Pollen-based biome reconstructions for Latin America at 0, 6000 and 18 000 radiocarbon years

The biomisation method is used to reconstruct Latin American vegetation at 6000±500 and 18 000±1000 radiocarbon years before present (C yr BP) from pollen data. Tests using modern pollen data from 381 samples derived from 287 locations broadly reproduce potential natural vegetation. The strong temperature gradient associated with 5 the Andes is recorded by a transition from high altitude cool grass/shrubland and cool mixed forest to mid-altitude cool temperate rain forest, to tropical dry, seasonal and rain forest at low altitudes. Reconstructed biomes from a number of sites do not match the potential vegetation due to local factors such as human impact, methodological artefacts and mechanisms of pollen representivity of the parent vegetation. 10 At 6000±500 C yr BP 255 samples are analysed from 127 sites. Differences between the modern and the 6000±500 C yr BP reconstruction are comparatively small. Patterns of change relative to the modern reconstruction are mainly to biomes characteristic of drier climate in the north of the region with a slight more mesic shift in the south. Cool temperate rain forest remains dominant in western South America. In 15 northwestern South America a number of sites record transitions from tropical seasonal forest to tropical dry forest and tropical rain forest to tropical seasonal forest. Sites in Central America also show a change in biome assignment to more mesic vegetation, indicative of greater plant available moisture, e.g. on the Yucatán peninsula sites record warm evergreen forest, replacing tropical dry forest and warm mixed forest presently 20 recorded. At 18 000±1000 C yr BP 61 samples from 34 sites record vegetation that reflects a generally cool and dry environment. Cool grass/shrubland prevalent in southeast Brazil, Amazonian sites record tropical dry forest, warm temperate rain forest and tropical seasonal forest. Southernmost South America is dominated by cool 25 grass/shrubland, a single site retains cool temperate rain forest indicating that forest was present at some locations at the LGM. Some sites in Central México and lowland Colombia remain unchanged in their biome assignments, although the affinities that


Introduction
Biomisation is an objective method to reconstruct broad vegetation types based on the 5 assignment of pollen taxa to one or more plant functional types (PFTs) (Prentice et al., 1996a). The method is based on the assumption that a pollen spectrum will have different degrees of affinity to different biomes that can be quantified by a simple algorithm. Biome reconstructions from pollen data at 6000±500 14 C yr BP and the last glacial maximum (LGM) at 1000±18 000 14 C yr BP have been produced for most regions of the world under the auspices of the BIOME 6000 project . The validity of the method in reconstructing biomes at different time intervals has been demonstrated for Africa (Jolly et al., 1998a;Elenga et al., 2000), Australia (Pickett et al., 2004) Beringia (Bigelow et al., 2003;Edwards et al., 2000), China (Yu et al., 1998(Yu et al., , 2001, Eastern North America (Williams et al., 2000), Eurasia (Tarasov et al., gations, the method has been applied down-core down to a 450 000 year pollen record from the high plain of Bogotá (Marchant et al., 2002b). As Colombia is biogeograph-373 ically complex, encompasses high altitude, temperate and tropical floras reflecting a range of environmental space including transitions from hyper-humid to semi-arid climates, these analyses provided a suitable test-bed for the wider geographical focus presented here. In addition to reconstructing vegetation patterns, and investigating factors that can 5 explain observed changes, data on past biomes contributes to testing of climate and vegetation models (Prentice et al., 1992;Haxeltine and Prentice, 1996;Peng et al., 1998;Marchant et al., 2006;Braconnot et al., 2007). Vegetation models can be used to portray output from Global Circulation Models (GCMs) as maps of potential vegetation (Claussen and Esh, 1994;Foley et al., 1996;Prentice et al., 1996b;Williams et 10 al., 1998) that can be used in the development of models that couple biosphere, atmosphere and oceanic components (Braconnot et al., 2007;Claussen, 1994;Harrison et al., 2003;Texier et al., 1997) and testing of biogeochemical dynamics (Peng et al., 1998). There has been growing interest has focused how atmosphere-biosphere interactions have operated under the changing environmental conditions since the LGM, 15 particularly in trying to understand the response of ecosystems to different types of environmental forcing (Jolly and Haxeltine, 1997). Transformed pollen data can further be used in conjunction with other data types, such as on lake status (Jolly et al., 1998b) and archaeological evidence (Piperno et al., , 1991a, to better understand the causal factors driving vegetation change over the recent geological past. 20

Latin American region
Latin America comprises the area from 35 • N to 65 • S, and from 35 • W to 120 • W extending from México to islands off southernmost South America from eastern Brazil to the Galapagos Islands. Latin America is characterised by strong environmental gradients associated with 100 • of latitude, approximately 7000 m of altitude and the transition 25 from oceanic-to continentally-dominated climate systems (Fig. 1). Physiographically, Latin America is characterised by stable cratons associated with the interior and areas of active mountain building, particularly associated with the Andes. This environmental variability is reflected by an incredibly diverse biogeography, ranging from the highly diverse rain forest of the Chocó Pacific (Colombia) to the cold deserts of the high Andes, from the hot semi-desert areas of México to the cold moorlands of Tierre del Fuego (Fig. 2). Descending an altitudinal gradient there is a transition from páramo (cool grass/shrubland) to high Andean forests (cool mixed and cool temperate rain 5 forests) and lower Andean forest (warm evergreen forest) (Fig. 3). Complicating this potential vegetation distribution is the factor of human impact with the majority of the vegetation in Latin America being impacted on by the vegetation (Ellis and Ramankutty, 2008). The timing of early human settlement in Latin America is a contentious subject, although it seems from the early Holocene there was considerable cultural diversity 10 and adaptation to a series of different environments (Gnécco, 1999). Human-induced impact has had a direct influence on vegetation composition and distribution through land-use practices and the introduction of alien taxa and cultivars to the Latin American flora. For example, in excess of 100 plants were under cultivation prior to the European conquests in the 15th century (Piperno et al., 2000). 15 1.1.1 Latin America climate Cerverny (1998), Eidt (1968) and Metcalfe et al. (2000) have reviewed Latin American climate. Given the broad geographical scope, Latin America is characterised by a variety of climates that relates to its global position, shape of the landmass, location and height of the Andes, offshore currents, general hemisphere air flow and proxim- 20 ity of large water bodies (Fig. 1). Four dominant circulation regimes influence Latin America: the Inter-Tropical Convergence Zone (ITCZ), the prevailing westerlies, the semi-permanent high pressure cells located over the South Pacific and South Atlantic Oceans and the trade winds. Perhaps most dominant is the annual oscillation of the meteorological equator (ITCZ), this migrating some [10][11][12][13][14][15] • latitude about the equator 25 ( Fig. 1). The ITCZ reaches its northernmost location in June, this bringing high rainfall for northern South America and the Caribbean, with January and February recording the dry season (Cerveny, 1998). However, due to the influence of the westerlies from 375 the Pacific, and the sharply rising topography of the Andes, the ITCZ has a sinusoidal profile over northwestern South America (Fig. 1). In southern South America the prevailing westerlies south of 40 • S are particularly important in controlling the moisture regime. The topographic barrier of the Andes contributes to the creation of two large semi-present anticyclones, one over the South Pacific and one over the South Atlantic, 5 the southeast trade winds associated with this latter system brings abundant moisture to the Amazon Basin (Cerveny, 1998). Due to the large size of South America, and the highland ranges that fringe much of the continent there is often a rapid transition from relatively moist coastal areas to a dry interior reflecting the transition from oceanic-to continental-dominated climate systems. For example, due to the proximal location of 10 the Pacific-based moisture source and steeply rising ground, precipitation is highest (>15 000 mm yr −1 ) in the Chocó Pacific region. Exceptions to this scenario are areas located between the anticyclones, e.g. the Peruvian coast, where relatively arid conditions prevail. One of the main environmental gradients in Latin America is associated with the 15 Andes. The Andes are characterised by a diurnal climate (Kuhry, 1988); at a given location differences in monthly temperature are small (<3 • C) although daily fluctuations may be large (20 • C), especially during the dry seasons. Climatic changes with altitude can be summarised as a lapse rate (Barry and Chorley, 1990). Applying a lapse rate of 6.6 • C 1000 m −1 (Van der Hammen and González, 1965;Wille et al., 2001), this altitudi- 20 nal rise equates to a temperature change of more than 30 • C. Also associated with the Andes are steep gradients of moisture availability. Rainfall is high on the eastern slopes of the Andes; the concave nature acting as a receptacle for moisture transferred by the southeast trade winds from the Atlantic Ocean, in part receiving moisture generated by the Amazonian forest (Fjeldså, 1993). Low rainfall is recorded within rain shadow ar- 25 eas, such as on the lower slopes of the Magdalena Valley and the inter-Andean plains (Kuhry, 1988). These climate gradients result in rapid transitions from mesic to xeric vegetation types, e.g., cool high-altitude grasslands change to "temperate" forests at mid-altitudes and diverse tropical rain forests within a few kilometres (Fig. 3). In south-ern part of Latin America rainfall is largely controlled by the persistence and strength of the westerly winds (Gilli et al., 2005). In recent years there has been increased interest in large-scale temperature-driven surface pressure oscillations in the Pacific Ocean termed the Southern Oscillation, and its assimilated oceanic aspects, El Niño and its antithesis La Niña (Cerveny, 1998). 5 The climate of the tropical Pacific basin, extending from the western Americas across to Australia, New Zealand, and northeast Asia, oscillates at irregular time intervals (3 to 7 years) between an El Niño phase, with warm tropical waters upwelling off Pacific coastal South America, and a La Niña phase, with cold tropical waters dominating. ENSO events are the largest coupled ocean-atmosphere phenomena resulting in cli- 10 matic variability on inter-annual time scales (Godínez-Domínguez et al., 2000). As climates, particularly rainfall patterns, are driven by temperature differences between land and ocean, the influence of changing oceanic sea surface temperatures (SST) on coastal South American environment can be dominant, and have a strong influence elsewhere (Marchant and Hoohiemstra, 2004). El Niño events primarily result in in- 15 creased precipitation along the Pacific coastal regions, decreased precipitation within lowland tropical moist forests of Central America (Cerveny, 1998) with increased precipitation in northern Central America (Metcalfe et al., 2000).

