the use of the base VOC emission factors of Guenther et al. (1995) that were developed from a literature survey of VOC emission studies and are a factor of 3±5 greater than those previously used.
Our results suggest that the annual global rate of total VOC emission could have been $ 30 AE 10% lower during the LGM than during the Early-Holocene and present potential conditions depending on the particular scenario for forest coverage. The reduction in VOC emission during the LGM could be even greater if there were a substantial (e.g. 5°C) reduction in the mean annual temperatures of the lowland tropics. This would reduce the VOC emission of during the LGM to 56% less than the present potential vegetation. However, emission of VOC from speci®c ecosystem classes between the LGM and the present potential vegetation are highly variable, re¯ecting the dierential eects of cooler LGM temperatures on the spatial distribution of different ecosystem classes. For example, during the LGM there would be a reduction on VOC emission from tropical rainforests (EC 1) due to aridity and cooling, but greater emission from steppic-steppe tundra (EC 20a) representative of the changes in areal coverage of these two ecosystem classes. In spite of the combination of uncertainties in areal coverage of ecosystem classes during the LGM, the exact climatic conditions and ecosystem class VOC emission rate, the general changes which occurred between the cool and dry LGM to a warmer more vegetated Holocene resulted in a substantial increase in VOC emission. In temperate latitudes the reduction in total annual VOC emission would have been driven by the replacement of forest ecosystems by either ice sheets or ecosystems (such as tundra and grasslands) that emit VOCs at negligible rates. The expansion of grasslands during the LGM relative to pre-sent day is supported by the physiological modeling work of Collatz et al. (1998). The decline in the areal coverage of tropical low latitude forests during the LGM relative to present potential vegetation would have resulted in a decline in VOC emission, because in most cases tropical forest emit VOCs at greater rates than the dry tropical and temperate ecosystems that replaced them (Table 3). The emission characteristics of the Ôsteppe-tundraÕ vegetation of the LGM remain somewhat problematic (Spasskaya, 1992;Velichko and Spasskaya, 1992;Kolpakov, 1995;Adams and Faure, 1997). By analogy with present-day ecosystems existing under what appear to be the most nearly similar conditions (cold winters, fairly cool summers, low vegetation cover), we have assumed à dry tundra' set of VOC emission characteristics for the steppe-tundra. However, this matter may require further debate and consideration.
As explained in more detail above, in order to address the uncertainty of vegetational distribution during the LGM, we repeated the VOC emission simulations using estimates of forest area both lower (LFA) and higher (HFA) forest areas than that of the baseline LGM MFA forest distribution ( Table 2). The relatively limited eects of the LFA and HFA scenarios on total annual VOC emission ()13% and +15%, respectively) is due to the fact that the change in area of the tropical vegetation (EC: 1±9 and 12) is relatively limited ()3% in the LFA and +5% in the HFA scenarios, respectively) ( Table 2). As tropical forests account for a large fraction of total VOC emissions, limited changes in the coverage of these ecosystem classes have a large impact on total annual VOC emissions. The changes in the relative contribution of isoprene, monoterpenes and ORVOCs to the total annual VOC emission are in¯uenced similarly by the LFA and HFA forest distribution scenarios.
