July 2026 Field Report – Advancing Biodiversity Monitoring by Integrating Machine Learning with Indigenous Rainforest Expertise

Acaté’s versatile new AI tool advances biodiversity data management by delivering highly accurate image recognition of Amazonian species. This technology directly supports wildlife-monitoring devices deployed by Indigenous ecology experts.

This exciting application of technology guided by Indigenous stakeholders opens the door to continual, efficient biodiversity monitoring in a large and globally significant conservation area, and beyond.

Global satellite-based platforms automate alerts for the loss of tropical forest cover at high resolution; however, no comparable tool exists to detect impacts on biodiversity. Yet, biodiversity inventories and monitoring are key tools for assessing ecosystem health and determining conservation priorities, and as such, are integral to safeguarding the planet’s future. Biodiversity monitoring in the Amazon on Indigenous lands is primarily conducted by outsiders, specifically biologists. However, this assessment is often itinerant, sporadic, or occurs only once. If carried out by trained local stakeholders/native inhabitants, biodiversity monitoring can be continuous and longitudinal rather than a snapshot in time.

The Matsés people of Peru, along with many Indigenous groups in the Amazon, possess a deep understanding of the local flora and fauna that exceeds current scientific knowledge. They can recognize and classify various habitat types based on geomorphology and vegetation, and they know which plant and animal species inhabit each environment. For instance, the Matsés identify and name 47 different habitat types within their rainforest ecosystems, a number that far surpasses what is recognized in Western science.

biodiversity machine learning AI

Illustration of some categories of the Matsés traditional habitat classification system.©Acaté

Since diverse ecosystems and soil types underpin the rich biodiversity of the Amazon rainforest, having a detailed understanding of these microenvironments is key for effectively studying its extraordinary biological diversity.

Rapid biological inventories and remote sensing are invaluable tools for assessing broad-scale environmental changes and identifying priority conservation areas. However, neither method can fully substitute continuous, on-the-ground monitoring. While remote sensing is effective at detecting changes in forest cover, it cannot confirm species presence, track population trends, or assess the ecological integrity of habitats. Rapid inventories provide crucial snapshots of biodiversity but only capture a moment in time; they may miss rare species, seasonal fluctuations, or subtle, gradual declines.

Our initiative enhances these approaches by establishing permanent monitoring infrastructure within Matsés territory, allowing us to generate continuous, verifiable data that tracks species composition over time. This ground-level evidence is vital for detecting early warning signs of biodiversity loss—signs that satellite imagery and short-term surveys often overlook.

By integrating the strengths of all three methods—using remote sensing for landscape context, rapid inventories for baseline species lists, and permanent on-the-ground monitoring for diachronic analysis—we create a more comprehensive and accurate picture of ecosystem health.

Matsés project leaders strategically place camera traps and acoustic recording devices in the rainforests near their communities. The monitoring devices, placed in key locations by experts, rapidly generate massive volumes of data files. However, manually processing and analyzing this data would be extremely time-consuming and require significant technical expertise. These limitations hinder many of the practical applications for biodiversity monitoring.

To overcome these barriers, Acaté developed a Python-based tool that fully automates data processing and accurately identifies recorded species. The tool selects individual frames for analysis and uses both publicly available recognition tools and our own machine learning algorithm, created by expert biologists, to identify species. This results in unparalleled accuracy in automated species identification for the fauna of the Matsés ancestral rainforest, enabling effective and continuous biodiversity monitoring. The analyzed data is then uploaded to the Matsés Ancestral Rainforest Territorial Protection app, providing the Matsés with easy access.

Acaté is an on-the-ground conservation organization that strategically focuses on strengthening rural and indigenous communities as the most effective natural climate solution. Our work centers on creating sustainable economic alternatives for indigenous and local communities that depend on forests, thereby halting deforestation at its source. We integrate field data, traditional knowledge, and technology to monitor ecosystems, enhance territorial rights, and develop bio-economies. This directly preserves biodiversity, secures vital carbon stocks, and enhances community resilience. Our outcomes-based approach demonstrates that sustaining the environment and combating climate change are inseparable from supporting the guardians of the forest.

Background

The Amazon Rainforest is the most biologically diverse terrestrial biome on Earth.

