Robust Prediction of Sea Surface Temperature Based on SSPGAN

Abstract

Sea surface temperature (SST) is an important parameter for monitoring ocean phenomena. Driven by ocean satellite Big Data, deep neural networks have achieved state-of-the-art performance in forecasting fields of oceanic phenomena. However, when there are some changes in the data set, the performance of these models drops significantly. In order to address this issue, this article develops a robust strong stability-preserving generative adversarial network model (SSPGAN) to forecast SST. The main objective of this article is to continue to produce accurate SST predictions even in the presence of minor data perturbations. The SSPGAN consists of a generator of a deep learning model and a strong stability preserving discriminator. The generator aims to generate SST that are close to the future true SST using SST from the past. The discriminator simultaneously tries to tell the generated SST apart from the real SST. The robust of the discriminator and the overall model is enhanced by the strong stability preserving network. Experiments confirm the SSPGAN's accuracy and robust for predicting SST in the presence of small data perturbations. In this study, we employ the Wasserstein metric to quantify the gap between the actual SST distribution and the generated SST with a perturbation. In addition, this article also analyzes the robust conditions of the deep neural network when the data are disturbed.

Keywords

• Generative adversarial network
• Lipschtiz
• prediction
• robust
• sea surface temperature (SST)