Operational nighttime sea fog monitoring for maritime navigation safety: Decadal hotspot dynamics using an explainable geostationary satellite-based analysis

Sea fog significantly impedes maritime operations in critical shipping corridors like the Taiwan Strait, where in-situ observations are sparse. To address the need for operational risk management, this study establishes a nighttime monitoring framework utilizing decadal Himawari-8/9 geostationary infrared observations from 2016 to 2025 during the peak-traffic season. We developed an interpretable machine learning model integrating spectral and textural indicators to distinguish fog from complex cloud covers, achieving a probability of detection (POD) of 81.66% and a true skill statistic (TSS) of 0.74 validated against vertical Lidar profiles. Unlike traditional black-box approaches, SHapley Additive exPlanations (SHAP) are employed to quantify feature contributions, revealing how thermal sensitivity in specific infrared bands facilitates the distinction between fog and low clouds. Spatial analysis identifies distinct risk regimes essential for Marine Spatial Planning (MSP): the northern sector maintains a persistently low-risk cold-spot regime, whereas the central strait and Penghu archipelago exhibit dynamic, interannually variable hotspot clusters. These findings challenge static monitoring schemes by characterizing the spatio-temporal evolution of fog risks. By offering a baseline for optimizing offshore transportation schedules and enhancing early warning systems, this framework provides actionable intelligence for maritime authorities to support safer and more resilient maritime logistics in high-traffic regions.