Stretching across nine South American countries, the Amazon’s rich biodiversity supports ecosystem resilience, nutrient cycling, and food webs crucial to natural and human systems on the continent. Globally, the Amazon plays a vital role in maintaining ecological balance and regulating the planet’s climate.

biodiversity integrated machine learning

Since time immemorial, the Matsés Indigenous People of the Peruvian Amazon have lived in their rainforest territory without extirpating a single species or compromising the environmental integrity of their lands. Their sustainable practices result from their historical (1) semi-nomadic settlement patterns, (2) strong territorial defense, 3) extensive ancestral ecological knowledge, and (4) traditional game management strategies, the latter three of which remain in effect today. However, this balance is threatened by extractive commercial activities and by the risk that new generations of Matsés will not learn to manage resources in their territory and overhunt for the sale of bushmeat, harvest timber for sale, or allow industrial agriculture enterprises to convert their rainforest to farms or cattle pastures.

In response to these threats and to ensure the continued conservation of the ecosystem, Acaté and the Matsés have initiated a biodiversity monitoring program using non-invasive methods (camera traps, audio recorders, and routine site surveys) to detect any declines in biodiversity and guide biodiversity conservation strategies. Acaté has previously partnered with the Matsés on a number of biodiversity initiatives, including an indigenous-led survey of amphibians and reptiles using smartphone cameras, the first systematic survey of endangered giant river otter populations thriving in the headwaters of Matsés ancestral territory, a series of eleven ecological books written by Matsés authors, a number of educational apps, and a successful indigenous-led project that led to the reintroduction of the threatened giant river turtle and restoring extirpated populations near frontier settlements.

biodiversity AI machine learning river turtles

Challenges and Pitfalls of Biodiversity Monitoring in the Amazon Rainforest

Biodiverse ecosystems are inherently complex, and assessments of biodiversity are highly resource intensive. Typically cross-sectional, such assessments provide a snapshot or glimpse in time rather than continuous observations over time, which would confer greater utility from a conservation perspective for noting population declines or detecting rare species. Biological surveys of Amazon rainforest confer challenges because many species are naturally elusive, occur in low population densities, inhabit the canopy or other difficult-to-access habitats, or may have been extirpated in parts of their range closest to human occupation.

In the Amazon, biodiversity surveys are typically carried out by experts in particular fields of biology (e.g., ornithologists, mammalogists) who can identify the species in the field. Considering the difficulty of access and the considerable limitations of personnel, time, and financial resources, scientific biodiversity inventories in Amazonia are few and incomplete. Except perhaps for a few university-supported research field stations, long-term biodiversity monitoring is essentially nonexistent.

The benefit of extended assessments over rapid inventories is clear; data presented just this week reported detection of as many as 40,000 insect species, many almost certainly new to science, in one region of northern Brazil studied continuously by a large team of entomologists for a period of 14 months. The area studied by scientists in this “unique megaproject” comprised 10,000 hectares; in comparison, Matsés ancestral lands in Peru encompass 1.2 million hectares characterized by diversified ecosystems that include lush upland forests, wet bottomlands, floodplains, swamps, and white sand forests that hold high levels of endemism.

Currently available tools used by scientists for biodiversity assessments outside of specimen collection and direct observation include camera traps, bioacoustic recording devices, and environmental DNA (eDNA). Each method has its own advantages and disadvantages, and it’s important to understand that there is no “one-size-fits-all” solution; no method is truly “plug-and-play”.

Camera traps have long been a staple for biodiversity assessments and are generally the easiest to implement. However, even a handful of camera traps placed in active areas will quickly generate a large volume of memory-intensive video files. These files require processing, condensing, and analysis for the monitoring data to be useful. Additionally, identifying animals requires the time of knowledgeable experts and can be challenging due to factors such as lighting conditions or capturing only brief or partial images (for example, just the tail of an animal).

Bioacoustics, or the passive recording and identification of acoustic sounds of birds, bats, frogs, and insects, and other taxa. Analyzing and interpreting raw acoustic data may seem conceptually straightforward at first glance, but it actually requires considerable expertise, advanced software tools, and validated reference libraries. For instance, after our initial trial run of bioacoustic devices in the rainforests near Iquitos, we ran the downloaded data through the most widely utilized bioacoustic software and avian reference libraries. The report indicated the detection of cassowary birds, even though these large terrestrial birds are native to rainforests of New Guinea and Australia—over 8,000 miles away, among other spurious results. This highlights the importance of developing and curating an appropriate reference library, which we are currently investing significant time in.

