The aims of this data inventory.

1) Predicting future changes in sinks and sources . There is presently a great deal of interest in the sinks and sources which may affect the rising CO2 level in the Earth's atmosphere. To make the most accurate predictions of future changes in carbon storage over coming decades and centuries, it will be necessary to bear in mind that many ecosystems that we see today are in a transitional form, resulting from high intensities of artificial disturbance. In other cases, a more intensive natural disturbance regime (e.g. by fire) may have been suppressed by humans. This inventory aims to reconstruct the 'potential' ecosystem carbon storage that would result, in a natural disturbance regime, without human influence. It will be of considerable relevance to the many cases in the present-day world (mainly the temperate zones) where humans are attempting to withdraw from forests and to afforest greater areas as a carbon sink. If the influence of logging and other disturbance is decreased in parts of the tropics over coming decades, these ecosystems will also tend towards a more 'natural' state whose average carbon storage is unknown. It is hoped that this inventory will focus attention on the need to discuss such 'post-anthropogenic' aspects of global ecosystem carbon storage.

2) Recent and geological history of the carbon cycle. Many ecologists and biogeochemists are interested in reconstructing the recent geological history of changing carbon storage on the Earth's land surface. In addition to the hope that this may add to the understanding of 'missing sinks' for carbon dioxide in the contemporary world, there is also the challenge of understanding the CO2 and climate oscillations that have occurred during the Quaternary Period (the last 2.4 million years of the Earth's history).

In the context of trying to improve understanding of the Quaternary carbon cycle, there has been a flurry of papers by different groups attempting to reconstruct LGM and Holocene land carbon storage (e.g. Prentice & Fung 1990, Adams et al. 1990, Prentice et al. 1992, Van Campo et al. 1993, Peng 1994). Various methods have been used to reconstruct land vegetation / ecosystem distribution for the prehistoric Late Quaternary, but a limitation that all of these estimates have in common is a fundamental dependence upon contemporary field-based data listings of per-unit-area carbon storage. In many of these calculations of long term changes in carbon storage, published figures of per-unit-area carbon storage are used uncritically to represent past ecosystems, despite the strong possibility that they represent anthropogenically degraded areas. Furthermore, many of the published figures for per-unit-area carbon storage are based on confusing and misleading vegetation definitions, and have been gathered using poor field sampling procedures. There is a need to sift through this mass of data and critically examine it with the specific task in mind. So far, amongst all the papers published on long-term changes in carbon storage, only Adams et al. (1990) have made any declared attempt to select data from the less anthropogenically altered sites. Even their selection of data was exceedingly ad hoc, and their selection procedure was not explained in any detail in their brief paper.

The present summary of data is a preliminary attempt at the task of providing a careful, reasoned set of per-unit-area values that might be used for calculating carbon storage changes in forest and other ecosystems on historical or geological time scales.

The collection of carbon storage data from the present-day world.

In addition to the variability caused by differences from one local set of site conditions to another, part of the reason that different authors tend to get such different results may lie in the methods which they have been using to add up carbon storage in their field sites. When ecologists measure total organic matter directly in the field, there is often uncertainty as to how far down into the ground one should go to gather up roots and organic carbon, or how much of the soil 'litter' layer to include in with soil organic matter. When they use indirect methods of calculation of vegetation biomass, based on such parameters of basal area of tree trunks per unit area of forest, ecologists can apply any one of several different algorithms which will yield differing results. The conversion from raw biomass or soil organic matter into pure carbon mass is also an area of disagreement, bringing with it another small set of errors (see below).

Some of the disagreement within the literature may also be caused by ambiguity in the definitions of vegetation types. Data on soil and vegetation carbon storage must be compiled from diverse sources, which may use different ways of defining each ecosystem type. For example, it is well to remember that one author's 'forest' may be another's idea of 'scrub', and what is called 'scrubland' in one part of the world is called 'desert' in other places. There is also the obvious but easily forgettable fact that the world's vegetation actually consists of continua and mosaics, which must usually be divided up or lumped together in order that one can work with them. If carbon storage follows a gradient across a particular vegetation zone, it is difficult to know which value to take as the overall average of this gradient. Certainly, if more data are gathered in the future it may one day be possible to plot carbon storage as a continuum against environmental factors or particular vegetation attributes. This has already been attempted to some extent on a very coarse global scale in a diagram given by Post et al. (1982), and for certain well-studied grassland regions (J. Guiot, Universite Aix-Marseille, pers. comm.). However, if this approach is to be generally applied and in a reliable way, much more work will need to be done.

Hence, in this data summary it has been necessary to divide up the world into manageable, workable biome categories, which in some cases are further subdivided where the dataset is sufficient to warrant this. The categories used here are tailored to fit in with the traditional ways of thinking in ecology, with the world divided according to major biomes such as 'tropical rainforest' and 'temperate deciduous forest'.

In trying to estimate global carbon storage, there are further problems due to the difficulty of selecting representative sites from within the spatial heterogeneity that is present in all vegetation. It is now beginning to appear that many of the earlier published estimates of present-day carbon storage (such as some of the IBP measurements used by Olson et al. 1983) were based on selective samples of vegetation stands which were unusually high in biomass, perhaps based on the assumption that these represented the 'true' natural vegetation undisturbed by humans or natural disturbance events.

In fact, it is difficult to know what the real reasons are for the differences in total carbon storage suggested by different authors for each particular vegetation type. No doubt the disagreements in the literature over the 'representative' carbon storage value for each ecosystem are the summative result of several error factors, each rather small in itself but together multiplying up into a much bigger error.

Conversion from organic matter to carbon storage. All vegetation and soil carbon storage data are ultimately calculated from raw organic matter, converted into carbon storage equivalents by a conversion factor. Different authors use different conversion factors, some using a figure of 0.50 or 0.51 and others using 0.45. Many published figures have already been converted into carbon storage by a conversion factor deemed appropriate by those who publish them, usually between 0.45 and 0.51. Such figures are derived from the proportion of carbon in cellulose (0.40) and the somewhat higher proportion of carbon in lignin (about 0.51), which together comprise most of the organic matter in plant tissue (J. Grace, University of Edinburgh, pers. comm.). Where raw dry weight biomass figures (not already converted into carbon storage equivalents) are given by the sources cited here, these are converted into carbon storage through multiplication by a 'compromise' figure of 0.475.

