Tropical forest–savannah ecotones are of considerable interest to biologists for several reasons. They typically exhibit high biodiversity, due to the mosaic of different ecosystems (i.e. habitat heterogeneity) resulting in high levels of beta diversity. They may play an important role in rainforest speciation (Bush 1994; Smith et al. 1997), whether via allopatric or sympatric mechanisms or both. Forest–savannah transitions are often influenced by numerous factors including nutrient supply, soil hydrology, natural disturbance and human activity, as well as direct climatic constraints, principally differences in seasonal and annual precipitation (Furley et al. 1992). A key uncertainty is the manner in which these vegetation boundaries will respond to future climatic change and anthropogenic influence (Houghton et al. 2001).

It is therefore unfortunate that vegetation transition zones, such as forest–savannah ecotones, have been neglected in recent conservation initiatives, as highlighted by Smith et al. (2001). For example, the ‘ecoregion’ (Dinerstein et al. 1995; Olson & Dinerstein 1998; Olson et al. 2001) and ‘biodiversity hotspot’ (Myers et al. 2000) schemes constitute two different approaches to prioritizing conservation efforts for the Earth's terrestrial biota, but both focus on species richness within a particular habitat type and neither consider the biological significance of biodiversity associated with transitions or boundaries between different kinds of habitat.

Hannah et al. (2002) cogently argue that these now traditional approaches to conservation are inappropriate for the twenty-first century, since they are predicated on the assumption that ecosystems are static and will not be affected by climate change. Hannah et al. (2002) advocate ‘climate change-integrated conservation strategies’ (CCS) that explicitly consider the likely ecosystem responses to climate change scenarios predicted for the next 100 years or more. In contrast to past conservation thinking, CCS would attach far greater importance to conserving forest–savannah ecotones, providing corridors spanning different vegetation types to allow species to shift their geographic ranges in response to a changing climate.

Although CCS take account of future climate change simulations (e.g. Malcolm et al. 2006), conservationists have rarely paid attention to proxy-based reconstructions of ecosystem responses to past changes in climate, whether over centuries, millennia or glacial–interglacial time-scales (Willis et al. 2005). This is most probably because conservation biologists are unfamiliar with palaeoecological techniques and literature and also because most palaeoecologists to date have not explored the implications of their findings for the conservation community, notwithstanding a few notable exceptions (e.g. Bush 2002).

What can a palaeoecological perspective offer conservation scientists and practitioners with respect to tropical forest–savannah ecotones? Firstly, palaeoecological data can demonstrate whether a forest–savannah ecotonal shift observed over recent decades merely constitutes a minor short-term oscillation or instead is part of a much larger millennial-scale biome replacement forced by climate change. Palaeovegetation records, coupled with independent palaeoclimate data (e.g. palaeo lake-level changes) and palaeofire (charcoal) data can reveal how resilient (or sensitive) forest–savannah ecotones have been to past changes in climate–fire regimes and thereby inform conservationists of the kinds of vegetation responses to be expected in the future.

Secondly, combined palaeoecological and archaeological approaches can potentially inform conservationists whether or not a particular forest–savannah mosaic or transition zone is essentially natural or is instead largely anthropogenic in origin and a legacy of land-use practices and human intervention over past centuries/millennia (e.g. changes to microtopography, hydrology, soil conditions and burning frequency). This is important for conservation planners. For example, if the biodiversity and structure of current ecosystems is a function of past centuries or millennia of intensive land management, traditional attempts to conserve this biodiversity by establishing protected areas such as national parks, and excluding human land use from them, would be misguided and could unwittingly change the very species composition that is supposedly being conserved.

Figure 1 MODIS image of the Bolivian Amazon and adjacent areas of Peru and Brazil, 28 July 2000, showing Noel Kempff Mercado National Park (NKMNP) and the Llanos de Moxos. Light areas with herringbone pattern to the north of NKMNP are zones deforested since the (more ...)

We consider the implications of these palaeovegetation and archaeological records for: (i) understanding the present forest–savannah mosaic and biodiversity patterns in these contrasting areas and (ii) adopting appropriate conservation strategies in the light of future climate change scenarios and ecological studies.

