The Marupiara dataset
The name Marupiara is derived from a specific epithet in the Indigenous Tupi language and is traditionally associated with the figure of the good or virtuous hunter. We compiled data from primary and secondary sources on wild tetrapods (mammals, birds, reptiles and amphibians) that are hunted for food by Indigenous, traditional, and small-scale farming peoples in all nine Amazonian countries. Primary data were contributed by researchers involved in 12 long-term and short-term studies conducted in 342 communities in Brazil, Peru and French Guyana between 1991 and 2024 (Supplementary Methods 4). Secondary data were obtained from 203 peer-reviewed articles, technical reports, postgraduate dissertations and theses reporting hunting studies in 290 communities in all nine Amazonian countries between 1965 and 2021 (Supplementary Methods 5).
The composition of hunted taxa, documenting the number of individual animals hunted per taxon, year and locality, which was used to model the TSOP, was recorded in 590 localities. Whenever available, we also included data on hunting effort—specifically the number of hunters surveyed and the number of recording days—which was used to model the HHR. Hunting effort data were available for 301 georeferenced localities.
We also compiled information on the number of hunters and consumers in each locality and year to calculate the hunters-to-consumers ratio, which we used to estimate the number of rural hunters in Amazonia based on regional population figures. Each locality was georeferenced, and we recorded the cultural identity of hunter communities where available. We also collected data on the biomass of animals hunted. Further details on the derivation and application of these metrics are provided below.
The final Marupiara dataset comprises 21,397 records of hunted taxa composition, representing 447,438 individual animals hunted across 625 georeferenced communities in rural Amazonia from 1965 to 2024.
Hunting monitoring schemes
Primary data obtained through hunting monitoring schemes were generally collected under the supervision of researchers and Indigenous, traditional and small-scale farming trained researchers. Three main data collection methods were utilized:
-
(1)
Written surveys. Standardized forms were used to record detailed information on hunted taxa, the number of individual animals hunted, and the estimated biomass per taxon.
-
(2)
Face-to-face interviews: In some communities, structured oral questionnaires were conducted with hunters. Trained interviewers asked hunters about their hunting activities and recorded their responses on paper or digital devices. These questionnaires captured the same information as the written surveys.
-
(3)
Direct monitoring: In select cases, researchers accompanied hunters during their activities, recording real-time observations on species composition and offtake.
All data collection efforts were meticulously documented to ensure consistency, accuracy and comparability across communities and time periods. The approach was designed to be both culturally respectful—honouring the knowledge systems and practices of participating communities—and scientifically rigorous, thereby ensuring the reliability and integrity of the data compiled in the Marupiara dataset.
Data selection and validation
Given the diversity of secondary data sources spanning nearly six decades, our dataset naturally varied in objectives, methodological rigour, reporting standards and cultural contexts. To ensure consistency, we focused on extracting comparable information across all studies.
We implemented a multi-step validation process to mitigate potential inaccuracies. Primary data served as a baseline for evaluating datasets and assessing methodological consistency, identifying discrepancies and evaluating the reliability of secondary sources. We prioritized studies that provided clear methodological descriptions, reproducible metrics or supplemental documentation and contacted the original authors or institutions when clarification was needed regarding sampling design and local conditions.
We scrutinized outliers and apparent inconsistencies by cross-referencing with more recent peer-reviewed sources or primary datasets from similar regions or time periods. When discrepancies were identified, we assessed whether they reflected genuine cultural differences, shifting hunting practices or methodological shortcomings. Records showing implausible biological or cultural values were excluded.
Throughout the validation process, we remained attentive to cultural and practical factors—such as hunting laws, local traditions and wildlife management strategies—which vary widely across regions and over time. To address this, we consulted local experts and researchers familiar with such nuances to confirm that data collection methods were culturally appropriate and to verify that the underlying assumptions of each dataset remained valid. Only datasets meeting our standards for scientific rigour and comparability were integrated into the analysis and annotated with metadata. We also excluded studies that reported less than four hunted taxa, as these offered limited insights into species composition. Records lacking clearly described or reliable hunting effort methodologies were removed from the final dataset.
