The BIOME 6000 experiment aims to quantify the importance of biogeophysical feedbacks in the climate system by comparing the performance of coupled and uncoupled climate-biosphere models, driven by the Earth's orbitally-induced change in seasonal insolation from 6000 yr BP to present. The existing, extensive coverage of paleodata describing the state of the terrestrial biosphere at 6000 yr BP should provide a decisive standard against which to evaluate the model results.
One major area of interest for GAIM so far has been climate-vegetation interactions. Sensitivity studies with AGCMs have suggested that large-scale changes to the contemporary vegetation may have important consequences for global climate. Other AGCM studies with a specifically paleoclimatological focus have been carried out recently. These studies have shown that the known greater extension of shrub- or grasslands in what is now the Sahara during the early to mid-Holocene, up to around 6000 yr BP, would have amplified the direct effect of orbital forcing on the monsoons. Similarly, the poleward extension of taiga in Canada and Siberia during the same interval would have amplified the effect of orbital forcing on growing-season temperatures in the north. These are examples of "biogeophysical feedbacks", which arise because vegetation mediates the exchange of water and energy between the land surface and the atmosphere. Changes in vegetation characteristics such as albedo, height, foliage cover and phenology exert a strong influence on these exchanges. So when external factors (such as the Earth's orbital configuration) change in such a way as to produce a change in climate, the response of vegetation patterns is not just a passive reaction, but may have further feedback effects on the climate.
The GAIM focus "Climate-Vegetation Interactions: A 6000 yr BP experiment" aims to characterize and quantify these biogeophysical feedbacks, using paleodata for 6000 yr BP as the standard. The experiment has two major components: coupled climate-vegetation modelling, and paleoecological data synthesis.
The modelling component relies on the existence of mechanistic models (biome models) that predict the distributions of broad-scale ecosystem types (biomes) as a function of the physical environment. The BIOME model of Prentice et al. (1992) has been used in this way to translate the results of AGCM experiments (e.g. for the LGM, 6000 yr BP, and possible future high-CO2 climates) into global vegetation maps. Such a model can also be asynchronously linked to an AGCM, allowing simulation of the equilibrium state of the coupled system. Coupled model studies of this kind are underway for 6000 yr BP.
A further proposed GAIM activity, currently in the planning stage and involving co-operation with the IGBP core projects PAGES and IGAC, aims to account for natural changes in atmospheric trace-gas composition during the last glacial-interglacial cycle as shown by ice-core paleorecords. This effort will entail coupling models of physical climate, atmospheric chemistry and terrestrial biosphere trace-gas sources and sinks, with an obvious initial focus on the LGM.
The data synthesis component relies on the existence of a very large body of pollen and plant macrofossil records from the past twenty thousand years, from sediments dated (usually by 14C) with a precision of a few hundred years. In order to make full use of these data for global change research, a community-wide effort is required (a) to assemble the data for all of the continents, and (b) to represent them in some compact and globally consistent form. Global consistency requires more than just a data format; it also requires an objective means of achieving comparability between different regions that have taxonomically distinct sets of plant taxa, so that the paleoreords can be assigned unambiguously to globally accepted, non-taxonomic vegetation categories such as biomes.
Development of data sets has had community-wide support among paleoecologists, on the understanding that the value and integrity of the primary data is respected. Major data synthesis activities for key times (including 6000 yr and the LGM) are already under way in every continent. In such syntheses, the primary data must be accessible and documented. Also for the development of global data sets, biome reconstruction should follow an objective procedure based on plant functional types (PFTs). A suitable procedure has now been developed, and tested with spatial networks of surface pollen-sample data from Europe, east Africa and eastern North America. The development of the required paleovegetation data sets thus appears feasible with existing techniques, and builds on a firm collaborative basis.
The BIOME 6000 project was organized from an initial 1994 workshop. Biome reconstructions are carried out using a standard methodology, based directly on primary data to the greatest extent possible, with co-operation and feedback from members of a number of regional working groups. The primary data are derived in part from existing pollen data bases, with the co-operation of the data base managers. However, independent efforts to bring in data are required for regions that do not yet have well-established data bases. Regional data workshops are a key part of the global strategy for data assimilation.
