A laser survey of 4,497 square kilometres of Amazonian rainforest documented 432 earthworks, 396 of them previously unknown. Turning that into a peak population of 1.25 to 3 million takes five multiplications — and the last one can move the answer fivefold.
Between 6 June and 1 July 2024, a Seneca V aircraft carrying a full-waveform laser scanner flew 4,430 kilometres of one-kilometre-wide strips over the Brazilian states of Acre and Amazonas, covering 4,497 square kilometres of mostly closed-canopy rainforest at an average of 31.2 laser returns per square metre.1 Enough pulses find gaps in the canopy and reflect off the forest floor to build a terrain model of the ground beneath. On that model the team identified 432 earthworks. Only 36 had been recorded before.
Those structures — ditched geometric enclosures known as geoglyphs, plus banked enclosures without ditches — are attributed to a society the researchers call the Aquiry, which built in the region between roughly 600 BC and AD 850. Individual enclosures typically run from 0.35 to 14 hectares; the largest known covers nearly 50.1
The survey's most directly testable result comes from ground that had already been cleared. Along the same flight lines in the Rio Branco block, satellite imagery from Google Earth, Bing and ESRI had previously revealed 30 earthworks in deforested areas. The laser found 156 in the same places — 5.2 times as many.1 That is a like-for-like comparison on open ground, and it is the single strongest measured claim in the paper: satellite-based counts of Amazonian earthworks are substantial undercounts even where there is no canopy to hide behind.
The laser is not complete either. The team's own sensitivity analysis found that west of 66°W, 41.9 per cent of the surveyed ground returned too few ground points to reveal an ordinary earthwork; east of that line the figure was 54.5 per cent. For large geoglyphs over a hectare, with deep ditches and high banks, the blind fraction falls to 11.3 and 17.8 per cent respectively.1
Two transect sets returned nothing at all. Roughly 800 kilometres of flying between Lábrea, Carauari and Boca do Acre, and a further 200 kilometres north of the Purus, produced no earthworks whatever.1 The authors are careful that this is ambiguous: either the forest was too dense for the laser to model the ground, or there was genuinely nothing there — which would support the long-standing idea of uninhabited buffer zones between Amazonian polities. The null result is used to draw the western boundary of the Aquiry area, so which reading is correct matters to the size of the region the estimate is applied to.
The population estimate does not run off the 396 new discoveries. It runs off a narrower, systematic subset: six parallel flight lines, each about 445 kilometres long, one kilometre wide and ten kilometres apart, covering 2,380 square kilometres. Those lines yielded 347 structures, of which 334 lay far enough inside the swaths to be counted.1
Of those 334, the team identified 120 ditched geoglyphs and 115 banked quadrilateral or multilateral enclosures as Aquiry work — 235 in total. The remaining 99 are read as later, post-AD 1000 structures. It is worth noting how the 115 banked enclosures were assigned: they have not been directly dated, and are attributed to the Aquiry because their geometry resembles the dated ditched enclosures rather than the rounder embankments of the later period.1 Just under half the Aquiry-period sample rests on that morphological judgement.
From there the arithmetic is explicit. 220 of the 235 fall in the dense Rio Branco block, and the swaths cover a tenth of that block's area, giving 2,200. That figure is applied as an average to seven comparably dense blocks: 15,400. The remaining 15 structures, in the sparse block near Lábrea, scale up to about 1,262 across the 98,000 square kilometres treated as low-density periphery. Total: 16,660 Aquiry-period earthworks.
A second, independent route reaches a similar place. Multiplying the 1,256 earthworks previously found by satellite across the same eight blocks by the measured factor of five, then assuming the forested 77.5 per cent of each block holds structures at the density of the cleared portion, gives about 26,400; adding a sparse periphery gives 29,500 for all earthworks of all periods.1 Across the seven core blocks the two methods differ by 23 per cent overall — though individual blocks diverge from minus 52 to plus 160 per cent. The paper's headline range of 24,000 to 30,000 earthworks spans both routes.
Converting 16,660 structures into a population requires two further numbers, and neither is measured by this survey.
The first is contemporaneity: how many of the enclosures stood in use at once. From 162 radiocarbon dates across 32 sites, calibrated on a mixed Southern and Northern Hemisphere curve, the densest overlap falls at AD 100–300. Kernel density modelling of construction dates puts 25 per cent of all Aquiry earthworks in that window; a phase model of simultaneous use puts the overlap nearer 60 per cent.1 The paper carries both, giving 4,165 to 9,996 enclosures active at the peak.
The second is people per enclosure: 300. That figure is not derived from the geoglyphs at all, because the enclosures show no sign of permanent occupation — they are read as ceremonial and political gathering places, without dwellings or defences.1 It is instead calibrated from mound villages that appear in the same landscape from the thirteenth century, four centuries after the geoglyphs were abandoned. Those villages average 1.28 hectares with 11.75 mounds apiece; if each mound was a multi-family longhouse holding at least 25 people, a village held about 294, rounded to 300.
Multiply out: 4,165 × 300 gives 1,249,500, and 9,996 × 300 gives 2,998,800. Hence 1.25 to 3 million.
The published range of 1.25 to 3 million is wide — a factor of 2.4 — but all of that width comes from one term, the contemporaneity fraction. The other inputs are carried as point values, and the paper is unusually explicit about what happens if they move.