Taxonomic reclassification
We extracted the list of mammal, bird, reptile and amphibian species for the 625 localities from the IUCN spatial database18, assuming that the taxonomic identity of all recorded hunted species aligns with the taxonomic and geographic distribution currently recognized by the IUCN18. We then reviewed all 1,789 original taxa from the Marupiara dataset to correct potential misclassifications, including outdated taxonomy, vernacular identifications, misidentifications or typographical errors. Taxonomic entries were updated to the most refined, species-specific refined classifications, with taxa categorized into species wherever possible.
Following this review, we identified 438 species, 51 higher taxa and an ‘undetermined’ category, resulting in at least 490 distinct species (Supplementary Table 1). The ‘undetermined’ category includes unidentified species or aggregations for which TSOP could not be reliably calculated. Higher taxa represent broader taxonomic groupings retained from the original sources, such as Mazama spp., which includes lower taxa like Mazama americana, Mazama nemorivaga and Mazama gouazoubira, which are already listed separately in the Marupiara dataset.
All allopatric species were aggregated at the genus level to improve analytical coherence, and taxa with very low sample sizes were grouped into broader categories at the family, order, or subclass level. This classification process resulted in 173 analytically focused hunted taxa, plus the ‘undetermined’ category, which were used in TSOP and related analyses (Supplementary Tables 1 and 2).
Taxon-specific body mass and density
We compiled 4,019 observations on body mass for 477 animal species, representing all 173 hunted taxa. These data were sourced directly from the Marupiara dataset, from primary data and from the scientific literature. Based on these records, we estimated the average body mass for each of the 173 taxa (Fig. 2, Extended Data Fig. 3, Supplementary Table 2 and Supplementary Data 3). Additionally, we collected from published sources a total of 2,024 observations on population density for 330 hunted animal species, representing 139 of the 173 hunted taxa. For these 139 taxa, we estimated the average number of individuals per 100 km² (Supplementary Table 2 and Supplementary Data 5). Details on the estimation procedures for both body mass and population density are provided in the Metrics Estimation section of the Methods.
Spatial modelling and raster manipulation
All spatial analyses were performed at 10 × 10 km raster resolution and WGS (World Geodetic System) 84 Datum using the terra58 and randomForest59 packages in R software60. Maps were produced in QGIS (https://qgis.org/en/site/), with the final edition in Inkscape (https://inkscape.org) and GIMP (https://www.gimp.org). We spatially predicted the 10%, 25%, 50%, 75% and 90% quantiles for all the spatial variables and calculated the mean, standard deviation and 90% confidence intervals from the pixel values of this five rasters stack.
Although relatively fine-grained given the vast extent of the Amazon biome, the 10 × 10 km resolution may still be insufficient to accurately capture fine-scale heterogeneity in habitat conditions, species distributions, and conservation needs. However, this scale provides a necessary balance between regional coverage and data availability, guaranteeing compatibility with widely used ecological and environmental datasets while allowing for broad-scale assessments that inform conservation planning. A finer resolution would significantly increase computational demands and face data limitations, particularly across such an extensive and heterogeneous region. The 10 × 10 km scale remains effective for identifying broader patterns, and its outputs can be complemented in the future by higher-resolution studies or downscaled using local ecological data to refine site-specific conservation actions.
Amazonian geographical boundaries
We defined the geographical boundaries of Amazonia according to the Amazon Network of Georeferenced Socio-Environmental Information (RAISG), which combines the Amazon biome, its river basins, and relevant administrative regions1. We rasterized this polygon at a resolution of 0.0083, resulting in a total area of 8,179,389 km2.
Amazonian regions are classified into: (1) Guiana Shield (GS); (2) northern part of western Amazonia (WAN); (3) central Amazonia (CA); (4) southern part of western Amazonia (WAS); (5) southern Amazonia (SA); and (6) eastern Amazonia (EA) (Fig. 1).
