A Reliable Offshore Sensing Framework for Social Ocean Systems Based on Self-Supervised Progressive Masked Denoising

Distributed acoustic sensing (DAS) serves as an embedded physical-layer sensing modality within critical coastal infrastructure, providing high-resolution spatiotemporal observations of marine geohazards (like marine seismic) to support the modeling, monitoring, and resilience assessment of computational social systems under oceanic disturbances. However, strong background noise from marine gravity waves, biological activity, and shipping traffic often overwhelms low-level seismic signals within DAS data. This article proposes a self-supervised denoising model, termed progressive masked autoencoder denoising network (PgMAD), which exploits the inherent spatiotemporal correlations among multichannel seismic recordings to perform high noise suppression without any annotated clean data. Specifically, it utilizes an easy-to-difficult reconstruction strategy that progressively uncovers the underlying signal features. This approach alleviates the computational burden typically imposed by fixed masking schemes. Real-world experiments on the DAS recordings of Shantou earthquakes reveal that PgMAD delivers a competitive boost in signal-to-noise ratio at an affordable training cost. DAS signals Denoising is essential not only for data quality but also for unlocking their utility in computational social systems, particularly in building resilient applications such as disaster early warning and critical infrastructure monitoring.