The Emptied Amazon · source record · paper

More than 10,000 pre-Columbian earthworks are still hidden throughout Amazonia

Vinicius Peripato; Carolina Levis; Guido A. Moreira; Dani Gamerman; Hans ter Steege; Nigel C. A. Pitman; Jonas G. de Souza; José Iriarte; Mark Robinson; André Braga Junqueira; Umberto Lombardo; Luiz E. O. C. Aragão; et al. · 2023

AI-verified citationSRC-PERIPATO-2023

identifier / locator: doi:10.1126/science.ade2541
url: https://doi.org/10.1126/science.ade2541
verification note: text retrieved and shown to the drafter on 2026-09-10 (run 2026-09-10-draft-pre-columbian-amazon-184613); passages checked by the verify verb
reliability notes: Peer-reviewed research article, Science 382(6666):103–109 (2023).
Text read from the accepted manuscript (author version) deposited at White Rose Research Online (https://eprints.whiterose.ac.uk/207653/1/Vinicius_et_al_2023_Science_AAM.pdf), reached via OpenAlex; page locators cite that PDF, not the journal pagination. The manuscript title reads 'Over 10,000…'; the deposit record gives the published title used here.
Internal inconsistency to note: the abstract gives the predicted range as 10,272–23,640 undiscovered sites; the main text (p. 7) gives 10,272–23,648.
Not independent of several ledger sources: shares authors with SRC-LEVIS-2017 (Levis), SRC-DESOUZA-2018 (de Souza, Iriarte), SRC-LOMBARDO-2025 (Lombardo, Iriarte), and draws its 'domesticated species' list from Levis et al. 2017. Treat as the same research network as those records for independence grouping.
Presence-only archaeological data with acknowledged sampling bias toward deforested, road-accessible areas; the model corrects for this with observability covariates, but the correction is itself modelled.
Authors declare no competing interests; funding from Brazilian public agencies (CAPES, FAPESP, CNPq) and ERC (PAST project).

Evidence drawn from this source

Context· moderateAMZ-E038

Comparator: a basin-wide 10–24k undiscovered-earthwork prediction beside the Aquiry 24–30k regional prediction — part 1

what the source states: The predictive model "predicts the number of yet undiscovered pre-Columbian structures as 10,272" to 23,648, with 95% probability, for Amazonia's 6.7 million km² extent.
what we infer: Editor's comparison: SRC-PARSSINEN-2026 predicts roughly 24,000–30,000 earthworks within a ~183,000 km² Aquiry area alone — more than Peripato's upper bound for the entire basin, over an area about twice Peripato's high-probability rectangle. The two are not strictly commensurable (different earthwork definitions, data, and estimators; Peripato's 2023 model predates the 2026 LiDAR), so this does not undermine what AMZ-C004 states the 2026 study predicts. It does show that earthwork totals differ several-fold between published models, which AMZ-R004-style dependency work should address. Zero diagnosticity for the truth of C004 as worded; logged as the comparator the report sought.
limitations: Different operational definitions of 'earthwork' and different survey data between the two studies.
Peripato's model predates and does not incorporate the 2026 LiDAR campaign.

Vinicius Peripato (2023)AI-verified citationAMZ-C004

Context· moderateAMZ-E039

Comparator: a basin-wide 10–24k undiscovered-earthwork prediction beside the Aquiry 24–30k regional prediction — part 2

what the source states: Nearly all the highest-probability cells (≥ 25% predicted probability) occur in a 94,713-km² "rectangle that overlays a significant portion of the Brazilian state of Acre."
what we infer: Editor's comparison: SRC-PARSSINEN-2026 predicts roughly 24,000–30,000 earthworks within a ~183,000 km² Aquiry area alone — more than Peripato's upper bound for the entire basin, over an area about twice Peripato's high-probability rectangle. The two are not strictly commensurable (different earthwork definitions, data, and estimators; Peripato's 2023 model predates the 2026 LiDAR), so this does not undermine what AMZ-C004 states the 2026 study predicts. It does show that earthwork totals differ several-fold between published models, which AMZ-R004-style dependency work should address. Zero diagnosticity for the truth of C004 as worded; logged as the comparator the report sought.
limitations: Different operational definitions of 'earthwork' and different survey data between the two studies.
Peripato's model predates and does not incorporate the 2026 LiDAR campaign.

Vinicius Peripato (2023)AI-verified citationAMZ-C004

Supports· moderateAMZ-E040

53 of 79 domesticated tree species significantly associated with modelled earthwork probability — part 2

what the source states: The occurrence and/or abundance of "53 of the 79 domesticated species showed significant association" with the predictive model of earthwork distribution.
what we infer: Corroborates AMZ-C008's tree-domestication leg with a different response variable (modelled earthwork probability rather than distance to known sites), but the species list is Levis et al. 2017's own and the authors overlap, so this is not independent of SRC-LEVIS-2017; the McMichael 2017 post-Columbian-management objection (AMZ-E013) applies equally here.
limitations: Association with a modelled probability surface, not with dated archaeological sites.
Same species list and overlapping authorship as SRC-LEVIS-2017; not an independent group.
Cannot separate pre- from post-Columbian management (see AMZ-E013).

Vinicius Peripato (2023)AI-verified citationAMZ-C033 AMZ-C008

Supports· moderateAMZ-E041

53 of 79 domesticated tree species significantly associated with modelled earthwork probability — part 5

what the source states: The 53 domesticated tree species significantly associated with earthwork occurrence probability are "likely suggesting past management practices" by pre-Columbian societies.
what we infer: Corroborates AMZ-C008's tree-domestication leg with a different response variable (modelled earthwork probability rather than distance to known sites), but the species list is Levis et al. 2017's own and the authors overlap, so this is not independent of SRC-LEVIS-2017; the McMichael 2017 post-Columbian-management objection (AMZ-E013) applies equally here.
limitations: Association with a modelled probability surface, not with dated archaeological sites.
Same species list and overlapping authorship as SRC-LEVIS-2017; not an independent group.
Cannot separate pre- from post-Columbian management (see AMZ-E013).

Vinicius Peripato (2023)AI-verified citationAMZ-C033 AMZ-C008