Data-driven global ocean modeling for seasonal to decadal prediction

Area of Science:

• Oceanography
• Climate Science
• Artificial Intelligence

Background:

• Accurate ocean dynamics modeling is vital for understanding climate variability and change.
• Traditional numerical models face challenges in multiyear global ocean prediction.
• Predicting extreme events like El Niño-Southern Oscillation and heat waves remains difficult.

Purpose of the Study:

• To introduce ORCA-DL, a data-driven 3D ocean model for seasonal to decadal global ocean dynamics prediction.
• To evaluate ORCA-DL's accuracy, physical consistency, and performance against state-of-the-art numerical models.
• To assess ORCA-DL's capability for skillful decadal predictions and climate projections.

Main Methods:

• Development of ORCA-DL, a deep learning-based three-dimensional ocean model.
• Utilizing a data-driven approach for simulating global ocean dynamics.
• Comparative analysis with existing state-of-the-art numerical ocean models.

Main Results:

• ORCA-DL accurately simulates 3D global ocean dynamics with high physical consistency.
• The model outperforms traditional methods in capturing extreme events like ENSO and ocean heat waves.
• ORCA-DL demonstrates stable emulation of ocean dynamics at decadal timescales.

Conclusions:

• ORCA-DL shows significant potential for efficient and accurate global ocean modeling and prediction.
• Data-driven models offer a promising alternative for enhancing climate variability and change predictions.
• The model's performance suggests its utility for future climate projections.