Forecasting Tropical Instability Waves Based on Artificial Intelligence

Abstract

AbstractThe trend of quickly increasing volumes of satellite remote sensing big data and the successful application of deep learning (DL) technology to many research fields inspire us to develop a deep neural network-based DL ocean forecasting model that is driven only by the time series of gridded sea surface temperature (SST) data. The model forecasted the SST pattern variations in the eastern equatorial Pacific Ocean, where a well-known prevailing oceanic phenomenon, tropical instability waves (TIWs) characterized by cusp-shaped waves, propagates westward between 5 \(^\circ \)S and 5 \(^\circ \)N. The model was trained and tested in two non-overlapping periods of four (2006–2009) and nine years (2010–2019). The model can make an 18 km \(\times \) 18 km gridded five-day SST forecast with a root mean square error of 0.29 \(^\circ \)C. The model also successfully captures the spatial-temporal propagation characteristics and the seasonal cycle and interannual variability of the TIW modulated by the El Niño-Southern oscillation. Thus, the data-driven DL technology could be a promising way to forecast complicated oceanic phenomena.