Latin America vegetation
For the purpose of this investigation the potential vegetation composition and distribu- 20 tion Latin America is classified at a coarse resolution with twelve biomes being identified ( Fig. 2) that summarise the 57 categories mapped by Hück (1960) and 45 by Schmithüsen (1976). The vegetation composition and distribution generally reflects the main climatic and topographic gradients described above. However, a series of caveats to this must be stressed. Firstly, the actual and potential vegetation can be 25 quite different, the former reflecting a long history of human interaction that has been particularly pronounced since the colonial period but has been influencing the vegetation for at least the last 5000 years (Marchant et al., 2004). In numerous areas this 377 interaction has completely transformed the potential vegetation to an agricultural landscape. Another factor complicating the relationship between climate and vegetation is the locally strong edaphic influence by substrate, topography or geographic character (Fig. 3). The strength of this influence is characterised by areas of tropical dry forest that forms on free-draining sandstones, e.g. the Llanos Orientales (Colombia); these 5 are located in areas where the climate regime would support tropical seasonal forest, or even tropical rain forest. The vegetation at such locations is relatively insensitive to climate changes as these must be of a greater magnitude than the influence imparted by the edaphic factor. Broad types of vegetation with similar composition and distribution (biomes) result 10 from a combination of plant functional types (PFTs). PFTs and biomes, which lie at the heart of the biomisation technique, allow the high floristic diversity of the Latin America pollen flora to link with the relatively coarse vegetation classification (Fig. 4).
PFTs group together species that have common character (Prentice et al., 1992). This grouping is based on common life form and phenology, combined with the geographic 15 distribution that is in part determined by climate (Woodward, 1987). An indication of the bioclimatic range of each PFT and plant physiological adaptation, to the given environmental condition, is presented in Table 2. The range of biomes identified within the Latin America, floristic description, main location and equivalent floristic units is portrayed in Table 3. The cool grass/shrubland biome incorporates a relatively wide 20 range of vegetation dominated by grasses, heath, cool temperate sclerophyll shrubs and cushion plants (Fig. 3). This biome is present in southern South America and at high altitudes along the Andes. In addition to the cool grassland, a warm grassland (steppe) is identified. Steppe is found predominately under the warm, dry climates of southeast and northeast Brazil, northwestern Argentina and coastal northern South 25 America. Warm temperate rain forest represents a mix of warm conifers such as Araucaria, Andean and Atlantic rain forest elements, whereas cool temperate rain forest contains cool conifers, such as Fitzroya, Andean and Valdivian rain forest elements. Dry forests are extensive in Latin America, specifically associated with areas located between the two semi-permanent anticyclones and influenced by the high seasonality of rainfall imposed by the annual migration of the ITCZ. For our classification we characterise the diverse dry vegetation formations (Fig. 4) as the tropical dry forest and xerophytic trees and shrub biomes. Xerophytic trees and shrubs is widespread in the interior of South America, along the southwestern Pacific coast and northeast 5 Brazil where it grades into steppe, additionally, there are patches in Colombia, on the Yucatán peninsula and in México (Fig. 2). Tropical dry forest is predominantly recorded in two main swaths either side of the Amazon basin, with an extension through Central America. The tropical seasonal forest biome is predominantly recorded to the north of Amazonia where it is interspersed with patches of dry forest; this reflecting a strong 10 edaphic influence. A large area of tropical seasonal forest is recorded away from the hyper-humid area of Brazil along the Atlantic coast. The tropical rain forest biome is present in three main areas: Amazonia, linear strips along the Atlantic coast and northeast South America extending into Central America. Forest associated with highland areas is divided into three biomes: warm evergreen forest, cool temperate rain 15 forest and cool mixed forest (Fig. 4). Warm evergreen forest is most extensive along the lowland Andes, adjacent to the tropical rain forest. Cool mixed forest has a more restricted distribution, occupying a highland position until temperature becomes limiting for a number of taxa. Warm mixed forest is characterised by a mix of Pinus and Quercus species and is mainly restricted to Central America. The desert biome is re-20 stricted to coastal Peru, due to the Pacific Ocean anticyclone, this area receives very little moisture, except when the area is subjected to El Niño events.

Data sources
Over the past five decades palynologists have collected numerous pollen-based 25 records from lakes and bogs ( is an online resource used to collate these data and facilitated the systematic interoperation presented here; indeed the majority of the pollen data used here are available through the LAPD. Additional 5 data were obtained from palynologists working in Latin America; all active palynolgists being given the opportunity to contribute data not currently lodged in the LAPD. Indeed, data from a number of sites in Argentina, Brazil, Costa Rica, México and Panama were made available specifically for this work. The majority of data from Colombia were prepared for this analysis directly from the original count sheets and are in preparation for 10 uploading to the LAPD. The majority of the data used in our analysis are complete raw pollen counts, this permitted all pollen taxa recorded by the original analyst to be allocated to PFTs and allowed the integrity of the data to be maintained throughout the analysis. Application of raw pollen data in other regions has been shown to help in differentiating between 15 biomes (Tarasov et al., 1998b). However, numerous pollen records are either not submitted to the LAPD, or, were not made available for this analysis. Rather than omitting these data, the pollen counts were digitised from published pollen diagrams (Table 1): digitising such data provides a spatially more complete reconstruction than available from presently archived data. The process of digitisation involved either back calcula-20 tion of the pollen counts if information on the pollen sum was present. If the pollen sum was not available the pollen percentage diagram was used as a count of 100 and values for the pollen taxa were abstracted at the time intervals used for our analysis. This scenario of combining data from different formats comes with a number of caveats that can have bearing on the results, and their interpretation (Marchant and Hooghiemstra, 25 2001). Firstly, the sub-set of pollen taxa in a count used to construct published pollen diagrams, and pollen sums that comprise it, often result from the bias of individual researchers', particularly on what are the reliable indicator taxa for a particular area and range of different vegetation types under investigation. This issue is particularly crucial in Latin America where the large numbers of pollen taxa encountered in the original counting are rarely depicted on published pollen diagrams. Furthermore, the level of identification achieved within pollen analysis, to a generic or family level, commonly comprises species that can be found in a range of different vegetation types, ecologies and growth forms (Marchant et al., 2002c). The majority of the samples for the 5 biomisation presented here are derived from sites close to the Andean spine. Primarily, this concentration reflects the sensitive response of the vegetation to climate change on the steep altitudinal gradients (Marchant et al., 2001b); the area forming an ideal location for palaeoecological research. Additionally, the comparative lack of data from the lowlands is fuelled by problems of access, suitable sites and strong river dynamics 10 that commonly result in sedimentary hiatuses (Ledru, 1998). This spatial bias of the location did not reduce the number of biomes we were able to reconstruct, because of the steep environmental gradients associated with 7000 m of altitudinal change found along the Andes (Fig. 4).