For soil organic matter, the proportion of lignin and lignin-like compounds is greater, so a conversion factor of 0.50 - 0.55 is generally used to derive carbon storage from dry organic matter. However, in the case of soils authors usually tend to present their data already converted into carbon mass.

Sources of vegetation data.The vegetation carbon storage figures presented in this inventory are for plant parts both above and below ground level, and they include all the 'living' (i.e. still functional, or at least connected to functional parts) pieces of plants, unless otherwise stated. Most of the data in the literature on forest biomass include only trees and woody vines that are over a certain size limit, usually defined in terms of their stem girth. For forest vegetation, the smaller woody plants, herbaceous plants or the understory biomass in general tends to be ignored by most authors as being either relatively insignificant, too difficult to measure, or simply irrelevant to the study. Some attempt has been made here to bring in estimates of litter and understory carbon storage, although there is far less information available in the literature on these. For simplicity, vertebrate biomass is ignored in this study, because all studies of land ecosystems show that it is equivalent to no more than a tiny fraction of 1% of plant biomass (Olson et al. 1983), much smaller than the intrinsic errors in assessing plant biomass alone. Probably, most microarthropod biomass below ground is included in with measurement of soil dead organic matter (see below).

A major source of data on carbon storage in vegetation is the classic study by Olson et al. (1983), which represents the compilation of a massive amount of information gathered mainly under the International Biological Programme (IBP) studies of the late 1960s and 1970s. Note that the aim of the Olson et al. study was to present data on the contemporary (around the year 1978) level of carbon storage in vegetation around the world, using broad vegetation and ecosystem categories irrespective of the subtleties of anthropogenic interference in many of these vegetation types.

There has been a recent resurgence of interest in finding representative values of actual and natural carbon storage in vegetation. This is because of the uncertainty surrounding rapid fluxes in carbon dioxide from vegetation and soils over the past few centuries, and its influence on atmospheric levels of carbon dioxide. Whilst major uncertainties remain, it does seem likely that understanding has advanced significantly since Olson et al.'s work at the beginning of the 1980's.

When Olson et al. published their carbon storage inventory, they also published a global vegetation map and accompanying description scheme that set out in broad terms the vegetation definitions that they were referring to. In the present inventory, the categories of global vegetation types used generally correspond to the structural-taxonomic ones of Olson et al. (1983), although in certain cases these categories have been 'lumped' where carbon storage values in two different vegetation types are generally very similar, and 'split' where there is evidence of a recognisable sub-type of vegetation with distinct carbon storage characteristics. A more refined scheme, or one based only on climate-zone characteristics, would not be useful in the context of the aim of this inventory, which is to present data for use against direct and indirect palaeoecological indicators of past vegetation cover types. There is little point in having a highly sub-divided scheme if there is actually no hope of being able to distinguish between the distribution of these minor vegetation categories in the past, since this is after all an inventory intended for the past world and not the present.

Soil and peat carbon. On a global scale, soils are a more important reservoir of organic carbon than the living vegetation that roots into them. Other than the 'living' plant parts within soils (which are here included with vegetation biomass), the major store of carbon is in the form of heterogeneous organic humic compounds that are derived by decay of plant materials. The living biomass of bacteria and fungi may also be a major carbon reservoir in soils, but in practice the methods of measuring soil organic carbon (through loss on combustion) mean that this reservoir is automatically included along with the humic substances.

Soil carbon data for the world's biomes have been summarised by Post et al. (1982), and substantially improved upon in Zinke et al. (1984). These figures represent the outcome of a database of thousands of standardised samples, taken from all around the world. For the convenience of ecologists taking a vegetation-related biome approach, the figures were 'slotted into' the vegetation scheme used by Olson et al. (1983) and the bioclimatic scheme of Holdridge (1967). Some might criticise this approach as an inappropriate way of presenting information on soils, which vary in their own ways not always related to the overlying vegetation. However, although soil type and soil carbon do not necessarily follow vegetation structure in any simple way, there is at least a noticeable relationship which Zinke et al. have noticed and emphasised. Certainly, it is much simpler to reconstruct vegetation and soil carbon from a single set of historical or palaeovegetation maps than to try to reconstruct soil distributions separately.

In fact, in the context of reconstructing past carbon storage it might be possible to reconstruct the distribution of the standard soil categories for the last glacial period or early Holocene from the information on palaeoclimate and vegetation conditions, combined with a knowledge of such factors as underlying geology and slope angles. However, this represents too great an undertaking for the present, and it would require the application of specialist knowledge that is beyond my own scope. For now, the only realistic way forward is to use a more ad hoc method, treating vegetation types and the soils underneath them as linked units.

Thus, in this inventory, the soil carbon storage figures represent the organic matter in the soils found under each vegetation type and do not refer to any sort of separate soil units (except in the case of peat bogs). The carbon storage figures for soils do not include the litter layer of fallen leaves and branches, which is instead dealt with separately.

Usually in field studies, the organic carbon is measured down to one metre depth in the soil profile, unless (as in peat bogs) large amounts of organic material clearly go down deeper than this. There is in fact considerable ambiguity in many studies on global or local carbon storage as to where the category of 'peat' ends and that of 'soil' begins. One gets the impression that without enough care and understanding they might easily be dealt with in either an overlapping or incomplete way, so that when an estimate for global soil carbon storage is added to an estimate for global peat carbon storage, some of the areas could be counted twice or not at all. One widely accepted definition for peat is a pure organic layer at least 20 cm in thickness, and this was used by the widely cited studies by Post et al. and Zinke et al.. However one can take as another example the study of Canadian peatland areas by Tarnocai (1980), which defines peatlands as having peat depths (i.e. an organic matter layer) greater than 40cm, and mineral wetlands as having an organic matter layer of less than 40cm. The recent wide-ranging study of northern peatlands by Gorham (1992) used a minimum figure of 30cm organic matter as its dividing line between peat and non-peat, so there is no sign of a true consensus emerging!

In peatlands there are also often areas of open water, which the definition does not include. In fact, the proportion of the total surface of a landmass covered by small lakes and pools (which is very substantial in the northern latitudes of Canada, Scandinavia and Siberia) and by streams and rivers, is something that needs to be allowed for more rigorously in global carbon storage calculations. Many previously published studies of prehistoric carbon storage have not even attempted to take this factor into account.

