This paper is 'in press' in Global and Planetary change. The text presented here is an older version of the manuscript (tables and figures are not included here). It is intended to give a 'taster' of the final published paper which should be out within the next couple of months; various parts have been condensed and updated since this earlier version.

A NEW ESTIMATE OF CHANGING CARBON STORAGE ON LAND SINCE THE LAST GLACIAL MAXIMUM, BASED ON GLOBAL LAND ECOSYSTEM RECONSTRUCTION

J.M. Adams (1,2) and H. Faure (1)

(1) Laboratoire de Géologie du Quarternaire, CEREGE, Europole de l'Arbois, B.P. 80, F-13545, Aix-en-Provence Cedex 04, France.

(2) MS 6335, Environmental Sciences Division, Bldg 1000, Oak Ridge National Laboratory, Oak Ridge, TN 37831, USA.

Introduction

Polar ice cores indicate that the levels of greenhouse gases in the Earth's atmosphere have closely paralleled the changing climate over at least the last 250,000 yr. (e.g. Neftel et al. 1988, Barnola et al. 1989, Chapellaz et al. 1990). There is presently widespread interest in understanding the controls on these changes in atmospheric composition. Work has concentrated on producing ocean models (reviewed by Broecker & Peng 1993) to explain the larger amounts of carbon sequestered in the glacial-age ocean. More recently, however, there have been various attempts to provide a more complete picture by estimating how land carbon storage might have changed between glacial and interglacial conditions (e.g. Faure 1989, Prentice & Fung 1990, Adams et al. 1990, van Campo et al. 1993).

One such study that included both of the present authors was that of Adams et al. (1990). The present study is intended as a follow-up to our earlier work, including a much expanded literature and palaeo-ecosystem database, and an expanded and more carefully considered carbon storage database. Much new source material on both palaeoenvironments and on ecosystem carbon storage has been published in last few years, and this study has taken such evidence into account to provide a more reasonable assessment of how organic carbon storage on land has changed since the last glacial maximum. It is also our intention to better discuss and quantify the sources of uncertainty which may contribute to the overall picture.

Approaches towards estimating the glacial-to-interglacial land carbon shift

In terms of quantifying the glacial-to-interglacial shift in carbon storage on land, most attention has so far concentrated on the changes in the carbon cycle during the last half of the most recent glacial-interglacial cycle (between the last glacial maximum around 20,000 calendar years ago and the Holocene interglacial that began around 11,000 calendar years ago) because this is the period of time for which the most data are available.

Three main methods.

Three main approaches have been used to estimate how the amount of organic carbon on land may have varied between glacial and interglacial conditions. Each of the three methods - ocean isotopes, GCM-based reconstruction, and palaeoenvironmental evidence - tends to give a different estimate of carbon storage shifts, and each has its own likely benefits and drawbacks (Maslin et al. 1993). Both the ocean isotope method and the GCM-based method have been placed in very serious doubt (e.g. Crowley 1994a,b), at least in their present state. In this paper the discussion will not consider the relative merits of estimates obtained from each of the three different sources, or suggest ways in which their differing results might be reconciled. However, given the considerable difficulties that are emerging with accepting both the GCM-based and ocean-isotope based methods, the relatively direct paleoevidence method offers the best hope of reaching a figure for the glacial-interglacial change in carbon storage.

The approach used in this paper is to base the carbon calculations on direct and indirect evidence of palaeovegetation cover. Here, the work will concentrate on a new and up-to-date set of estimates based on the reconstruction of global palaeovegetation using a range of sources of information about land palaeo-environments.

The procedure used in this study.

Stage 1: Providing palaeovegetation maps.

In the present-day world, there are clear relationships between particular vegetation types and the amount of carbon stored in the vegetation and soil. Thus, it seems reasonable to suppose that if one knew what the world's vegetation was like at a particular time in the past, one could estimate the total amount of carbon on land.

The choice of time slices.