Lastly, eDNA is the newest available and most expensive modality. Rigorous bioinformatics and well-validated reference libraries are needed to minimize the significant potential for false positives or negatives in results. eDNA arrays are expensive, and each test is consumable and cannot be reused like camera traps or bioacoustic devices. The approach minimally involves/engages Indigenous communities/local stakeholders, as the swabs are sent off to genomic laboratories in large metropolitan areas. The thrust of our initiative is to place monitoring directly in the hands of frontline and stakeholder communities who are protecting biodiversity. For this initiative, which is intentionally centered around local communities, we selected a combination of camera traps and bioacoustics to support indigenous surveys as the most versatile, cost-effective, reliable, and least invasive ways to detect the greatest variety of species. Further, the methodology is transparent and resulting data can be externally validated.

Tropical rainforests are home to many thousands of species. An important practical question that arises in respect to biodiversity assessments is which taxa or species to focus on.

Target species need to be accessible for cost-effective detection and the detection must be both quantifiable and verifiable. Importantly, species selected should be representative indicators of total ecosystem biodiversity. Various approaches have been utilized in this context. One popular method is to use a single “umbrella species,” often a charismatic or iconic species like the jaguar, as a proxy for overall biodiversity to guide ecosystem management. While this idea is simple and appealing, this “short-cut” is not supported by a number of scientific studies. Multi-species approaches are more promising. Similarly, focusing exclusively on one taxon, such as birds, as a standalone metric for total ecosystem biodiversity has been shown in studies conducted within different geographic areas to have limited utility. Again, adding species from other taxa improves effectiveness.

For this initiative, we recognize that relying on a single species is likely insufficient. Therefore, we focused on a range of indicator species, which include jaguars, pumas, short-eared dogs, giant armadillos, giant anteaters, red uakari monkeys, blue-headed macaws, harpy eagles, pale-winged trumpeters, and king vultures.

The inherent adaptability and versatility of our system is a major advantage; the selection of indicator species can be readily expanded or reduced, as needed, based on new data, to conduct scientific studies, or in response to new conservation pressures. The primary requirement for the species is that they are reliably detectable by camera traps or bioacoustics in a manner that can be quantified and validated. Our automated AI tool has the capacity to analyze one species or a combination of multiple species.

Field Implementation

Over the past two years, we have been training Matsés project leaders in the use of camera traps and bioacoustic recorders. The lead for this project is Guillermo Nëcca Pëmen Mënque, a Matsés artist and expert in rainforest ecology. He acquired his knowledge from his father and other elder relatives during hunting expeditions and while harvesting forest foods and collecting medicinal plants.

biodiversity AI machine learning

Guillermo is a very talented self-taught artist, and drawing on his understanding of rainforest animals, he has created hundreds of watercolor paintings for educational readers, ecological books, and apps that promote Matsés bilingual education and the intergenerational transfer of ecological knowledge.

The positioning of devices is guided by the Matsés’ extensive ecological knowledge of various habitat types within the forest. The Matsés are experts at identifying the presence and identity of animals based on direct observation and indirectly through tracks, calls, scent markings, signs of foraging or predation. It is notable that in the Matsés culture, both men and women are experts in rainforest knowledge and animal tracking.

biodiversity machine learning AI

Deployment is in the primary rainforest within easy walking distance of the Matsés villages and swiddens. The biodiversity monitoring teams are composed of Matsés men and women who reside in the nearest village to surveyed areas. This set-up is ideal for multiple reasons: (1) the residents are familiar with the surveyed area and can readily find mineral licks and other ecologically strategic camera trap placements where animals concentrate, (2) forest near villages is the most likely to be overhunted or avoided by animals due to the proximity of humans, (3) reduced travel distance enable the sites to be easily accessed and visited by monitoring teams, and (4) these nearby areas provide an effective early warning system as these are the areas that are most likely to be negatively impacted by humans, thus permitting timely remediation.

Depending on the models, both the camera traps and audio recorders have around 20 individual settings that need to be configured through their interface menus. Field teams must be able to operate trail cameras, bioacoustic recorders, and handheld GPS units effectively. These settings must be accurate for the specific conditions; otherwise, the recordings may not provide reliable species identifications. Conducting field tests with trial deployments of camera traps is also crucial to determine the optimal height and lighting conditions for effective visual capture and identification.

