Wave forecast information is fundamental for safety operation of working vessels for port construction. There are two different kinds of methods for wave forecasting. One is a numerical model of wind and wave interation. The other is an empirical model based on a statistical relationship between the weather and the wave data obtained in the past. The former method has often been used for wave hindcasting. The reliability of the models has been also discusseed to some extent. practical computation with these models, however, requires a special knowledge of both atmospheric and wave systems and a large investment in the computation. The latter method utilizes simple regression equations or relationships and does not need sophisticated knowledge in the process of practical computation. Because of the
advantage of the latter method, several statistical models have been proposed so far. The most popular conventional models for the latter method are the multiple regression model, multi-variable autoregressive model and a model with the use of contingency table.
　The reliability of these statistical models have been also examined to have found unfavorable behavior of the conventional statistical models.
　In this report, a new statistical wave forecasting model through Kalman filtering combined with principal component analysis is proposed. The reliability of the model is examined through simulation based on 5-year wave data and weather maps. The examination shows that the reliability of short-term wave forecasting is satisfactory enough for practical use.