The time slices studied here are intended to represent the extreme states of land carbon storage during a glacial-interglacial cycle. It selecting these time slices is important to estimate carbon storage for time slices when the vegetation is known in detail and in which there is a fair probability of an equilibrium in terms of broad scale ecosystem maturation. 18,000 radiocarbon (14C) years ago (about 21,000 calendar years ago) is the extreme of the glacial state by most indicators (e.g. global ice volume, dust flux to the oceans; see Crowley & North 1991), and the time when CO2 levels were at their lowest (although some indicators suggest that the global average aridity maximum was slightly shifted in time from the Last Glacial Maximum, to about 17,000-15,000 14C years ago, and by combining some rather loosely dated evidence for the most arid conditions in the general interval of 19,000-15,000 14C years ago, the LGM maps may in fact be more representative of this slightly later time frame). 8,000 14C years ago (9,000 calendar y.a.) is early enough to serve as an example of the moist, warmer-than-present early Holocene state, but late enough to be a time at which most ecosystems seem to have reached a rough successional equilibrium (e.g. south-east Asian rainforest areas were no longer dominated by early successional species as had been the case during the earliest Holocene). The 5,000 14C years-ago (5,900 calendar y.a.) value represents a later phase of the Holocene, at which climate and vegetation were somewhat more similar to its present state, yet still lacking large-scale agricultural impact (Tallis 1990). A 'present-potential' set of maps and values is also presented here, not because this represents what actually existed at any time in the past (it does not), but because it makes a convenient and relatively familiar 'standard' against which to compare the past.

At the most basic level, there is still considerable uncertainty over the distribution of vegetation at any particular time in the past, especially under the drastically different global climates of the last glacial period (COHMAP 1988). Various attempts are currently being made to collate pollen evidence from around the world (e.g. Prentice & Webb 1994), but the necessary plant fossil data are often sparse and ambiguous, especially for the LGM itself. In order to reconstruct global palaeovegetation distributions, one must use not only of pollen but all other sources of relevant data. The ecosystems maps upon which the carbon storage calculations in this paper are based were produced by J.M. Adams & H. Faure in a literature survey and in a system of voluntary data exchange with a large number of Quaternary experts from around the world, together forming the Quaternary Environments Network or QEN (Adams 1995, Adams & Faure in press, and at the World Wide Web site; QEN 1995). On the basis of this data gathering and exchange process, it has been possible to arrive at a preliminary set of global maps for three time slices since the Last Glacial Maximum. By incorporating multiple sources, including many different published reviews as well as direct comment from over a hundred specialists, these maps would seem to combine a greater amount and range of data and diverse expertise than any previous set of palaeovegetation maps.

Deducing the areas of global vegetation types.

On the basis of the QEN maps (Fig. 1a-g), areas of each of the 29 ecosystem categories (and of several sub-categories, making 35 categories in all) were calculated. Areas were calculated by tracing vegetation map polygons onto graph paper, counting the squares and multiplying the result by a scaling factor. The accuracy of this approach (which was repeated several times and the results averaged) was aided by the fact that the vegetation maps for each continental region were drawn onto the central parts of near-equal-area atlas base maps (Lambert Equivalent Azimuthal Projection). This method appears to give good overall agreement with the actual areas of continents and total global land area (error less than 3%) by reference to the 'actual' figures presented in the Times World Atlas (Times 1992) figures.

Open water areas, such as small boreal lakes, are subtracted from the total area of the vegetation biome in which they occur. Note that the beds of lakes are often very rich in organic carbon, and this itself may represent a substantial increase in continental carbon storage between glacial and interglacial conditions: however it will require a separate study to quantify this carbon reservoir. In the present study, estimation of lake areas has been attempted here through reference to very large-scale maps showing all lakes >/= 1kilometre in diameter. Estimated on the basis of the 'counting squares' method, the resulting 'sectors' of equal lake coverage during the Holocene are shown in Fig. 2 a and b. The lake areas were then subtracted from the total area of the vegetation type within which they occurred.

The results for each of the main ecosystem types (each corresponding to the maps presented in Fig. 1) are presented in Table 1. The resulting 'preferred' (considered most likely on available evidence) global area of each vegetation type, for each time slice, is given in the unbracketed values in Table 1 with some further modifications for lake and swamp areas (see below) included in these totals.

