AUTHOR=Ning Pengfei , Zhang Cuicui , Zhang Xuefeng , Jiang Xiaoyi 

TITLE=Short- to Medium-Term Sea Surface Height Prediction in the Bohai Sea Using an Optimized Simple Recurrent Unit Deep Network

JOURNAL=Frontiers in Marine Science

VOLUME=Volume 8 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/marine-science/articles/10.3389/fmars.2021.672280

DOI=10.3389/fmars.2021.672280

ISSN=2296-7745

ABSTRACT=Global warming has intensified the sea level rise problem and caused severe coastal ecological disasters in shallow waters, e.g. the Bohai Sea. The prediction of sea surface height (SSH) is of much significance to monitor the sea level changes. However, the instability of SSH due to complex physical dynamical phenomena challenges existing methods in precise SSH prediction. In this paper, we develop an optimized Simple Recurrent Unit (SRU) deep network for the short-to medium-term SSH prediction using AVISO data. Thanking to the parallel structure of SRU, the computational complexity of deep network can be reduced to a large extent and makes the short-to-medium SSH prediction more efficient.To avoid over-fitting and gradient-disappearing, we utilize a skip connection strategy for model optimization, which significantly improves the prediction accuracy. Finally, the experiments are carried out in the Bohai Sea to evaluate the proposed model. Extensive experiments demonstrate that our framework (i) significantly outperforms the existing deep learning methods including BP, RNN, LSTM, GRU on the challenging 1 day, 5 days, 20 days, and 300 days SSH prediction, (ii) can predict the short-term SSH trend (in the next one or two days) real time, (iii) achieves medium-term SSH prediction in the next 5-20 days fast, and (iv) shows great potential for applications in medium-to long-term SSH forecasts. To our knowledge, this is the first work, which investigates SRU deep model for short-to medium-term SSH prediction.