NKMNP is a 15230km² biological reserve in NE Bolivia on the Precambrian Shield near the SW margin of Amazonia, adjacent to the Brazilian States of Rondônia and Mato Grosso (figure 2). The following description summarizes detailed accounts by Killeen & Schulenberg (1998) and Killeen et al. (2003). NKMNP is of global importance, designated a UNESCO World Heritage Site in 2000, due to the diversity of pristine ecosystems that it contains: terra firme evergreen rainforest, riparian and seasonally flooded evergreen forest, seasonally flooded savannah wetlands, upland (cerrado) savannahs and semi-deciduous dry forest. In no other part of South America do such dramatically different ecosystems occur in such close proximity to one another (Killeen 1998). Although local (alpha) diversity (species richness) within any one ecosystem is not particularly high, the fact that these ecosystems are clearly distinct from each other (i.e. very high habitat heterogeneity or beta diversity), and well represented on the landscape, means that the total regional diversity for NKMNP is very high. These diversity patterns apply not only to the vegetation, but also to the birds, mammals and scarab beetles (Killeen & Schulenberg 1998).

Figure 2 Vegetation classification of Noel Kempff Mercado National Park (derived from Landsat TM data) (Modified from Killeen & Schulenberg 1998). Laguna Chaplin and Laguna Bella Vista are marked with circles. Translations of local Bolivian/Brazilian terms (more ...)

The juxtaposition of these different ecosystems, and consequently high habitat diversity, is due to the ecotonal position of the park, between humid Amazon rainforests to the north, semi-deciduous Chiquitano dry forests to the south and upland cerrado savannahs to the east. The overriding control on this vegetation ecotone, at the regional scale, is climate; mean annual precipitation is approximately 1500mm per annum (pa) and there is a four month dry season between June and September when mean monthly precipitation is less than 30mm. Precipitation falls mainly in the austral summer, originating from a combination of deep-cell convective activity over the Amazon basin and southerly extension of the intertropical convergence zone (ITCZ) due to peak insolation in the austral summer (Bush & Silman 2004). Although the mean annual temperature is 25–26°C, temperatures frequently decrease to 10°C for several days at a time during the dry season (June–August) when cold dry Patagonian air masses (surazos) reach the area.

At finer spatial scales, geomorphology and disturbance become key controls upon the ecosystem mosaic. The most conspicuous geomorphologic feature is the Huanchaca Plateau, a table mountain between 600 and 900m above sea level (a.s.l.), composed of Precambrian sandstone and quartzite, that dominates the eastern half of the park (Litherland & Power 1989). This is not only dominated by edaphically derived upland (cerrado) savannahs but also contains patches of evergreen and deciduous forests where soils are sufficiently deep and nutrient rich. The adjacent lowland peneplain to the west (200–250ma.s.l.) is blanketed by Tertiary and Quaternary alluvial sediments, which are covered by humid evergreen rainforests that form an ecotone with semi-deciduous dry forests at the southern border of NKMNP. Two black-water rivers originating on the Precambrian Shield define the boundaries of the park: the Río Iténez to the north and east (forming the Brazilian border) and the Río Paraguá to the west (Killeen & Schulenberg 1998). The latter are bordered by seasonally flooded evergreen forests and savannah wetlands, the spatial arrangement of which is controlled by local edaphic, topographic and hydrological conditions.

Fires are a key feature of the study area, especially in the savannahs. Seasonally flooded savannah wetlands outside the park are often burned annually by cattle ranchers (Killeen 1991), although natural fires caused by lightning are common within the park, especially in the upland savannahs of the plateau escarpments, which typically burn once every 3 years on average (Killeen et al. 2003). Flooding is another important type of disturbance that creates distinct types of forest and savannah ecosystems, which vary in community composition according to the depth and duration of flooding. Changes in the seasonal distribution and intensity of precipitation can alter fire and flooding regimes which can in turn cause marked changes to the vegetation mosaic of NKMNP.

In contrast, the cerrado savannahs of the Huanchaca plateau are highly diverse, especially with respect to their avifauna (Bates et al. 1998; Da Silva & Bates 2002), comparable with any other savannah region in South America. Killeen & Schulenberg (1998) suggest this high biodiversity likely reflects the long history of isolation of this edaphic table mountain savannah, possibly spanning millions of years. The seasonally inundated savannahs have markedly lower plant diversity than the upland cerrado savannahs, although this is probably unrelated to their history but due to their seasonal flood and drought regime, which is no doubt inhospitable for many species.