It is difficult to relay the degree of confidence to be placed in the palaeovegetation areas presented in these maps. It is in a general sense possible to give an impression of how seriously one takes the uncertainties in the information, by giving 'upper and lower limit' scenarios for the range of vegetation areas and carbon storage values that might result from different assumed vegetation distributions at the LGM, 8,000 14C years ago and 5,000 14C years ago (Table 1). In the 'low carbon' scenario, most of the rainforest present in the 'medium carbon' scenario is assumed to have been replaced by savanna and grassland vegetation, more towards the view that certain authors such as Clapperton (1993) have suggested in the light of existing data. Also, large areas of what may have been mostly woodland and monsoon forest in South-east Asia at the LGM are given over to savanna and grassland. For the 'high carbon' vegetation distribution, a relatively large proportion of the present area of rainforest was already in place at the LGM, as Colinvaux (1987) has suggested for Amazonia and possibly other areas. Likewise, given the uncertainties in present data, relatively large areas of woodland and monsoon forest, and non-desert vegetation, are allowed in this 'high carbon' scenario.

Stage 2: Adding carbon storage values onto the maps, using a carbon storage database/review.

a) Per-unit area carbon storage.

To estimate total global carbon storage, it is necessary to have realistic estimates of how much carbon would have been present in the soils and vegetation of an average hectare of each particular ecosystem. These estimates can only be obtained by studying vegetation as it exists in its present, often very altered, form.

To improve the likelihood of arriving at realistic estimates for the past, it is necessary to bear in mind the ways in which the detailed structure of past vegetation and soils might have differed from the present. One must consider, for instance, the ways in which forestry and shifting cultivation have altered the carbon storage of forest ecosystems by affecting the density of trees and their overall sizes (Brown & Lugo 1992). Since these human influences were often absent or much reduced before the mid-Holocene, one must search for data from sites that have not recently been disturbed by humans, and use this preferentially. On the other hand, there are other disturbance factors that might be underestimated from a purely present-day perspective on the world. These include the influence of wildfires, which are often controlled by foresters to protect valuable timber resources. The broad scale disequilibrium which results after a disturbance event on the scale of a glaciation is another factor that should be considered (Schlesinger 1990); the carbon in ecosystems that replaced polar deserts did not appear overnight.

No-analogue factors.

The most intractable problem in estimating per-hectare carbon storage values for past vegetation is in allowing for 'no-analogue' factors which could have altered ecosystem distribution, structure and carbon storage in ways that can only be approximately understood. For instance, many vegetation types in the past contained combinations of plant species that do not normally grow together in the present-day world. The 'steppe-tundra' which covered much of the high latitudes at the LGM is a notable example (Tallis 1990, Ritchie & Cwynar 1982), but other striking instances appear in many places in the plant palaeontological record.

There are various aspects of the physical and biological environment of the past which can been seen as being qualitatively dissimilar to the present, in this sense providing a no-analogue environment which might have altered ecosystem carbon storage in unknown ways. A much-discussed example is the influence of lower-than-present CO2 levels in the atmosphere of the Holocene, and especially during the last glacial maximum. As mentioned in the Introduction to this paper, for most of the last 10,000 calendar years, carbon dioxide levels stood at around 270-280 ppm (Alley et al. 1993), around a third less than at present. At 21,000 calendar years ago (18,000 14C y.a.) during the last glacial maximum, the CO2 level was even lower, at about 200 ppm (Alley et al. 1993). A considerable amount of discussion and speculation has focussed on models of past ecosystem responses to lower carbon dioxide levels (e.g. Robinson 1994). Previous attempts to deduce past CO2-induced changes in carbon storage as a result of back-extrapolation from the results of closed chamber experiments on raised CO2 effects on well-fertilized and disease-free crop plants (Esser 1984, 1987) could well be in error, because these conditions are so far from those experienced by most natural plant communities. It is extremely difficult to know what effect rising CO2 is having on the natural ecosystems of the present-day world (Koerner & Arnone 1992, Allen et al. 1987, Koerner & Arnone 1992, Mooney & Koch 1995, McConnaughy et al. 1993), and even more difficult to back-extrapolate to suggest what the effect of lower CO2 levels was in the prehistoric past. Given that the rapid rise of 80-90ppm CO2 (of the same order as the glacial-interglacial increase) over the last 200 years has provided no clear evidence of a direct-CO2 effect on productivity or biomass (Adams & Woodward 1992, Shiel & Phillips 1995), it seems unwarranted at present to even attempt to make any quantitative allowance for the effect of past changes in CO2 level. This may mean that the calculations presented here underestimate the true size of the glacial-to-interglacial increase in land carbon storage, which may be expected to have affected LGM ecosystems more severely than Holocene ones.