Uncalibrated radiocarbon dates available from the original stratigraphic analysis were 15 used to select samples representing the time period used here; these were. On a site-by-site basis, a linear age-depth model was applied to the pollen data. The validity of this model was assessed at each site taking into account sedimentary hiatuses and dating problems such as age reversals and dates with large standard errors; a summary of this dating control is provided in Table 6 following the COHMAP 20 scheme (Webb, 1995;Yu and Harrison, 1999). Multiple samples (≤3) were selected when more than one sample fell within the age range allowed for each time period. These data were compiled, to produce a site vs. taxa matrix that was then checked to standardise nomenclature, e.g., the combined file contained many synonyms such as Gramineae and Poaceae, and Mysine and Rapanea. Synonymous taxa were com-25 bined using the nomenclature of Kewensis (1997) and the International Plant Names Index (IPNI) (1999). Aquatic and non-arboresent fern taxa were removed from the matrix as they commonly reflect local hydrological conditions rather than local climate envelope. Marker additions and exotic spikes such as Lycopodium were also removed.

381
A total of 381 samples from 287 locations derived from core tops (<500 14 C yr BP), surface samples, pollen traps and moss polsters comprise the modern data set (Table 1). For the time period 6000±500 14 C yr BP, 255 samples derived from 127 pollen records comprise the data set (Table 1). For the time period 18 000±1000 14 C yr BP, 61 samples derived from 34 pollen records comprise the data set (Table 1). The data 5 sets to undergo analysis comprised 515 pollen taxa for the modern calibration, 493 for the 6000 14 C yr BP reconstruction and 232 for the 18 000 14 C yr BP reconstruction. The taxonomic diversity of the Neotropical phytogeographical realm can be demonstrated by taking the modern biomisation as an example: the number of pollen taxa for the production of our biomes is greater than Africa (364) (Jolly et al., 1998a), Europe (41) 10 (Prentice et al., 1996b), Russia and Mongolia (98) (Tarasov et al., 1998a) and China (68) (Yu et al., 1998). Prentice et al. (1996a) and Prentice and Webb (1998) have documented the steps involved in the biomisation technique. First, a conceptual framework for biomes and 15 PFTs in Latin American vegetation was developed by investigating the relationship between potential biomes and three environmental gradients. The environmental gradients considered were moisture availability (α: Priestley-Taylor coefficient of plant available moisture), temperature (MTCO: mean temperature of the coldest month) and seasonal warmth (GDD: growing degree-days). To enable a definition of the biomes 20 to be based on bioclimatic data, rather than qualitative assessment, the climate space encompassed by Latin America was plotted against climate data set of Leemans and Cramer (1991) with relationship between biomes at individual site locations and macroscale climate changes (α and MTCO) investigated in two-dimensional space (Fig. 5). The twelve biomes identified within Latin America (Table 2) are designed to incorpo-25 rate the range of major vegetation types and ensure consistency with previous areas to undergo the process within the BIOME 6000 community.

Biomisation
Similarly to the biomes, but on a finer ecological resolution, the spatial distribution of PFTs is determined by environmental controls on plant growth form and ecological tolerance (Woodward, 1987). In Latin America the dominant environmental gradients are temperature, primarily associated with altitude, moisture availability and seasonality. PFT definitions were modified from the classification originally developed for the 5 BIOME 1 model (Prentice et al., 1992(Prentice et al., , 1996a taking into account schemes developed for other regions, particularly those that abut the Latin American region or contain similar floristic elements (Jolly et al., 1998a;Pickett, et al., 2004;Takahara et al., 2001Elenga et al., 2000Thompson and Anderson, 2000;Yu et al., 2000). Five main groups of PFT were distinguished: these containing tropical (non-frost tolerant), coniferous 10 (needle-leaved), temperate (frost tolerant), xerophytic (drought tolerant), and frost and drought tolerant taxa (Fig. 6). This latter group is present in cold dry conditions of southern South America and the high Andes. A sixth "miscellaneous" group represents various life forms with restricted diagnostic value. The Latin American flora was divided into 25 PFTs (Table 3). The PFTs, although being ecological distinct, can be 15 multiply assigned to the biomes ( Table 4). The classification is based on the original scheme devised for the Biome 3 vegetation model (Prentice et al., 1992) and modification through regional applications to pollen data. Where possible the scheme devised for Latin America conforms to existing classification and definitions. However, some of the specific vegetation types in Latin America were not adequately covered by the 20 existing range so two new PFTs (heath and cushion plants) were added. To aid in the separation of the African forest/savanna boundary, Jolly et al. (1998a) subdivided the tropical raingreen trees PFT (Tr) into three groups. In the case of Latin America, it was decided that the overlap (taxa being multiply assigned to the PFTs) between the PFTs would be too great, and the distinction somewhat minimal. Furthermore, the tropical 25 xerophytic trees and shrubs PFT encompass many taxa that would be assigned to the driest tropical raingreen category. Therefore, the Tr PFT was subdivided into "wet" (Tr 1 ) and "dry" (Tr 2 ) tropical raingreen trees.
The cornerstone of research concerned with the composition and distribution of Latin

383
American vegetation, be it in a contemporary time frame, or one that aims to work in the past, is a good understanding of the ecology and distribution of the taxa concerned. The Latin American pollen taxa were assigned to one or more PFTs depending on the modern ecological range of the most important (i.e. most abundant) taxa responsible for producing the pollen identifiable within the modern data set. These assignments 5 were made following reference to the known biology of plants from several floras (Rzedowski, 1983;Schofield, 1984;Wingenroth and Suarez, 1984;Kahn and de Granville, 1992;Maberly, 1993;Seibert, 1996), botanical and palynological studies (Beard, 1955;van der Hammen, 1963van der Hammen, , 1972Wijmstra and van der Hammen, 1966;Eiten, 1972;Cleef and Hooghiemstra, 1984;Hooghiemstra and Cleef, 1984;10 Pires and Prance, 1985;Cuatrecasas and Barreto, 1988;Brown and Lugo, 1990;Bush, 1991;Dov Par, 1992;Kappelle, 1993;Duivenvoorden and Cleef, 1994;Witte, 1994;Armesto et al., 1995;Harley, 1995;Kappelle, 1995;Kershaw and McGlone, 1995;Veblen, et al., 1995;Colinvaux, 1996;Grabherr, 1997;Hooghiemstra and van der Hammen, 1998) and personal communication with modern ecologists 15 and palaeoecologists. Much of this information has been collated into a dictionary on the distribution and ecology of parent taxa of pollen lodged within the Latin American Pollen Database (Marchant et al., 2002c). The resultant taxon vs. PFT assignments are presented in Table 5. Due to the high intra-generic diversity, and also the wide range of ecology's exhibited by the parent taxa present within some genera, a number 20 of taxa were multiply assigned to a number of PFTs; where possible, pollen taxa were assigned to the PFTs within which the parent taxa are most common. Thus, the identified PFTs from Latin America are described by the suite of pollen taxa assigned to them, in turn the biomes are distinguished by the suite of constituent PFTs. A number of pollen taxa belong to more than one PFT, and, as is the case 25 with the potential vegetation, most PFTs contribute to more than one biome. Two problems can arise here for our analysis that can be circumvented by manipulation of the input matrices and output biome affinity scores. First, pollen samples can have equal maximum affinity with more than one biome; this commonly occurs when the PFTs characteristic of one biome are a subset of another biome. Assigning the biomes so that subsets always come first in the analysis solves the problem. A second problem arises where multiple samples encompass the age boundaries. Multiple samples from a single site may have maximum affinity to a number of different biomes; the chance of this is high when the score of the "best" biome is close to that of the next "best". 5 In such cases, the "majority" biome is mapped. For example, site A contains eight samples within the time frame of 6000±500 14 C yr BP, five samples have the greatest affinity to biome 1, two samples to biome 2 and one sample to biome 3. The result is that biome 1 is mapped for site A at 6000±500 14 C yr BP.