Palaeoecological data may help explain some of these biodiversity patterns and address some of these hypotheses/questions. In particular, why do the rainforests of NKMNP exhibit lower alpha diversity than those in more central parts of Amazonia? Are observed forest–savannah shifts merely minor short-term oscillations, or instead part of a long-term, millennial-scale trend? If the latter, then which is the predominant process—forest encroachment into savannah or savannah encroachment into forest? To what extent have Palaeo-Indians contributed to the current forest–savannah mosaic and species assemblages? How are the ecosystems of NKMNP likely to be affected by future climate change scenarios?

Figure 3 Laguna Chaplin summary pollen percentage diagram, showing the most common taxa from the full complement of 290 pollen types. Dots on the curves denote less than 0.5%. Curves showing ×10 exaggeration are depicted for selected taxa which have low (more ...)

Figure 4 Microscopic and macroscopic charcoal concentrations for Laguna Chaplin. Pollen assemblage zones are the same as in figure 3. (After Burbridge et al. 2004.)

Throughout the glacial period, between ca 50000 and 11400 calendar years before present (yr BP), the dominance of Poaceae (more than 40%) and Cyperaceae pollen (approx. 15%), together with Mauritia/Mauritiella palm pollen, suggests that low-lying areas which are today covered by seasonally flooded rainforest were instead covered by seasonally flooded termite savannahs and savannah marsh. Consistent presence of Paullinia/Roupala, Celtis, Machaerium-type and Erythroxylum pollen suggests that semi-deciduous dry forests may have occupied higher ground presently covered by terra firme evergreen rainforest. However, absence or negligible abundance of pollen of Curatella americana and Anadenanthera, which are important elements in present-day termite savannahs and dry forests, respectively, shows that these glacial-age ecosystems were floristically distinct from those that occupy the region today. The charcoal record shows that fires were more prevalent than today around these lakes through most of the glacial sequence, as would be expected in landscapes dominated by savannah and dry forest. These Pleistocene (glacial-age) plant communities are consistent with reduced precipitation (e.g. longer or more severe dry season) and lowered atmospheric CO₂ levels, both of which would have favoured expansion of C₄ grasses and sedges and drought-adapted savannah and dry forest tree species. Continuous presence of pollen of the predominantly Andean genus Podocarpus throughout the glacial sediments corroborates other palaeoclimatic evidence (e.g. Stute et al. 1995) for an Amazon-wide cooling of approximately 5°C at the Last Glacial Maximum (LGM; approx. 21000yr BP).

A mosaic of seasonally flooded savannahs and semi-deciduous dry forest communities also occupies the park throughout most of the Holocene (present interglacial period, 11500yr BP to present), but presence of C. americana, Anadenanthera, Astronium spp. and Gallesia pollen shows that, in contrast to the savannah and dry forest communities of the Pleistocene, these Holocene communities are floristically similar to those growing in the region today. Abundance of charcoal demonstrates the continued importance of fire in the park. Although it is unwise to attach too much significance to the high abundance of grass and sedge pollen that dominates most of the record, as it could conceivably have originated from shoreline aquatic vegetation rather than neighbouring savannahs, our interpretation of seasonally flooded termite savannahs is corroborated by grass and sedge cuticle analyses (Metcalfe 2005).

Humid evergreen rainforest, characterized principally by high percentages of Moraceae pollen (more than 40%) and low percentages of herbaceous pollen, only expanded southward from more central parts of Amazonia to dominate the park within the last two millennia, surrounding Laguna Bella Vista ca 1500yr BP and Laguna Chaplin ca 660yr BP. The initial onset of this rainforest spread probably began ca 3000yr BP at Laguna Bella Vista and 2000yr BP at Laguna Chaplin, when Moraceae pollen percentages began to rise and herbaceous pollen percentages began to fall. These dates suggest that rainforest spread southward between Laguna Bella Vista and Laguna Chaplin at a rate of ca 100–120m per year. As expected, this rainforest expansion was accompanied by a reduction in fires, as reflected in the charcoal record. Although Moraceae pollen is continuously present throughout the entire sequence, modern pollen rain trap samples collected from 1ha permanent vegetation study plots in each of the different ecosystems in NKMNP demonstrate that it only signifies local closed canopy rainforest when above 40% (Gosling et al. 2005) and that values of 25% or below are consistent with local semi-deciduous dry forest, savannah or long-distance transport from distant rainforest populations (Gosling 2004).

