School of Information Technology, Engineering, Mathematics and Physics

Drone-Based AI Monitoring for Climate Change & Wildlife Conservation

An Innovative USP-Led Project Enhancing Environmental Surveillance Across the Pacific - (2024 – 2025)

Drone-Based AI Monitoring for Climate Change & Wildlife Conservation

An Innovative USP-Led Project Enhancing Environmental Surveillance Across the Pacific

The University of the South Pacific is proud to present a transformative research initiative titled “Drone-Based Environmental Monitoring: Tracking Climate Change and Wildlife Conservation.” This project integrates AI-powered image recognition, drone technology, and GIS mapping to monitor and respond to environmental shifts impacting Pacific ecosystems.

🌱 The Problem

Climate change and human activity are causing widespread disruption in ecosystems—affecting biodiversity, altering wildlife behavior, and increasing deforestation and habitat loss. Traditional environmental monitoring is often costly, slow, and limited in reach, especially in remote island environments.

🛩️ The Innovation

To overcome these challenges, a team of USP researchers and collaborators have developed an AI-enhanced drone-based system to:

  • Monitor climate-sensitive ecosystems and wildlife populations
  • Detect land degradation, water level changes, and habitat disturbance
  • Deliver real-time geospatial visualizations using GIS tools
  • Support data-driven policy-making for conservation and climate resilience

This system uses AUTEL Evo II Dual 640T V3 drones, capable of capturing thermal and high-resolution visual data, which is then processed by YOLOv8 AI models trained using Python and MATLAB.

How It Works

  1. Data Collection: Drones capture imagery across coastlines, forests, wetlands, and known wildlife zones.
  2. AI Image Processing: YOLOv8 detects environmental patterns (e.g., deforestation, erosion) and wildlife (e.g., birds, mammals).
  3. GIS Mapping: Real-time mapping of changes, enabling spatial tracking of habitat transformation.
  4. Performance Evaluation: Results assessed using mAP@0.5, F1-score, ROC curves, etc.
  5. Field Validation: Drone field surveys and AI predictions are compared to manual observations for model refinement.

📈 Initial Key Outcomes

  • Model precision: 74.4%, recall: 71.2%, mAP@0.5: 79.9%, F1-score: 0.68
  • Accurate identification of forest loss, animal presence, and ecosystem stress indicators
  • Visual overlays on maps help prioritize critical conservation zones

🧪 Methodological Highlights

  • Use of pre-annotated image datasets and field-captured drone images
  • Grad-CAM and edge detection methods for improved AI interpretability
  • Custom pseudocode pipeline for environmental monitoring, adaptable to future extensions
  • Open-source training datasets under development to support regional research

This project directly contributes to the Pacific’s climate resilience and biodiversity conservation by offering a scalable, affordable, and non-invasive solution for real-time environmental surveillance. It supports:

  • National biodiversity strategies
  • Rapid disaster response
  • Educational opportunities in AI, GIS, and environmental science
  • Community engagement and data-sharing through workshops and presentations

For collaboration or information requests, contact
📧 mansour.assaf@usp.ac.fj

With a total budget of €25,500, the project spans drone acquisition, model training, stakeholder workshops, and deployment (2024 – 2025). The team aims to present its findings in international conservation conferences and publish in peer-reviewed journals.

  • Expand dataset to include thermal imagery and rare species classification
  • Improve detection accuracy under different weather/light conditions
  • Collaborate with regional conservation bodies and NGOs
  • Scale monitoring to additional Pacific Island nations