A hybrid model for significant wave height prediction based on an improved empirical wavelet transform decomposition and long-short term memory network

Abstract

Due to strong non-linearity, ocean surface gravity waves are difficult to directly and accurately predict, despite their importance for a wide range of coastal, nearshore, and offshore activities. To minimize forecast errors, a hybrid combined improved empirical wavelet transform decomposition (IEWT) and long-short term memory network (LSTM) model has been proposed. Data from National Data Buoy Center buoys deployed in the North Pacific Ocean are taken as an example to verify the models. Wave forecasts using the LSTM, EWT-LSTM, and IWET-LSTM models are compared with the observations at 6, 12, 18, 24 and 48 h forecast windows. Consequently, IEWT-LSTM is superior to EWT-LSTM or LSTM models, especially for larger waves at longer long forecast windows.