1999 年 100 巻 p. 221-227
This paper concerns with statistical forecasting of the sea surface movement in winter, which generates in Uchiura (Funka) Bay, Hokkaido. Traditionally, the above movement is often ragarded to be stationary and then an autoregressive model is used to construct the predictor. However, our statistical researches using measured data show the possibility that the movement does not always keep stationarity. From the above reason, in this paper, we consider geographical characteristics around this bay and then propose a predictor for forecasting sea surface movements by taking account of the possibility of nonstationarity. To discuss the validity of our predictor coupled by a locally stationary autoregressive model and Holt-Winters' method, forecasting accuracies are compared numerically among the predictors constructed by an autoregressive model, a locally stationary autoregressive model, a time varying coefficients autoregressive model and Holt-Winters' method. The results showed that our predictor gave the best forecasts among the above predictors.