A spatiotemporal deep learning model for sea surface temperature field prediction using time-series satellite data. (October 2019)

• Record Type: Journal Article
• Authors: Xiao, Changjiang; Chen, Nengcheng; Hu, Chuli; Wang, Ke; Xu, Zewei; Cai, Yaping; Xu, Lei; Chen, Zeqiang; Gong, Jianya
• Is Part Of: Environmental modelling & software. Volume 120(2019)
• Publication Date: 2019-10
• Subjects: Big data modelling -- Spatiotemporal deep learning -- ConvLSTM -- Sea surface temperature (SST) -- Field prediction -- Time-series satellite data; Environmental monitoring -- Computer programs -- Periodicals; Ecology -- Computer simulation -- Periodicals; Digital computer simulation -- Periodicals; Computer software -- Periodicals; Environmental Monitoring -- Periodicals; Computer Simulation -- Periodicals; Environnement -- Surveillance -- Logiciels -- Périodiques; Écologie -- Simulation, Méthodes de -- Périodiques; Simulation par ordinateur -- Périodiques; Logiciels -- Périodiques; Computer software; Digital computer simulation; Ecology -- Computer simulation; Environmental monitoring -- Computer programs; Periodicals; Electronic journals; 363.70015118
• DOI: 10.1016/j.envsoft.2019.104502
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: Sea surface temperature (SST) is a vitally important parameter of the global ocean, which can profoundly affect the climate and marine ecosystems. To achieve an accurate and holistic prediction of the short and mid-term SST field, a spatiotemporal deep learning model is proposed which can capture the correlations of SST across both space and time. The model uses the convolutional long short-term memory (ConvLSTM) as the building block and is trained in an end-to-end manner. Experiments using 36-year satellite-derived SST data in a subarea of the East China Sea demonstrate that the proposed model outperforms the persistence model, the linear support vector regression (SVR) model, and two LSTM models with different settings, when judged using multiple statistics and from different perspectives. The results suggest that the proposed model is highly promising for short and mid-term daily SST field prediction accurately and conveniently. Highlights: A spatiotemporal deep learning model is proposed to predict SST field. The model can capture spatiotemporal correlations of time-varing SST fields. 36-year satellite-derived time series daily SST data are used. The model outperforms persistence, SVR, and LSTM models on multiple statistics.