Integration of Multi-Modal Field Data for Statistical Analysis of Wave Overtopping

Conventionally, overtopping prediction at coastal defence structures is based on empirical methods that yield biased estimates since their database is constructed for specific wave and structures characteristics, which limits their application. Knowing that overtopping is a highly non-linear phenomenon, strongly affected by local wave characteristics and bathymetry, this study proposes localized statistical analysis of overtopping using multi-modal data. The Yeondaepu breakwater in Jeju is selected as a study area given its high vulnerability to wave overtopping. Field data were obtained from a KMA operated Gueom buoy and locally installed wave gauges, complemented by CCTV-based video monitoring. To address the inherent uncertainty in overtopping predictions, the field data is integrated and analyzed to establish quantitative relationships between significant wave height, wave direction, time period and overtopping characteristics. Using synchronized measurements and vision-based detections, we performed correlation analyses to identify statistically significant drivers of overtopping and to determine threshold conditions under which overtopping initiates. The collected multi-modal data is used to train a YOLO-based detection model for identifying overtopping events. This integrated observational and statistical framework supports improved site-specific overtopping assessment under real environmental forcing.