Although the relative contributions of atmospheric CO₂ concentrations, precipitation, temperature and geomorphology as controls of glacial-age vegetation in NKMNP are difficult to discern, it is clear that a precipitation increase was the primary driver of this late Holocene rainforest expansion, as CO₂ concentrations and temperature remained relatively constant through this period. Although rainforests reached northern NKMNP (Laguna Bella Vista) by 3000yr BP, the absence of fossil pollen data further north means that the timing and geographic origin of the onset of this rainforest expansion beyond NKMNP is uncertain. However, we can be confident that this advance began not earlier than 5000yr BP because Lake Titicaca (which lies at a roughly similar latitude to NKMNP on the Bolivian/Peruvian Altiplano) experienced a Mid-Holocene lowstand (approx. 90m below present) between 8000 and 5000yr BP (its lowest water level of the past 25000 years) and rose to modern levels by 3000yr BP (Cross et al. 2000; Baker et al. 2001; D'Agostino et al. 2002). It is possible, though, that some of this forest expansion in the lowlands of NKMNP was only indirectly driven by climate and that local geomorphologic changes played a more direct role. For example, increased precipitation may have led to higher rates of deposition and aggradation in the flood plains, thereby changing the hydrological conditions to facilitate expansion of riparian forest in the floodplains and on the levees.

Figure 5 The δ13C values of soil profiles of 15 soil pits dug along an ecotonal transect between open savannah (grassland) and tall forest in the central part of the Huanchaca plateau, NKMNP. (Modified from Killeen et al. 2003.)

Conversely, the soil stable carbon isotope data are consistent with the hypothesis that the cerrado savannahs on the Huanchaca plateau owe their high biodiversity to their long geological history, which possibly extends back uninterrupted beyond the Quaternary. Upland savannahs probably persist on the plateau, not because precipitation is insufficient to support forest, but more likely because soils are insufficiently deep or fertile to support forest.

The pollen data suggest that the scattered seasonally flooded savannah patches below the plateau are most likely ancient relicts of a much greater, possibly continuous distribution that covered most low-lying areas prior to their replacement by seasonally flooded rainforest.

Ecological studies of permanent, one hectare, vegetation study plots from several types of forest in NKMNP have revealed consistent increases in plant turnover (i.e. recruitment and mortality), above-ground biomass, basal area, stem density and liana development over the past decade (Phillips et al. 2002, 2004; Baker et al. 2004). It might be tempting to hypothesize that these ecological trends can be explained by the recent spread of rainforest species into NKMNP over recent millennia and that they signify long-term rainforest succession, especially as recruitment leads mortality (Phillips et al. 2004). However, this cannot be the explanation because these recent changes are a consistent feature, not only of NKMNP, but also of dozens of old-growth forest sites, with different Quaternary histories (Mayle et al. 2004), across the entire Amazon basin. None of the alternative explanations suffice either, such as secondary succession of forest following indigenous population collapse after the Spanish Conquest (Wright 2005), or responses to climatic perturbations such as the El Niño Southern Oscillation, because they too cannot account for similar ecological responses across the entire Amazon basin. Although the underlying causal factor and mechanism is still controversial and unresolved (Wright 2005, 2006; Lewis et al. 2006), Phillips et al. (2004) make a forceful argument that the only conceivable driver that could synchronously affect the entire Neotropics in a similar way is the anthropogenic rise in atmospheric CO₂ concentrations, causing increased canopy photosynthetic rates, possibly augmented by increasing sunlight.

Figure 6 Climate predictions of five leading global climate models (GCMs) for NKMNP during the twenty-first century, compared with the twentieth century average data calculated from the Climatic Research Unit (CRU) global meteorology data set (New et al. 1999 (more ...)

These feedbacks are both biophysical and biogeochemical, and their effects include reductions in evapotranspiration and reductions in carbon storage in vegetation and soil. The net effect on a drying and warming climatic trend would be acceleration, but significant uncertainty exists in quantifying such feedbacks, especially over the longer term (Davidson & Janssens 2006; Meir et al. 2006).

