TideNet: A unified time-aware framework for sparse maritime trajectory recovery

Maritime trajectory recovery plays a critical role in ensuring navigational safety, optimizing logistics, and improving environmental monitoring in the maritime industry. However, maritime trajectory recovery faces unique challenges that fundamentally differ from urban mobility scenarios. Unlike vehicles constrained by road networks, vessels navigate open seas without fixed pathways while being influenced by dynamic environmental factors like tides and currents. These challenges are compounded by severe data sparsity caused by satellite transmission failures, signal occlusion, and irregular reporting intervals. Existing trajectory recovery methods, primarily designed for structured urban environments, fail to preserve time-constraint movement patterns and struggle to generalize across intricate sparsity patterns for maritime data, particularly under high data sparsity. In this paper, we propose TideNet, a unified framework that integrates a fine-grained time-dependent transition pattern with a multi-step transition mechanism for complex sparsity pattern capturing. The proposed model learns long-term movement probabilities while preserving time-sensitive transition dynamics. Extensive experiments on two real-world datasets demonstrate that TideNet significantly outperforms state-of-the-art baselines, achieving an average of 9.1% lower recovery error than existing methods.