Further uncertainty exists regarding the climate of the LGM, speci®cally the annual temperature regime. Temperature exerts signi®cant leverage on the emission of VOCs on an instantaneous (Guenther et al., 1993), continental (Constable et al., 1999a) and global  scales. The paleoclimatic reconstructions noted by Broecker (1985), among others, suggest that not only was the mean annual global temperature cooler during the LGM, but that tropical zones had substantially cooler-than-present lowland tropical temperatures. To simulate the eect of this putative tropical land surface cooling on VOC emission, the simulations were repeated with a 5°C reduction in the mean annual temperature for areas containing ecosystem classes 1±9. The resulting impact on total annual VOC emission was 38% less than simulated for the LGM under baseline climatic conditions. For this reason, variation in total VOC emission during the LGM is in¯uenced to greater magnitude by the reduction in mean annual temperature of the tropical zones ()38%) than by variation in total global forest cover ()13.9% and +15.2% for the LFA and HFA forest cover scenarios, respectively) Figs. 2±4. This conclusion highlights that fact that improved accuracy in estimation of VOC emissions during the LGM requires improved estimates of global temperature distribution patterns rather than improved maps of vegetation distribution. Distributional and climatic uncertainty may aect the simulations at the global scale, but the biological controls (baseline emission rate, vegetation seasonality, canopy structure) over VOC emission may be similarly in¯uenced as do the existence of`no analog' species or ecosystem classes under present potential vegetation conditions. The assumption that vegetation during the LGM-emitted VOCs at rates comparable to present conditions is unveri®able, but considerable uncertainty exists around the actual present-day VOC emission rates from the world's vegetation . In spite of this uncertainty the simulations presented represent a valid approximation of the total annual VOC emissions from vegetation during the LGM. The in¯uence of`no analog' conditions under present day conditions for those during the LGM is problematical (e.g., lower concentrations of atmospheric CO2 or extinction of important VOC emitting species). With respect to the reduced concentration of CO2 during the LGM ($285 ul l À1 ) relative to current levels (365 ul l À1 ) (ref), it is possible that the reduction in carbon supply could alter the production of carbon-rich VOCs. The response of isoprene emission to elevated CO2 is highly variable with oak exhibiting increased emissions, whereas aspen displayed a reduction in isoprene emission (Sharkey et al., 1991). Studies examining the eect of double growth CO2 concentration on monoterpene emission may be less problematic as they demonstrated no change in monoterpene emission rate (Lincoln and Couvet, 1989;Constable et al., 1999b). Assuming these relationships hold for conditions during the LGM, the 80 lmol mol À1 decline in atmospheric CO 2 concentration may have had a negligible eect on monoterpene emission from vegetation. The in¯uence of CO 2 concentration on ORVOCs is uncertain as the diversity of ORVOC compounds makes if dicult to assign uniform CO 2 responses to a range of metabolic processes. However, as ORVOCs contribute signi®cantly to the total VOC emission it is necessary to clarify the control that CO 2 supply exerts on emission of these compounds. While the direct eects of CO 2 on VOC emission and production may be limited, CO 2 could indirectly aect VOC emission through an alteration of the leaf area index (LAI, unit area foliar surface/unit area ground surface). In experiments doubling of CO 2 has, at least temporarily, increased the LAI of ponderosa pine (Tingey et al., 1996) and loblolly pine (Tissue et al., 1997). Assuming that broadleaf species respond similarly and that a negative relationship occurs when CO 2 concentration is reduced it is possible that the VOC emissions simulated here overestimate those that may have occurred during the LGM. The potential for now extinct species or subspecies to have emitted VOC at greater or lesser rates than those assumed by analogy with present day vegetation classes is dicult to resolve. For example within the genus Quercus there is a phylogenetic separation between those species that emit isoprene and those that emit monoterpenes (Loreto et al., 1998). Should a similar situation exist between extant and extinct genotypes between present potential vegetation and the LGM the VOC emission simulated from a speci®c ecosystem class simulated here could be erroneous. In the pollen record, it is often dicult to tell closely related tree species apart, and thus shifts in species composition of vegetation that might result in changes in VOC emission cannot be detected and quanti®ed.
