Environmental impact of geometric earthwork construction in pre-Columbian Amazonia

Significance

The discovery of extensive geometric earthworks beneath apparently pristine rainforest across southern Amazonia has fueled debate over the scale of environmental impact caused by ancient human societies. Whereas some claim that these sites are evidence of vast deforestation by populous pre-Columbian (pre-A.D. 1492) societies, others propose a model of small-scale, localized clearance. We tested these contrasting hypotheses by reconstructing environmental change in a region of geometric earthworks in northeast Bolivia over the last 6,000 y. Our unexpected findings reveal a surprising third scenario, in which earthwork builders took advantage of a naturally open savanna landscape, which existed under drier-than-present climatic conditions before ∼2,000 y ago. This finding suggests lower environmental impact, less labor, and possibly a smaller population than previously assumed.

Abstract

There is considerable controversy over whether pre-Columbian (pre-A.D. 1492) Amazonia was largely “pristine” and sparsely populated by slash-and-burn agriculturists, or instead a densely populated, domesticated landscape, heavily altered by extensive deforestation and anthropogenic burning. The discovery of hundreds of large geometric earthworks beneath intact rainforest across southern Amazonia challenges its status as a pristine landscape, and has been assumed to indicate extensive pre-Columbian deforestation by large populations. We tested these assumptions using coupled local- and regional-scale paleoecological records to reconstruct land use on an earthwork site in northeast Bolivia within the context of regional, climate-driven biome changes. This approach revealed evidence for an alternative scenario of Amazonian land use, which did not necessitate labor-intensive rainforest clearance for earthwork construction. Instead, we show that the inhabitants exploited a naturally open savanna landscape that they maintained around their settlement despite the climatically driven rainforest expansion that began ∼2,000 y ago across the region. Earthwork construction and agriculture on terra firme landscapes currently occupied by the seasonal rainforests of southern Amazonia may therefore not have necessitated large-scale deforestation using stone tools. This finding implies far less labor—and potentially lower population density—than previously supposed. Our findings demonstrate that current debates over the magnitude and nature of pre-Columbian Amazonian land use, and its impact on global biogeochemical cycling, are potentially flawed because they do not consider this land use in the context of climate-driven forest–savanna biome shifts through the mid-to-late Holocene.

Evidence for the existence of large and socially complex societies in Amazonia before the arrival of Europeans (pre-A.D. 1492) is emerging from a growing number of archaeological sites across the Amazon Basin (1–9). The scale of these societies’ environmental impact and its potential legacy in modern Amazonian forest ecosystems are hotly debated. Whereas some argue for a relatively limited and localized human influence (10–13), others have described pre-Columbian Amazonia as a “cultural parkland,” which was widely impacted by human disturbance (14–16). It has been proposed that deforestation and biomass burning before the collapse of native Amazonian populations following European contact in A.D. 1492 occurred on a scale large enough to contribute to an early anthropogenic influence on the global carbon cycle (17), and was a significant forcing of Holocene climate perturbations (18). It has also been suggested that this pervasive historical human influence on Amazonian forest ecosystems should change our views on their resilience to human impacts and influence our approach to their conservation (19). However, in many regions, a lack of appropriately scaled paleoecological data means that we have no paleoenvironmental context in which to place these societies and assess their environmental impacts.

Paleoecological studies (20–23) conducted in aseasonal western and central Amazonia suggest stability of the rainforest biome throughout the Holocene and appear to show little evidence for significant deforestation or biomass burning by their pre-Columbian inhabitants. However, the presence of regionally extensive pre-Columbian geometric earthworks underlying the seasonal southern Amazonian rainforests (SSAR) (Fig. 1A) (2, 4, 15, 24, 25) is suggestive of large-scale historical deforestation by substantial populations. These earthworks, uncovered by modern deforestation, are thought to represent only a fraction of the total, which lie undiscovered beneath the intact SSAR (3), a subregion of ∼1,000,000 km² (SI Text, Site Descriptions), which constitutes one-fifth of the Amazon basin. This practice of geometric earthwork building appears to have developed and diverged across southern Amazonia over a long time period from prehistory to European contact (2) (SI Text, Geometric Earthworks).

Fig. 1.