Biomes were reconstructed from pollen data at sites with surface sample, trap and 10 radiocarbon-dated core-top data. The results were used to produce a modern pollenderived biome dot map (Fig. 7); for each site a colour dot records the reconstructed biome with the highest affinity score. These were compared site by site, with the potential modern vegetation distribution (Fig. 2). The biomisation procedure was applied to the fossil datasets without modification. Results for all sites and periods are pro- 15 vided in Table 6 which allows a site-by-site comparison through time and a comparison between the modern reconstruction and potential vegetation.

Modern pollen vs. potential biome reconstruction
Visual comparison shows that the biomes reconstructed from modern pollen data 20 ( Fig. 7) accurately reflect the broad features in the potential vegetation map (Fig. 2). In particular the modern reconstruction correctly reproduces the transition from relatively mesic vegetation types, around the coastal areas of South America, to the more xeric biomes towards the interior. For example, in eastern Argentina there is a transition from steppe to xerophytic woods and scrub. Warm temperate rain forest is an impor-25 tant biome in the southern and southeastern Brazilian highlands, with tropical dry forest 385 being reconstructed towards the interior. Notable from this region is the large number of different biomes being reconstructed in a relatively small area. In part this reflects the variability of potential vegetation, not portrayed in our relatively coarse resolution vegetation map (Fig. 2). For example, a site recording tropical rain forest reflects the sites' lowland position where it is characterised by moist gallery forest with a number of 5 typical rain forest taxa present. Steppe is correctly reconstructed from the grasslands of south-eastern Argentina and dry forest in central Argentina, mirroring the transition to "drought-deciduous thorn forests" of central Argentina (Schmithüsen, 1976). Steppe is assigned farther west at approximately 1000 m in the Andes, southernmost South America and northeast Brazil. The vegetation of southern South America is dominated 10 by cool temperate rain forest. The failure of the analysis to pick up the transition from cool temperate rain forest to cool grass/shrubland as one progresses east along Tierre del Fuego stems from the pollen spectra having a relatively large amount of Nothofagus pollen. Moving northwards from southern South America there is a transition to cool mixed forest, cool grass/shrubland and steppe; these latter assignments are par- 15 ticularly associated with eastern flanks of the Cordillera de los Andes. The concentration of sites along the Andes results in a wide range of reconstructed biomes being geographically adjacent to each other when mapped in two-dimensional space (Fig. 7). This phenomenon is most apparent in the northern Andes where the altitudinal, and therefore climatic gradients are at their steepest. Despite these rapid en-20 vironmental changes, the biome assignments reflect the changing vegetation patterns. There is a clear altitudinal transition: low altitudes (<300 m) being mainly assigned to the tropical rain forest, tropical dry forest, tropical seasonal forest and steppe. Sites located at mid altitudes are described by a number of different biomes including tropical seasonal forest, warm mixed forest, cool mixed forest and cool temperate rain forest. 25 Within this wide range of warm temperate rain forest and tropical seasonal forest are commonly assigned at lower elevations (Fig. 7, Table 6). Many of the sites at high altitude have a high affinity to the cool grass/shrubland biome. The line of the Andes can be easily seen by the cool grass/shrubland biome assignments, these being commonly recorded at sites above 3800 m. The warm temperate rain forest is assigned at lower elevations and is analogous to Andean forest, being dominated by Podocarpus, Quercus and Weinmannia and comprises a different forest type to that assigned in southern and southeast Brazil or southern South America. A mixture of biomes presently characterises the Amazon Basin with only four sites recording the tropical rain forest biome; 5 two of these are in coastal locations. Tropical seasonal forest is recorded in four locations; this representing a slightly drier type of forest than tropical rain forest, containing some deciduous taxa. A number of "Amazonian" sites record warm temperate rain forest, these assignments responding to the presence of Andean floristic elements within lowland vegetation. There are a number of sites that record tropical dry forest, this 10 being relatively widespread, e.g. on Easter Island, lowland Colombia and the Brazilian interior. Warm temperate rain forest describes the majority of the sites in the Panamanian and southern Costa Rican isthmus with warm mixed forest, being commonly at higher altitudes. This is an area where the comparison between the observed and predicted biomes shows a discrepancy; the possible reasons behind this will be discussed 15 fully. Warm mixed forest is correctly assigned to the highlands of central and southern México as is tropical dry forest on the Yucatán peninsula of southern México.
Investigating the correspondence between the pollen-based reconstruction and the potential vegetation for individual biomes provides a check on the methodology, particularly the construction of the matrices. The cool grass/shrubland biome is accurately 20 reconstructed at the majority of sites. The sites that do not match the potential vegetation commonly result from the inclusion of high altitude arboreal pollen, this resulting in assignments of cool mixed forest and cool temperate rain forest. The other common assignment is towards steppe; the dominance of the pollen spectra by Poaceae, and lack of shrubby taxa, result in the assignment to the steppe biome. Indeed, the 25 affinity scores to the cool grass/shrubland and steppe biome at most sites, where one of these biomes is dominant, is normally quite close. For the cool mixed forest biome 66% of sites accurately reconstruct the potential vegetation. The 34% of "wrong" assignments mainly result in either a reconstruction of cool grass/shrubland, thought to 387 represent possible forest clearance, and the dominance of the vegetation by grassland, or cool temperate rain forest biome due to the numerous shared taxa between these two biomes. 75% of cool temperate rain forest biome reconstructions match the potential vegetation at the site. The remaining 25% of the sites mainly show either warm mixed forest or warm temperate rain forest assignments. For the tropical dry 5 forest biome some 90% of the sites accurately reflect the potential vegetation at the site. For the remaining 10% of "wrong" assignments the common result is towards a cool temperate rain forest or steppe. For the tropical rain forest biome 85% of the sites accurately reflect the potential vegetation. For the sites that do not match, a common reconstruction is warm temperate rain forest. This can be explained by the number of 10 Andean elements being present within lowland tropical forests with a couple of sites reconstructing the closely related biome of tropical seasonal forest. This facet of the pollen data is also exemplified by a number (35%) of the tropical seasonal forest sites recording the warm temperate rain forest biome. Warm evergreen forest is correctly assigned at 80% of the sites. Warm temperate rain forest is assigned correctly at 78% 15 of sites. 75% of the sites that do not reconstruct this biome "correctly" lead to assignments of tropical rainforest and tropical seasonal forest at low altitudes (<500 m) and cool mixed forest at high altitudes (>2000 m) .
The generally correct biome assignments, in relation to a map of potential vegetation confirm the robustness of our application of the biomisation method to Latin America. 20 Where the match between pollen and potential vegetation reconstructions is relatively low (tropical seasonal forest and warm temperate rain forest), then a common forcing factor, that of "high altitude" plants presently growing at low altitudes appears important. For other sites where the reconstructed biome does not match the potential vegetation map a series of different explanations, particularly local site-specific factors such as 25 human impact, can be invoked, these will be discussed fully. Taking our modern pollen to potential vegetation calibration, and the design of the matrices that drive it, we reconstruct vegetation at past time intervals with "cautious confidence".

6000 14 C yr BP biome reconstruction
The biomes reconstructed at 6000±500 14 C yr BP (Fig. 8) show relatively small patterns of change compared to the present. 85% of the sites retain the same biome assignment as present (Table 6). In southeastern Brazil, the majority of the sites that were previously assigned to the warm temperate forest biome remain unchanged. A number 5 of sites (e.g. Serra Campos Gerais, Rio São Francisco, Aguads Emendadas) record tropical dry forest at 6000±500 14 C yr BP replacing tropical seasonal forest (Laguna Angel, Laguna Chaplin) record at the present. Sites assigned to tropical rain forest and tropical seasonal forest today mainly remain unchanged at 6000±500 14 C yr BP.