The project outlined here fills a need which arises from the increasing maturity of two hitherto only tenuously connected disciplines within global change science.
Earth system modelling aims to elucidate the dynamic interactions between the physical climate system and the biosphere. Whereas climate modelling until recently was solely concerned with modelling the physical dynamics of atmosphere and oceans, earth system modelling attempts to include the major biogeochemical cycles and their contribution to the state of the climate and the biosphere. This modelling activity reflects the recent recognition that climate-biosphere interactions are important in the regulation of planetary metabolism, and that successful prediction of the consequences of human modification of climate and the biosphere on the time scales of "global change" is unlikely to be achieved unless such interactions are explicitly included. The emergence of earth system modelling has been made possible partly by recent developments in physical climate modelling and, especially, by rapid developments in large-scale modelling of ecosystem structure and function.
Quaternary Paleoecology aims at an understanding of the nature and causes of changes in ecosystems during the past 2-3 million years of earth history, based on analysis of geological proxy records of ecosystems and the physical environment. From its roots as a descriptive historical science, Quaternary Paleoecology during the past 10-20 years has evolved a predictive agenda based on advances in our understanding both of the causes of natural environmental change and the mechanisms by which organisms, species and ecosystems respond to their environment. Analyses of large-scale spatial and temporal patterns in ecosystem composition have been made possible by (a) the continuing accumulation of proxy records, dated by radiocarbon and other means, by paleoecologists around the world; (b) the progressive development, with community-wide support, of continental-scale data bases facilitating data extraction, plotting and mapping; and (c) synthesis activities carried out by a smaller number of individuals for the purposes of mapping, analysis and comparison with paleoclimate model simulations.
The link between these endeavors is the fact that Quaternary paleodata, properly analyzed, can test various aspects of the performance of earth system models under conditions different from today.
The best-established examples of the application of Quaternary paleodata to test models come from the period around and after the last glacial maximum (LGM), ca 18000 14C-yr BP (corresponding to ca 21000 astronomical years before present, according to the U-Th calibration of the 14C clock), and especially the Holocene epoch, starting at 10000 14C-yr BP and extending through the so-called "climatic optimum" ca 6000 yr BP (see Box) to the present. There are vastly more terrestrial paleodata for these intervals than for any earlier time in geological history. Although there is still much that is not understood about higher-frequency climate variations since the LGM, the broad outlines of the transition from LGM to 6000 yr BP to present have an accepted explanation in terms of changes in the Earth's orbital configuration: direct (radiative forcing) effects of orbital variations combined with consequent (but lagged) changes in continental ice-sheets, atmospheric composition and sea-surface conditions produce the low-frequency "envelope" of climate change since the LGM. This explanation is convincing because even simple "snapshot" experiments with atmospheric general circulation models (AGCMs), forced by these changing boundary conditions, correctly predict the large-scale qualitative aspects of paleoenvironmental patterns at different times since the LGM.
The period around 6000 yr BP offers the simplest case to analyze because by that time the large continental ice sheets had gone and atmospheric CO2 and other trace gases has reached their more-or-less stable late-Holocene (pre-industrial) levels, leaving only the still very different-from present orbital configuration. Snapshot experiments for 6000 yr BP have, for example, consistently and correctly predicted higher than present temperatures in the high latitudes and higher than present precipitation in today's arid northern subtropics, as well as many more specific regional patterns. The realism of these predictions has been evaluated with paleodata of various kinds, with paleoecological data and lake-level records playing a primary role.
There are differences among the predictions of different AGCMs, just as there are for high-CO2 scenarios. There also remain data-model discrepancies, especially in the magnitudes of simulated climate anomalies. These differences and discrepancies presumably are due to inexact or incomplete formulations of processes, and/or to omission of significant processes. So far the comparisons of paleodata and paleoclimate simulations have been mostly general and qualitative, and the discrepancies have not usually been analyzed in detail. However as models develop, the paleodata will be used increasingly as a standard against which model performance in paleoclimate simulations is assessed.