Natural disturbance of vegetation. Humans are not the only disturbance factor tending to reduce carbon storage in vegetation and soils. Many areas of the world are subject to periodic fires or storms which disrupt natural ecosystems, and (as already mentioned) large grazers are sometimes very destructive of vegetation at their natural population densities. If one uses only data from areas that have escaped recent natural disturbances - perhaps due to protection from humans against fires or natural grazers - then the carbon storage values observed at present will tend to be higher than would normally have been the case in the 'natural' state

However, it is perhaps too easy to get an exaggerated picture of the importance of some natural disturbance processes. Such dramatic events as the felling of coastal rainforests by hurricanes may be catastrophic on a local scale, but most areas of tropical forest are never subject to hurricanes (Bose et al. 1994). Even in those areas that do experience violent storms, the really severe damage may occur only once every few centuries. A few centuries may give more than enough time for the rainforest vegetation to approach its maximum potential biomass; consider the rapid regeneration of forest over Maya temples and volcanic islands (however, the evidence that many large neotropical forest emergent trees are between 400 and 1,300 years old, suggests that even very infrequent disturbance on this timescale would be enough to suppress carbon storage; Chambers et al. 1998) Chambers J.Q., Higuchi N. & Schimel J.P. 1994. Ancient trees in Amazonia. Nature v.391 p.135-136).

Even in a relatively windy climate such as Britain's, a really destructive storm such as that which hit Kent and Surrey in October 1987 had not occurred before in at least the previous 150 years, and this storm left most areas of high forest largely intact (it was the isolated trees, ridgetops and anthropogenically created woodland edges which suffered badly; personal observations by the author). In North America, the indications from historical and meteorological records are that hurricane, tornado damage and fire damage to the pre-colonial temperate forests was very infrequent (Whitney 1994). Except for certain coastal and dry marginal areas where these events were liable to recur more than once a century, an average patch of forest would have been destroyed by a severe disturbance event only once every several hundred to several thousand years (Tallis 1990, Whitney 1994). Tornados can cause quite severe damage locally in forests in parts of the eastern USA, but the return time for a tornado damaging a particular precise point is of the order of several centuries or more. Even then, many large trees will survive (though in a damaged state) being hit by a tornado (personal observations by the author). Indian populations at the time of European colonisation may have been sufficient to suppress forest biomass in some areas, but in the mid Holocene and earlier it seems that agriculture was sparse or non-existent throughout the American forests (Whitney 1994, and see Appendix 1 of this thesis). 19th century pictures and photographs (e.g. Sears 1994) of colonial 'virgin' forests in the eastern USA seem to support the impression that the American forests were infrequently disturbed (though perhaps returned to this state by the early genocide of the previous Indian populations); such images generally show a dense canopy with a considerable proportion of moribund and standing dead trees.

There are also some areas of the world (e.g. the lower montane forest belt of the Andes) where landslides are very common due to a combination of rapid weathering and high rainfall combined with tectonic uplift of the landscape; however, even in these particularly unstable areas it is hard to imagine that any individual patch of forest would be swept away by a landslide more often than once every few centuries or even every few thousand years. Colinvaux (1994) suggests that lowland tropical forests in general might also be subject to relatively high frequencies of disturbance due to shifting river channels, but again it seems unlikely that the true incidence of these events is more than once every few centuries for a given patch of forest (though see the results of the study by Chambers et al., cited above).

On the other hand, the importance of certain other disturbance factors may have been under estimated in most estimates of carbon storage. For example, in the boreal forests and in other forest types dominated by resinous trees, there is a natural fires are often started by lightening during periods of drought (Tallis 1990, Whitney 1994). The return period of fires in natural boreal spruce forests may be as little as 80-100 years (Wein & McClean 1983), presumably enough to suppress overall carbon storage (although the severity of fire is more significant than its frequency; a crown fire will be much more destructive than a ground layer fire). In recent times, large areas of forest (e.g. in parts of Canada) have been deliberately protected from fires, and thus measurements of carbon storage from within these protected areas may be unrepresentative of the early-to-mid Holocene state (Apps et al. 1993). Apps et al. (1993) suggest a higher return rate of destructive events than Wein & MacLean (1983), concluding that in fact the true overall state of biomass of boreal forests uninfluenced by man would be far lower than is normally supposed, because lightening-induced fires and insect outbreaks are so frequent.

The true significance of natural fires in suppressing boreal forest biomass remains a controversial area. S.P. Payette, (Univ. Laval, pers. comm., July 1994), and other Russian and Canadian boreal forest ecologists whom I have spoken to, all feel that Apps et al. are likely to be incorrect in this view. The available record of estimates of forest fire frequency since 1918 (Auclair et al. 1996) suggests that there was about a ten-fold decline in the volume of wood lost to forest fires in the USA between about 1920 and the 1960's onwards. However, for Canada there appears to be less of a decline, and for the former USSR there is no clear trend in loss of wood to fires over the same period. Since the latter two regions contain most of the boreal forest mass in the world, this may suggest that the decline in fire disturbance of boreal forests is more a USA-based than a global phenomenom. The same summary graph does show however that there was a dip in fire losses during the period between 1950 and 1970 when much of the important early work on forest biomass was being carried out in the USSR and Canada, perhaps tending to lead to inflated estimates in studies which looked at forest areas where forest fires had already been suppressed for several decades. Unfortunately, there has not yet been sufficient time for this matter to be taken up and discussed in the published literature.

There is evidence that natural crown (canopy) fires occurred on the timescale of centuries in the warm temperate pine forests of the southeastern USA, especially on the relatively drought-susceptible sandy soils of the coastal plain (Christensen 1978), where some areas of natural scrub-savanna seem to have been maintained by occasional fires. In these areas, it would seem, there would indeed have been a significant effect of natural fire frequency in supressing carbon storage. It is not clear whether occasional fires have ever been frequent enough to suppress carbon storage in other parts of the eastern USA forest zone. Christensen suggests that ground fires, or even the occasional crown fire on the timescale of centuries, were important even in the moist cove forests of the southern Appalacians.

