Abstract
The motion of waves in water causes the slow movement of drifting sea targets—a phenomenon usually ignored in target-drift prediction models for maritime search and rescue (SAR). This study examined the wave-induced drift’s influence on field-observation experiments involving two common, differently sized SAR targets—an offshore fishing vessel (OFV) and a person in the water (PIW)—using parameter stepwise calibration and machine-learning (ML) methods. The sample of wave-induced drift velocity was obtained by gradually separating current-induced (CI) drift’s and wind-induced (WI) drift’s influence from the target-drift velocity using the least-square method and AP98 model. A force analysis method and three ML methods, long short-term memory (LSTM), back-propagation (BP) neural network, and random forest (RF), were used to fit the wave-induced drift velocity by combining eight different parameter schemes. Finally, the drift trajectories considering the influence of waves were fitted and verified based on 2 independent samples respectively. Compared with the force analysis method, the accuracy of the ML methods in the verification test was higher. In addition, the results show that for OFVs, considering wave-induced drift’s influence in the ensemble-trajectory prediction could improve the simulation accuracy. However, for a PIW, no significant improvement was observed. This result also indicates that wave-induced drift may not be simply ignored in large SAR targets’ drift prediction.