Machine-Learning-Based Microseismic Monitoring of Nearshore Wind–Wave Dynamics: Observations from Hainan Island, China

Amid global warming, coastal regions face escalating risks from extreme meteorological events, yet conventional oceanographic monitoring struggles to deliver sufficient resolution under harsh conditions. This study exploits short-period seismometers for passive nearshore wind–wave monitoring via microseismic noise generated by air–sea–solid Earth interactions. A six-station microarray deployed along Hainan Island, China, from 14 September to 15 October 2024, captured signals across three distinct regimes: Tropical Storm Soulik, Quiescent Period, and a sustained strong-wind period. Power spectral density, time–frequency, and polarization analyses revealed regime-dependent microseismic responses—wind–wave dominance during Soulik, tidal modulation in calm intervals, and combined influences during moderate winds. Strong correlations emerged between spectral energy and both wave height and wind speed. A random forest model fusing spectral and tidal features predicted significant wave height (mean absolute error, MAE = 0.117 m) and wind speed (MAE = 0.861 m/s) on independent test data, outperforming traditional empirical spectral models (MAE = 0.133 m for wave height). Secondary microseisms (0.2–0.8 Hz) proved highly sensitive to wind–wave dynamics, whereas long-period signals were modulated by coastal reflections and tides. This cost-effective approach provides a complementary framework to conventional methods, enhancing real-time marine hazard monitoring and studies of Earth system coupling.