Amazonian peoples and rural hunters
To spatially determine rural Amazonia, we included only regions 2–3 h away from a small city or town (100,000–250,000 urban inhabitants) as defined in the urban–rural catchment area raster61 (Extended Data Fig. 1). Amazonian peoples were defined as all individuals living in rural Amazonia, frequently self-declared as Indigenous, traditional or small-scale farming peoples. To generate the Amazonian peoples population size (APPS) raster, we first removed all urban areas61 from the raster of the total human population (people per pixel for 5 time points between 2000 and 2020 at 5-year intervals)62 within Amazonia boundaries (Extended Data Fig. 1). This spatial filtering resulted in an estimated rural population of 10,870,022 individuals (APPS), corresponding to 34.2% of the 31,783,941 individuals living (total human population size) in Amazonia.
To obtain the rural hunters population size (RHPS) raster, we multiplied the APPS raster by 0.178 (median) and by 0.168, 0.173, 0.183 and 0.187 (the 0.10, 0.25, 0.75 and 0.90 quantiles), which represent the estimated proportion of hunters to consumers based on data from 72 localities of the Marupiara dataset with both information of number of hunters and number of consumers (Extended Data Fig. 1). See ‘Metrics estimation’ details on estimation. Based on these calculations, the total number of rural hunters in Amazonia was estimated at 1.93 ± 0.08 (1.83–2.04) million.
Wild meat trade and urban consumption data were excluded to maintain a focus on rural hunting practices and their ecological and nutritional implications. Including trade and urban consumption would have introduced complexities related to market dynamics, transportation and intermediate processing, which are outside the scope of this study’s traditional-focused approach. Additionally, reliable trade and urban consumption data are often sparse or inconsistent in Amazonia, making integrating them into the models challenging without introducing significant uncertainty. Although the spatial exclusion of urban areas may inadvertently omit individuals who rely on wild meat as part of their diet, the inclusion of peri-urban populations in our dataset allows for a more reliable assessment of the effects of urbanization on HP.
Spatial variables
To analyse the factors shaping wild meat harvest in Amazonia, we selected the following set of environmental and anthropogenic spatial variables (Supplementary Methods 1–3):
Enhanced vegetation index
A dimensionless index that describes the difference between near-infrared and red reflectance of vegetation cover, normalized by their sum, corrected for some atmospheric conditions and canopy background noise, that can be used to estimate the density of green cover on an area of land, especially in areas with dense vegetation63. We included values for enhanced vegetation index (EVI) from 2000, 2005, 2010, 2015 and 2020.
Annual gross primary productivity
A MODIS-Terra digital database expressed in kgC m−2 year−1 (ref. 64) represents the total amount of carbon compounds produced by the photosynthesis of plants in an ecosystem over a given period of time65. We included values for annual gross primary productivity (GPP) in 2000, 2005, 2010, 2015 and 2020.
Annual net primary productivity
A MODIS-Terra digital database expressed in kgC m−2 year−1 (ref. 64) represents the carbon uptake plants retain in an ecosystem after accounting for plant respiration (net increase in biomass)65. We included values for annual net primary productivity (NPP) in 2000, 2005, 2010, 2015 and 2020.
Soil fertility
A digital database covering Amazonia, represented as the sum of exchangeable base cation concentration66.
Proportion of flooded areas
A digital database of the proportion of flooded areas in the Amazon region to upland forests, built on the raster manipulation and combination of the products provided by Hess et al.67 and Lehner & Dohl68.
Elevation
A digital topographic database scaled in metres based on the NASA Shuttle Radar Topography Mission69.
Height above the nearest drainage
A digital terrain model normalized to the elevation in metres of the drainage network70.
Historical distribution of Indigenous family languages
A categorical digital database of the distribution of the Indigenous territories during the early accounts in Amazonia, built from the combination of the maps provided by Loukotka71 and Eriksen72 and classified into their respective Indigenous family language (Supplementary Methods 2).
Current distribution of Indigenous and non-Indigenous peoples
A categorical digital database built from the RAISG database of the distribution of the Indigenous lands73, classified into Indigenous peoples (regions inside Indigenous lands) and non-Indigenous peoples (regions outside Indigenous lands). The current peoples’ cultural identity (Indigenous or non-Indigenous) was compiled in the Marupiara dataset from primary and secondary studies.