In general, it appears that natural broad-scale disturbance events would not have been frequent enough to rival the effects of man in lowering the vegetation carbon storage of many areas of forest. It seems more appropriate to look to 'protected' old-growth areas of temperate and tropical forest as representative of the late Quaternary character of these biomes. However, this assumption of stability may turn out to be unjustified for many boreal forests, with their greater susceptibility to burn events, and for tropical woodland areas with their high natural populations of large herbivores.

Making allowance for the past.

'Naturalness' in vegetation carbon storage; can the present be representative of the past? There of course is no prospect of gaining direct access to the world of the past. The most direct information that one can obtain about the per-unit-area carbon storage of past ecosystems is gathered from the present-day world. Yet it is obvious that there are certain important ways in which the ecology of the present world differs from even the recent geological past. Most importantly, the vegetation that exists at present has largely been modified and degraded by modern populations of humans, through agriculture or wood cutting. When making estimates about the actual present-day carbon storage it is important to consider how this differs from the state that once existed in the past, 5000 years or more ago (and might exist once again if humans were to vanish off the face of the Earth). Yet the published data on per-unit-area carbon storage have often made little effort to distinguish different degrees of degradation in the carbon storage of vegetation, even though such data could also be very useful for calculating future carbon fluxes under different scenarios.

No-analogue communities in the past and future. When dealing with past vegetation it is always necessary to bear in mind that very often, the plant communities (the species that tended to occur together) were different from those existing at present. Sometimes, even the boundaries between what we perceive as discrete 'biomes' were blurred in the past, by particular species crossing over from one biome to another. Many examples of these no-present-analogue communities are described in the palaeobotanical literature, and perhaps the most obvious instance is the steppe-tundra that covered the northern latitudes during the Last Glacial Maximum (21,000 years ago), combining species of both steppe and tundra environments into a single vegetation type. Other examples include the abundant occurrence of certain montane trees in lowland tropical forests during the LGM, and during the Holocene the combination of temperate tree species that do not have closely overlapping ranges at present (Tallis 1990). In terms of reconstructing the carbon storage of such ecosystems, the unfamiliar combinations of species are disconcerting. For instance, for the steppe-tundra, was the carbon storage value more like that of present-day steppe, or present-day tundra, or like neither? The possible causes of no-analogue communities are many and various, including changes in carbon dioxide level (see below), climatic parameters, herbivore abundance, and the ongoing processes of broad-scale succession (Tallis 1990). In many cases, one can only hope to make progress in reconstructing the actual carbon storage of the unusual communities and mixed biomes of the past by assuming that their carbon storage was similar to that that of the present-day biome which they most resemble, or a simple mean of the two or three biomes which they seem most similar to at present.

We are also likely to lose our present-day familiar assemblages of species and see new ones appear if global climate change occurs over the coming decades and centuries. One of the lessons of the palaeoecological record for the future is that species assemblages and even biomes as we see them in present-day world are actually a fairly transitory phenomenon on the timescale of millennia. As climate changes, each plant species will move independently according to the climatic opportunities open to it, and the migration routes available. This will make estimation of future potential carbon storage even more difficult than it would be if the plants had just stayed still.

Direct carbon dioxide effects. A particularly striking problem in terms of understanding both past and future ecosystem carbon storage is the possible influence of direct-CO2 effects. In future decades and centuries, CO2-levels will be much higher than they are now. In past centuries and millennia, CO2 levels were much lower than they are at present.

For most of the last 10,000 years, carbon dioxide levels stood at around 270-280 ppm (Alley et al. 1993), around a third less at present. 21,000 calendar years ago during the last glacial maximum, the CO2 level was even lower, at about 200 ppm (Alley et al. 1993). Since CO2 is a key factor in the growth of all green plants, these changes in its concentration must have had some effect on plant growth and carbon storage. The difficulty is in estimating how large this effect actually was. Many experiments have been performed on plants growing at higher-than-present levels of CO2 (usually double either the pre-industrial or the present-ambient CO2 level), in order to predict the future effects of the present phase of rapid CO2 increase. The plants are grown in closed or open-topped chambers, and growth-response models are produced on the basis of these results (e.g. Allen et al. 1987). However, the results in terms of CO2-responses are complex and sometimes strongly conflicting, and no-one is sure how such limited (and patently artificial) experiments could translate into the future long-term functioning of real ecosystems on a global scale (Koerner & Arnone 1992, Mooney & Koch 1995, McConnaughy et al. 1993). Generally, the strongest responses to CO2 changes are found in closed systems with crop plants growing for a single season under high nutrient levels without any herbivores or pathogens being present. Experiments on more realistic vegetation microcosms (e.g. tree seedlings and saplings on unfertilized soils, enclosed saltmarsh vegetation) usually reveal a significant positive response, often a 30-40% increase in biomass acculumulation rate, though this tends to decline over a period of several years. The experiments often suggest that even a doubling of the present CO2 level has little or no detectable effect on biomass carbon storage beyond an initial burst of growth and an increase in turnover rate of leaves and roots (Mooney & Koch 1995, Wullshetger et al. 1995). There is at present no simple and consistant quantitative pattern. In certain cases, however, there has been a very strong and lasting CO2 fertilization effect despite the plants having been grown quite a nutrient-deficient soil and exposed in open-topped chambers to the normal array of pests and diseases.

The wide variability of the results in CO2-doubling experiments is worrying, as is the fact that no-one has yet had the chance to take a CO2-fertilized plant community to equilibrium over the timescale of decades or even centuries that we know is important in ecological processes. Many effects in the longer term (e.g. nutrient cycling, diseases and herbivory, internal shading of the growing forest canopy) might either magnify or cancel out the CO2 fertilization effect on carbon storage observed in shorter term experiments lasting several years (Wullschetger et al. 1995).

Despite certain bold attempts to model CO2 effects on the present-day and future biosphere (Esser 1984, 1987), there are grounds for considerable scepticism that at our present state of knowledge we can even approximately quantify the effects of the present anthropogenic phase of CO2 increase on broadscale ecosystem processes (Mooney & Koch 1994, Wullshetger et al. 1995, Amthor & Koch 1996).