The most specific and quantitative comparisons of paleoclimate simulations and paleoclimate that have been performed to date rely on statistical inverse modelling procedures (e.g. transfer functions, analog methods) to reconstruct climate variables from paleodata. However, there are good reasons to rely more on "forward modelling" procedures in which the comparisons are made in terms of properties closer to the data. For example, pollen assemblages are translated into biomes rather than climate. Then modelling the biome distribution as a function of climate can be done more mechanistically, using biome models.
A systematic data-model comparison approach, using biome modelling as the tool to facilitate comparisons with paleoecological data, is already being planned for the Paleoclimate Modelling Intercomparison Project (PMIP: Joussaume and Taylor, 1995). About twenty climate modelling groups participate in PMIP and are producing LGM and 6000 yr BP simulations under a standard protocol. More ambitiously, it should be possible to compare model results obtained with and without the inclusion of further processes, not normally included in AGCMs, including the reciprocal interactions among vegetation, surface hydrology and climate.
The global nature of this new paleovegetation mapping enterprise means that the paleoecological data must not simply be left in the form of abundances of taxa, so one of the challenges for BIOME 6000 is to find a way to translate the taxon information into a common language. It is necessary to use the concept of "plant functional types" (PFTs) as advocated by GCTE. Plant taxa must be assigned to PFTs and the data must be interpreted in terms of biomes, which are defined as combinations of PFTs.
Preliminary work carried out prior to the workshop, with data from Europe, North America and East Africa, had already shown that it is indeed possible to translate pollen spectra into biomes by means of an objective algorithm based on fuzzy logic. The key to the method is the idea that plant taxa identified in the paleorecord can all be assigned a priori to one or more PFTs, based on knowing the basic biology (e.g. leaf form, habit, phenology) and the present bioclimatic distribution of the species included. A taxon may potentially belong to more than one PFT if it includes several species that represent different PFTs, or individual species that can behave as a different PFT in different environments. Fuzzy logic enters the picture because the problem is neither a strictly statistical one, due to the past existence of plant assemblages that lack modern analogs, nor is it susceptible to ordinary Boolean logic, because it is rarely possible to be certain from the paleorecord whether a given taxon was locally present or not (long-distance transport is a common complication). The method therefore relies on ranking "affinity scores" that represent the weight of information contained in each pollen spectrum that would support the assignment of that spectrum to a given biome. The steps in the method are as follows:
Each pollen taxon is initially assigned to one or more PFTs on the basis of the known biology of the species it represents.
The PFT assignments are checked by comparing the bioclimatic distribution of each taxon in turn with an expected bioclimatic distribution, based on its assignment to one or more PFTs. If discrepancies are found then the PFT assignments must be corrected. The product is a corrected PFT x taxon matrix .
Next, biomes must be defined in terms of their characteristic PFTs, yielding a biome x taxon matrix.
The above matrices are manipulated to yield a taxon x biome matrix, indicating which pollen taxa may occur in each biome.
Affinity scores for any given pollen spectrum and biome are calculated as the sum of pollen values for taxa that may occur in that biome. Prior to this calculation, the pollen values have to be transformed to increase the signal-to-noise ratio. Good results have been obtained using a square-root transformation after subtraction of a uniform "threshold" pollen percentage (1% or less).
The pollen spectrum is assigned to the biome with which it has the highest affinity, subject to a tie-breaking rule by which any biome whose list of characteristic taxa is a subset of another biome's list of characteristic taxa is given precedence.
The method has numerous advantages for worldwide application, including its basis in PFTs (so that taxa with similar bioclimatic ranges in different biogeographic provinces can be treated as equivalent); an ability to produce well-founded biome reconstructions even when the pollen spectra have no known modern analogs; insensitivity to quantitative changes in taxon abundance due to human impacts on the landscape; and the lack of any absolute requirement for comprehensive data sets on modern (surface-sample) pollen distribution. This last point distinguishes the method from earlier objective vegetation reconstruction methods based on statistical analysis of surface sample data. The problem with such methods is that they break down when the appropriate modern analogs for fossil samples either do not exist, or are not represented in the surface-sample data set. The present method allows surface-sample data (in so far as they have been collected) to be reserved for a validation of the biome reconstruction method.
