Large-Scale Coastal Marine Wildlife Monitoring with Aerial Imagery

Area of Science:

• Marine biology
• Wildlife conservation
• Deep learning applications

Background:

• Effective monitoring of coastal marine wildlife is vital for biodiversity conservation and environmental management.
• Traditional in situ methods for surveying extensive marine habitats are often logistically challenging.
• Advanced technologies are needed to improve the efficiency and accuracy of wildlife monitoring.

Purpose of the Study:

• To investigate the use of aerial imagery and deep learning for automated detection, classification, and enumeration of marine-coastal species.
• To develop and train a deep learning model for accurate population metrics of marine mammals.
• To assess the impact of emergent threats on marine wildlife populations.

Main Methods:

• A dataset of high-resolution aerial images of southern elephant seals (Mirounga leonina) and South American sea lions (Otaria flavescens) was curated and annotated.
• A deep learning framework was developed and trained on this dataset for automated animal identification and classification.
• The model's performance was evaluated using F1 scores, achieving between 0.7 and 0.9.

Main Results:

• The developed deep learning model successfully automated the detection and classification of southern elephant seals and South American sea lions.
• The model achieved high F1 scores, indicating reliable performance in population enumeration.
• This methodology provided crucial data on the impact of the H5N1 avian influenza outbreak on these species in 2023.

Conclusions:

• Aerial imagery combined with deep learning offers an efficient and accurate method for monitoring coastal marine wildlife populations.
• Automated population metrics derived from this approach support ecological dynamics analysis and conservation efforts.
• This technology is valuable for understanding and responding to emergent threats affecting marine ecosystems.