And if there are problems in knowing how the future CO2 increase will affect plants, there is even less understanding of the biological effects of the 80ppm change in CO2 levels between glacial and interglacial conditions. It appears that so far no experiments at all have been run to explore how plants might have coped on a year-to-year timescale at a continuous mean level of 200ppm CO2 in the world of the LGM, or at 280 ppm in the pre-industrial Holocene (F.A. Bazzaz, Harvard University, pers. comm. 1995). Observations of transient photosynthetic rates of the leaves of crop plants under short-term CO2 depletion in growth chambers have been used to argue that there would have indeed been significant differences in water use efficiency under the lower LGM CO2 levels, favouring C4 plants over C3 plants (Johnson et al. 1993), but these were transient effects on very artificial high-nutrient systems; and in any case their results were not translated into effects on overall biomass. Combining the data from various sources of evidence, one can perhaps glean a tentative picture of how a lowered CO2 might have affected the LGM vegetation as compared to the Holocene (Robinson 1990). Robinson (1990) has back-extrapolated from the biomass effects of raised CO2 levels in closed chamber experiments, to suggest that at 200ppm CO2 the experimental grassland 'communities' which she was studying would have stored 20% less carbon than corresponding types growing in the present-day (350ppm) world, under otherwise identical climatic and soil conditions. A 20% depletion certainly seems quite a major effect, but there is a great need for caution both in interpreting back-extrapolation, and in accepting the results of a few (highly artificial) closed-chamber experiments as relevant to the global history of vegetation.

Robinson (1994) and Boreshkov (1994, published in Russian according to E. Lioubimtseva, Moscow State University, pers. comm.) have each suggested on theoretical grounds - from back-extrapolation of curves of plant physiological responses to CO2 concentration - that the peculiar combination of species in the 'steppe-tundra' vegetation which existed across Eurasia at around the LGM was largely a product of lower CO2 levels. Many other no-analogue communities have been described from the world of the LGM, and it is possible that they too might have been partly the product of the changed ecological relationships which existed under lower CO2 conditions.

Various other observations have been suggested as indicating that the difference in CO2 levels caused significant differences in plant ecology between the LGM and Holocene. For example, C4 plants may have been more abundant in many tropical plant communities during the LGM. Aucour et al. (1994) have suggested that the dominance of C4 plant species in a peatbog in Burundi during the LGM was due to lower CO2, favouring these plants over the less CO2-efficient C3 species. The plants in this bog were growing under conditions (lower temperatures, but apparently almost constant water table conditions relative to the interglacial) that would instead have been expected to favour a shift towards C3 plants relative to the present (R. Bonnefille pers. comm.). This shift towards C4 plants under conditions that would be expected to particularly favour C3 plants seems difficult to explain without invoking some sort of direct-CO2 effect on plant ecology. Similar tends are noted for glacial-interglacial changes in the proportion of C4 plants growing around the shores and shallows of high-altitude lakes on Mount Kenya and Mount Elgon in east Africa (Street-Perrott et al. 1997).

Other observations have been taken as indicating the effect which lower CO2 levels were having on plant growth during the Last Glacial Maximum. It is generally accepted that stomatal indices (the relative frequency of stomatal guard cells in leaf surfaces) responded to past Quaternary CO2 changes, although this does not in itself show what effect (if any) this had on biomass and vegetation structure. From a study of stomatal density and delta-13C of subfossil Pinus flexilis, Van der Water et al. (1994) have suggested that water use efficiency by this tree species was 15% lower under LGM CO2 levels than under Holocene CO2 levels. Thus, it does at least seem plausible that the 80ppm change in CO2 from LGM to Holocene conditions would have had a significant effect (at least several %) on overall carbon storage by vegetation. However, this does not mean that there is necessarily enough evidence to confidently extrapolate major CO2 effects on vegetation for the LGM, as several recent modelling studies have done.

If the 70-80ppm LGM-to-Holocene shift in CO2 levels was significant in terms of plant ecology and carbon storage, one would expect to find at least some signs that the subsequent 80-90 ppm increase that has occurred over the past 200 years has also had noticeable effects on plant growth. Although there seem to have been effects on stomatal density in various plant species (e.g. Woodward 1987), there is no firm evidence for any significant changes in global plant growth rate or biomass resulting from direct-CO2 effects on plant physiology (Adams & Woodward 1992, Wullschetger et al. 1995). Such evidence would in fact be difficult to obtain convincingly, but it certainly does not leap out from the data on tree rings and other indicators of plant growth and biomass.

The extensive and carefully standardised findings of Phillips & Gentry (1994) on tropical tree turnover rates initially implied that something dramatic was happening at present in forest communities throughout the tropics, perhaps as a direct result of the rising CO2 levels (though it is important to note that Phillips & Gentry found an increase in tree growth rate and death rate, not in biomass). Recent analysis of Philips & Gentry's work (Sheil & Philips 1995) suggests in any case that their result may be no more than a statistical artefact caused by changes in sampling intervals and by sampling error. In experiments on artificial tropical plant communities fertilised with twice-the-present CO2 levels, Koerner & Arnone (1992) found no response in terms of plant biomass or leaf area, and similar results have been found in many cases for experiments on high-latitude ecosystems (see discussion in McConnaughay et al. 1993).

As mentioned above, various growth-response models (e.g. Esser 1984, 1987) have been used by other authors to model past and future carbon storage (e.g. Freidlingstein et al., Prentice et al., Peng et al., Jolly et al., Bird et al.). Such physiological modelling attempts suggest that during the LGM the direct-CO2 effect on both soils and vegetation would have been been very major, perhaps exceeding the effects of climatic differences on global land ecosystem carbon storage (e.g. Friedlingstein et al. 1992, Peng 1994, Peng et al. in press, Bird et al 1994, Farquahar 1997). Yet at present (considering the rather unclear results obtained so far from the various CO2 fertilisation experiments on artificially constructed ecosystems and on semi-artificial enclosed ecosystem studies), such heavy reliance on extrapolated models seems unwarranted. The vegetation growth response models (such as the much-used model of Esser 1984, 1987) utilise a CO2 effect based on back-extrapolation from short-term photosynthetic responses, or from growth experiments which inevitably offer a drastic simplification of a very complex world. For this reason, extrapolation from their results to 'real' ecosystems on a global scale, equilibriating over centuries, may be unwarranted. Recent reviews (e.g. Wullschetger et al. 1995, Amthor & Koch 1995) voice considerable scepticism that useful beta (direct-CO2 fertilization) factors can be forecast for a future CO2 doubling; even greater scepticism should surely be applied to published attempts to quantify similar effects for the recent geological past.

