Using AI combined with local knowledge.

Logo da Parceira

Given the growing demand for açaí from the Amazon, it is urgent to quantify the impacts of extractive activities on biodiversity. We combined artificial intelligence (AI), network ecology, and local knowledge to measure the frugivory interactions of açaí, which are fundamental to the sustainability of the production chain and ecosystem services. The use of AI speeds up the scientific work by automatically excluding videos without animals of interest.

A combination of AI, network ecology, and local knowledge.

Methodology

Partnership in the Middle Juruá (Brazilian Amazon) involving scientists, communities, the private sector (Microsoft AI for Good Lab), the third sector (Instituto Juruá), and public agencies (Rio de Janeiro Botanical Garden, FAPERJ). We collected data using camera traps in the canopy of açaí palm trees, applying the MegaDetector V5 model integrated with time-lapse identification.

Using the MegaDetector V5 and Time-lapse for biodiversity monitoring.

Results

4 TB of data (84,614 videos, 34,000 hours) were analyzed, resulting in 4,316 videos with 500 frugivory interactions of 48 species. Extractive practices altered the species composition, with potential impacts on the sustainability of açaí. The AI (MegaDetector V5) automatically identified 44,703 of the 80,298 videos without animals, reducing human effort by half. The collected data helped train the MegaDetector V6 model, which then detected 67,054 videos without animals and reduced effort by 82%, an improvement of 32 percentage points. The project also resulted in a community mini-guide and an AI prototype for the restoration of the Atlantic Forest.

4 TB of data analyzed, 500 frugivory interactions recorded.

Supporting Material

Forest
AGRICULTURE ENERGY Industry INFRASTRUCTURE LIVESTOCK