An animal scurries through the thick underbrush of the Colombian forest during the dead of night. Hidden deep within the foliage, a camera trap snaps a picture. An AI tool known as MegaDetector identifies and classifies the image, transforming a process that once took days into mere minutes. This image is then sent to wildlife experts for analysis.
Could this picture of genus Dasyprocta, a rodent-like animal known as an agouti, help solve the ongoing deforestation crisis that has plagued the Amazon for years?
The Amazon is the world’s largest rainforest, spanning nine countries, including Colombia and Brazil. According to NASA’s Earth Observatory, it covers approximately 2.3 million square miles, or 6 million square kilometers of land. The Amazon’s significance to the planet’s overall health—absorbing carbon dioxide, regulating weather patterns, and providing a habitat for countless species—cannot be overstated.
Despite this, deforestation remains an urgent issue. Research by Amazon Conservation shows that nearly 2 million hectares (almost 5 million acres) of the Amazon were deforested in 2022, a 21 percent increase from the previous year, as land was cleared for agriculture, cattle grazing, or illegal mining operations. In Colombia alone, almost 1 million hectares of forest were lost in 2022, primarily in northwestern areas like Caqueta. If deforestation continues unchecked, it could permanently disrupt the planet’s ecosystem, according to international environmental experts.
How can such a vast and complex problem as reversing large-scale deforestation be addressed? Enter AI. Utilizing the power of data, machine learning, cloud technology, data science, and applied analytics, experts are developing innovative, collaborative programs to make recognizing deforestation patterns easier and equip policymakers with the tools to take action against deforestation.
“We need to use technology and innovation to think creatively and address the problems we face,” said Diego Ochoa, director of external affairs at the Alexander von Humboldt Institute in Bogotá, Colombia. “We have powerful tools at our disposal to drive societal change.”
Ochoa and his team at Humboldt have partnered with several other organizations in Colombia, including the CinfonIA Research Center at Universidad de los Andes, Instituto Sinchi, and Microsoft AI for Good Lab to create Project Guacamaya. This project uses top-tier AI models to monitor deforestation and protect the ecosystem’s biodiversity. The algorithms they have developed allow data to be processed in a fraction of the time compared to manual analysis.
Project Guacamaya (MACAW in English) employs a combination of new AI models for satellite analysis and adapted AI models from existing Microsoft projects for camera trap analysis and bioacoustics. The data is stored in the cloud, and the group uses the virtual machines and computational power of Microsoft Azure to design and train the models.
The multimodal AI project employs a three-pronged approach to monitor the Amazon and develop solutions, combining satellite data, hidden cameras, and acoustic records to provide aerial and terrestrial insights. This approach supports Colombian conservationists and accelerates the pace of analysis, helping all parties gain a clearer picture of the ecosystem. It also enables the development of critical reports for governmental bodies and conservation organizations, which can then take steps to mitigate deforestation.
“For instance, there is a public institution in Colombia called IDEAM responsible for creating maps that pinpoint deforestation each year,” said Pablo Arbeláez, director of the Center for Research and Formation in Artificial Intelligence at Universidad de los Andes. “They manually create these maps until they are perfect because they serve as the official statistic for deforestation in Colombia. This manual process takes months.”
“If we can produce models with 90 percent accuracy instead of 80, we can reduce a months-long effort to weeks,” Arbeláez added. “Officials can make this information available to the public much sooner. That would be a game-changer because, for example, if you plan to implement a conservation project, you need to know how much carbon you’re capturing through reforestation. This information is crucial for measuring and converting the captured carbon.”
Project Guacamaya uses high-resolution satellite images to train AI models in tracking the Amazon rainforest’s evolution over time. Image by Planet Labs and SINCHI.
The project begins with high-resolution daily satellite images provided by technology partner Planet Labs PBC. Project Guacamaya develops AI models to track these images over time, highlighting areas where illegal deforestation or mining is occurring. These satellite images give Colombian authorities and scientific organizations the ability to monitor the Amazon from space, providing a comprehensive overview. One telltale sign of impending deforestation or mining is the presence of unauthorized roads. Spotting these in satellite imagery allows authorities to be alerted to the activity.
