Deep Learning Model for Sea Surface Salinity Forecast in the Tropical
Pacific Ocean during ENSO Events
Sea surface salinity (SSS) in the eastern tropical Pacific Ocean
significantly influences the process of sea-air interactions and
exhibits a strong response during the analysis of the El Niño-Southern
Oscillation (ENSO). Recently, satellites have provided long-term SSS
data, and deep learning methods can achieve spatial-temporal forecasts.
We developed a satellite-data-driven deep neural network (DNN) model to
achieve reasonable forecasts of SSS fields associated with the ENSO
using a series of past satellite SSS data. Our model achieved short- to
medium-term forecasts for SSS fields from 6 to 96 days, with an error of
less than 0.2 pss. Consistent with the Climate Change Initiative (CCI)
SSS Anomaly (SSSA), the SSSA appears approximately 4 months earlier than
the filtered Sea Surface Temperature Anomaly (SSTA) during ENSO events.
Moreover, the SSSA index forecasted by the DNN also showed strong
negative relationship with the Niño3.4 SST index during ENSO events.