Spatial Forecasting of Significant Wave Height from Terrestrial Seismic Noise in the Sicily Channel

Terrestrial microseisms provide a continuous record of ocean-wave forcing and have increasingly been used to reconstruct contemporaneous sea-state conditions. Here, we investigate whether their recent temporal evolution also contains predictive information on future significant wave height, H s . We analyse broadband seismic observations from western Sicily and the Egadi Islands together with Copernicus Marine Service wave fields and independent buoy measurements. Spatial correlation analysis shows that seismic amplitudes are most strongly associated with H s in the upper secondary-microseism band, with station-dependent sensitivity patterns across the Sicily Channel. We then develop a causal multi-station deep-learning framework that uses the preceding 48 h of three-component, multi-frequency observations from five permanent seismic stations to reconstruct and forecast spatial H s fields at lead times of 0, 1, 3, 6, 12, and 24 h. In nowcasting mode, predictions correlate strongly with independent buoy observations at Mazara del Vallo and Palermo, with Spearman coefficients of 0.91 and 0.87, respectively, while reproducing the dominant spatial structure of the reference wave field. Forecast median domain-wide RMSE increases from approximately 0.22 m at 1 h to 0.31 m at 24 h. Crucially, dedicated forecast models remain more skilful than persistence of the contemporaneous seismic nowcast throughout the 1–24 h interval, with an improvement of approximately 15% at 24 h. These results show that sparse terrestrial seismic networks contain exploitable information on the subsequent evolution of regional wave fields and could provide a complementary observational constraint for short-term marine forecasting.