Short-term declines in rainfall not only affect evapotranspiration and productivity but also respiration rates in soil, thus modifying the net emissions of CO₂ to the atmosphere. Over the longer term, pollen and stable carbon isotope data show that the drier climate of the Mid-Holocene favoured savannah and semi-deciduous forest over humid evergreen rainforest in NKMNP. Any future change in climate towards drier conditions would be expected to cause similar vegetation changes, through increased mortality and subsequent species replacement. These changes in soil and vegetation imply large net emissions of CO₂ (through reductions in carbon storage), as well as reductions in evapotranspiration and increases in albedo, although their effects are quantitatively uncertain (Meir & Grace 2005). However, new datasets are emerging that may help improve the modelling of long-term responses (decadal to century scale) of rainforest trees to drought; for example, mortality rates have been found to at least double after both experimental and El Niño-associated drought (e.g. Nakagawa et al. 2000; Nepstad et al. 2004; Meir & Grace 2005), and potential rates of vegetation change have been calculated on this basis (e.g. Condit et al. 1998). However, it is important to note that while the Mid-Holocene vegetation may be considered a reasonable analogue for the future, it is an imperfect analogue because CO₂ levels in the twenty-first century are predicted to be at least double those of the Mid-Holocene (ca 270p.p.m., Indermühle et al. 1999). The effects of a future drier climate may be partially compensated for by the concomitant increase in atmospheric CO₂ concentration, which would have the physiological effect of allowing a greater rate of photosynthesis per unit of water loss, enabling trees to survive a longer dry season. Although this effect is incorporated into many biosphere models, its long-term permanence and, indeed, the uniformity of response across species has been questioned (Idso 1999; Körner et al. 2005).

Fully coupling biosphere processes to a GCM has only been achieved by a few modelling groups (Cox et al. 2000; Friedlingstein et al. 2003, 2006) and while current models show a positive feedback with climatic change, uncertainty in the relevant ecological and physiological mechanisms translates into modelled differences in climatic and vegetation change scenarios (Meir et al. 2006). In addition, the compounding effect of anthropogenic influences is particularly important for the time frame relevant to conservation. Deforestation is expected to reduce evapotranspiration, further reducing rainfall regionally (Costa & Foley 2000), with the additional potential for climatic effects further afield (Avissar & Werth 2005). Coupled with other land-use changes and increased fire frequency, a significant positive feedback with climatic warming and drying is possible (Nepstad et al. 2004). Thus, fully coupling land–atmosphere feedbacks into a climatic change scenario may well show that current GCM model outputs are under-estimates of climatic change in the Amazon region. As a result, an intensified drying and warming trend is likely over both short and longer time frames, leading to conversion of humid evergreen rainforests in NKMNP to an ecosystem with a lower leaf area index and with seasonal vegetation growth (i.e. semi-deciduous dry forest and savannah).

If the Mid-Holocene forests serve as a reasonable analogue, one would predict that future climate warming and drought would lead to progressive replacement of rainforest species by the more drought-adapted and fire-adapted tree species of the Chiquitano dry forest that lies immediately south of NKMNP. However, seed source availability is a crucial and under-recognized constraint on predicted vegetation change. In the case of NKMNP, unfortunately, much of the Chiquitano dry forest is unprotected and currently experiences among the highest rates of deforestation in the world (Steininger et al. 2001a,b). To allow for this rainforest–dry forest ecotonal shift to occur, there is an urgent need to protect the tracts of semi-deciduous dry forest that lie immediately south of NKMNP, establishing a ‘latitudinal landscape corridor’. This would not be an easy task as most of the land is privately owned and a development corridor will inevitably emerge along the road between Santa Cruz, Bolivia, and Cuiabá, Brazil (Killeen et al. 2003).

If the GCM climate simulations for the twenty-first century are correct, then it is possible that many rainforest species will become locally extinct in NKMNP within the next 100 years, especially as many such species must already be close to their climatic thresholds in this ecotonal region. Rainforest species on the Huanchaca plateau would be expected to be under greatest threat due to their geographic isolation within forest patches surrounded by savannah. Even in lowlands, below the plateau, movements of bird and animal rainforest species northward to more central parts of Amazonia would be highly limited due to the extensive deforestation that has already occurred in neighbouring Rondônia (Skole & Tucker 1993; see figure 1), although the humid forest habitats along the Río Iténez could potentially serve as a biodiversity corridor allowing northward migration (Killeen et al. 2003).

