Prediction of Sea Surface Temperature off the Southern Korean Coast using Spatial-Temporal Neural Network

Youngjin CHOI1, Young-Min Park1, Young-Kwang Ju1, Seok Jae Kwon2, Young-Taeg Kim2, Heung-Bae Choi1 and Gwang-Ho Seo2, (1)GeoSystem Research, Corp., Gunpo, South Korea, (2)Korea Hydrographic and Oceanographic Agency, Busan, South Korea
Fast and accurate prediction of sea surface temperature(SST) is very important in operational oceanography. A novel hybrid deep neural network(DNN) architecture was used to predict the spatial-temporal changes of SST directly from numerical simulation results. Since the hybrid deep neural network is a mixture of Convolutional Neural Network(CNN) and Long Short Term Memory(LSTM), it can adequately recognize spatial informations and time-series patterns as well. The KOOFS(Korea Ocean Observing and Forecasting System), an operational ocean forecast system of KHOA(Korea Hydrographic and Oceanographic Agency), is operated to establish the training data sets of DNN. The proposed forecast method is proven to achieve reasonable prediction performance compared with numerical simulations. The result indicates that the DNN framework can be applied to the spatial-temporal prediction based on any data sets, including satellite imageries.