Bearing all these perplexing uncertainties in mind, I have not attempted to estimate direct-CO2 effects on biomass in the 'recommended' values given here. It is frustrating to have to admit that, given the possible importance of this factor, it is actually almost anyone's guess as to how much lower the per-unit-area biomass would have been due to direct-CO2 effects in the past. Analagous problems should perhaps be admitted for forecasting direct-CO2 effects on future ecosystem carbon storage

Overall, given the current state of the evidence for longer term direct-CO2 effects in natural and semi-natural ecosystems, one gets the impression that there is presently a great deal of wishful thinking, in terms of both the forecasting and reconstruction of direct CO2 effects on ecosystem processes. It is assumed by many vegetation modellers that strong direct-CO2 effects on biomass 'must' be present in natural ecosystems in both the past and the future, even if the available evidence offers only tentative support. It is also generally assumed that the question 'must' be answerable and quantifiable by the current relatively short-term experimental approaches; there is little thought given to the (rather defeatist) possibility that questions concerning the magnitude of direct CO2 effects might actually be unanswerable by present experimental and monitoring techniques, due to the spatial and temporal scaling difficulties involved in quantifying direct CO2 effects over broad areas on the timescale of decades to centuries. The problems are surely even greater for attempts to quantify such direct-CO2 effects for thousands of years in the past, when it is so difficult to disentangle the effects of past climate changes. It is not my intention here to discourage the important and necessary work on direct-CO2 effects (if the experiments and monitoring studies are not done, we will certainly not know the answer), but only to add a small amount of healthy scepticism to the interpretation of results, and to suggest that some of the more adventurous modellers are more careful, by adding and emphasizing all the necessary caveats to accompany their bold extrapolations.

Direct CO2 effects on soil carbon. Similar (but even greater) problems to those which occur with understanding direct-CO2 effects on past vegetation carbon storage values also apply to soil carbon storage. Some clues to the way in which lower past CO2 levels might have affected soil processes can be gleaned indirectly from discussion in the literature on the effects that future raised CO2 levels might have on soils (e.g. Wullschleger et al. 1995). Under lower-than-present CO2 levels during glacial periods, and during the preindustrial Holocene, a lower photosynthetic rate of vegetation could have meant changes in the net flux of primary production reaching the soil as dead leaves, roots, branches etc. Quite possibly, even where there was no change in living plant biomass, the result of lower CO2 in terms of slower turnover time of plant organs such as roots would constitute a substantial reduction in the amount of organic carbon to the soil, without any corresponding increase in microbial decomposition rates. For example, some experiments on CO2 fertilisation in artificial tropical microcosms have found that lower (present-day) CO2 levels give slower turnover rates of fine roots than under CO2 levels above 600ppm (Mooney & Koch 1994). Many experiments on both tropical and temperate plants (Mooney & Koch 1994) also indicate that at lower CO2 levels, the root mass is reduced much more than the aboveground material; this might have implications for the supply rate of organic matter directly into the soil from dead root material.

Subtle changes in the carbon-to-mineral ratios in the plant materials reaching the soil surface could also have had far reaching effects on the levels of long-lived carbon in soils. There might for instance have been a lower ratio of carbon to minerals in the soil litter (due to relative carbon starvation of the plants), promoting more rapid fungal and bacterial decomposition. This would in turn have given soils that were poorer in carbon, giving a lower global soil LGM carbon storage than one would expect simply from mapping the past ecosystem distribution and applying present-day soil carbon storage values. However, such scenarios are only a matter of pure speculation on my part. In truth we can have very little idea of what the effect might have been. Experimental systems that manipulate CO2 levels seem to give little clear indication of what we should expect to have happened to soil carbon storage. For example, in their experiment on several artificial tropical ecosystems exposed to high CO2 levels, Koerner & Arnone (1992) found a decrease in soil carbon at higher-than-present CO2 concentrations (i.e. lower CO2 = more soil carbon), suggesting an effect opposite to that which would generally be expected.

Various more ambitious experiments are currently under way around the world to simulate the responses of particular ecosystems to raised CO2 levels of 600ppm or more (Wullschleger et al. 1995), but the short-term results in terms of soil carbon storage seem equivocal. Furthermore, there is no relevant evidence in the literature on the effects on soil carbon of these past increases in CO2, either the preindustrial-to-the-present or the LGM-to-the-preindustrial-Holocene. Ecologists have enough trouble struggling to understand the effects of past or future changes in CO2 levels on vegetation growth, and they appear to know even less about the long-term effects on soil carbon density (indeed the problem does not even seem to have been explicitly discussed within the literature). All that one can say is that there may have been a significant effect from the low CO2 levels, lowering LGM soil carbon storage relative to Holocene carbon storage, but that we do not know how large this influence was.

Everything taken together, it does seem quite likely that the direct physiological effect resulting from an 80ppm glacial-to-interglacial or a preindustrial-to-present change in CO2 did cause a substantial change in both biomass and in soil carbon storage. However, it also seems quite likely that it had almost no effect on these ecosystem attributes. The evidence, at present, is simply inconclusive and it is unfortunate that this is not more openly admitted and discussed by many of the modellers when they put forth their global extrapolations.

Human intervention. It is generally accepted that the intensity of human interference in most ecosystems has increased enormously over the past several millennia, and especially the last 3,000 years (Tallis 1990). Sifting through reports of carbon storage data for soils and vegetation as a source of data for an earlier mid-Holocene or late-glacial state, it is important to focus on sites in areas that have apparently not fallen under the plough or axe within recent centuries. Where no such data are available, old relatively undisturbed sites must be studied. However it is also important to bear in mind that all ecosystems are subject to natural disturbance. In this sense the 'oldest' undisturbed sites are not necessarily representative of the preanthropogenic state which existed in the past, if they have been artificially protected from all major disturbance factors. The aim is to find a representative point somewhere between these two, but the process of finding it may require a fair amount of intelligent guesswork.

In a sense, one is searching for 'natural' vegetation, but the very concept of 'naturalness' is itself elusive and confused. For example, if humans have lived in an area for hundreds of thousands of years, are they a natural feature of the ecosystem? Archaeological evidence in many parts of the world shows that there must have been at least some direct or indirect human influence throughout the late Quaternary. In some areas (e.g. Africa, Australia), humans seem likely to have been modifying the vegetation by burning for tens of thousands of years, and possibly more than a million years in the case of Africa (Tallis 1990). Historical records show that African, Australian and North American aborigonals used fire as an important aid to hunting at the time of first documented contact with Europeans (and so had in all probability been using it for many thousands of years beforehand) (Stewart 1956).