The Holocene palaeodata show that savannah ecosystems would also be expected to expand at the expense of rainforest as climate becomes more arid in the future. From a conservation standpoint, this would be a positive outcome, as it is the savannahs of NKMNP, rather than the rainforests, that contain the most endangered species (Da Silva & Bates 2002), and the Neotropical savannah biome is under far greater threat (from cattle-ranching, agriculture and human population expansion) than the Amazon rainforests (e.g. Silva et al. 2006). The Huanchaca plateau probably contains the largest protected tract of undisturbed cerrado (upland) savannahs (approx. 42000ha) anywhere in the Neotropics (Killeen 1998; Da Silva & Bates 2002), which together with the seasonally flooded savannahs below are home to some of the most threatened savannah megafauna on the continent, such as the pampas deer (Ozotoceros bezoarticus), the marsh deer (Blastocerus dichotomus), the maned wolf (Chrysocyon brachyurus), the greater rhea (Rhea americana) and the giant anteater (Myrmecophaga tridacyla). In the absence of hunting, future climate change could actually favour expansion of these endangered species.

The Llanos de Moxos (figure 1) is a 160000km² expanse of seasonally inundated savannahs, interspersed with a complex mosaic of forest ‘islands’ and riverine gallery forests (figure 7a), occupying the extremely flat Beni–Mamoré–Iténez basin in SW Amazonia (10°40′–16°25′S), situated between the Precambrian Shield to the east and the Andes to the west and south. Numerous white-water rivers and hundreds of shallow, flat-bottomed lakes cover the landscape. The spatial distribution of vegetation is closely related to micro-topography. Flat, low-lying areas, which are seasonally inundated for several months a year during the rainy season, by flooding from rivers and ponding of rainwater, are covered by completely open, tree-less savannah (locally referred to as pampa). Conversely, forest islands (islas) are restricted to raised areas (mounds) which are sufficiently high to escape flooding.

Figure 7 Earthmounds, vegetation and lakes of the Llanos de Moxos. (a) Remote-sensing image showing the forest–savannah mosaic of the Llanos de Moxos, palaeofluvial features and gallery forests (red) trending SW–NE, with intervening open savannahs (more ...)

The extent to which this forest–savannah mosaic is natural or a human artefact has been debated since the 1960s. Beck (1983) argues that these savannahs are the legacy of centuries, or even millennia, of deforestation by pre-Conquest indigenous peoples, and subsequently large-scale cattle ranching over the last several centuries. However, we argue that, at a regional scale at least, the seasonally flooded savannahs are natural, resulting from the unique suite of edaphic, hydrological and climatic conditions of the basin. Heavy, shallow soils with impermeable clay pans are inundated with stagnant waters at high temperature for several months of the year, resulting in anaerobic conditions and physiological heat stress, followed by a severe dry season. Very few woody species are able to endure this combination of abiotic factors.

However, at a fine spatial scale there is extensive archaeological evidence showing that pre-Conquest indigenous peoples made significant modifications to this savannah environment, principally by constructing a vast array of different types of earth mounds (Denevan 1966; Erickson 1995, 2000). As these artificial landforms were above the flood-level, they provided suitable habitats for colonization by trees once they were abandoned.

However, while some forest islands do indeed appear to cover artificial earth mounds, closer examination reveals a variety of forest island types that differ in species composition, morphometry, soil type and physical setting; these clearly must have had different modes of origin. Many of the smallest forest islands, 10m or less in diameter, have been created by termites rather than humans. Such islands are regularly spaced apart, sometimes less than 50m, and form on earth mounds created by eroding termite mounds. Such termite savannahs are not only characteristic of seasonally flooded areas of NKMNP (Killeen 1998), discussed earlier, but are also abundant in many parts of the Llanos de Moxos (Denevan 1966; Beck 1983; Hanagarth 1993). Aerial photography and, more recently, LANDSAT imagery, have demonstrated that many forest islands are eroded relics of natural levees or terraces of abandoned river channels, and therefore constitute fragments of former gallery forest (Braun 1961; Hanagarth & Sarmiento 1990). Corroboratory evidence comes from their underlying sandy soils, which contrasts with the clays of the surrounding savannah (Langstroth 1996).

It is therefore clear that a single model cannot be used to explain the origin of these forest islands. While some are demonstrably artificial, whereby trees have colonized abandoned Palaeo-Indian earthworks, others have formed by entirely natural processes. Furthermore, the degree to which some supposedly artificial mounds are entirely anthropogenic in origin, or modified natural mounds, is also often unclear.