Map of site locations. (A) Present-day precipitation across Amazon basin during the three driest months of the year (26); solid black line delimits modern extent of Amazonian forest; approximate extent of SSAR/geometric earthwork region is within the <125-mm isohyet. Over 400 geometric earthworks have been discovered in eastern Acre, Brazil (location no. 3 in A), and many more across Iténez (location no. 1 in A, detail in B) and Riberalta (location no. 2 in A), northeast Bolivia, and the Upper Xingu (location no. 4 in A), Brazil. Previously published lake-sediment pollen records showing evidence for late Holocene biome shifts are represented by white circles: (a) Bella Vista, (b) Chaplin (27), (c) Yaguarú (28), (d) Carajas (29). Previously published lake-sediment pollen records in central and western Amazonia, which show stability of the forest biome and limited or no human impact over mid-to-late Holocene, are represented by gray circles: (e) Siberia (30), (f) Chalalán and Santa Rosa (22), (g) Gentry, Parker, Vargas, Werth (23). Black crosses represent soil-pit sampling locations/transects discussed in the main text, which show little evidence of pre-Columbian forest disturbance (20, 21). (B) Study area showing locations of LO and LG (detail in C), main rivers, and biome distribution. (C) Laguna Granja and Granja del Padre ring ditch.

Here we investigated the environmental impact of geometric earthwork building by using paleoecological techniques to reconstruct vegetation change and human land use on both a local and a regional scale around a pre-Columbian earthwork site in northeast Bolivia.

Study Area

Our study was conducted in Iténez province in northeast Bolivia at the geologically defined boundary between the terra firme (nonflooded) humid evergreen rainforest on the uplands of the pre-Cambrian Shield (PCS) and the seasonally flooded savannas of the adjacent Beni sedimentary basin (31). Although the forest–savanna boundary of our study area is locally controlled by geological, edaphic, and hydrological conditions, it is situated well within the SSAR, the southern ecotone of which is controlled by precipitation at a broad regional scale (Fig. 1A). Archaeologically, the Iténez region is characterized by extensive networks of artificial earthworks covering an estimated area of 12,000 km² (24). On the terra firme land of the PCS, these include dense clusters of ring ditches, ranging from circular forms, up to several hundred meters in diameter, to kilometer-long curvilinear ditches up to 3-m deep and 4-m wide. The function of these ring ditches is unknown. One possibility is that they acted as defensive features around settlements and were perhaps enhanced by the construction of palisade walls built from tree trunks (24). However, no archaeological evidence of post holes or wood remains has yet been found to confirm this hypothesis (32). Other suggested functions for geometric earthworks include: drainage (5), ceremonial/religious (3), or formal organization of space according to social hierarchy (1, 15).

To gain a temporal understanding of the nature and scale of human impact on vegetation in the ring-ditch region of northeast Bolivia, we analyzed fossil pollen and macroscopic charcoal from radiocarbon-dated lake sediment cores to reconstruct paleovegetation and fire history over the past ∼6,000 y (Methods). The pollen source area represented by fossil pollen in lake sediments is determined by lake surface area (33). We therefore selected a pair of lakes of contrasting size to capture vegetation on two distinct spatial scales (Fig. 1B). The smaller of the two lakes, Laguna Granja (LG), reflects local-scale vegetation at an adjacent ring-ditch site (Fig. 1C) and is nested within the regional-scale pollen catchment of the much larger lake, Laguna Orícore (LO), which reveals broader, biome-scale vegetation changes on the PCS (SI Text, Pollen catchment area).

During the rainy season, both lakes receive flood waters (and hence pollen input) from the surrounding seasonally flooded savannas of the Beni basin, as well as from the adjacent San Martín river, which drains rainforest, dry forest, and savanna regions south of Iténez. However, despite this floodwater input, we found that the surface-sediment pollen assemblages of both lakes were dominated by taxa indicative of the terra firme humid evergreen rainforest that covers the PCS. Our interpretation is based upon extensive modern pollen rain studies conducted in the Bolivian Amazon, which enabled us to distinguish between the characteristic pollen signatures of savanna, seasonally dry tropical forest, and humid evergreen rainforest (34–37). Most of the pollen entering the two lakes must therefore originate via wind dispersal from the vegetation of the PCS, with only a background input from the Beni basin and riverine flood waters. Analysis of the fossil pollen and macroscopic charcoal from the deeper lake sediments confirms that the paleoecological records from our study lakes are representative of changes in the ring-ditch region in northern Iténez and not the region to the south drained by the San Martín river (SI Text, Pollen and charcoal source).