Another factor that must be allowed for in arriving at the ecosystem carbon storage figures is the role of anthropogenic disturbance. Many of the sites used in the standard database of Olson et al. (1983) were from areas known to be strongly degraded by human activity. Returning to the original reference sources for stand age in temperate forest (Cannell 1982, Brown & Lugo 1992), one finds much higher average biomass accumulations in forest stands over 100 years old (Adams 1995), even allowing for storm and fire disturbance rates (Huston 1994). The carbon in litter and dead standing trees is also neglected in most literature studies (including Olson et al.), but these are often a major carbon reservoir in non-anthropogenically disturbed forests (Harmon & Hua 1992, J.S. Olson pers. comm. 1996). The generally used soil carbon database of Zinke et al. (1985) also requires some critical re-appraisal. Examination of their data suggests that their very high values for desert soil organic carbon storage (60 tC/ha or more) are biased by a handful of values from the North American 'desert' zones and the southern Russian dry steppes, which do not qualify as true desert on a global scale because of their dense vegetation cover (Adams 1995). Schlesinger (1922) has also found much lower values (around 22 t organic C/ha) for the warm temperate semi-desert zones of North America. Furthermore the uncritical application of Zinke et al.'s 'tundra' and 'steppe' carbon storage values to the LGM steppe-tundra realm has seriously biased many previous studies of glacial-age carbon storage. Here, on the basis of a wide range of palaeofloristic evidence (e.g. high percentage of Artemisia, occurrence of desert halophytes) and the very poor soil development evident in preserved steppe-tundra sections (Zelikson pers. comm., Jalut pers. comm. 1994, QEN review 1995), a low soil carbon content approximating to a present-day dry steppe or semi-desert is used.

On the basis of the accumulated results of this reconsideration, discussed in detail by Adams (1995), a revised set of per-hectare carbon storage values is presented in Table 2, with emphasis on 'pre-anthropogenic' ecosystem storage (including litter and dead tree carbon storage data) and with a re-examination of the bioclimatic zones to which the Zinke et al. data actually correspond. We regard these as 'preferred' values, which are used for our palaeo-carbon storage reconstructions. Given the overall uncertainties in allowing for sampling and past differences in disturbance regime, the per-hectare carbon storage values given in Table 2 are varied by 30% to give 'high' and 'low' boundaries that are used below to calculate the limit scenarios, with a wider error margin (50%) allowed for the no-analogue steppe-tundra ecosystem.

Estimates of peat mass changes.

Global peat at the LGM.

Klinger (1991) suggests that a great deal of peat was present in high-latitude peatlands at the LGM, and that this carbon has since re-entered the global carbon cycle. A problem with this hypothesis seems to be a lack of good well-dated data sources and a priori it does also seem fairly implausible, considering the strong evidence of drier-than-present conditions across the world's landmasses. Dry conditions in the mid- and high-latitudes during the last glacial are indicated by the prevalence of such open-ground plants as Artemisia, skeletal poorly developed soils, and the rarity of preserved peats (or indeed any carbon-rich material at all) of that age. Where peats of LGM age do occur on the continents, they tend to form only isolated lenses, with no indication of anything like the great sheets of peat of the present-day world having existed at that time (West 1978, and E. Zelikson, V. Astakov, C. Kreminetski pers. comm.). Localised peat bogs dating to around the LGM itself have been found in many tropical mountain areas around the world (where they provide useful pollen evidence), but these are nothing compared to the vast peatlands that make up most of the present-day Holocene mass of peat.

Klinger (1991) also emphasizes the possibility that peats could have accumulated on the exposed shelf areas of tropical south-east Asia during the last glacial, pointing to undated records of peats being trawled up from the sea bed there. Peats of the lowland Congo and Amazon Basins and Malesia appear to date back no further than the early Holocene, wherever they have been sampled and dated at the base of the column (e.g. Morley 1981, Radjagukuk 1985). Ziegler (pers. comm., Univ. Chicago, May 1994) has supplied us with with a list of published records of submerged shelf peat, collected from the literature on South-East Asia. Most of the records are undated, and those for which there are dates (radiocarbon or otherwise) all fall around 10,000-11,000 14C years ago or around 30,000-36,000 years ago, not anywhere close to the LGM.