Placing this technology in the hands of Indigenous rainforest experts immediately proved successful, with detection, in a short time, of repeated captures of highly elusive and enigmatic species such as the Short-Eared Dog (Atelocynus microtis) shown below. Known to the Matsés as nimëduc opa, the species is rarely glimpsed by the outside world and has been called the Ghost Dog of the Amazon. A unique creature endemic to the Amazon Rainforest, the Short-Eared Dog is perhaps the world’s least understood canid species.

Camera traps and bioacoustic data are collected at regular intervals. The batteries are replaced, and the memory chips are swapped out and sent to our office in Iquitos on scheduled flights. Once the chips arrive, the data is downloaded and processed using a Python-based program we developed to help us manage the large volume of data generated through the continuous monitoring. This program cleans, compresses, and re-names the video files.

Next, the data is run through a Python-based AI tool custom-coded by Acaté. The pilot version, developed over the past year, takes the raw videos and selects key frames for identification. These frames are then run across known biodiversity libraries and analyzed against our expertly curated library. The algorithm is designed to prioritize our smaller, but expertly curated library specific to the Matsés rainforests, over external reference libraries. This is how each still image is analyzed, and the aggregate score of all the still images yields an identification and a certainty score for the identification.

Many of the camera trap videos are triggered as the result of wind, insects, and bats that cannot be identified, so it is crucial to train the AI not to hallucinate species IDs in these empty videos. This is an iterative process; as the libraries grow, the accuracy and precision of the outputs will increase.

The last step in the process is that species IDs, audio and visual files are uploaded into the Matsés Ancestral Territory mobile application, a pioneering GIS system and mobile app developed over two years ago by Acaté in collaboration with the Matsés people to support them in the protection of their land and preservation of their rich cultural heritage.

biodiversity AI machine learning

For uncounted generations, the Matsés fought to defend their land against the encroachment of outsiders intent on resource extraction. It is for this reason that their forests, which are among the most biodiverse and carbon rich in the world, remains intact today.©Acaté

The Matsés Ancestral Territory App was the logical progression of the landmark Matsés Indigenous Mapping Initiative. In the latter initiative, the Matsés Indigenous people with the assistance of Acaté Amazon Conservation set out to demarcate and map, for the first time, the Matsés ancestral lands in Peru. Teams of Matsés elders and youth set out across their territory and deep into the headwaters to locate and georeference with GPS units sites of historical events, natural resources, current and past villages, hunting camps, trails, rainforest habitat types and other sites of cultural and ecological significance. In total, after five years of intensive work, the Matsés recorded and georeferenced over 12,000 culturally significant data points.

At the start of the Matsés Indigenous Mapping Initiative in 2015, the Matsés’ rich knowledge of their lands and ancestral history resided only in the memory of their living elders. Much of this knowledge is now available within the app along with the recorded voices of the elders.©Acaté

Through the Matsés Indigenous Mapping Initiative, over 150 Matsés were introduced and trained in basic computation, 45 in GPS usage, and 8 in GIS software. ©Acaté

The Matsés Ancestral Territory app features an interactive map with thousands of locations documented by the Matsés themselves in the Matsés Indigenous Mapping Initiative, bringing their history and traditions to life in an appealing format for the younger generations. The development of the Matsés Ancestral Territory app was timely due to the shifting and exceedingly complex situation of land tenure and management in parts of their ancestral territory. The Matsés currently control about half of their ancestral territory in Peru, and legal changes are actively being applied to or proposed for parts of their ancestral lands outside their titled and concession areas. This app makes this information accessible at their fingertips, as well as translations of their legal rights in Peruvian law into the Matsés language.

Matsés Ancestral Territory app biodiversity machine learning

Benefits and Applications

Why is this initiative important?

The implications of this proof-of-concept initiative are profound; for the first time, it opens the door to continuous biodiversity monitoring by Indigenous stakeholders themselves and makes that data accessible to them at their fingertips. The methodology is efficient, replicable, and transparent. Further, our platform is highly versatile and can be automated to quantify valuations of species frequency, population densities, as well as behaviors such as feeding or territorial marking. The power of machine learning is that, properly trained, it can detect and identify patterns faster and more accurately than human observers. For certain endangered or threatened species, such as giant river otters or jaguars, we can apply AI to identify and track individuals through their individually-distinct pelt patterns.