Current distribution of family languages
A categorical digital database built from the RAISG database of the distribution of the Indigenous lands73, classified into their respective Indigenous family languages (Supplementary Methods 3). Regions outside the Indigenous Lands are tentatively classified as Latin or German languages. The cultural identity of the current people was compiled in the Marupiara dataset from primary and secondary studies.
Proportion of habitat loss
A digital database built on the raster manipulation of annual land cover mapping provided by the MapBiomas74. We included measures for the Proportion of habitat loss in 1985, 1990, 1995, 2000, 2005, 2010, 2015 and 2020. We manipulated the MapBiomas land use rasters to obtain rasters with the proportion of habitat loss.
Urban–rural catchment areas
Urban–rural catchment areas (URCA) is a digital database of the 30 urban–rural catchment areas showing the catchment areas around cities and towns of different sizes, in which each rural pixel is assigned to a defined travel time category61.
Hunting recording time span
A variable that controls the effort to record hunting in each study, that is, the time range in days in which hunted animals were recorded. This metric was only used to model HHR. We included this variable since we assumed that different time spans of hunting surveys could have different accuracies for data on the animals hunted. We obtained measures of the hunting recording time span from the Marupiara dataset.
Raster manipulation and processing
We reprojected all spatial variables’ rasters to WGS 84 Datum and 10-km resolution, cropped them to the geographical boundaries of Amazonia1, and replaced missing pixels with the interpolated value from the neighbour pixel. However, even performing this gap-filling technique reduced the original spatial inference from 81,790 to 80,459 10 × 10 km cells (Extended Data Fig. 1).
We extracted the values of all these spatial variables for the 301 georeferenced localities with HHR measures and 590 with TSOP measures (Fig. 1a).
Temporal variation of spatial variables
None of the digital spatial variables fully cover the 1965–2024 period of the Marupiara dataset, preventing a year-by-year evaluation of their spatial and temporal effects on HHR and TSOP. To address this, we matched HHR and TSOP records to the closest available values of spatial variables in years where spatiotemporal data were available. This approach was feasible for modelling the effects of EVI, NPP and GPP on HHR and TSOP by incorporating measurements at five-year intervals. We obtained proportion of habitat loss data for 1985, 1990, 1995, 2000, 2005, 2010, 2015 and 2020. For EVI, NPP, and GPP, data were available from 2000 onwards at the same five-year intervals. HHR and TSOP values from 1965 to 1987 were assigned the 1985 habitat loss data; from 1988 to 1992, the 1990 data; from 1993 to 1998, the 1995 data, and so on, with post-2018 records assigned the 2020 data. A similar process was applied to EVI, NPP, and GPP, though for a shorter period: records from 1965 to 2002 were matched to the 2000 values, continuing at five-year intervals up to 2020. Despite the Marupiara dataset spanning 1965–2024, most hunting studies occurred around 2006 ± 9 years (90% quantiles: 1995–2017), aligning well with the available temporal coverage of the spatial variables.
Overall individual animals HHR and overall individual animals HP
After cropping the Marupiara dataset to the geographical boundaries of Amazonia, we obtained georeferenced observations with information on the number of individual animals hunted per hunter per day—overall individual animals HHR. We extracted the values of all spatial variables for the 301 georeferenced localities (both from primary and secondary data) with HHR measures. We determined the overall individual animals HHR as the total number of animals hunted of all taxa in each locality divided by the sum of hunters accountable for catching those animals during the monitored period in days. This includes days when hunters neither went hunting nor harvested any animals.
Firstly, we spatially predicted the overall individual animals HP by performing random forest models, where the overall individual animals HHR across Amazonia was a function of the enhanced vegetation index, gross primary productivity, net primary productivity, net primary productivity quality control, soil fertility, proportion of flooded areas, elevation, height above the nearest drainage, historical distribution of Indigenous family languages, current distribution of Indigenous and non-Indigenous peoples, current distribution of languages, proportion of habitat loss, urban–rural catchment areas, and hunting monitoring time span. We ran 30 models to the entire Amazonia with 70% of observations each and took the central and 75% and 90% quantiles of these 30 overall individual animals HP spatial models. The ranking importance of each predictor in the full model (that is, with 100% of the observations) is given in Supplementary Data 1. Individual animals HP (individuals HP) reflects how the overall individual animals HHR varies with environmental and anthropogenic factors (Fig. 1b,c).