Furthermore, there are a priori grounds for suspecting that peats would have been less abundant than at present in most areas at the LGM. A recent global study of present-day peatlands (Lottes & Ziegler 1994) has shown that the seasonal water table fluctuation in many areas of the moist tropics is enough to prevent peat from accumulating under present climates. For example, there seems to be very little peat in the central Amazon Basin - despite its moist climate and abundance of swampland - because the water levels fluctuate markedly in response to seasonal inputs from the periphery of the basin (Lottes & Ziegler 1994). Since all the signs are that LGM moisture conditions would have been significantly more seasonal or variable than at present, even in those areas where rainforest is known to have survived, it seems very unlikely that there would have been any substantial accumulation or survival of tropical peat under LGM conditions.

On the basis of the available evidence, our peatland estimates assume that there was a negligible or very minor peat reserve at the LGM.

Calculation of Holocene peatland buildup and spread.

Peat mass seems to have built up both horizontally and vertically during the Holocene in many regions of the mid and high latitudes. Neustadt (1984) suggested that horizontal spread has been dominant during the later Holocene (as seems to have been the case in European peatlands Tallis 1990), but later work suggests that much of the area of peat bog now present in west Siberia already existed in the early-to-mid Holocene (Velichko et al. 1995, and this volume). In North America (Nichols 1969), a progressive spread at the periphery of the zone is well supported, but this is less the case in the interior of the continent. A precise reconstruction of aerial rate of spread of peatlands during the Holocene is presently impossible, considering the lack of systematically gathered data within the literature.

The area presently covered by peat bogs, as opposed to other forms of carbon-containing ecosystems such as tundra or closed forest, remains a further uncertainty that is difficult to allow for in calculating carbon storage. It is of course important to subtract the area of land surface covered by peat bog from that covered by better-drained vegetation types, or one would end up counting the same area twice. There is in fact no clear and consistently used definition in the literature of exactly how one should distinguish peat bogs from other ecosystems. Here we have made a rough estimate of present percentage peat coverage in North America (10% peat bog coverage for the total area of each of the areas covered by boreal and tundra ecosystems) from the maps of Tarnocai (1990), and made a similar estimate for northern Eurasia (50% coverage by bogs for the large west Siberian and Arctic peat basins, a figure suggested by Kreminetski [pers. comm.]. For the areas to which this calculation applies see the areas indicated in the maps of Gore 1983, and Lottes & Ziegler 1994). These estimates for peat coverage have allowed the subtraction of the estimated areas of peatland from the non-peatland areas in our calculations, reducing the estimates of carbon held in each of these other ecosystems by the appropriate amount). Note that these estimates of peatland cover are within the general range found in other recent studies (e.g. by Gorham 1992). As mentioned above, for earlier stages during the Holocene, the peatland cover was probably slightly less than at present.

For the tropics, the actual area and depth of peatland remains largely unknown (Walter 1971, Radjagukuk 1985). For this reason it has been necessary to exclude tropical peatlands from our estimates (instead ascribing to these areas a 'normal' forest and soil cover). This probably leads to the present study underestimating the carbon that has accumulated in the tropics since the early Holocene, although the underestimate will be compensated to some extent by the thinner forest cover which tropical swamp areas tend to have (Walter 1971).

Given the inherent uncertainties and complexities of estimating the detailed time course of northern-latitude peat deposition from its beginnings in the early Holocene or at the end of the late glacial, a simplified scenario must be used here. Rather than attempting to reconstruct the rate of spread, we have assumed here only a slightly reduced peatland area for the 8,000 and 5,000 14C year-ago time slices. By subtracting too much area, this may tend to underestimate vegetation/soil carbon storage at each of these two time frames. Here, for simplicity, a simple linear buildup of global peat with time (following intitation 14C 10,000 years ago) is assumed, giving an addition of 46 GtC (with plausible error scenarios of +/- 30%) for each thousand 14C years that has passed since the beginning of the Holocene. For the two Holocene time intervals which are examined here, 8,000 years ago and 5,000 years ago, this gives a 'preferred' total peat reservoir of 92 Gt and 230 Gt respectively.

In addition to their peat, peatlands have some living vegetation on the top. Boreal peatlands generally have quite a sparse vegetation cover, apart from the surface layer of sphagnum mosses. Where a tree cover does occur, it is generally sparse and stunted (Olson et al. 1983). A fairly low mass (20 tC/ha) in vegetation (Olson et al. 1983) has been added to the peat carbon estimates in Table 3 to allow for this.

Results:

Global estimates for the LGM-to-Holocene shift in soil and vegetation carbon storage.