Results and Interpretation

L. Orícore.

In zone LO-1 (5700–2000 calibrated years before present, or cal yr B.P.) (Fig. 2), LO shows low arboreal pollen levels (≤30%) and high Poaceae (grass) abundance (up to 50%), which—together with peak abundance of drought-tolerant taxa, such as Anadenanthera, and maximum charcoal concentrations—indicate a drier and regionally more open environment compared with the present (34–37). At the zone LO-1/LO-2 boundary (∼2000 cal yr B.P.), total arboreal pollen percentages begin to increase, with especially marked increases in abundance of evergreen taxa, such as Brosimum and Alchornea, whereas Poaceae, Anadenanthera, and charcoal levels decline, signaling an expansion of closed-canopy humid evergreen rainforest (34–36). By ∼1700 cal yr B.P., total arboreal pollen percentages had reached modern values, indicating that the regional terra firme PCS landscape had reached a level of forest cover comparable to the modern closed-canopy rainforest. This record of a savanna landscape changing to evergreen rainforest ∼2000 cal yr B.P. is consistent with paleoecological studies from other parts of southern Amazonia, which were dominated by savanna and seasonally dry tropical forest until ∼3000–2000 B.P., when humid evergreen rainforest began to expand southward (27–29). These records of expanding evergreen rainforest coincide with increasing lake levels at Lake Titicaca (38, 39) (an Andean site that receives most of its precipitation from lowland Amazonia), suggesting that rainforest expansion occurred as a result of increasing precipitation across southern Amazonia. The latter reflects a strengthening of the South American summer monsoon, driven by increased insolation linked to precessional orbital forcing (27). Further independent, multiproxy evidence of mid-to-late Holocene climate change in the study area is presented in SI Text, Mid-Late Holocene Climate Change.

Fig. 2.

LO pollen and charcoal records. Pollen data presented as percent abundance (of the total terrestrial pollen sum) and total concentrations per cubic centimeter. Charcoal presented as particles per cubic centimeter. “Total Arboreal” is the sum of all tree taxa. Dominant tree taxa (> 2% abundance) are shown. Plotted against calibrated radiocarbon years B.P. and individual radiocarbon dates shown on right (Fig. S1 and Table S1).

L. Granja.

LG provides a record of local-scale human impact at a ring ditch site on the PCS, on the outskirts of the modern day town of Bella Vista (SI Text, Site Descriptions). Ground mapping and airborne light detection and ranging (LiDAR) surveys conducted by H.P. over an area of 200 km² around LG have documented numerous earthworks enclosing areas up to 200 ha. Two ring ditches, one located 100 m east of the lake and the second located 1 km away, have been radiocarbon dated to between ∼800 and 500 cal yr B.P. (40). The ring ditch nearest to the lake (Fig. 3) is 150 m in diameter and 2 m deep, and is connected to the second ring ditch by a linear ditch that bisects an area of 150 ha. Excavations of this nearest ring ditch uncovered 15 separate burials and associated ceramics.

Fig. 3.

Aerial photo of ring ditch next to LG (taken 2008 by H.P.). Open ground around the ditch reflects modern, small-scale clearance for cattle pasture.

