Predicting surf zone injuries along the Delaware coast using a Bayesian network

More than 2000 surf zone injury (SZI) events, including 196 spinal injuries and 6 fatalities, were recorded at the five most populated beaches along the 25 miles of Atlantic-fronting Delaware coast from 2010 through 2017. The episodic nature of SZI indicates the importance of linking the environmental conditions and human behavior in the surf zone to predict days with high injury rates. Higher order statistics are necessary to effectively consider all associated factors related to SZI. Two unique Bayesian networks were constructed to model SZI and predict changes in injury rate (proportion of injuries to bathers) and injury likelihood (probability of at least one injury occurrence) on an hourly basis. The models incorporate environmental data collected by weather stations, wave gauges, and researcher personnel on the beach as prior (e.g., historic) information to infer relationships between provided parameters. Sensitivity analysis determined the most influential parameters related to injury rates were significant wave height, foreshore slope, and water temperature. Log-likelihood ratio scores indicate the network predicts SZI likelihood during any specified hour with more skill than prior predictions with the best performing model improving prediction 69.1% of the time (log-likelihood ratio = 69.1%). Issues persist with predicting SZI that have a log-likelihood ratio ≪ − 1 (