AI tools can be used in various ways to monitor plant and animal biodiversity, track effects of policies such as those that create wildlife corridors, and assess adverse events, such as pollution and deforestation. The United Nations’ Global Ocean Observing System has used robotic technologies to monitor ocean conditions and support the management of sustainable fisheries. Organizations monitor marine biodiversity and assess the impacts of climate change on ocean health by collecting relevant data such as temperature, salinity and pollution levels. On land, remote sensing robots are put to diverse uses, such as monitoring deforestation, tracking wildlife populations and assessing the health of ecosystems. For instance, the organization Amazon Conservation has employed drones and satellite-based robots to track illegal logging activities and monitor deforestation in the southern Peruvian Amazon. In Africa, TrailGuardAI is using ground robots and camera traps to monitor wildlife poaching in protected areas.
AI for Good in focus: Machine learning supporting ecology
ML and conservation ecology is revolutionizing wildlife monitoring and ecological research. On 21 November 2022, the AI for Good platform hosted Dr. Devis Tuia, Head of the Environmental Computational Science and Earth Observation Lab in Sion. The webinar addressed the challenge of harnessing vast amounts of unstructured digital data in wildlife monitoring, demonstrating how AI-powered drone imaging systems and camera traps can automate animal detection and population estimation across expansive savannah landscapes with up to 90 per cent recall, a significant improvement over manual methods.
In order to address the challenge of detecting small animals in aerial imagery (averaging just 30.18 pixels) and navigating complex backgrounds, the webinar demonstrated the use of crowdsourced annotation platforms, which processed 26,000 images in a matter of days, as well as curriculum-based deep learning models pretrained on ImageNet. These models reduced false positives by 95 per cent, allowing rangers to analyse 300 photos rather than 1,500 during population surveys.
The webinar also illustrated how open-source tools such as AIDE (AI Interface for Ecological Data) can combine citizen science with ML, therefore allowing ecologists to create custom detection models without requiring programming skills.The webinar additionally highlighted how platforms such as Wildbook and iNaturalist can harness citizen science and social media data for ecological monitoring. This democratization of AI tools enables conservation practitioners to independently apply advanced ML methods, encouraging a critical synergy between ecological expertise and technical innovation. The webinar featured case studies on species identification in Namibia and migratory bird monitoring in West Africa, where AI cut analysis times from weeks to hours. It also introduced new and emerging methodologies such as taxonomy-aware models and spatio-temporal species distribution modelling to improve the precision of ecological decision-making. The webinar presented a scalable, replicable framework for global conservation efforts, tackling critical issues in real-time biodiversity monitoring and threat detection, such as poaching.