In zone LG-1 the high ratio of Poaceae to arboreal pollen (Fig. 4) (6100–2500 cal yr B.P.) shows that, before occupation, the environment around LG was an open savanna, in contrast to the dense evergreen rainforest that exists there today (notwithstanding scattered clearings made for cattle pasture and slash-and-burn agriculture), an interpretation reinforced by modern pollen-vegetation studies (34–37). The LG-1/LG-2 zone boundary marks the onset of agriculture and anthropogenic burning at LG at 2500 cal yr B.P., as indicated by a sharp charcoal increase and the presence of maize pollen (Zea mays L.). Although the charcoal record shows variations, with two distinct periods of more intensive anthropogenic burning at ∼2500–1600 cal yr B.P. and ∼700–500 cal yr B.P., the almost continual presence of maize pollen shows that the site was occupied throughout zones LG-2 and LG-3. The second charcoal peak (700–500 cal yr B.P.) coincides with peak Poaceae pollen percentages and ring ditch construction/occupation (∼800–500 cal yr B.P.) (40), which indicates, either a more intense period of burning associated with the ring ditch construction, or burning closer to the lake itself (SI Text, L. Granja charcoal). The decrease in arboreal types at this horizon is driven mostly by the disappearance of taxa such as Alchornea, Celtis, and Cecropia, indicating clearance of the gallery forest immediately around the lake (34–36). Open savanna/grassland vegetation persists throughout zone LG-2 and expansion of dense-canopy rainforest into the terra firme around LG does not occur until ∼500 cal yr B.P. (zone LG-2/3 boundary) following a decline in human activity at the site, possibly linked to the introduction of Old World diseases associated with the arrival of Europeans in the Americas in A.D. 1492 (41).

Fig. 4.

LG pollen and charcoal records. Pollen data presented as percent abundance (of the total terrestrial pollen sum) and total pollen concentrations. Charcoal presented as particles per cubic centimeter. Zea mays plotted as no. of grains per ∼0.4 cm³. “Total Arboreal” is the sum of all tree taxa. Dominant tree pollen types (>2% abundance) are shown. Dashed line indicates timing of regional forest expansion shown in LO. Plotted against calibrated years before present, with individual radiocarbon dates shown on the right (Fig. S2 and Table S1).

The timing of this local forest expansion at LG (∼500 cal yr B.P.) contrasts with the much earlier regional-scale pattern of forest expansion seen at LO (∼2000 cal yr B.P.). The most parsimonious explanation for the persistence of this open landscape at LG is that early settlers maintained open ground on a local scale by suppressing forest expansion around the settlement. The maintenance of this opening is not seen in the LO record because the much larger pollen catchment area of LO means that it is insensitive to detecting openings on the scale of the ring-ditch site (33). The LG pollen record therefore demonstrates that construction of the earthworks and agricultural activity around LG occurred in an open environment, maintained since the first occupation of the site. Rather than having to rely upon stone axes and burning to clear dense-canopy rainforest, a feat that would have been hugely labor intensive and highly impractical before the introduction of steel tools (42), our data show that the inhabitants took advantage of an existing open savanna landscape. Although the movement of earth to construct the ring ditches would have required a considerable labor force, the maintenance of an open landscape locally around LG, by the suppression of tree growth associated with the climate-driven rainforest expansion, would have required far less human effort (and potentially a lower population density) than that needed for the removal of dense-canopy rainforest.

Implications.

It is intriguing that most geometric earthworks found beneath terra firme tropical forest have been concentrated in seasonal southern Amazonia [i.e., northeast Bolivia (4, 24), eastern Acre state (3, 25), and the upper Xingu region of Brazil (15)], rather than wetter, less seasonal parts of the basin (Fig. 1A). Late Holocene climate-driven rainforest expansion has been documented in other parts of southern Amazonia (27–29), coinciding with rising lake levels in the tropical Andes (38, 39), which demonstrates that the forest expansion at LO reflects a broad-scale vegetation response to increasing precipitation across the southern Amazon. However, the geographic scale of this biome shift in the context of large geometric earthwork construction has hitherto not been considered. Our discovery raises the strong possibility that other geometric earthworks across southern Amazonia were also built in a savanna/open woodland that subsequently became covered by closed-canopy forest. Although dating of two of the numerous ring ditches at the Bella Vista site has shown them to be relatively late features (built and occupied between ∼800 and 500 B.P.), geometric earthworks in Pando (northeast Bolivia) and eastern Acre (northwest Brazil) have been dated to as early as 2000 B.P. (4, 43). The recent discovery of ancient shell middens in the central Beni basin have demonstrated a human presence in the Bolivian Amazon since the early Holocene (∼10,000 y B.P.) (44).

