Examining the Effect of Time Lags of Meteorological Variables on a Wave Forecasting Model

Oh, J. and Suh, K.-D., 2018. Examining the Effect of Time Lags of Meteorological Variables on a Wave Forecasting Model. In: Shim, J.-S.; Chun, I., and Lim, H.S. (eds.), Proceedings from the International Coastal Symposium (ICS) 2018 (Busan, Republic of Korea). Journal of Coastal Research, Special Issue No. 85, pp. 1186–1190. Coconut Creek (Florida), ISSN 0749-0208. Since the last decade, wave forecasting using machine learning such as artificial neural networks (ANN) has been conducted. In this paper, a hybrid model is used, which was developed to forecast significant wave heights and periods by combining the empirical orthogonal function (EOF) analysis and wavelet analysis with the neural network (abbreviated as EOFWNN model). The EOF analysis enables the model to reflect the relationship between spatially distributed meteorological variables and waves to wave forecasting model and to forecast waves at multiple stations simultaneously. The wavelet analysis enables the ANN to deal with non-stationary data. The model is employed to forecast real-time waves at eight stations (Gangneung, Wangdolcho, Genkainada, Tottori, Fukui, Sakata, Aomori, Rumoi) in the East/Japan Sea for various lead times using the past observed wave height or period data and the past and future meteorological reanalysis data. In this study, to investigate the time lag effect of the meteorological data, the results of various time lags are compared. The performance of the model is evaluated by correlation coefficient, normalized root mean square error, and index of agreement. The results of the forecasted significant wave period showed high accuracy. The values of NRMSE for 24 hour lead time were between 0.038 and 0.069. The comparison of the results depending on the time lags of the meteorological variables showed slight difference in model accuracy but no significant difference in phase shift of the results.