Although developed and tailored to the species of the rainforests of the Matsés Ancestral Territory in its present application, the technology is readily adaptable to other areas, in Amazonia, and beyond.

While the methodology enables potential for novel remote biodiversity monitoring applications, importantly the approach is firmly centered on the ground, in the forests, operated and led by Indigenous stewards and local operators together with scientists. As current trends for conservation monitoring become increasingly remote with satellites or aircraft based monitoring, this comes at the risk of becoming ever more disconnected from the on-the ground realities where deforestation is occurring, and from the local communities who far too often get marginalized in terms of international funding opportunities and engagement. The world looks simple from 300 miles up, but on-the-ground realities are far more chaotic, where local communities and operators on the ground – who are the heart and soul of conservation – are struggling from lack of resources and opportunities. Without local partnership and engagement, conservation efforts are doomed to failure.

biodiversity machine learning AI

This biodiversity monitoring initiative leverages AI and recent technological advancements in an environmental forward application to help inform and support on-the-ground conservation. AI, as with other technologies, are tools, not conservation solutions nor a substitute for on-the-ground conservation initiatives.

Biological diversity initiatives such as this add economic value to conservation and thereby become a source of pride and opportunity for young people. Due to their recent isolation, the elder Matsés still possess undiminished traditional knowledge. However, acculturation of the Matsés to the national cultures proceeds rapidly every year. Young Matsés often look to the world outside their territory for entertainment, prestige, and economic success and generally show less interest in traditional culture and knowledge. While Matsés elders continue to take pride in their identity as Matsés and wish to pass on their wisdom, increasingly it is harder find a young person willing to listen. This initiative not only helps transmit natural history knowledge, but it will also make evident to young people the importance and value of their natural environment.

biodiversity machine learning AI

From a global conservation perspective, species diversity is a commonly accepted metric for prioritizing conservation efforts and for the emerging global market for biodiversity credits. Gaining a better understanding of the territory’s diversity might translate into better external financial support for on-the-ground conservation projects and result in new kinds of job opportunities for the Matsés, both in larger and smaller communities, that are compatible with their culture.

biodiversity machine learning AI

For those that donate and support our on-the-ground conservation initiatives, this App is a portal to see for themselves verified outcomes of intact rainforests and healthy biodiverse ecosystems. Through the App, supporters can glimpse the wildlife and hear the sounds of the very rainforests they are helping to protect, not only reinforcing their funds are well spent but conferring a tangible connection for those who in their lifetimes may not have the opportunity to visit the rainforests they passionately support.

We are grateful for our funders, whose support and vision made this ongoing work possible. Our team is continuing to advance this initiative together with the help of a wonderful group of talented volunteer tech interns from Egypt, Madagascar, South Africa, and Bangladesh. If you are interested in supporting our work to reach its full potential please contact us.

If you missed it, take a look at our October 2025 Field Report for an update on our highly successful Integrated Aquaculture program, which received a special recognition award from the United Nations Food & Agriculture Organization (FAO). Stay tuned shortly for exciting new developments as we advance in the challenging next stage to construct an on-site fish propagation laboratory in the remote Matsés territory!

Acaté Amazon Conservation is a non-profit organization based in the United States and Perú that operates in a true and transparent partnership with the Matsés people of the Peruvian Amazon to maintain their self-sufficiency and way of life. The Matsés safeguard a critical conservation corridor and shield some of the last remaining uncontacted tribes in isolation from unwanted encroachment from the outside world. Acaté works to protect their forests and way of life through supporting on-the-ground initiatives that are led by the Matsés indigenous people.

Operating on the frontlines of conservation, Acaté’s initiatives over the past decade have included the first indigenous medicine encyclopedia as well as projects with original methodology in sustainable economic development, traditional medicine, medicinal agroforestry, nutritional diversity, regenerative agriculture, integrated aquaculture systems, biodiversity inventory, education, native language literacy, participatory mapping, and protection of uncontacted tribal groups in isolation. All of our initiatives are developed with, led and implemented by the Matsés indigenous people. Donations are tax-deductible and go directly to fund these on-the-ground initiatives that operate with unparalleled transparency.

All content and images copyright 2026 Acaté

Leave a Reply

  • (will not be published)

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>

Time limit is exhausted. Please reload CAPTCHA.