Present-day 'natural' vegetation may also be lacking another indigenous component of the system, in the form of dense populations of natural herbivores which would have grazed the vegetation and kept its biomass down. For instance, non-anthropogenically influenced elephant and rhino populations may have a very destructive effect on the woody vegetation where they live (Kortlandt 1982). Over large areas of the world (e.g. Eurasia, North and South America), most of the herbivores which would have existed during the last glacial phase are now completely extinct (possibly driven extinct as a result of human hunting) (Martin & Klein 1986). These past populations of herbivores might have had an important role in maintaining glades and other open areas within forests and woodlands, thus reducing overall biomass and soil carbon storage. For instance, one can speculate that by analogy with present-day Africa, the forest, woodland and steppe-tundra elephants that existed in North and South America and Eurasia during the LGM and up until the earliest Holocene were important in keeping reducing woody cover and creating patches of bare ground (Martin & Klein 1986, Owen-Smith 1988, Tallis 1990). However, in at least some areas the 'natural' herbivores may now have been replaced in approximately equal measure by herded domesticated animals (Owen-Smith 1988).

Humans can also play a more direct role in other, quite surprising ways. For instance, the subtle influence of indigenous peoples of South America in encouraging the establishment of groves of useful forest trees is becoming increasingly clear (L. Rival pers. comm., G. Mombiot pers. comm.; 1994. work as yet unpublished), and in Central America fruit trees are still unusually abundant in the forests of the lowlands abandoned by the Maya several hundred years or more ago (F.A. Street-Perrott, pers. comm. 1995). Yet it is doubtful that overall carbon storage would have been much affected by such processes, except perhaps in the initial rebound phase after a dry episode when the forest is gradually spreading back again over large areas of grassland.

If the aim of one's work is only to reconstruct carbon storage for particular time intervals during the late Quaternary, the dilemma about 'naturalness' should in principle vanish. If for example humans were burning the vegetation 18,000 years ago, then all well and good, one can incorporate that influence into the calculation for vegetation carbon at the LGM. Thus there is no need to be concerned about whether it is 'natural' or not. Yet in practice, such questions of 'naturalness' remain all-important because one must often guess at how strongly this effect has varied over time without human influence in the 'modern' sense. In fact, the inventory compiled here may tend to make automatic allowance for such pre agricultural disturbance factors, because the areas that are currently referred to as 'natural' do sometimes retain a low element of disturbance by indigenous human populations.

The importance of relatively recent intensification of the human disturbance regime is only just beginning to be appreciated. As Harmon & Hua (1992) have found, temperate forests that have not been felled for several centuries accumulate surprisingly large quantities of carbon in dead and moribund trees. Yet, almost all of the temperate forests we see in the present world, from which published carbon storage values are derived, have been subject to wood extraction certainly for hundreds and more probably thousands of years. The same may be true of many areas of tropical forest; Brown et al. (1991) and Brown & Lugo (1992) find evidence of a subtle but very significant depletion of the standing biomass of rainforests and dry forests over much of south-east Asia as a result of centuries of shifting cultivation and selective logging. They also suggest (Brown & Lugo 1992) that many forests in relatively accessible parts of Amazonia which had been thought of as being pristine have in fact been selectively logged during the past few centuries, so that biomass inventories from these forests give a misleadingly low impression of the carbon stock of more 'natural' precolonial forest.

In this inventory, the figures given for the temperate and tropical forest zones are all for stands known to be over a century old since the last clear cutting or major disturbance event. Thus, many of the vegetation types which are presented as if they are still more-or-less natural (in whatever way one chooses to define 'natural', whether in historical terms or not - see above) according to IBP data sources are undoubtedly significantly altered from their natural state, in terms of species composition, age structure of trees, and the amount of organic debris. For example, it is almost a truism to state that virtually all of the surviving temperate forest communities of Europe and North America have been greatly altered by their history of woodcutting and arboriculture; over large areas of forest it is hard to find any trees approaching old age (pers. obsv.n by the author). Likewise, in many areas around the Mediterranean Basin, humans are known to have greatly increased the frequency of destructive fires during the past few millennia, changing the woody cover from mainly deciduous forest to evergreen scrub and scrub-woodland (Laval et al. 1991, Willis & Bennett 1994).

For such reasons it has been necessary here to go back to some of the primary sources of data, and also to use data sources which have been published since Olson et al.'s (1983) work, to critically select those particular forest sites where there is reasonable historical evidence of a low intensity of human disturbance over the last century or more. This might give the superficial appearance of overlap in the use of the literature; some of the data sources and results cited in detail here will already have been included in Olson et al.'s survey of the literature. Their inclusion as separate citations here is partly intended to emphasise those sources which are likely to more closely reflect 'mid Holocene' rather than recent forms of anthropogenic vegetation.

For many other biomes, similar uncertainties remain within the Olson et al. figures. Again, data has been selected from sites which appear to have a long history relatively free from anthropogenic disturbance (although not necessarily free from natural disturbance, such as natural grazing or fires, see below).

Broad-scale disequilibrium in soil and vegetation carbon storage. Just as small-scale disturbances tend to throw back the process of carbon accumulation in ecosystems, one can imagine global-scale changes in both the past and the future having a similar effect. There is some circumstantial evidence of this from Quaternary vegetation history. Disequilibrium in species migrations and in soil development may have been important in producing some of the 'no-present-analogue' species assemblages (discussed above) that occurred during the late glacial and early Holocene (Adams & Woodward 1992), when the Earth's climates and ecology were changing fastest. Ecological disequilibrium in vegetation, particularly in forest vegetation, may have prevented maximum vegetation carbon storage from being reached for thousands of years after the climate initially became suitable for it in many areas of the world, in both temperate and tropical environments (Adams & Woodward 1992).