The following estimates are obtained from the reconstructed ecosystem maps (Fig. 1 & Table 1), together with the preanthropogenic carbon storage values suggested here (Table 2), with due allowance for lake areas (Fig. 2 a,b), topography and the rate of peat build-up during the Holocene. From the global calculated carbon storage totals (Table 3) on the basis of 'preferred' vegetation areas and 'preferred' per-hectare carbon storage values, one arrives at estimates of the glacial and Holocene total land ecosystem carbon (Fig. 3, Fig. 4) and the size of the resulting shifts in carbon. From a very low value (slightly over 900 GtC) at the LGM, carbon storage more than doubles to about 2,400 Gt by 8,000 years ago. It reached a slightly higher value at 5,000 years ago, and then (due mainly to anthropogenic effects and low-latitude climate drying) declined slightly from this high point to the present-actual situation (variously estimated by different authors to be around 2000-2500 Gt). Taking for purposes of a rough comparison the figures used by Olson et al. (1983) for vegetation carbon and those of Post et al. (1985) for soil and peatland carbon, it seems that the 'present-day' (i.e. 1980) land organic carbon storage is somewhere around 2130 Gt (allowing 560 Gt in present vegetation from Olson et al. 1983, 1115 Gt in non-peat soils in Post et al 1985, and 461 Gt in present peatlands from Gorham 1993). This would represent a net decline of around 600 Gt from the situation around 5,000 years ago, although this is merely an ad hoc estimate based on somewhat different data sets and different studies which are not directly comparable.

In these calculations, an overall error limit of around +/- 30% has been suggested in the per-hectare soil and vegetation carbon storage values. When multiplied by the limits of the vegetation map configurations, this gives the wider outer limbs of the error bars shown in the diagram (Fig. 2), around +/- 40-50% depending on the time slice involved. Whilst this is not a true statistical exercise, the 'outer limbs' on Fig.2 are intended to indicate what we regard as the chance of that the 'true' figure for per-unit-area carbon storage falls outside this limit (perhaps 1 in 20).

The resulting estimates of the change in carbon storage. Subtracting the estimated carbon storage at each time slice from the carbon storage of the next time frame gives the change (either positive or negative) in overall carbon storage on land. Allowing for these error limits alone, one obtains a range of possible values for the land carbon storage change between the LGM and either of the two Holocene time slices (see Table 4), varying from around 900 Gt to 1900 Gt.

The limits of uncertainty in terms of the carbon storage changes between one time slice and another would be wider if one were to allow a switch from a 'low per-hectare carbon' scenario at one time slice and a 'high per-hectare carbon' value during another time slice. However, there seems no reason for accepting that this would have occurred; if the estimate of per-hectare carbon storage for rainforest (for example) is too high for the mid-Holocene, then it will also presumably tend to be too high for LGM rainforest.

Carbon storage underneath ice sheets

During the closing stages of isotope stage 3, several thousand years before the LGM, there had been vegetation and soils in many areas that later became occupied by the larger LGM ice sheets (Tallis 1990, van Andel & Tzedakis 1996), and as discussed above, peatlands may or may not have been widespread in these areas. The eventual fate of most of the pre-LGM soil and vegetation carbon remains speculative. Franzen (1994) has suggested that this subglacial store of peat and other organic carbon provided a major part of the 'extra' carbon which entered the ocean-atmosphere-biota system following the LGM.

Franzen (1994) also mentions that palaeosols and peat deposits have been over-lain or over-ridden by ice sheets and preserved almost intact 'over large areas' in northern Scandinavia, although he does not give detailed evidence for this in his brief paper. In fact, most sedimentological and pedological studies of both ancient and recently-produced glacial debris appear to regard it as a truism that it is highly sterile and lacking in organic matter and weathered nutrients (e.g. see papers in van Husen & Schluechter 1993, or such textbooks as West 1987 or Williams et al. 1993).

At present, the hypothesis that there was a major store of organic carbon underneath the world's ice sheets (or frozen into permafrost in the periglacial zones) at the LGM remains tentative. The idea would merit much further study, but at present it seems that in most areas there is plenty of evidence to the contrary (e.g. the above-mentioned lack of organic content of glacial deposits almost everywhere), and a shortage of good evidence in its favour.

The general time course of events in land carbon storage since the LGM.

