Purely satellite data-driven deep learning forecast of complicated tropical instability waves

Area of Science:

• Oceanography
• Artificial Intelligence
• Remote Sensing

Background:

• Oceanic phenomena forecasting traditionally relies on complex physical numerical models.
• These models require consideration of numerous natural processes for accurate predictions.
• Time-series observational data inherently contains the rules governing these processes.

Purpose of the Study:

• To develop a novel deep learning model for forecasting sea surface temperature (SST).
• To utilize satellite remote sensing data for a purely data-driven approach.
• To specifically forecast SST evolution related to tropical instability waves.

Main Methods:

• A deep learning model was developed using solely satellite remote sensing data.
• The model was trained and tested on a 9-year period (2010-2019).
• The model focuses on forecasting sea surface temperature evolution.

Main Results:

• The developed deep learning model accurately forecasts the sea surface temperature field.
• The model demonstrated high efficiency in its predictions.
• Forecasts were validated over a 9-year testing period.

Conclusions:

• Satellite data-driven deep learning models show significant potential for oceanic forecasting.
• This approach offers a viable alternative to traditional physics-based numerical models.
• The study highlights the effectiveness for phenomena like tropical instability waves.