Abstract
Significant wave height forecasting is a key problem in ocean data analytics.This problem is relevant in several maritime operations, such as managing thepassage of vessels or estimating the energy production from waves. In thiswork, we focus on the prediction of extreme values of significant wave heightthat can cause coastal disasters. This task is framed as an exceedanceprobability forecasting problem. Accordingly, we aim to estimate theprobability that the significant wave height will exceed a predefined criticalthreshold. This problem is usually solved using a probabilistic binaryclassification model. Instead, we propose a novel approach based on aforecasting model. A probabilistic binary forecast streamlines information fordecision-making, and point forecasts can provide additional insights into thedata dynamics. The proposed method works by converting point forecasts intoexceedance probability estimates using the cumulative distribution function. Wecarried out experiments using data from a buoy placed on the coast of Halifax,Canada. The results suggest that the proposed methodology is better thanstate-of-the-art approaches for exceedance probability forecasting.