TSOP and taxon-specific individual animals HP
We obtained 590 georeferenced observations on the number of animals hunted of each taxon in each locality of the Marupiara dataset (Fig. 1a). We extracted the values of all spatial variables for these georeferenced localities. We assigned a value of zero (absence) when the hunted taxon was predicted to occur in that locality according to the IUCN spatial database, but no animals were hunted there. We then spatially predicted the individual animals HP for the 174 hunted taxa covered in our dataset. To accomplish that, we first calculated the TSOP of the 174 hunted taxa in the 590 localities as the number of animals hunted of a taxon in each locality divided by the total number of animals hunted in the same locality. We ran spatially explicit random forest models for the 174 taxa using their TSOP as functions of the same environmental and anthropogenic factors used to estimate the overall individual animals HP, except for the hunting recording time span not considered in the TSOP random forest models.
To prevent underestimation of the TSOP of some taxa, we removed localities that contained data for both the lower and the corresponding high taxa when modelling for the lower taxon. For example, localities that contained hunting data for both M. americana and Mazama spp. were removed from the analysis when the TSOP of M. americana was modelled.
Each of the 174 TSOP rasters was clipped to its respective geographic distribution obtained from the IUCN database18. To ensure that the proportions of the 174 TSOP rasters summed to 1 in each spatial cell, we normalized each raster by dividing it by the sum of all 174 TSOP rasters, improving the accuracy of TSOP spatial predictions.
We then multiplied the overall individual animals HP raster by each of the 174 TSOP rasters. This allowed us to spatially predict the individual animals HP for each of the 174 hunted taxa, resulting in a taxon-specific individual animals HP. The sum of the 174 taxon-specific individual animals HP equals the overall individual animals HP.
Taxon-specific and overall animal biomass HP
By multiplying the 174 taxon-specific individual animals HP rasters by their respective estimated median and 0.10, 0.25, 0.75 and 0.90 quantiles of taxon-specific average body mass, we built the 174 taxon-specific animal biomass HP rasters (see Fig. 3). By summing all the 174 taxon-specific animal biomass HP rasters, we got the overall animal biomass HP raster (Fig. 1c).
Overall HP
Overall HP regarding individual animals and animal biomass reflects how the overall individual animals HHR and animal biomass HHR vary with environmental and anthropogenic factors throughout Amazonia (Fig. 1b,c).
Taxon-specific and overall individual animals offtake
We spatially predicted the individual animals offtake–the annual number of animals hunted per taxon–for the 174 hunted taxa across Amazonia by multiplying the 174 rasters of taxon-specific individual animals HP (the estimated number of animals hunted per taxon per hunter per day in each pixel) by the RHPS raster (the estimated number of rural hunters in each pixel). To avoid overestimations in areas with high rural hunter density, we applied an upper limit on the number of animals hunted per taxon per pixel: For 139 taxa (primarily species-specific or allopatric genera), we truncated the maximum offtake values based on their taxon-specific average density, defined as the average number of individuals per 100 km² or 10 × 10 km (Supplementary Table 2 and Supplementary Data 5).
We then calculated the number of animals hunted per taxon per year in Amazonia by summing the values from the 174 taxon-specific individual animals offtake rasters. By aggregating these, we generated the overall individual animals offtake raster (Fig. 1d), from which we derived the total number of animals hunted per year by summing the pixel values across the entire Amazonia region.
Taxon-specific and overall animal biomass offtake
The spatial prediction of the taxon-specific animal biomass offtake (the animal biomass in kg extracted of the 174 hunted taxa across Amazonia) was accomplished by multiplying each one of the 174 rasters of taxon-specific individual animals offtake by their respective estimated median and 0.10, 0.25, 0.75, and 0.90 quantiles of taxon-specific average body mass. By summing all the 174 taxon-specific animal biomass offtake rasters, we got the raster of the overall animal biomass offtake (Fig. 1d). From the sum of values of the pixels of overall animal biomass offtake raster, we calculated the total animal biomass extracted (in kg) per year in Amazonia.