“The capacity to analyze massive amounts of data is critical here because the Amazon spans multiple countries and millions of square kilometers,” Arbeláez said. “To have a global view of the Amazon basin, you truly need the power of AI and these new models.”
The next step focuses on biodiversity. Camera boxes have been placed throughout the Colombian Amazon. Anytime the cameras detect movement, they take a picture. A single camera can capture up to 300,000 photos, and manually reviewing each image would take years.
AI programs expedite this process. The Department of Biological Sciences at Universidad de los Andes contributed 110,000 images collected over the past four years. AI helps researchers train a model that identifies which images contain animals and which do not. Now, only 10 percent of images need manual validation, freeing up valuable time for other research. Quickly determining what images have been captured is crucial, as anything unusual could indicate ecological shifts that require attention.
“It’s like a snapshot in time,” Ochoa said. “Monitoring allows us to see how plants and animals live and thrive. But if we see the wrong species in the camera, it serves as a bioindicator. For example, if we spot a bird that typically lives in the savannah but is now in the Amazon, this is a red flag. We can notify authorities, ‘We’re detecting a species not normally found in this area. The ecosystem is changing.’”
“This technology and the data it generates enable us to inform policymakers, especially when dealing with invasive species,” Ochoa added.
The final component of Project Guacamaya is sound. Using bioacoustics, researchers capture sound from the Amazon and use an audio AI model to differentiate between bird and non-bird sounds to classify various species. Guacamaya consolidates more than 100,000 sounds into 1,000 recordings, establishing a baseline and training a model to detect and classify the different sounds. This technology enables reliable bird identification, which is significant given Colombia’s over 2,000 bird species.
“The use of this technology allows us to reduce research costs,” Ochoa said. **“Monitoring biodiversity for three months manually is something no one can afford. This technology makes it possible in a cost-efficient way.”
Protecting the Amazon rainforest is a large-scale issue that requires collaboration. Project Guacamaya involves numerous groups working together, each bringing their expertise to the table, with the goal of creating models that can be used throughout the region for the greater good.
“Our goal is to bring everyone together and work towards a common goal because each partner can play a complementary role in the project,” Arbeláez said. “To protect the world’s largest rainforest from climate change, all the countries with a stake in the Amazon need to work together. We hope regional leaders and all our Latin American neighbors will take this project as their own.”
In Brazil, where deforestation and mining in the Amazon have been most severe, a collaboration between Microsoft, environmental organization Imazon, and nonprofit Fundo Vale Foundation called PrevisIA is also addressing the problem using technology.
The PrevisIA platform utilizes Microsoft’s AI to analyze data and satellite images to forecast and monitor deforestation in Brazil’s Amazon. Imazon uses satellite images stored in the Azure cloud, where AI algorithms detect unauthorized roads and other risk factors for deforestation. The results are visualized on an interactive map, highlighting high-risk areas. This information supports decision-making to protect the rainforest.
“We use PrevisIA to anticipate risk areas and implement actions to prevent deforestation,” said Carlos Souza, a senior researcher with Imazon. “When people go into the field to work, the forest is still there, as opposed to getting a deforestation alert and finding the forest already gone. That’s the real motivation.”
As the data and results from Guacamaya and PrevisIA grow and improve, the goal is for other countries to replicate these projects to protect the Amazon.
**“All the
models we develop will be open source,”** Lavista Ferres said. “Once we finish these models, we can share them with countries like Ecuador and Peru to see how they can use them.”
In Colombia, the hope is that by demonstrating AI’s power and presenting clear data, the public will better understand the importance of protecting the Amazon for both present and future generations.
“One of our aims is to transform society,” Ochoa said. **“Colombia is a mega-diverse country, but our economy relies on raw materials. The data and information generated by this project will help inspire our society to value biodiversity.”
“This project isn’t just about monitoring or artificial intelligence or even publishing scientific papers,” Ochoa emphasized. **“We need to inspire society. Using technology, innovation, and AI, we hope to inspire Colombians to rethink their relationship with biodiversity and nature, bringing them closer to what we have in these areas.”
Project Guacamaya uses high-resolution satellite images to train AI models in tracking the Amazon rainforest’s evolution over time. Image by Planet Labs and SINCHI.