Steppe continues to be reconstructed in southeastern Argentina, as today. However, 10 a number of sites had substantially more arboreal components at 6000±500 14 C yr BP than today, for example Empalme Querandíes and Lake Valencia show a transition from steppe to tropical dry forest. An expansion of steppe is recorded at sites previously assigned to cold mixed forest on the Cordillera de los Andes. Unlike sites in southern-most South America that similarly record steppe, these sites also contain 15 significant amounts of Alnus and Podocarpus indicative of "parkland" at this time. On closer inspection of the affinity scores, sites record an increased affinity to cool temperate rain forest, primarily due to increased amounts of Nothofagus pollen, although this was not sufficiently numerous to produce a cool temperate rain forest assignment. Southernmost South America continues to have a mixture of cool mixed forest, cool 20 grass/shrubland, steppe and cool temperate rain forest biomes, the latter being dominant. Along the southern Andean spine, the assignments do not differ greatly from the modern assignment. There are broadly similar assignments to the present at Colombian sites although there is a slight increase in the number of cool mixed forest and cool temperate rain forest biome assignments relative to cool grass/shrubland of the 25 present day. The sites where this occurs (e.g. La Primevera, and Páramo de Peña Negra) are located at high altitude sites and may reflect either a lowering of the forest line or increased distribution of Andean forest that predates early human impact. Sites 389 in coastal northern South America show a transition to tropical dry forest and tropical seasonal forest from steppe and tropical dry forest respectively, both indicative of a relatively mesic environment. A number of sites on the Yucatán peninsula show a clear distribution of warm evergreen forest at 6000±500 14 C yr BP changing from the warm mixed forest and tropical dry forest reconstruction for the present day. These transi-5 tions are not recorded everywhere, for example sites located in the Mexican highlands retain the same biome assignment at the present -warm mixed forest.

18 000 14 C yr BP biome reconstruction
Vegetation at 18 000±1000 14 C yr BP was substantially different from the present-day, or that reconstructed at 6000±500 14 C yr BP (Fig. 9). The intensity of this vegetation 10 transformation is demonstrated by 82% of the sites change the biome assignment relative to the two previous periods. In Amazonia, tropical seasonal forest and tropical dry forest is recorded instead of tropical rain forest or tropical seasonal forest reconstructed for the present. A site on the present southern Amazonian boundary (Laguna Chaplin) records tropical seasonal forest. Sites in southern South America nearly all 15 sites show a transition from cool mixed forest biomes and cool mixed forest to cool grass/shrubland and cool grassland. However, within this homogenous reconstruction a number of sites have a relatively high affinity to the cool temperate rain forest biome, due to a mix of Donartia and Nothofagus pollen: this explains why the northernmost site in this cluster records cool temperate rain forest. Sites in southeastern Brazil record a 20 transition from tropical dry forest to tropical seasonal forest and cool grass/shrubland. Sites in Amazonia record mainly tropical seasonal forest, warm temperate rain forest or steppe; this combination indicating relatively mesic forest. In the Colombian lowlands, tropical dry forest continues to be assigned whereas Colombian highland locations reflect a marked change from cool temperate rain forest and cool mixed forest to the 25 cool grass/shrubland biome. Sites in Central America show a change from tropical seasonal forest to tropical dry forest, e.g. El Valle. The Mexican highland sites remain unchanged, continuing to support warm mixed forest with the pollen records being 390 dominated by Pinus and Quercus.

Discussion and conclusions
Previous applications of the biomisation method in Africa (Jolly et al., 1998a;Elenga et al., 2000), China (Yu et al., 1998(Yu et al., , 2001, Australia (Pickett et al., 2004), Eastern North America (Williams et al., 2000), Eurasia (Tarasov et al., 1998a), Europe (Prentice et 5 al., 1996a, b;Tarasov et al., 1998a, b;Elenga et al., 2000), Japan (Takahara et al., 2001) and Western North America ( Thompson and Anderson, 2000) demonstrate that technique is able to translate multi-site pollen data to coarse resolution vegetation reconstructions that works well over a range of vegetation types. The Latin American results presented here provide a further test of this ability. The ability of the biomisa-10 tion method to reconstruct biomes derives in part from the relatively coarse vegetation classification (Fig. 2); which conceals significant intra-biome variation; for example, we do not distinguish subtypes of the warm evergreen forest biome which contains Araucaria in southern and southeastern Brazil and Podocarpus in the northern Andes. The success of the biomisation technique is in part be due to reconstructions being carried 15 out at a regional scale, allowing the methodology to be adapted to the local flora, bioclimatic gradients and pollen spectra. For example, the treatment of Quercus pollen in Latin America is quite different from that in a European context. Similarly, in Africa Podocarpus is assigned to the warm temperate broad-leaved evergreen PFT (Jolly et al., 2001), although this taxon is a coniferous needle-leafed tree, in Latin America it is 20 assigned to cool and intermediate temperate conifers. This regional focus also allows the pollen to plant functional type allocations to be based on good ecological information concerned with environmental tolerances to growth limits and an understanding of how representative the pollen is of the surrounding vegetation. This is particularly important as the pollen taxa identified to the generic level (the taxonomic level usu-25 ally identified to) exhibit considerable plasticity in their growth form and environmental tolerance. For example, within the genus Cordia, commonly a woody shrub of open 391 thorn woodland, of the northern Andes (Cleef and Hooghiemstra, 1984) two species of Cordia are herbs in cerrado (Pereira et al., 1990;Sarmiento, 1975), the genus is also present (C. lomatoloba and C. sagotii) in Amazonian terra firme forest and Guyanese lowland rain forest (Steege, 1998). Furthermore the specific nature of pollen production, dispersal and incorporation into a sedimentary environment exhibits considerable 5 variability that is part dependent on site characteristic. All these factors have a bearing on the results and need to be considered in the designing of the input matrices into the biomisation process and interpretation of results. Biomes are mainly accurately reconstructed for the present-day even though large areas of Latin America are covered with vegetation that has been altered by a long 10 history of human land use (Behling, 1996;Binford et al., 1987;Fjeldså, 1992;Gnécco and Mohammed, 1994;Gnécco and Mora, 1997;Marchant et al., 2004;Northrop and Horn, 1996). One possible reason for this may relate to the nature of the "modern" samples. Within our analysis the modern samples are largely derived from sedimentary columns rather than surface trap pollen data, and hence they may stem from the 15 last 500 years and be reflective of a period prior to intensive human-induced change. However, the signal of vegetation clearance does impact on the modern reconstruction as shown by the large number of sites recording cool grass/shrubland, particularly at lower and mid-altitudes that should support cool mixed forest or cool temperate rain forest. These assignments are thought to result from human impact with the pollen 20 spectra being dominated by Poaceae and hence recording more open vegetation. To quantify the nature of this impact, it is possible to tailor the biomisation methodology to include elements of the pollen spectra, such as agricultural and ruderal taxa, that may indicate human impact . The ability to reconstruct potential, rather than actual vegetation, may also relate to the type of impact; although spatially 25 relatively widespread, forest clearance is often only partial with many localised patches of forest and secondary vegetation remaining. This results in the floristic composition of the remaining vegetation, in palynological terms at least, closely reflecting the original vegetation composition. For example, the forest surrounding the Fúquene-II site is a successional type of forest whereas the natural vegetation would be a Andean forest type dominated by Quercus and Weimannia mixed with Croton, Oreopanax and Phyllanthus (Van Geel and Van der Hammen, 1973). In addition to the relatively coarse potential vegetation and biome classification, mapping the highest biome affinity score to each site as a single dot also allows the method to be relativity robust. Although this 5 is suitable for the relatively coarse reconstructions necessitated by the continental/subcontinental scale, when investigating a small area, more information can be preserved from the analysis. Indeed at a regional scale information on sub-dominate biomes can be kept (Marchant et al., 2001a), new more defined biomes (Bigelow et al., 2003) or at a site-specific scale where the affinity scores in all the biomes can be retained (Marchant et al., 2001b(Marchant et al., , 2002b. Despite the overall agreement between potential and reconstructed biomes a number of locations show anomalies. Due to the floristic and structural similarities between warm and cool grasslands (Tarasov et al., 1998a), grass-dominated biomes can be particularly difficult to distinguish from one another. Differentiation is possible by the 15 other plants within steppe and cool grass/shrubland, although there remains a high affinity score to the cool grass/shrubland at low altitudes with the reverse for the steppe biome at high altitudes. Another facet is that some lowland sites show reconstructions of highland biomes, e.g. sites in central Panama and Amazonia recording warm temperate rain forest. This result is driven by the presence of genera that are typical of 20 montane vegetation, e.g. Hedyosmum, Podocarpus and Quercus. A possible explanation for the presence of these highland elements is that they are relictual; relatively isolated today they were previously much more widespread under the glacial climate norm of the Quaternary. This suggestion is supported by the similarity, at a generic level, of the flora in highland Brazil and the northeastern Andes, and the isolated patches of 25 savanna within Amazonian forest and the Brazilian cerrado. Furthermore, it is interesting that the presence of highland elements appears to be greater when the moisture levels are high. For example, within the Chocó Pacific region, where rainfall exceeds 15 000 mm yr −1 , montane elements appear more common than within Amazonia (Gen-393 try, 1986).