The lasting effects of anthropogenic disturbance on soil carbon storage have already been touched upon here. Likewise, there is evidence from well-dated studies that there may be a natural disequilibrium in soil carbon lasting thousands of years following climatic changes or other more localised disturbances in the environment (Schlesinger 1990). In addition to the examples of increasing carbon storage cited by Schlesinger, another more recent example found by Schwartz (1991) shows how the 'imprint of the past' can persist in a soil's carbon reservoir for thousands of years. Schwartz found that central African savanna soils still contain some carbon at depth that bears the isotopic imprint of forest vegetation. From dating this carbon it seems that these areas were covered by forest during the early Holocene, and that this relatively small deep soil reservoir still persists thousands of years later after the forest has retreated. One should ideally take into account the possibility that the soil carbon density we observe in natural sites nowadays might differ greatly from the levels back at 8,000 years ago or 5,000 years ago, when the soil carbon might not have had as much chance to equilibrate with the vegetation conditions. A similar situation could exist in a future scenario of climate change. Likewise, at the LGM there might have been some form of disequilibrium in soil development that would have affected its carbon content. One should consider however that the slide into glacial conditions took thousands of years, and that at least part of the climatic amelioration towards interglacial conditions began well before the start of the Holocene proper.

Despite such concerns about disequilibrium in soil carbon storage following past or future environmental change, it seems that at many sites most of the humic carbon entering such soils does so within the first millennium after formation (Schlesinger 1990, and see discussion in Adams & Woodward 1992), and very often within the first few decades. Thus the longer-term lag will probably not be especially great as a proportion of the total carbon in a soil. This general picture of a very rapid initial build-up of soil organic matter following a change in circumstances is found in a great many studies from around the world, in many different sorts of vegetation.

To take just one fairly representative example, in the classic Rothampstead experiments in England where arable land was allowed to revert to deciduous temperate woodland, soil organic carbon increased 300-400% from around 20 t/ha to 60-80 t/ha in less than a century (Jenkinson & Rayner 1977). The rapidity with which organic carbon can build up in soils is also indicated by examples of buried steppe soils formed during short-lived interstadial phases in Russia and Ukraine. Even though such warm, relatively moist phases usually lasted only a few hundred years, and started out from the skeletal loess desert/semi-desert soils of glacial conditions (with which they are inter-leaved), these buried steppe soils have all the rich organic content of a present-day chernozem soil that has had many thousands of years to build up its carbon (E. Zelikson, Russian Academy of Sciences, pers. comm., May 1994).

However, there is some circumstantial evidence that the slowness of soil development may have retarded vegetation colonisation of many formerly glaciated or barren areas, for as long as hundreds or even thousands of years. Possible clues to the importance of this effect include the surprisingly slow rate of recolonisation of deglaciated landscapes by local tree populations after sudden warming events in northern Europe and North America at around the beginning of the Holocene (Pennington 1977). As was mentioned above, Magri (1994) has found slow exponential rises in pollen input to an enclosed lake basin in central Italy, for relatively constant species composition, taking thousands of years during which the sites were apparently being recolonised by vegetation following disturbance events. She suggests that this pattern might be due to lags in vegetation build-up resulting from slow soil maturation and nutrient limitation (Magri 1994). It is also important to bear in mind that carbon storage could have been affected by more subtle and undetectable differences in vegetation structure that might have persisted in many ecosystems that formed on previously barren surfaces in the high and low latitudes following the last glacial cold period.

There can be no doubt that disequilibrium in carbon storage is especially significant in the case of peat build-up. There is abundant evidence that this process can continue at more-or-less the same rate for many thousands of years, adding incrementally to a waterlogged column of almost pure organic matter which can reach many metres in thickness (Clymo 1984).

Trying to bring together all the factors that can affect such time-related changes in carbon storage is a virtually impossible task, and one has to stop somewhere. It seems most reasonable to suggest that in the late Quaternary a broad state of equilibrium in vegetation and non-peat soil carbon storage with the then existing climate had been reached by around 8,000 years ago (early Holocene), with carbon values broadly equivalent to those for present-natural ecosystems. At least, the evidence does not strongly point to most terrestrial ecosystems being a long way from a general 'equilibrium' (albeit an equilibrium representing the average of many small disturbance events, against the background of a continually fluctuating climate) in terms of biomass and soil carbon (although the story for peatlands is a very different matter). This is based on the assumption that in most regions of the world where forest recolonisation had taken place there had been enough time at least for several tree generations to pass, and also on the observations such as Schlesinger's (1990) that soil carbon maturation curves tend to 'plateau out' after a couple of thousand years. Even so, it is necessary to bear in mind that North America 8,000 years ago still had very large and rapidly retreating ice masses which seem to have given a broad band of relatively immature vegetation and soil carbon storage in the zone around their perimeters, where recolonisation and ecosystem development was still taking place (Harden et al. 1992). Given that the previously published prehistoric global soil carbon storage estimates for the Holocene focus on time intervals where climate and general vegetation structure had already been relatively stable for well over a millennium (at 8,000 years ago or later), it seems reasonable to suggest that in most areas soil organic carbon was more-or-less in the equilibrium state that we would define on the basis of undisturbed soils we see today.

The resulting errors. Many of the factors discussed above bring with them uncertainties which one must be aware of in trying to quantify past or future changes in carbon storage.

Because of the difficulties of allowing for no-analogue factors such as changed CO2 levels or different combinations of climatic parameters, the estimates given here are based only on the likely potential steady-state carbon storage under 'present-day' conditions. For the 'present-day' world, the potential role of natural disturbance factors are not at all easy to allow for, although (as I have argued above) in terms of the biomass of many forest types certain factors can be dismissed as relatively unimportant in terms of affecting steady-state carbon storage (e.g. large-scale wind disturbance events) whilst others remain to trouble us (e.g. carbon dioxide effects, the influence of natural and anthropogenic burning, the ambiguities in definition of vegetation types). For working purposes, an error limit of +/- 30% is suggested. Note that this figure is not based on any actual statistical calculation, due to the difficulties of quantifying the effects of the various uncertainty factors. A +/- 30% error would give a total range, for a 'preferred' estimate of 100 tC/ha, of around 60 tCha (extending from 70 to 130 tC/ha); clearly a wide error bar. However, in a changed no-analogue world of the past or the future fact it might not be broad enough. In the end, each reader must read the linked data tables and text to form his or her opinion on whether I have properly allowed for such factors in the final values that I recommend for potential ecosystem carbon storage.

Visit the QEN Atlas of Quaternary global ecosystem maps