What would be the in¯uence of lower global VOC emissions in terms of atmospheric chemistry? There are some clues to this from a modeling experiment by Guenther et al. (1999a,b) they used a global three-dimensional chemistry and transport model (CTM) to characterize the impact of changing isoprene and monoterpene emission on distributions of the chemical compounds responsible for the oxidation capacity of the troposphere. They found that predicted distributions of ozone, CO, NOx, peroxymethacrylyl nitrate (MPAN), OH, H 2 O 2 , HO 2 and other compounds were sensitive to the magnitude of biogenic VOC emissions. For example factor of 2 increase in biogenic isoprene and monoterpene emissions resulted in a 10±20% decrease in OH, a 15±35% increase in H 2 O 2 and a 5±15% increase in O 3 near terrestrial surfaces. The model also predicted that this factor of 2 increase in emission results in a 6±10% increase in CO throughout the southern hemisphere (up to 18 km AGL), and a 5±15% decrease in OH at heights between 7 and 14 km AGL in the tropics. While we recognize that there are other major uncertainties associated with global CTM predictions of trace gas distributions, the analysis of Guenther et al. (1999a,b) suggests that the predicted changes in biogenic isoprene and monoterpene emissions between the present time and the LGM could indeed have resulted in a signi®cant perturbation to the chemistry of the EarthÕs atmosphere.
Although the uncertainty associated with the simulations is considerable, the lower total annual VOC emission during the LGM could serve to partially explain lower atmospheric CH 4 levels recorded in ice cores. Decreased VOC emission during the LGM could limit the atmospheric concentration of CH 4 by allowing reaction of hydroxyl radicals with CH 4 due to a decreased supply of the more reactive VOCs. If this situation were to prevail, the warming potential of CH 4 in the atmosphere would be minimized and surface temperature would remain cooler relative to current conditions. Without running complex atmospheric chemistry models, and identifying seasonal VOC emission rates calculated from palaeotemperature reconstructions, it is speculative to discuss the actual implications of these changes in emission rates for atmospheric chemistry or climate (which could have been oset by or add to the eects of VOC emissions from the oceans which seem to have changed greatly between glacial and interglacial conditions). However, as the experiments by Guenther et al. (1999a,b) (above) indicate, assuming the basic chemistry of the atmosphere during the LGM is similar to those of the preindustrial atmosphere there is the potential for VOC emission changes to aect the radiative balance of the atmosphere. The results of this study show that there is a clear potential for major changes in VOC emission between glacial and interglacial conditions. Indirectly by aecting atmospheric haze content and methane concentrations, these changes could account for some of the temperature and rainfall distribution dierences between the LGM and those of the Holocene.
Also to be considered is that greater VOC¯uxes should produce more low-level ozone, perhaps counteracting the cooling eect of haze production through the greenhouse gas properties of ozone (Hansen et al., 1997). This remains a further challenge for atmospheric chemical and climate modeling.

Conclusions
The diculties of reconstructing and considering the implications of long-term changes in VOC emission are considerable, but this does not mean that the factor should be ignored. The indications are, from changes in global vegetation distribution and tropical temperatures since the LGM, there have been quite large changes in VOC emission from vegetation. These were changes in both overall quantity, and in the relative importance of dierent classes of compounds. Such changes present a potentially very signi®cant ÔunknownÕ in the climate system; something which could have aected both haze production (and thus the global heat balance) and atmospheric methane chemistry. If such an unknown factor exists, even if it cannot be precisely quanti®ed, it must be properly allowed for in de®ning the error limits and range of scenarios for reconstructing past climates.
Though the scenarios we have compared are the extremes of the system, it is necessary to bear in mind that most of the last 100,000 years has been spent in intermediate vegetation states, cooler and drier than presently, but not as cold and dry as the LGM. Furthermore, there have been many sudden regional (and possibly global) transitions in climate, involving warming by several degrees in mean annual temperature in as little as a few decades (e.g. see review by Adams et al., 1999). Such changes might have been either am-pli®ed or damped by changes in vegetation cover, producing rapid changes in the eux of VOCs. Even a pre-existing vegetation cover, such as tundra or open conifer woodland, could perhaps have responded to a sudden climate warming with a dramatic increase in VOC output, without any shift in species ranges. In at least some areas, such as European Russia, rapid warming events such as the end of the last Glacial phase were followed by a very rapid spread of early successional forest taking only a few hundred years. Many early successional species are known to be strong sources of VOCs.
Following on from our preliminary study, further work is required to explore these potential changes in VOC emissions and their implications in greater detail, using the increasingly detailed knowledge of factors controlling VOC emission rates, and using atmospheric chemical modeling techniques.