There is a striking contrast between the cold, dry world of the LGM-Late Glacial, with a very low total land organic carbon storage, and the much warmer and moister interglacial in which forests and organic soils were abundant. It seems most likely that a huge amount of carbon (at least 900 Gt, and possibly as much as 1900 Gt) that was 'missing' from the land system at the LGM, had appeared on land by around 8,000 14C years ago (9,000 calendar y.a.). By 5,000 14C years ago (5,900 calendar y.a.), ongoing cooling and greater aridity of the climate would have reduced carbon storage in vegetation and soils, this being offset by continuing peat buildup in the high latitudes. Only in the past 3,000 years or so does human influence seem likely to have significantly depleted carbon storage on a global scale (e.g. see Tallis 1990). Given the scale of past deforestation and forest degradation, it is likely that hundreds of gigatonnes of carbon have been released from vegetation, debris and soils, but this may have been partly offset by continuing peat buildup occurring at a rate of around several tens of GtC each millennium, to give a global land carbon storage value only slightly below that of the early-to-mid Holocene.

The usefulness of data from scattered time slices during the last 20,000 calendar years is limited somewhat by the fact that they cannot indicate the speed of change during the times between them. In the absence of the necessary quality of data to connect the time intervals precisely, the task has been attempted here in a general way (Fig. 4), based on a general knowledge of global events (Fig. 5) gathered through the Quaternary Environments Network (Adams 1995, Adams & Faure in press, QEN 1995) and through the literature that this draws upon.

As discussed above, most of the calculated glacial-to-interglacial increase (perhaps around 900-1000 Gt of it) may have been concentrated into a period of several thousand years during the rapid glacial-interglacial transition, between around 13,000 and 8,500 calendar years ago. This would give an average uptake of around 0.20 GtC/yr. into the land system; a tiny amount compared with the annual flux exchanged in photosynthesis and respiration each year (around 50 GtC; Olson et al. 1983), but representing a sustained unidirectional sink ocurring at a faster rate than the accumulation of carbon as CO2 in the atmosphere during deglaciation (Alley et al. 1993). At times, the land system may well have been taking up carbon at double or triple this 'average' rate (around 0.40-0.60 GtC/yr.). Consider for example the centuries which followed the ending of the Younger Dryas, when forest vegetation almost everywhere in the world was advancing rapidly towards its Holocene state, with renewed warm and moist conditions in Europe and in the monsoon belt.

Comparison with other studies.

Several previous attempts have been made to arrive at estimates of the glacial-to-interglacial increase in land carbon storage. T.J. Crowley (1995 a,b) has performed a series of calculations for LGM carbon storage based on a 220-site pollen database compiled by Webb et al. (1995). Peng et al. (1994) and Peng (1993b) have produced a set of Northern Hemisphere/global estimates based on the vegetation maps of Grichuk (1992) and the climate reconstruction maps of Frenzel (1992). The Adams et al.. (1990) estimate is limited by the use of a relatively narrow literature survey of palaeoenvironmental and ecosystem carbon storage information, which has necessitated this further study. Excluding the study by Adams et al., of which the present study is a development using a basically similar methodology, these studies can be said to suffer from various problems;

1) Inappropriate reconstruction of ecosystem distribution. Whilst a considerable element of doubt remains concerning the detailed distribution of ecosystems in the early Holocene and during the LGM, certain salient features are now effectively beyond dispute because of the wealth of information available from different sources in the fossil and palaeoenvironmental record. Several previous studies do not appear to have made use of an appropriately wide range of the sources which can be used to indicate past vegetation (not including, for example, such important sources of information as dated animal ecological indicator fossils and dated soil/sedimentary features) producing palaeovegetation maps which conflict strongly with the consensus picture of environmental change since the LGM that exists in the Quaternary community at large.

The recent study of Crowley (1995 a,b) is based only on LGM-dated plant fossil evidence and thus extrapolates over large distances because such 'ideal' data are relatively scarce. This has given rise to reconstructions of palaeovegetation which in some areas are strongly contradictory to a range of other sources of evidence (e.g. sedimentary and palaeozoological) which occur in abundance in the areas between the isolated plant fossil data points (for example, Crowley finds a considerably reduced Australian desert area and a major SW Amazonia forest expansion, which is contrary to large amounts of evidence from each region. A belt of forest across southern Siberia is also suggested). Overall, this approach appears to have given rise to the reconstruction of more forest and less desert than is plausible for the LGM world.