The key to the method's robustness lies in the application of the PFT concept, which in turn implies that taxa can be assigned to PFTs. This first step immediately overcomes what has been one of the major barriers to more widespread use of paleoecological data in global change research, viz. that the data are normally presented in the form of abundances of taxa whose ecological significance is only known to researchers who have specialized in the region in question. Such basic information is often not clear from floras, which have generally been written from a taxonomic rather than an ecological or biogeographical perspective. Global application of this approach thus requires active participation in the project by paleoecologists with a knowledge of the flora and vegetation of "their" regions.
Paleoecological data give information about past ecosystems. They are complementary to other terrestrial paleodata sources (such as lake-level records, ice-core records, paleosols, fossil dunes) that give information about other climatically influenced processes at the Earth's surface. The most common sources of paleoecological data are pollen and plant macrofossil records, which indicate the composition of past vegetation. There are already regional- to continental-scale compilations of paleovegetation data covering most regions of the world, in the form of "time slices" including 6000 and 18000 yr BP among others. Several such compilations originated during the 1980's as part of or in association with the COHMAP project, and have subsequently been maintained and updated to a lesser or greater degree by the original investigators. Other compilations have been started during the last few years.
Primary data repositories (pollen and plant macrofossil data bases) are a more recent development. Pollen data bases store all of the original pollen counts from entire time-series records, together with primary dating information, and provide software that facilitates the extraction of secondary products including interpolated chronologies (age models), graphic representation of time series, and data summarized for time-slices. There are established pollen data bases for Europe (the European Pollen Data Base in Arles, France) and North America (the North American Pollen Data Base in Springfield, Illinois). There are also efforts under way to set up comparable pollen data bases for other continental regions, including Africa, Japan, China, Latin America and S Asia/Australasia, and to set up a plant macrofossil data base for North America. However some of these efforts are relatively new and still far from comprehensive. Thus, somewhat different strategies have to be followed for BIOME 6000 to compile the data from different continents, depending on the degree to which data have been provided to primary data repositories. In the future, it is to be hoped that data will routinely be lodged in data bases, thus adding enormous value to the data and obviating the need for much of the "one-off" effort in data assembly that is required for BIOME 6000.
Figure 4.1: The first specific task was to generate a provisional listing of sites, region by region, where paleoecological data should be available with confident dating to either 6000 yr BP ca 500 yr, or 18000 yr BP ca 1000 yr (14C-years). For the most part the sites represent pollen records, but in the arid western USA pollen sites are few but many detailed, 14C-dated floristic records have been obtained from macrofossils in packrat middens. An initial site map was provided, based on the COHMAP project archives at Brown University. It was found that in some regions the coverage of paleoecological data has improved dramatically during the past 5-10 years, both in terms of numbers of sites and in terms of "plugging" spatial gaps. Particularly important improvements were noted for South America, China, SE Asia and Siberia. Data provided after the meeting by several members of the regional working groups allowed the production of Figure 4.1, which gave a first (and certainly incomplete) view of the data available.
The key interpretative method used by BIOME 6000, known as "biomization", translates assemblages of pollen and plant macrofossils into an assignment of biome by (a) first assigning all taxa to plant functional types (pfts) based on their known biology and bioclimatic limits, (b) calculating the "affinities" of the assemblage for all biomes based on a priori definitions of biomes in terms of functional types. The method does not rely on the existence of a reference data set of surface (modern) assemblages for calibration; on the other hand, if such data are available, then they can be very useful to test the performance of the method in reconstructing present biome distributions. The current state of the project, including both regional tests of the biomization procedure is as follows:
Europe (including North Africa and the Middle East). The biomization method was developed initially for this region. It was successfully tested against a large surface sample data set, and applied to a digitized pollen data set to construct a 6ka biome map. This map shows starkly the northward and upward expansion of forests at 6ka relative to present, the near-absence of "Mediterranean" type vegetation around the Mediterranean at 6ka, and the great extension of temperate deciduous forests to the north, east and south. New versions are being produced based on original pollen counts archived in the European Pollen Data Base (EPD), including improved coverage of North Africa and the Middle East.
