Data-driven global ocean model resolving atmospherically forced ocean dynamics.

Area of Science:

• Earth System Science
• Climate Modeling
• Artificial Intelligence in Geosciences

Background:

• Traditional numerical weather models struggle with subseasonal to seasonal (S2S) climate predictions.
• Accurate simulation of ocean-atmosphere interactions is crucial for extending forecast timescales.
• Deep learning (DL) offers potential for more efficient and accurate climate modeling.

Purpose of the Study:

• To introduce KIST-Ocean, a novel deep learning-based global ocean general circulation model.
• To evaluate the simulation skill and efficiency of KIST-Ocean.
• To demonstrate the model's capability in reproducing key ocean-atmosphere dynamics relevant to climate phenomena.

Main Methods:

• Development of a deep learning-based global three-dimensional ocean general circulation model (KIST-Ocean).
• Comprehensive evaluation of KIST-Ocean's performance against established benchmarks.
• Analysis of the model's ability to simulate specific ocean responses like wave propagation and vertical motions.

Main Results:

• KIST-Ocean demonstrates robust ocean simulation skill and high computational efficiency.
• The model accurately reproduces atmospherically forced ocean dynamics, including Kelvin and Rossby waves.
• KIST-Ocean successfully simulates vertical motions driven by wind stress curl, crucial for phenomena like ENSO.

Conclusions:

• Deep learning models, like KIST-Ocean, can effectively capture essential ocean-atmosphere relationships.
• These findings enhance confidence in DL-based global weather and climate models.
• KIST-Ocean provides a foundation for developing integrated Earth system models for advanced long-range climate prediction.