Correcting for the earthworks the laser could not see raises the range to 1.6–3.8 million. Changing the house assumption moves it much further in the other direction. If the mound-village mounds held single families rather than multi-family longhouses, the community figure drops from 300 to about 60, and the population falls to 250,000–600,000 — or 315,000–760,000 with the canopy correction applied.1 The authors judge the single-family reading “extremely unlikely” but do not attach a probability to it.
Read that exhibit as a whole and the shape of the problem is clear. The gap between the two blue rows — detection — is modest. The gap between blue and orange is a factor of five, and it turns on a single interpretive judgement about house occupancy in a village type that postdates the civilisation being counted by four hundred years. That, rather than the laser data, is where the estimate is soft.
Acre state today holds 830,000 people across 164,000 square kilometres, a density of 5.1 per square kilometre.1 The paper's modelled Aquiry densities exceed that across a comparable area, and substantially exceed it in the 85,000 square kilometre core.
The wider claim is conditional and the authors state it as such: the Aquiry area is about 2.7 per cent of Greater Amazonia, so if comparable densities were found elsewhere, basin-wide precolonial population estimates would need revising upward.1 Independent work already points the same way — a 2023 study estimated more than 10,000 earthworks remain hidden across Amazonia as a whole,5 and laser surveys of the Bolivian Llanos de Mojos found a dense, planned settlement system where none had been mapped.8
One comparison needs handling carefully. The commonly cited range for the whole Amazon basin in 1492 is 5–10 million.6 Three million in 2.7 per cent of the basin would be 30 to 60 per cent of that total. But the two numbers are separated by roughly 1,200 years: the Aquiry peak is dated AD 100–300 and the geoglyphs were abandoned between AD 850 and 950, six centuries before contact. They are not competing estimates of the same thing, and treating them as though they were is the most common error in coverage of this paper.
How representative the base is. The 235 observed Aquiry-period earthworks are 1.4 per cent of the 16,660 the estimate assumes, drawn from swaths covering 1.3 per cent of the cultural area. The extrapolation treats the Rio Branco block as the average for seven blocks; the paper's own satellite cross-check shows individual blocks departing from that average by anywhere from −52 to +160 per cent.
Whether the banked enclosures are Aquiry. 115 of the 235 — just under half — are assigned to the period by their geometry, not by dating. The paper says solitary quadrilateral and multilateral embankments have not previously been dated.
Where the people lived. No dwelling sites are counted. The population is inferred from ceremonial centres that the paper itself describes as lacking permanent occupation, via a community size calibrated on villages built four centuries later.
What the empty transects mean. Roughly 1,000 kilometres of flight lines returned no earthworks. Dense canopy and genuine absence are not distinguished, and the boundary drawn from that result sets the area the density is applied to.
A minor internal inconsistency. The paper gives an area-wide density of 7.1–16.5 people per square kilometre. Dividing its own population figures by its own 183,000 square kilometre area gives 6.8–16.4. The core-area densities of 13.6–32.6 reproduce exactly. The discrepancy is small and does not affect any conclusion, but the exhibit above plots the figures as printed.
Why it ended. The abandonment between AD 850 and 950 is dated but unexplained. The paper notes it coincides with declines elsewhere in the Americas and offers no mechanism.
The dispute is not about whether the Amazon held large precolonial populations — that is now well supported by several independent lines of evidence. It is about whether this particular chain of inference can carry a figure as specific as three million.
Two methods built on different data — laser transects and archived satellite imagery — converge within 23 per cent on the earthwork count. The community-size constant is independently corroborated by ethnographic work with Apurinã collaborators in the same region, who describe ceremonies drawing 200 to 400 people.1 Michael Heckenberger of the University of Florida, not involved in the work, called the estimates “plausible and extremely important”.9
Neither of the two constants is constrained by the survey. Heckenberger's own caveat is that the occupation sites needed to support a population of this size have not yet been found around the earthworks.9 A 2025 study that did attempt formal constraint — combining construction-labour and carrying-capacity models for the neighbouring Casarabe culture — still returned a tenfold range, 10,000 to 100,000 people, or 2 to 20 per square kilometre.7 On that precedent, a 2.4-fold published range here may understate the real uncertainty.
Every step below is reproduced from the figures printed in the paper. All eight reproduce the published outputs to within rounding.
| Step | Basis | Operation | Result | Status |
|---|---|---|---|---|
| Structures in six swaths | 2,380 km² | — | 347 | Observed |
| Usable for extrapolation | 347 | −13 edge cases | 334 | Observed |
| Attributed to Aquiry | 334 | 120 ditched + 115 banked | 235 | Observed / assigned |
| Rio Branco block, in full | 220 | × 10 (swath share) | 2,200 | Extrapolated |
| Seven core blocks | 2,200 | × 7 | 15,400 | Extrapolated |
| Low-density periphery | 15 | × 10 × 8.41 | 1,262 | Extrapolated |
| Aquiry earthworks, total | 15,400 + 1,262 | — | 16,660 | Extrapolated |
| In use at the peak | 16,660 | × 25–60% | 4,165–9,996 | Modelled |
| Peak population | 4,165–9,996 | × 300 people | 1.25–3.00m | Modelled |
| Alternative house reading | 4,165–9,996 | × 60 people | 0.25–0.60m | Modelled |