Somewhat similar problems may have occurred in the LGM mapping effort by van Campo et al. (1993). In this case, a heavy reliance on the out-of-date CLIMAP (1976) land surface maps seems to have resulted in some suprising choices of land ecosystem cover that lead to a large LGM carbon storage. For instance, a major expansion of tropical forest to cover much of the Indian subcontinent is shown for the LGM, contrary to strong evidence for a regional aridity trend (e.g. Cullen 1981). van Campo et al. also compare the LGM carbon storage with the present-actual (human influenced) vegetation distribution including large areas of croplands, rather than comparing it with a state more representative of the bulk of Holocene time.

Peng et al. (1995) have based their reconstruction of global LGM carbon storage on a single published review map source, that of Grichuk (1992), for the Northern Hemisphere. They then extrapolated to the whole global land surface on this basis. Apart from the evident limitations of using the history of one Hemisphere to deduce that of another, there are also serious doubts to be had about the relevance of Grichuk's maps. These maps show relatively large areas of dense and forested vegetation in southern Europe and central Eurasia where there is abundant palynological, zoological and sedimentological evidence to show sparse steppe, semi-desert and desert-like conditions. Various alternative maps in the same volume (Velichko et al. 1992) in fact show reduced vegetation coverage, in agreement with the range of evidence that has come forth since then and which is available in Adams & Faure (in press), Adams (1995) and the QEN (1995) atlas Web site. In fact, the origonal purpose of Grichuk's maps was apparently to show broad biogeographic 'source areas' (containing possibly very small and scattered refugia) and not vegetation coverage as such (E. Zelikson pers. comm.), so Grichuk's maps may in fact be inappropriate for the purpose that they have been used for by Peng et al.

2) Incorrect assignment of per-unit-area carbon storage values. Most previously published studies on LGM-to-interglacial carbon storage have uncritically used the published databases of Olson et al. (1983) for vegetation, and Zinke et al. (1984) for soils. As explained above, Olson et al. aimed to produce a present-day (circa 1978) atlas of carbon storage, and much of their data is based on vegetation that is heavily influenced by humans and hence depleted in biomass (J.S. Olson pers. comm. 1998). Furthermore, previous assessments of natural land carbon storage do not include the considerable amount of carbon held in dead standing trees, fallen woody debris, and in litter (Harmon & Hua 1992).

Amongst the soil values, the mis-assignment of high carbon storage to some 'desert' soils in the Zinke et al. database has led previous studies to assign implausibly high carbon storage to the desert zones of the LGM (P. Zinke pers. comm. 1997), in which temperate desert soil carbon content rivals that of forest ecosystems. As pointed out above, re-examination of the data in Zinke et al.'s database suggests that the figures are biased by a small number of samples, of which some were placed in densely vegetated scrub-chaparall zones of the USA, and the steppic zones of Russia (Adams 1995 & INQUA Carbon Commission Homepage in press). In addition, several studies (e.g. van Campo et al. 1993, Peng et al. in press) have assumed that the major 'steppe-tundra' biome of the LGM had a very high carbon storage value resembling that of present-day moist tundra or chernozem steppe. This was clearly not the case; buried soil profiles, plant community composition and faunistic communities all suggest that the steppe-tundra soils were poor in organic matter, resembling the soils found in semi-arid zones in the present-day world (E. Zelikson pers. comm., A.A. Velichko pers. comm., Adams 1995 & INQUA Carbon Commission Homepage).

These combined factors have led to several estimates with much lower values for the LGM-to-Holocene increase in land carbon storage. The range of estimates that has been obtained using the same general approach for reconstructing palaeovegetation clearly shows how important the details of methodology are in affecting the final result. If the most evident problems are allowed for, one could in fact close most of the gap between these smaller estimates and the larger one of over 1000 GtC presented here. Putting aside as probably unrealistic the estimate of Klinger (1991) of a large decrease in glacial-to-interglacial carbon storage, one has figures for an increase varying from around 469 Gt for the 'lowermost' estimate of Peng et al., to the previously published figure of 1350 Gt for Adams et al. (1990). Overall - from the perspective of the extensive extra input of recent data for both the ecosystem maps and the more representative use of carbon storage data in the present study - amongst the previously published work the estimate of Adams et al. (1990) seems the more likely to approximate to the true figure.

Conclusions.