Gaining a sound understanding of the historical role of humans in shaping Amazonian landscapes, and the extent to which Amazonian forests were resilient to historical disturbance, is critical to informing policy makers on sustainable Amazonian futures (8, 12, 16, 19, 45). However, the debate so far has considered only the extent of past human impact and land use in what is assumed to have been a forested landscape, based upon the modern distribution of forest. Our study demonstrates that current debates over the magnitude and nature of pre-Columbian Amazonian land use are potentially flawed because they do not consider this land use in the context of climate-driven ecosystem dynamics through the mid-to-late Holocene (27, 29). Our findings show that, to determine the scale of environmental impact associated with pre-Columbian land use, the type of vegetation cover (e.g., closed-canopy forest versus open woodland or savanna) antecedent to settlement must first be known. This is a vital prerequisite to drawing inferences regarding population size and subsistence strategies.

Our findings have implications too for the debate over the scale of pre-Columbian impacts upon biogeochemical cycling. It has been speculated that post-Columbian secondary forest regrowth and the reduction in biomass burning across the Neotropics following indigenous population collapse caused sufficient carbon sequestration to reduce atmospheric CO₂ concentrations (17, 46) and amplify Little Ice Age cooling (18). It has also been suggested that observed biomass gains among old-growth lowland Neotropical forests over recent decades may be because of continued recovery from pre-Columbian anthropogenic disturbance (47). Our data support an alternative scenario, whereby pre-Columbian terra firme agriculture and construction of vast and numerous earthworks did not necessitate large-scale deforestation. On a regional scale, forest expansion in the SSAR occurred largely in response to increasing precipitation over the past 2,000 y, rather than solely as part of a vast postdisturbance forest recovery over the last 500 y following indigenous population collapse. However, superimposed upon this broad-scale, climate-driven, biome shift, humans did have significant environmental impact on a more local scale by suppressing climate-driven forest expansion around occupation sites. In the case of LG, its occupants held back rainforest expansion around the site for ∼1,500 y until a probable population collapse ∼500 cal yr B.P., possibly associated with European contact, allowing afforestation to occur. This new scenario of pre-Columbian land use fits neither of the opposing views of the current debate, which pits a model of short-lived, small-scale forest disturbance against a model of large-scale deforestation. Instead, we posit a more complex interaction between climatically driven ecosystem changes and long-term human land-use strategies. This new model does not preclude the possibility that Amazonian geometric earthwork building cultures may also have impacted their environment in more subtle ways, for example, by altering the floristic composition of forest through management of economically valuable taxa and modifying soil conditions. Further paleoecological work and close integration with archaeological data are required to better understand the nature of pre-Columbian land-use strategies in southern Amazonia and their possible involvement in the development of the rainforest ecosystem of this region following forest expansion in recent millennia.

Methods

The LG core was subsampled for pollen analysis at 5-cm intervals between 0- and 110-cm depth (0 to ∼3300 cal yr B.P.) and at 10-cm intervals between 110- and 150-cm depth (3300 to ∼6100 cal yr B.P.). The LO core was subsampled at 1-cm intervals between 0- and 22-cm depth (0 to ∼2550 cal yr B.P.) to capture the savanna–forest transition at high resolution, and at 2- to 3-cm intervals between 22- and 40-cm depth (∼2550 to ∼6700 cal yr B.P.). Samples were prepared using standard laboratory protocol (48) and separated into a fine (<53 μm) and a coarse fraction (>53μm) for optimal recovery of large cultigen pollen (49). Calculation of pollen concentration confirmed that observed changes in pollen percentage abundance in the paleo record were not the result of changes within a closed sum (SI Text, Pollen concentration values). The pollen in the fine fractions was counted to the standard 300 terrestrial grains. The coarse fractions were scanned up to a standardized equivalent count of 2,000 Lycopodium grains, representing ∼0.4 cm³ of sediment scanned. Fossil pollen was identified with reference to the collection of over 1,000 tropical pollen specimens housed at the University of Edinburgh and University of Reading. Maize pollen grains were distinguished from those of other wild grasses according to the morphological criteria described in Holst et al. (50). Where possible, members of the Moraceae family were identified to genus using pollen reference material and morphological descriptions from Burn and Mayle (51). LG was subsampled for macroscopic charcoal analysis at every 0.5-cm interval from 0- to 110-cm depth, and in the remainder of the core where charcoal abundance was consistently low, sampling resolution was lowered to 5-cm intervals. LO was subsampled at 0.5-cm intervals from 0- to 40-cm depth. See SI Methods and Figs. S1–S5 for more detailed descriptions.