Aquascan: Graph-Based Learning for Distributed Marine Sensing

Marine monitoring faces unprecedented challenges as climate change and human activity reshape the oceans. Traditional tracking methods struggle with the scale and complexity of modern maritime sensing needs.

We explore the application of Graph Neural Networks (GNNs) for predicting marine life trajectories through a comparative study against Kalman filters in distributed sensor networks. To support this investigation, we introduce the Aquascan framework, which uses low-cost drifting sensors to collect distributed marine data in noisy, partially observable conditions.

Our experiments across multiple prediction horizons show GNNs significantly outperform constant-velocity Kalman filters: GNNs maintain over 95% Area Under the ROC Curve (AUC) across all horizons while Kalman filters degrade from 97% to 69% AUC. The performance gap widens under challenging conditions. At 5 km sensor spacing, GNNs achieve 92.8% AUC versus 66.9% for Kalman filters. GNNs' superior performance appears to stem from leveraging network topology and reasoning about non-detections to infer entity presence in coverage gaps.

Our results suggest that graph-based approaches offer substantial advantages over traditional independent tracking methods in distributed marine monitoring.

Keywords: Graph Neural Networks · Trajectory Prediction · Marine Monitoring · Heterogeneous Graphs