The Colombian Amazon covers roughly 400,000 square kilometers of some of the most biodiverse land on Earth. Monitoring it is an enormous challenge. Until recently, government agencies and conservation groups relied on monthly satellite composites to identify deforestation. By the time an alert was issued, loggers and miners had often cleared the land and moved on.
Project Guacamaya, named after the iconic macaw of the region, changed that timeline from months to days. Built through a collaboration between Microsoft's AI for Earth program and researchers at Colombian universities, the system uses deep learning models to analyze satellite imagery and generate daily deforestation alerts. It also accelerates species identification by a factor of ten, helping conservation biologists catalog the biodiversity that is disappearing.
The Problem with Monthly Alerts
Brazil's PRODES and DETER systems have set the standard for satellite-based deforestation monitoring, but their coverage is focused on the Brazilian Amazon. Colombia, Peru, Ecuador, and Bolivia have their own monitoring needs, and until Project Guacamaya, the tools available were slower and less granular.
In Colombia specifically, the Institute of Hydrology, Meteorology and Environmental Studies (IDEAM) produced deforestation reports on a quarterly or monthly basis. This reporting cadence was adequate for understanding trends but useless for law enforcement. Illegal deforestation in the Colombian Amazon is often driven by cattle ranching, coca cultivation, and illegal gold mining. These operations are mobile and fast moving. A clearing can expand by hundreds of hectares in a few weeks.
How the AI Works
Project Guacamaya processes imagery from multiple satellite sources, including Sentinel-2 (from the European Space Agency) and Landsat (from NASA and USGS). The system uses convolutional neural networks trained on labeled examples of deforestation, degradation, and intact forest. Cloud cover, which plagues optical satellite imagery in tropical regions, is handled through multi-temporal compositing and synthetic aperture radar data that can see through clouds.
When the model detects a new clearing or a significant change in forest cover, it generates an alert with geographic coordinates, an estimated area, and a confidence score. These alerts are delivered to environmental authorities, park rangers, and indigenous community monitors. The shift from monthly to daily alerting means enforcement teams can reach a deforestation site while the activity is still underway, rather than arriving to find a fait accompli.
Counting Species Ten Times Faster
Deforestation monitoring is only half of the project. Guacamaya also applies AI to biodiversity assessment. Traditionally, species identification from camera trap images and field recordings is done by expert biologists who manually review thousands of photographs and audio clips. This process is slow and limited by the availability of trained taxonomists.
The project's classification models can identify bird species from audio recordings and mammals from camera trap images with accuracy approaching that of human experts. The result is a tenfold increase in the speed of species inventory work. In pilot deployments, researchers cataloged more species in weeks than previous surveys had found in months.
This matters for conservation planning. Understanding which species are present in a given area, and how their populations are changing, is fundamental to setting conservation priorities and designing protected areas. The faster this data is available, the faster governments and NGOs can act.
Open Source for the Whole Amazon
All of the tools developed under Project Guacamaya are open source, published under permissive licenses on GitHub. This was a deliberate choice. The Amazon spans nine countries, and each has different institutional capacity and technical resources. By making the code, models, and training data publicly available, the project enables adaptation and replication across the basin.
Microsoft provided cloud computing credits through its AI for Earth program, allowing researchers to train and deploy models on Azure without building local infrastructure. The program has given more than $75 million in grants and cloud resources to environmental projects worldwide since 2017.
The Stakes
The Amazon is approaching what scientists call a tipping point. Research published in Nature in 2023 estimated that 10 to 47 percent of the Amazon could transition from forest to degraded savanna by 2050 under current trends. This transition would release billions of tons of stored carbon, accelerate global warming, and destroy ecosystems that support roughly 10% of all known species on Earth.
Technology alone cannot prevent this outcome. Deforestation is driven by economic incentives, weak governance, and complex land tenure systems. But technology can shift the information balance. When an illegal clearing is detected in hours instead of months, when species inventories are completed in weeks instead of years, the window for effective action opens wider.
Project Guacamaya is one piece of that puzzle. A daily satellite alert, processed by AI, sent to a ranger on a phone in the middle of the forest. It is not glamorous work, but it is exactly the kind of work that will determine whether the Amazon survives the coming decades.
Sources: Microsoft AI for Earth, Project Guacamaya case study; IDEAM Colombia deforestation monitoring reports; European Space Agency Sentinel-2 mission documentation; Flores et al., "Critical transitions in the Amazon forest system," Nature (2023); Microsoft AI for Earth program overview.