All these forms of landscape engineering indicate that the Moxos savannahs supported intensive fishing and farming industries, supplying, and maintained by, a much larger human population than exists in this part of Bolivia today. The chronology for these various earthworks is limited, but dates from basal units indicate that these societies modified the landscape extensively over at least the past three millennia and collapsed around the time of the Spanish Conquest (Erickson et al. 1991; Langstroth 1996).

If the majority of lakes are natural in origin, then how can their distinctly rectangular shapes and uniform orientation along a SW–NE axis be explained? Plafker (1964, 1974) has hypothesized that they were caused by fracturing of the underlying Precambrian basement rocks, which affected the alignment of superficial jointing in the overlying unconsolidated Tertiary and Quaternary sediments, causing them to subside and form shallow lake basins. However, Clapperton (1993), Dumont & Fournier (1994) and Langstroth (1996) have queried this mechanism as it is unclear how basement fracturing can be propagated to the surface through more than 800m of soft unconsolidated sediments. These authors suggest that these oriented lakes instead constitute palaeo-deflation basins that formed under more arid climatic conditions of the past. Dumont & Fournier interpret the existence of indurated clay caps beneath some of the lakes as additional evidence of a formerly drier climate. Furthermore, the orientation of the long axis of the lakes is almost exactly perpendicular to the axis of the Holocene dunes of the Piraí basin in Santa Cruz, consistent with an aeolian mechanism (Livingstone 1954; Carson & Hussey 1962; Dumont & Fournier 1994).

In the absence of empirical palaeovegetation data from the Beni, insights into the age of the forest–savannah mosaic may be gleaned by considering species' endemicity. Although not recognized as a centre of endemicity, the Llanos de Moxos is home to a few notable endemic vertebrates such as the Beni anaconda (Eunectes beniensis), the blue-throated macaw (Ara glaucogularis), and the Beni and Olalla's titi monkeys (Callicebus modestus and Callicebus olallae, respectively). The Beni anaconda is confined to the savannah wetlands, while the macaw and titi monkeys are restricted to forest islands. The latter are most significant because they are not found in larger contiguous forests or extensive gallery forests. The presence of endemic vertebrate species suggests that the forest–savannah mosaic that characterizes the Beni has at least a long Quaternary history, although the spatial configuration of these two ecosystems may well have changed over time.

This conservation assessment was based solely upon ‘conservation status’ and ‘biological distinctiveness’. The extent to which the ‘biological distinctiveness’ of the Beni is anthropogenically derived was therefore not considered. This brings into question the suitability of standard biodiversity-based conservation approaches for regions such as the Beni that have a strong legacy of human intervention in the landscape. Essentially ‘virgin’ forest–savannah environments, such as those within NKMNP, are clearly ideal for traditional conservation approaches, whereby the ecosystems are designated as a national park from which human activities are excluded. By contrast, any plans to conserve the forest–savannah mosaic of the Llanos de Moxos in a similar way, by excluding all human land use, may end up inadvertently changing the vegetation mosaic and species composition in unforeseen ways. Several important tree species may well have been managed by native peoples over centuries or millennia and are harvested to this day for a variety of uses. For example, the tree Sorocea, which grows on earth mounds, is commonly used by the Sirionó Indians to make beer (Mann 2000), while the palm Mauritia flexuosa, which grows in the surrounding seasonally flooded savannahs, produces up to 5000 edible fruits per year from a single tree and 10–60t of fruit from a single hectare (Erickson 2000). Although such tree species may have become established on earth mounds by natural means, such as bird dispersal (Langstroth 1996), rather than planting, native people could have encouraged expansion of such species over the other non-utilized species. The relative abundance of such taxa may change once people are excluded.

Furthermore, several tribes (e.g. Moxo, Movima, Canichana, Baure and Itonama) whose ancestors helped create the anthropogenic landscape of earthworks, fish weirs, causeways, etc. still occupy the Beni basin today and receive some form of protection via ‘indigenous lands’. Any conservation policy that aimed to supposedly conserve the landscape of the Llanos de Moxos by excluding the very people whose ancestors helped create it would surely be misguided and counter-productive. A more appropriate conservation strategy may be one that incorporates traditional indigenous land-use practices and local economies, given the evidence that the Llanos de Moxos has supported dense human societies over several millennia. Ultimately, biologists need to engage with anthropologists and archaeologists to adopt a conservation strategy that takes account of not only its biological value, but also its rich archaeological heritage. The level and type of human land use that this biota has sustained over past millennia needs to be understood in order to determine what type and intensity of land use is sustainable for this environment in the future.

