Uncertainty in marine weather routing

Abstract

Weather routing methods are essential for planning routes for commercial shipping and recreational craft. This paper provides a methodology for quantifying the significance of numerical error and performance model uncertainty on the predictions returned from a weather routing algorithm. The numerical error of the routing algorithm is estimated by solving the optimum path over different discretizations of the environment. The uncertainty associated with the performance model is linearly varied in order to quantify its significance. The methodology is applied to a sailing craft routing problem: the prediction of the voyaging time for an ethnographic voyaging canoe across long distance voyages in Polynesia. We find that the average numerical error is $0.396\%$, corresponding to $1.05$ hours for an average voyage length of $266.40$ hours. An uncertainty level of $2.5 \%$ in the performance model is seen to correspond to a standard deviation of $\pm 2.41-3.08\%$ of the voyaging time. These results illustrate the significance of considering the influence of numerical error and performance uncertainty when performing a weather routing study.