Study of LSTM model in sea surface temperature prediction of the Yellow Sea cold water mass area

Marine science is an observational science and observational data is a prerequisite for understanding the ocean. Satellite remote sensing technology provides a new means for observing the important parameters of Sea Surface Temperature (SST). This work innovatively explores the Long Short-Term Memory network (LSTM), a deep learning model, for the analysis and prediction of the SST in the specific research area of the Yellow Sea cold water mass. The research results show that LSTM can effectively analyze and predict SST. The accuracy of area-averaged SST prediction depends on the length of training observation time sequence. The increase in sequence length enhances prediction accuracy. Given training data of a dozen years, the accuracy of a 30-day length prediction can reach up to 95.87%, which meets the requirements of ocean forecasting. The study also revealed that specifying an observation length, the area-averaged prediction accuracy of a 30-day length prediction for different locations of the cold water mass make no difference.