WhaleGuard: A Neural Network-Based Movement Prediction Platform for North Atlantic Right Whales

WhaleGuard is an intelligent platform developed with support from the RBC Borealis Let's SOLVE It Undergraduate Mentorship Program. It aims to predict the 72-hour future movement trajectories of North Atlantic right whales using deep learning technology, to proactively protect this endangered species with fewer than 350 individuals remaining. The core of the project is to address the problem of delayed response in traditional passive monitoring, provide decision support for maritime management and fisheries, and promote the application of AI technology in marine conservation.

The North Atlantic right whale has fewer than 350 individuals remaining and is one of the most endangered large whale species. It faces multiple threats including ship collisions, fishing gear entanglement, marine noise pollution, and changes in prey distribution caused by climate change. Traditional conservation measures rely on passive monitoring; by the time whales are detected, they may already be in danger, lacking early warning capabilities and making it difficult to effectively prevent risks.

The WhaleGuard project is developed by the student team Neural Network Navigators, focusing on providing 72-hour whale movement predictions. This time window is sufficient to issue navigation warnings and adjust fishing plans in advance while ensuring the practical value of the predictions. The project requires integrating knowledge from marine biology, physical oceanography, and deep learning to transform complex ecological data into actionable conservation insights.

Predicting whale movement requires integrating three types of data: 1. Historical observation data (visual surveys, passive acoustic monitoring, satellite tag tracking) as the basis for model training; 2. Environmental data (sea water temperature, salinity, chlorophyll concentration, ocean current speed), since right whales move following prey distribution; 3. Human activity data (shipping routes, fishing areas, marine construction) for assessing potential risks.

Whale movement prediction is a spatio-temporal sequence problem; the model needs to learn historical patterns and combine current environmental data to infer future distribution. Possible methods include: RNN/LSTM/GRU to capture temporal dependencies, CNN to process spatial environmental data, spatio-temporal graph neural networks to model individual interactions; introducing attention mechanisms can identify key environmental factors and improve model interpretability.

WhaleGuard's output targets stakeholders such as maritime managers and fisheries operators. It displays the probability distribution of whale presence using heatmaps (overlaid with shipping/fishing area boundaries), provides a risk assessment matrix, and recommends temporary speed limit zones, fishing gear adjustments, or alternative routes. The system adopts a human-machine collaboration model: the algorithm provides probability estimates, and human decision-makers make final decisions by integrating multiple factors.

As an undergraduate project, WhaleGuard allows students to experience the complete machine learning project cycle and cultivate a sense of mission for using technology to serve social welfare. The project open-sources its code and documentation to promote knowledge sharing. In the future, it can be extended to the conservation of other marine species, representing the AI for Good direction and helping to address pressing challenges such as biodiversity loss.