It is unclear how a future drier climate might affect the Llanos de Moxos. On one hand, reduced seasonal inundation might make hydrological conditions suitable for tree populations to expand beyond the raised mounds to the surrounding low-lying plains. However, the species composition of these forests could change significantly, with more drought-tolerant tree taxa from the Gran Chaco deciduous woodlands to the south spreading north into the Beni and out-competing semi-deciduous and rainforest species. On the other hand, irrespective of the direction of future climate change, the riverine gallery forests could serve as important migration corridors for species interchange between the different types of tropical forest bordering the basin (e.g. between rainforest to the north and Chaco dry woodland to the south). Regardless of the anthropogenic legacy within these gallery forests, they should clearly be protected from future deforestation to maintain them as intact corridors.

The forest–savannah environments of NKMNP and the Llanos de Moxos in lowland Bolivia differ markedly from one another in a variety of ways: their abiotic controls; biota; geomorphological settings; current land use; and environmental histories (particularly with respect to human impact). Consideration of their respective palaeoecological and archaeological histories demonstrates markedly differing legacies of past human disturbance versus climatically driven vegetation change, which in turn call for differing CCS that reflect their contrasting histories.

NKMNP contains largely ‘virgin’ ecosystems. The palaeoecological records show no evidence that they have experienced any significant human impact over the past 50000 years. Fossil pollen and stable carbon isotope data demonstrate that present-day rainforest encroachment into savannah is significant, but is a part of long-term trend (over the last three millennia) of southward expansion of rainforest through the park, replacing dry forest and savannah communities. This ecotonal shift appears to have been an entirely natural process caused, either directly or indirectly, by increasing precipitation.

In contrast, there is abundant archaeological evidence that the forest–savannah mosaic of the Llanos de Moxos in the Beni basin has been altered extensively by Palaeo-Indians over centuries and millennia prior to the Spanish Conquest, largely by building a variety of different kinds of earth mound that persist today. Although some forest islands are clearly natural features, such as gallery forest fragments, others are demonstrably artificial, whereby Palaeo-Indians have built earthworks, which, once abandoned, have become colonized by trees. Therefore, at least some of the forest expansion in the Llanos de Moxos during the Late Holocene was not climatically driven, but instead was facilitated by humans creating edaphically suitable conditions for trees to grow.

The contrasting legacies of past human impact between these two forest–savannah transition zones have important implications for conservation planners. On one hand, the traditional approach of creating a national park, from which human land use is excluded, appears to be appropriate for the ecosystems of NKMNP which have no discernible legacy of past human disturbance. On the other hand, this approach would be inappropriate for the Llanos de Moxos, where much of the landscape owes its very character to past human activity. The dilemma for conservationists is determining the type and intensity of land use that is appropriate for this anthropogenically modified landscape.

How will the vegetation in these two ecotonal regions be affected by the drier and warmer climatic conditions predicted for the twenty-first century? The Late Holocene rainforest expansion in NKMNP is likely to be reversed, as more drought-tolerant species are favoured, resulting in semi-deciduous dry forest and savannah communities, akin to those that dominated NKMNP during the Early/Mid Holocene. In the Llanos de Moxos, future increased drought could result in drought-tolerant tree taxa from the Gran Chaco deciduous woodlands to the south spreading northward and out-competing semi-deciduous and rainforest species in future.

Landscape corridors are needed to facilitate these latitudinal shifts in species ranges. The riverine gallery forests that traverse the Beni basin, generally from south to north, should be targeted as natural corridors allowing species interchange between the Chaco dry woodlands to the south and the humid Amazonian rainforests to the north. In order to facilitate northward spread of semi-deciduous dry forest species into NKMNP, it is imperative that the protected area network is expanded to form a contiguous latitudinal landscape corridor that connects the rainforests of NKMNP with the poorly protected Chiquitano dry forest to the south.

We hope that knowledge of the long-term dynamic history of forest–savannah environments, considered in the context of ecological observations and modelled future climate change scenarios, can help conservationists develop the most appropriate conservation strategies for these biodiverse and often-neglected habitats.

We are grateful to Kathy Willis and Lindsey Gillson for inviting us to contribute this paper, and William Denevan and the two reviewers, Keith Bennett and Kathy Willis, whose comments and suggestions improved the paper.