Author(s): Jihee Oh, Kyung Duck Suh
Linked Author(s): Jihee Oh
Keywords: Forecasting, significant wave height, EOF, wavelet analysis, artificial neural networks
Abstract: Many studies have been established for the prediction of waves using hybridization of artificial neural networks (ANN) and other techniques to provide effective modeling. In this study, a hybrid empirical orthogonal function analysis (EOF) -wavelet transform-ANN (EOFWLNN) model is introduced and employed to forecast significant wave height simultaneously at 8 locations (Gangneung, Wangdolcho, Genkainada, Tottori, Fukui, Sakata, Aomori, Rumoi) in the East Sea of Korea using 30-min interval wave height observed data and 6 hourly meteorological data. Experiments have been conducted from October 2010 to February 2011 with training and test being conducted for 120 day period and 3 day period, respectively. The EOFWLNN model with 3 decomposition level is employed to make forecast for various lead times. The lead times are fixed as 1, 3, 12 and 24 hours. Model performance is evaluated using R, RMSE and Index of agreement. Most forecasting results show satisfactory performance except Aomori. R and I_a in Aomori are lower than other stations, but for the values of RMSE, it seems to be quite satisfactory. Although the model efficiency decreases as lead-time increases, the model results show good predictions
Year: 2017