Notwithstanding some of the anomalies mentioned, the biomisation method applied to Latin American pollen data can reconstruct large-scale vegetation patterns despite many pollen taxa having different ecological interpretations under different environmental settings (Grabandt, 1980), representation of parent vegetation by pollen likely to be 5 subject to inter-annual variability (Behling et al., 1997b), and tropical vegetation being difficult to reconstruct through pollen assemblages (Bush, 1991;Mancini, 1993;Bush and Rivera, 1998;Behling et al., 1997). These factors demonstrate the importance of basing the input matrices for the biomisation process on all the available ecological information that allowing for the multiple assignment of the pollen taxa to the PFTs. Compared to the present, the sites at 6000±500 14 C yr BP record either the same biome or one indicating more xeric vegetation. Dry environmental conditions in southern Brazil extend from the early Holocene until approximately 4500 14 C yr BP when 15 there was an increase in arboreal taxa (Alexandre et al., 1999). Maximum aridity in southeast Brazil was reached between approximately 6000 and 5000 14 C yr BP, prior to the transition to a modern climate (Behling, 1997a). The driest phase in central Brazil is at approximately 5000 14 C yr BP; relatively moist climate conditions similar to today setting in after 4000 14 C yr BP (Ledru, 1993;Marchant and Hooghiemstra, 2004). Al- 20 though fire has been proposed as being responsible for late Holocene variation in the forest/savanna boundary in Brazil (Vernet et al., 1994;Desjardins et al., 1996;Horn, 1993), this relative aridity is thought to reflect an extended dry season during this period (Behling, 1997b). An extended dry season may explain why Araucaria-dominated forest were still restricted in their distribution relative to the modern day, not signifi- 25 cantly increasing in range until approximately 3000 14 C yr BP (Behling, 1997a). From our analysis the temporal perspective is missing, hence, we are unable to indicate if the vegetation reflects a stable dry period, or a period where there are alternating periods of dry and humid climates linked for example to El Niño activity (Martin et al., 1993;Sifeddine et al., 2001). A relatively dry phase is also recorded in northwestern Argentina between 7500 and 5800 14 C yr BP (Schäbitz, 1991). Although pollen assemblages do not lend themselves well to distinguishing moisture from temperature 5 changes, stable hydrogen isotope analysis on mosses show the vegetation of southern South America is highly sensitive to changes in moisture regime (Pendall et al., 2001). The predominance of steppe in southeastern Argentina agrees with the reconstruction by Prieto (1996): steppe characterising the area between 7000 and 5000 14 C yr BP.
Locally high moisture levels at sites closer to the Atlantic Ocean (Prieto, 1996) may 10 explain why sites under strongest maritime influence (Aguads Emendadas, Cerro La China) changes from steppe to tropical dry forest as the local environment is able to support more arboreal taxa. In soutwestern Patagonia a sustained increase in Nothofagus pollen has been detected from around 6800 yr BP thought to result from locally increased moisture levels (Villa-Martinez and Moreno, 2007) Locally increased moisture 15 levels in this part of Latin America during the mid Holocene are though to stem from intensification of the southern Westerlies (Gilli et al., 2005). Farther west, cool temperate rain forest assignments indicate a similar climatic regime and the maintenance of Valdivian rain forest (Villigran, 1988). A dry phase is also recorded at many Andean sites, for example, in northern Chile desiccation 20 of the Puna ecosystem is recorded between 8000 and 6500 14 C yr BP (Baied and Wheeler, 1993;Villigran, 1988). In lowland Chile, the period of maximum aridity occurred between 9400 and 7600 14 C yr BP with drier than present conditions continuing until 5000 14 C yr BP (Heusser, 1982), this could explain the increased presence of steppe at sites along the southern Andes. On the central Peruvian Andes, a dry 25 warm climate was encountered between 7000 and 4000 14 C yr BP (Hansen, Seltzer and Wright, 1994). δ 18 O measurements from an ice core record taken from highland Peru show that mid-Holocene climatic warming and drying was recorded from 8200 to 5200 14 C yr BP with maximum aridity between 6500 to 5200 14 C yr BP (Thompson et   395 al., 1995). Farther north on the Bolivian Andes, a dry phase is recorded from approximately 5500 14 C yr BP (Abbot et al., 1997). The slight increase in the number of arboreal biome assignments at northern Andean sites can be interpreted as an up-slope shift of forest line. This conforms to the suggestion based on pollen data by van Geel and van der Hammen (1973) that the vegetation zones in the northern Andes were 5 several hundred of meters higher than the present at approximately 6000 14 C yr BP.
Relatively dry conditions have also been indicated for lowland Colombia for the mid-Holocene although the onset of dry conditions varied considerably between sitesoccurring between 6500 and 4500 14 C yr BP (Behling et al., 1999). Added complexity is caused by steep environmental gradients associated with non-climatic factors. For 10 example, the presence of the tropical dry forest biome in lowland Colombia, e.g. the catchment of El Piñal, results from a combination of strongly seasonal conditions at present and locally strong edaphic influence (Behling and Hooghiemstra, 1999). Farther north, the assignment of Lake Valencia to the tropical dry forest is in agreement with the site-specific interpretation that more arboreal taxa (Bursera, Piper and 15 Trema) were present after approximately 10 000 14 C yr BP due to the onset of a more humid climate (Bradbury et al., 1981): these tropical raingreen taxa indicative of a seasonal climate with relatively dry conditions. This appears to be a regional signal as early Holocene evergreen forests of northern Venezuela were replaced by semi-deciduous elements during the mid-Holocene (Leyden, 1984). Enhanced precipitation over Cen-20 tral America being accompanied by a northward shift of the ITCZ, enhanced southerlies and cooler equatorial sea surface temperatures (Harrison et al., 2003). Low lake levels in central Panama also indicate that environmental conditions at this period were more xeric (Piperno et al., 1991b;Bush et al., 1992) whereas sites on the Yucatán peninsula show a shift to warm evergreen forest where the warmer conditions that characterise 25 the early Holocene persisted until approximately 6500 14 C yr BP (Brown, 1985). This result may stem from locally high moisture levels as a result of maritime influence, a similar mechanism having being proposed to explain a comparable shift in coastal Brazil and Argentina. Despite the majority of the evidence for a mid-Holocene dry pe-riod, there still remains a debate about the intensity, and even the occurrence, of this.  suggests that most savanna areas were characterised by increased rainfall between 7000 and 6000 14 C yr BP although there is considerable variation in the timing of the onset of more humid conditions so it may be that such a mesic period falls outside our temporal window. 5 One of the main mechanisms used to explain moisture shifts is fluctuations in the Southern Oscillation and the migration of the ITCZ (Martin et al., 1997). Martin et al. (1997) suggests that during the mid-Holocene, the ITCZ was located farther north than its present-day position (Fig. 1) -this would produce a summer rainfall deficit and increased winter precipitation; in short greater seasonality. Rather than changes 10 in the median position of the ITCZ, changes in the character of the ITCZ oscillation, such as greater latitudinal range for annual migration, can be invoked to explain vegetation changes (Behling and Hooghiemstra, 2001). However, due to the topographical influence of the Andes and the convergence of westerly and easterly winds, the ITCZ has a sinusoidal profile over northern South America (Fig. 1). Therefore, moisture 15 changes over northeastern South America are likely to result from the importance of convective moisture sources; reduced precipitation, particularly in mid latitude western South America, following reduced intensity of westerly climate systems. It is also possible that episodic dry events that presently occur in South America in relation to sea-surface temperature anomalies of the Pacific Ocean (ENSO) were more frequent 20 in the mid-Holocene (Markgraf, 1998). This later suggestion may also have led to the increased fire frequency indicated in southeast Brazil (Alexandre et al., 1999).
