Research Project Overview

Our research, published in Marine Mammal Science, leverages computer vision, machine learning techniques, and high-resolution satellite imagery to classify species of Antarctic pack-ice seals from space.

Key Findings

1. High Classification Accuracy: Our Random Forest model achieved an accuracy of 97.4% for crabeater seals and 98.0% for Weddell seals, demonstrating the efficacy of machine learning in wildlife monitoring.

2. Spatial and Habitat Metrics: By integrating spatial arrangement and habitat characteristics, our model improves species identification, even in areas where multiple species coexist.

3. Cost-Effective Monitoring: The use of high-resolution satellite imagery offers a non-invasive, cost-effective method for large-scale wildlife surveys, making it feasible to conduct extensive monitoring of seal populations across Antarctica.

4. Conservation Implications: This methodology provides essential data for understanding seal populations and their responses to environmental changes, aiding in the development of effective conservation strategies.

Methods

We employed a combination of very high-resolution satellite and aerial imagery, along with a Random Forest model incorporating various spatial and habitat metrics. Additionally, we utilized computer vision convolutional neural networks to enhance the detection and classification process. This semiautomated approach allows for detailed species classification based on seal group patterns and environmental features.

Implications

This work represents a significant advancement in pan-Antarctic pack-ice seal monitoring, offering a scalable and efficient solution for tracking population dynamics. This approach is crucial for ongoing conservation efforts, especially in the face of climate change and human activities such as krill fishing.

For more detailed information, please refer to the full paper here.