Proportion of edible wild meat to undressed biomass
We estimated the overall annual production of edible wild meat in Amazonia (see ‘Metrics estimation’ for details) using data on the proportion of consumable meat contained in animal carcasses reported in the literature75,76. Edible yield proportions were calculated separately for each of the 20 key hunted taxa and then pooled by major taxonomic groups: mammals (0.63 ± 0.11), birds (0.73 ± 0.06), chelonians (0.47 ± 0.14) and caimans (0.45 ± 0.04). These values were then multiplied by the estimated total undressed animal biomass offtake for the corresponding taxa or groups. Based on this approach, we estimated that edible wild meat represents approximately 58.5% of the total undressed biomass harvested annually across Amazonia.
Available wild meat per rural inhabitant
Therefore, we multiplied the predicted raster of the daily animal biomass offtake by 0.585 to spatially predict the daily overall edible wild meat produced across Amazonia. Then, we divided the overall edible wild meat produced raster by the APPS raster, which includes the number of rural inhabitants per spatial cell, generating a raster of the available wild meat per rural inhabitant in each spatial cell (Fig. 4).
Wild meat nutritional composition
Using the scarce data from the literature on the nutritional composition in meat for 26 taxa (22 species and 4 genera; Supplementary Table 9) in Amazonia77,78,79,80,81,82,83,84,85,86,87, we estimated the average amount of energy and macro- and micronutrients in wild meat in Amazonia (Extended Data Table 1). We used 265 observations overall from these literature sources (for example, 59 for protein, 58 for total fat, 44 for energy, 28 for iron, 7 for zinc, 2 for selenium, 20 for vitamin B1, vitamin B2 and vitamin B3 and 7 for vitamin B12 (Supplementary Table 9). Out of the 20 dominant taxa, 12 had species-specific data on nutritional composition (Supplementary Table 9).
A limitation of our study stems from the reliance on approximate rather than species-specific nutritional composition data. This constraint is primarily due to the limited availability of detailed nutritional data in the existing scientific literature88. To address this challenge, we adopted the food-matching technique, a methodology endorsed by the Food and Agriculture Organization (FAO) for such scenarios89.
Daily amounts of energy and nutrients furnished by wild meat
We then produced 10 rasters of the daily amounts of energy and nutrients furnished by wild meat (that is, one for energy and nine for nutrients) for each spatial cell by multiplying the overall edible wild meat produced raster by the estimated average values of energy and macro- and micronutrients contained in Amazonian wild meat (Extended Data Table 1). See ‘Metrics estimation’ for details on estimation.
Percentage of dietary requirements furnished by wild meat
To estimate the nutritional needs of micronutrients of the Amazonian peoples we used DRIs. DRIs are a set of recommendations for nutrient intake based on the latest scientific evidence and intended to guide the amounts of nutrients that are needed to maintain health. Due to the lack of specific nutritional data for the targeted population, such as weight and food consumption, we had to rely on general references measured in grams per day instead of grams per kilogram per day of a given nutrient. Consequently, we selected the estimated average requirement (EAR) values, measured in weight per day, as our primary choice. This approach was taken because the EAR reflects the average daily nutrient intake estimated to meet the needs of half of the healthy individuals within a specific age and gender group90,91. In cases where there was no EAR available, we used the adequate intake (AI) or recommended dietary allowance (RDA) as a guide for determining the appropriate recommendation of a nutrient. For total fat, in instances where DRIs were unavailable, we utilized the acceptable macronutrient distribution range (AMDR). Considering the midpoint of the range, we converted these values to a percentage of energy. Finally, for energy (that is, calories), we applied the estimated energy requirements (EER)92. The DRI values are presented in the Supplementary Tables 10 and 11.
First, we constructed a raster of the population size for each sex-age group (that is, children, women and men) by multiplying the AMPS raster with the proportion of each group relative to the total rural population in Amazonia. These proportions were derived from data available for north Brazil through the Brazilian population census93. Using this approach, we established spatially explicit demographic distributions reflecting rural Amazonia’s population structure.