This regression of the forest during the mid-Holocene (8000 to 6000 14 C yr BP) in the southern tropical zone of Latin America is opposite to full forest development in Africa (Servant et al., 1993;Jolly et al., 1998a) and this spatial relationship between 25 Latin American and Africa warrants further investigation (Marchant and Hooghiemstra, 2004). A particular target for the investigation could be the impact and feedbacks of vegetation changes on climate. For example, large changes in African vegetation about the Sahel are suggested to have been important in influencing Indian monsoon 397 dynamic (Doherty et al., 2000). Such a phenomena of vegetation feedbacks on the climate system appears weaker in South America than in Africa although it is likely to have had an impact as yet unqualified. Certainly Latin America would benefit from targeted model applications in the same way that has been applied to Africa (Kubatzki and Claussen, 1998;Doherty et al., 2000). This modelling of climate dynamics Latin 5 American represents a special challenge for climate models and modellers (Valdes, 2000) primarily due to the steep environmental gradients and rapid transition from one biome to another (Fig. 2 , 2008). Ice caps were present on the southern tip of South America which spread onto the plains and the coastal area (Heine, 1995). Evidence from glacial moraines also indicates considerable expansion of Andean glaciers (Hollis and Schilling, 1981;Villagran, 1988;Birkland et al., 1989;Seltzer, 1990;Thouret et al., 1997). Most of southern South America was characterised by an erosional environment; locations that would later accumulate sediments were glaciated, or subject to fluvial activity (Heine, 2000). This situation is recorded by ice cores from the high Andes that contain large amounts of dust about the LGM, this being derived from 398 surrounding deflating desert areas (Thompson et al., 1995). This cold, arid environment is clearly reproduced by the vegetation which shows a transformation from the cool temperate rain forest to the cool grass/shrubland biome. tive to the present day; not as a discrete forest type but as a parkland type vegetation mosaic (Villagran, 1988), not forming closed forests until the early Holocene (Schäbitz, 1994;Heusser, 1995). This vegetation is evidenced with the analysis presented here by the northernmost site (Laguna Six Minutes) recording cool temperate rain forest. However, it is unlikely this represents closed forest persisting in the area, trees be- 15 ing present within a woodland/steppe vegetation mosaic (Villagran, 1988). The rate of spreading of this forest into the Holocene would probably have been strongly dependent on the density of the parent plants from the initial seeding fraction (Huntingford et al., 2000). The maintenance of cool temperate rain forest taxa, albeit at relatively low levels, may result from high moisture levels as recorded by high lake stands at this 20 time (Markgraf et al., 2000). These may reflect outbreaks of polar air and subsequent generation of low-pressure systems in the western Atlantic; combined with lower temperatures this situation would lead to a positive water balance. Indeed, the presence of relatively local moisture sources would have been important at the LGM and allow us to explain regional patterns of biome change outside the influence of the ITCZ migrations 25 (Markgraf et al., 2000).
Considering the sites along the northern Andes, it is clear from the vegetation that climate was colder during the LGM, reductions up to 12 • C may have been reached at very high altitudes (Thompson et al., 1995). A substantial temperature depres-399 sion during the last glacial period is mirrored by a significant impact on the vegetation composition and distribution. From our analysis it is apparent that the tree line was significantly lower at the LGM, concordant with a suggested lowering of vegetation zones by approximately 1000 to 1500 m relative to the present-day position (Monslave, 1985;Wille et al., 2001). At lower elevations in western Colombia, a more conserva-5 tive depression of the vegetation has been suggested from Timbio (Wille et al., 2000). Indeed, the spatial character of the cooling and drying in the Neotropics is still under debate (Markgraf, 1993;Colinvaux, 1996;Hooghiemstra and Van der Hammen, 1998;Farrera et al. 1999;Boom et al., 2002). Greater temperature change at high altitudes compared with those at low altitudes and at the sea surface (CLIMAP, 1976) can be ex-10 plained in terms of changes in lapse rate Peyron et al., 2000;Wille et al., 2001) or compression of vegetation belts (Van der Hammen and Absy, 1994). The lapse-rate gradient is partly influenced by atmospheric moisture levels (Barry and Chorley, 1990). As precipitation was reduced at the LGM, an overall steeper lapse rate, particularly at higher altitudes where moisture reductions would have been high- 15 est, seems likely (Wille et al., 2001). The extent to which lapse-rate changes can be used to explain spatially different signals from the data must be used with caution, particularly as most palaeoclimatic reconstructions have been carried out with some kind of modern analogue-driven transfer function (Farrera et al., 1999). These reconstructions commonly do not take into account non-climatic parameters which would impact 20 on vegetation composition and distribution such as volcanic activity (Kuhry, 1988), fire (Cavelier et al., 1998;Rull, 1999), UV-B insolation (Flenley, 1998) or atmospheric composition, in particular changing CO 2 levels (Woodward and Bond, 2004). For example, concentrations of CO 2 reduced to glacial levels (200 ppmV) have been shown to have a very significant impact on tropical vegetation (Jolly and Haxeltine, 1997;Boom et al., 25 2002;Marchant et al., 2002b).
In south-east Brazil vegetation at the LGM was characterised by tropical dry forest and tropical seasonal forest; this latter vegetation type may have been restricted within deep valleys and along waterways; site-specific records from southeast Brazil indicate open grasslands (campo limpo) with forest elements being retained as gallery forest (Behling, 1997a). Some model reconstructions of global vegetation patterns have indicated that there was an increase in warm evergreen forest in Brazil at the LGM at the expense of tropical seasonal forest (Prentice et al., 1993). This pattern of change is supported by the data presented here where plants generally found at high altitudes to-5 day were more common in Amazonia at the LGM. Clapperton (1993) used geomorphic data to infer a very sparsely vegetated landscape on the Brazilian Highlands, possibly relating to our reconstruction of steppe for a site in eastern Brazil. However, the cool grass/shrubland biome appears to be a common type of vegetation at this time. Cold climates in eastern South America could have resulted from the incursion of polar cold fronts that would occasionally reach northwards of the equator (Ledru, 1993;Behling and Hooghiemstra, 2001). This phenomenon could combine with equatorward shift of polar high-pressure areas and mid-latitude cyclones resulting in displacement and compression of the subtropical anticyclone between mid latitudes westerlies (Dawson, 1992). This climatic regime would result in more restricted migration of the ITCZ and 15 pronounced aridity that would have been compounded by lower sea surface temperatures and associated reduction in atmospheric moisture.
At altitudes of approximately 3000 m in northern Peru vegetation about the LGM comprised a mixture of wet and moist montane forest elements with open woodland (Hansen and Rodbell, 1995); this vegetation association having no modern analogue. 20 Although the Andes remained relatively moist at the LGM, particularly in the northern part where the concave shape of the mountain chain entrap moisture from the rising air (Fjeldså et al., 1997), it is not certain what occurred in the lowland areas to the east of the Andes (Colinvaux, 1989;Colinvaux et al., 1997;Bush et al., 1990;Thouret et al., 1997). In the Colombian lowlands two sites are characterised by the tropical dry 25 forest biome, this agrees with the suggestion from a pollen study at Rondonia that very open savanna characterised the catchment at the LGM (van der Hammen and Absy, 1994). Similarly, sparse vegetation cover would have been present on the Plateau of Mato Grosso (Servant et al., 1993) and is likely to have extended along the coastal 401 areas of Guyana and Surinam (Wijmstra, 1971) -this scenario is supported by our analysis, i.e. a site in lowland Panama recording tropical seasonal forest. Although the majority of the area presently covered by drier types of tropical forest would probably have been replaced by more open woodland at the LGM, environmental changes in savannas at the LGM appear to have been spatially complex. Whether the drier, 5 cooler, conditions resulted in restricted range forest refugia cannot be answered from the available evidence although the vegetation appears heterogeneous as a mosaic of Andean, savanna and tropical rain forest taxa combined. Indeed, this reiterates the suggestion by Colinvaux et al. (2001), now widely accepted within the palaeoecological community, that plants responded to Quaternary climate changes as individuals 10 not as biomes. Therefore, to fully investigate vegetation response to climate change is necessary to retain information contained within the affinity scores to the sub-dominant biomes , or to carry out the analysis at the PFT level. Indeed this approach would allow investigations into which elements of the vegetation were particularly sensitive to environmental change. Expansion of savanna could have been 15 aided by reduced CO 2 concentrations and the resultant competitive advantage attained by C 4 grasses over C 3 plants (Haberle and Maslin, 1999;. Within highland México, warm mixed forest continues to be reconstructed due to the presence of Pinus and Quercus-dominated forests. Although the same biome is recorded at all these periods, it unlikely to be analogous to present day mixed forest; 20 this was characterised by sparsely forested temperate scrub (Binford et al., 1987). Indeed, a strong aridity signal is directly recorded by low lake levels in central México due to reduced northern excursion of the ITCZ, trade wind circulation, and ensuing reduced oceanic-land moisture transfer (Markgraf et al., 2000) that would have been reflected in ecosystem response. For example, forest on the Pacific side of the Central America contained a mosaic of high and low altitude forest species; a similarly novel type of forest has also been shown for Mera, Ecuador (Liu and Colinvaux, 1985) and Peten, Guatemala (Leyden, 1984). Of the two sites that record the warm evergreen forest biome at this period a site in Guatemala was dominated by Chenopodiaceae, Juniperus, Pinus and Quercus.
