Sea Surface Temperature Nowcasting with 3-channel Convolutional LSTM

Accurate prediction of sea surface temperature (SST) is of great importance for ocean-related industries such as fisheries. In this paper, we propose a generative adversarial model for SST nowcasting. Our model combines a modified Convolutional Long Short-Term Memory model (convL-STM) as a generator, and a multi-layer Convolutional Neural Network (CNN)as a discriminator. The model utilizes spatial correlation as well as temporal correlation within the previous sea states, including not only sea surface temperature but also ocean current. Experiments show that our model is capable of generating consistent and accurate regional SST predictions.