Then, we spatially predicted each sex-age group’s minimum daily dietary requirements for seven micronutrients, two macronutrients, and energy. This was achieved by multiplying the average reference values for each nutrient or energy requirement with the corresponding sex-age population raster. The resulting minimum daily dietary requirements rasters for children, women, and men were then summed for each nutrient and energy to predict the overall minimum daily dietary requirements for Amazonian peoples across all spatial cells. As a result, these requirements were directly proportional to the number of rural inhabitants per pixel, ensuring that the analysis accurately reflected localized nutritional needs across the region.
Finally, the daily nutrient amounts furnished by the estimated wild meat production in Amazonia were evaluated against the daily micronutrient and macronutrient requirements to determine their adequacy for a nutritionally balanced diet for Amazonian peoples. This was done by comparing the corresponding spatial cell of a given nutrient/energy between the daily amounts furnished by wild meat and the minimum daily dietary requirements of Amazonian peoples. This process allowed us to spatially explicit the levels of energy and nutrients supplied by wild meat to Amazonian peoples.
Metrics estimation
To enhance the accuracy of our metrics (such as taxon-specific body mass, taxon-specific density, proportion of hunters to consumers, proportion of edible wild meat to undressed biomass and wild meat nutritional composition), we run each metric 1,000 times, drawing each estimated quantity from a normal statistical distribution defined by its corresponding mean and standard error. After obtaining 1,000 values for each metric, we took the mean, standard deviation and 90% quantiles of these 1,000 values to produce a 90% confidence interval.
Research collaboration and ethics
The community-based hunting monitoring initiatives that contribute primary data to the Marupiara dataset were developed and implemented through close partnerships with Indigenous and traditional peoples. These initiatives should not be seen as conventional academic research, where external researchers extract data on biodiversity use. Instead, they are grounded in sociopolitical realities relevant to Indigenous and traditional communities and are designed to strengthen their territories and livelihoods, empower communities, support wildlife conservation and management, enhance food security, and protect cultural practices—always with respect for their priorities and autonomy. The data collected directly informs community-led decisions on sustainable wildlife and territorial management, with meaningful community engagement throughout the monitoring cycle. Results are transparently shared at the community level and used to guide practical actions on the ground. Each project is tailored to the specific needs of the communities involved, ensuring that Indigenous and traditional knowledge is respected, safeguarded and valued. Our collaborative approach includes training Indigenous and traditional hunters and researchers, fostering long-term capacity building and education. Wildlife monitoring programmes are conducted under government regulation, ensuring ethical data handling and confidentiality. All research activities are formally approved by Indigenous and traditional communities, as well as by relevant academic or governmental institutions overseeing Indigenous Lands, parks, and extractive reserves.
Data-sharing agreements were established among communities, researchers, and technicians, enabling informed local decision-making and advancing research on wildlife use, management, and conservation in Amazonia. Free, prior, and informed consent (FPIC) was obtained—either orally or in writing—from all communities participating in those initiatives, ensuring ethical engagement and respect for rights, welfare, and autonomy. While some early initiatives predated formal non-Indigenous and local ethics committees (for example, before the Nagoya Protocol, 2010), agreements were always culturally adapted, ranging from oral consensus to written contracts detailing research objectives, participant rights, and data use (Supplementary Methods 4).
Six independent ethics committees reviewed and approved all primary data methods, with the 12 contributing initiatives receiving clearance from institutional review boards across Brazilian, Peruvian, and French Guianese Amazonia (Supplementary Methods 4). These approvals, secured through universities and research institutions, guaranteed compliance with international ethical standards for research involving Indigenous and traditional peoples.
The primary data collection methods were approved by the main representative organizations of Indigenous peoples and traditional communities in the Brazilian Amazon—respectively, the Coordination of Indigenous Organizations of the Brazilian Amazon (COIAB) and the National Council of Extractive Populations (CNS). Both organizations have formally endorsed the content of this article and have committed to participating in the Evaluation Committee for future research utilizing the Marupiara dataset (Supplementary Methods 6 and 7).
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