We have presented vegetation reconstructions throughout Latin America at 6000 ±500 14 C yr BP and 18 000±1000 14 C yr BP using an objective method based on biomes, constituent PFTs that are described by a set of unique pollen spectra. As a unified methodology has been applied to the pollen data, this reconstruction 5 of biomes provides an objective basis for interpreting large-scale vegetation dynamics, and the environmental controls on these over the Late Quaternary and can be used as a dataset for model-data comparisons at 6000 and 18 000 yr BP. Changes at 6000±500 14 C yr BP, although relatively small, indicate a transition to more xeric vegetation. The changes at 18 000±1000 14 C yr BP are more homogenous and indicative 10 of a cooler, drier climate. These reconstructions are consistent with numerous sitespecific interpretations of the pollen data. The success of the reconstruction has in part been determined by the coarse resolution of biome definitions, and using the most dominant biome for description and interpretation of the results. To develop understanding of vegetation response to environmental change, and possible feedbacks, 15 information that is presently redundant should be retained and the results combined with climate/vegetation modelling initiatives. It is apparent from the relatively sparse coverage, in comparison to Europe and North America, that the Late Quaternary vegetation history of the Neotropical phytogeographical realm remains still relatively poorly resolved despite its importance in model testing, developing biogeographical theory 20 (Tuomisto and Ruokolainen, 1997), and understanding issues concerned with biodiversity and human-environment interactions. It has been shown that environmental change is rarely spatially uniform and as such necessitates an even greater number of sites to determine more precisely this complexity and the driving mechanisms behind this. New sites, located in key areas, combined with the application of a range of prox-  Absy, M. L., Cleef, A., Fournier, M., Servant, M., Siffedine, A., Da Silva, M. F., Soubies, F. Suguio, K., Turcq, B., and van der Hammen, T.: Mise enévidence de quatre phases d'ouverture de la forêt dense dans la sud-east de l'Amazonie au cours des 60 000 dernières années, Première comparison avec d'autres regions tropicales, Competes Rendues Academie Science Paris, 313, 673-678, 1991. 5 Alexandre, A., Meunier, J. D., Mariotti, A., and Soubies, F.: Late Holocene phytolith and carbonisotope record from a latosol at Salitre, South-Central Brazil, Quaternary Res., 51, 187-194, 1999. Almeida, L.: Vegetación, fitogeographica y paleoecologíca del zacatonal alpine y bosques montanos de la región central       , 25, 997-1005, 1998a. Jolly, D., Harrison, S. P., Damnati, B., andBonnefille, R.: Simulated climate and biomes of Africa during the Late Quaternary: comparisons with pollen and lake status data, Quaternary Sci. Rev., 17, 629-657, 1998b.   25 temperate, and arid environments in Argentina, Palynology, 7, 43-52, 1983. Markgraf, V.: Late Pleistocene and Holocene vegetation history of temperate Argentina: Lago Morenito, Bariloche, Dissertationes Botanicae, 72, 235-254, 1984. Markgraf, V.: Late Pleistocene faunal extinction in southern Patagonia, Science, 228, 1110-1112, 1985a: Paleoenvironmental changes at the northern limit of the sub-Antarctic Nothofagus forest, latitude 37S, Argentina, Quaternary Res., 28, 119-129, 1987. Markgraf, V.:  Climates since the Last Glacial Maximum, edited by: Wright, H. E., Kutzbach, J. E., Webb III, T., Ruddiman, W. F., Street-Perrot, F. A., and Bartlein, P. J., University of Minnesota Press, USA, 568 pp., 1993. Markgraf, V.: Younger Dryas in southern South America?, Boreas, 20, 63-69, 1991 Paleoenvironments and paleoclimates in Tierra del Fuego and southernmost 15 Saporito, M. S.: Chemical and mineral studies of a core from Lake Patzcuaro, México, MSc. thesis, University of Minnesota, Minneapolis, Minnesota, USA, 482 pp., 1975. Sarmiento, G.: The dry plant formations of South America and their floristic connections, J. Biogeogr., 2, 233-251, 1975. Schäbitz, F.: Untersuchungen zum aktuellen Pollenniederschlag und zur Holozänen  Vegetationsentwicklung in den Anden Nord-Neuquéns, Argentinien, Bamberger Geographische Schriften, 8, 1-131, 1989. Schäbitz, F.: Holocene vegetation and climate in southern Santa Cruz, Argentina, Bamberger Geographische Schriften, 11, 235-244, 1991. Schäbitz, F.: Holocene climatic variations in northern Patagonia, Argentina, Palaeogeogr. 10 Palaeocl., 109, 287-294, 1994 , 9, 137-152, 1990. Yu, G., Xudong, C., Jian, N., Cheddadi, R., Guiot, J., Huiyou, H., Harrison, S. P., Ci-xuan, H., Jolly, D., Manhong, K., Zhaochen, K., Shengfeng, L., Wen-yi, L., Liew, P. M., Gunagxu, L., Jinling, L., Liu, K-B., Prentice, I. C., Guoyu, R., Changqing, S., Sugita, S., Xiangjun, S., Lingyu, T., Van Campo, E., Yumei, X., Qinghai, X., Shun, Y., Xiangdong Y., and Zhuo, Z.: Palaeovegetation of China: a pollen data-based synthesis for the mid-Holocene and last  Table 1. Characteristics of the sites from Latin America: specifically detailing location, potential vegetation around the sites, sample type, age range of sediments, number of C14 dates, data type, principle analysts and associated references. Dating control (DC) codes are based on the COHMAP dating control scheme (Webb, 1985;Yu and Harrison, 1995). For sites with a continuous sedimentation (indicated by C after the numerical code), the dating control is based on bracketing dates as follows. 1: both dates within 2000 years of the selected interval, 2: one date within 2000 years the other within 4000 years, 3: both dates within 4000 years, 4: one date within 4000 years one date within 6000 years, 5: both dates within 6000 years, 6: one date within 6000 years the other within 8000 years, 7: bracketing dates more than 8000 years from the selected interval. For sites with discontinuos sedimentation (indicated by D after the numerical code), the dating control is based on single dates 1: indicated a date within 250 years of the selected interval, 2: a date within 500 years, 3: a date within 750 years, 4: a date within 1000 years, 5: a date within 1500 years, 6: a date within 2000 years, 7: a date of more than 2000 years from the selected interval.    Markgraf (1989Markgraf ( , 1991Markgraf ( , 1993 Rodgers and Horn (1996) 435         454 49 forest (j, k), dominated by mangrove (l), tropical seasonal forest (m), cool temperate forest (j, k). Plate 0 shows the multi-vegetated lays within cloud forest (cool temperate rainforest), some of these taxa such as Rhipsalis and Bromeliacae are also found in very dry ecosystems. The bottom plate (l) shows the importance of edpahic factors on controlling vegetation; in this case local hydrology where rill channels allow trees to grow in areas that would be dominated by grassland.   Figure 8. Biome reconstruction at 6000 ± 500 14 C yr BP from radiocarbon dated fossil pollen. Fig. 8. Biome reconstruction at 6000±500 14 C yr BP from radiocarbon dated fossil pollen.