Africa (including the Arabian peninsula). The method has been tested against surface samples separately in West Africa, East Africa, and southern Africa. It has now been applied to a comprehensive 6ka pollen data set for the whole continent, based on original pollen counts obtained either directly (from a large number of scientists) or via the new African Pollen Data Base, which was inaugurated with DIS support in September 1996. In Africa, much interest has focused on the large early/mid-Holocene northward shift of the monsoon belt into what is now the Sahara desert; the data document this precisely, in terms of vegetation changes at specific locations.
Former Soviet Union and Mongolia (excluding western Beringia). Great progress has been made following a highly successful DIS-sponsored workshop in Hörby in 1996, which included paleoecologists from many different laboratories in several of the FSU countries. A full scientific report from that workshop is available. Since the 1996 workshop, a visit of Pavel Tarasov (Moscow) to Lund has led to the biomization method being successfully tested against a very large data set of surface samples, and applied to a now probably almost exhaustive 6ka data set of pollen records (mainly original counts, some digitized), supplemented by tree megafossil records in N. Siberia. The maps document the northward shift of the forest-tundra boundary at 6ka. They also contrast the substantial northward shifts of forest belts in the western part of the region with relatively little change in the eastern part. The map also makes it clear that (in contrast with eastern North America, see below) there was no large extension of steppe vegetation at the expense of forest in Central Asia; the boundary remained more-or-less stable, and in the western part steppe vegetation was actually reduced.
Pacific Asia. A workshop held in Beijing in 1995 started efforts to synthesize pollen data from China. There is a great deal of high-quality data and many active Chinese scientists in this field. Through a visit of De. Ge Yu (Nanjing) to Lund and a collaboration with Prof. Sun Xiangjung (Beijing) we have been able to make a preliminary test and a preliminary 6ka synthesis map for China. The map shows clear expansion of forests at 6ka into what is now steppe vegetation in the interior of China, a large upward shift of the treeline around the Tibetan plateau and northward shifts of the forests belts in eastern (especially northeastern) China. These are the broad conclusions based on a limited data set of digitized pollen counts. Data synthesis for key time periods (including 6ka) is already underway in Japan.
North America. The biomization method has been tested against a large data set of surface samples for eastern North America, and applied to 6ka and LGM pollen data sets producing results consistent with earlier interpretations, including the expansion of prairie at the expense of forest in the continental interior and the greater latitudinal extension of temperate deciduous forests at 6ka. A small workshop in Eugene in 1996 established a planning framework for the completion of the 6ka paleodata set for the region as a whole (USA, Canada, Greenland, Alaska and western Beringia) and biomization. Macrofossil analyses from packrat middens will be a major data source for the arid western USA. Tree megafossil records will be included in the synthesis for Canada, drawing on extensive 6ka synthesis work already carried out in Canada.
Australia and southern Asia. A comprehensive pollen data base for the "Asia/Pacific" region, including SE Asia, Australia, New Zealand and the Pacific islands, is led by Dr. Geoff Hope (Canberra) and is well advanced. A visit by Dr. Habiba Gitay to Lund in 1995 allowed an initial test of the biomization method for this region. In addition, Dr Raymonde Bonnefille (Pondicherry) is now assembling the available 6ka paleoecological records from the Indian subcontinent.
Central and South America. The new Latin American Pollen Data Base (led by Vera Markgraf, Boulder) will facilitate data synthesis. A biomization plan is being developed with Prof. Henry Hooghiemstra (Amsterdam).
In addition to pollen data and conventional plant macrofossils (seeds, leaf remains, etc.), 14C-dated megafossil remains such as in situ fossil wood have the potential to provide direct information about the presence of trees. This source of information is especially important in recording the former northward extent of forests in the present-day tundra regions of northern North America and Eurasia. A project of the Canadian Geological Survey has used tree megafossil data as a supplement to the relatively few available pollen diagrams from the Canadian Arctic, to map the paleodistribution of forest types in Canada at 6000 yr BP, and analogous work is underway in the Eurasian Arctic.
