AI-Driven Regional Downscaling for High-Resolution Oceanic and Atmospheric Forecasting

Accurate high-resolution forecasting of oceanic and atmospheric states remains a critical challenge. This study introduces an AI-based regional downscaling framework employing a U-Net deep learning architecture, trained on coarse-resolution simulations. By embedding physical constraints, the model effectively bridges scales, capturing fine-grained dynamics unresolved in traditional approaches.

The framework significantly enhances computational efficiency, reducing forecast times from hours to seconds per region while maintaining high accuracy. Its integration with data-parallel computing units enables scalable multi-region applications. Applied within a coupled ocean-atmosphere-wave-tide system, the model excels in reproducing extreme events and mesoscale dynamics.

This work highlights the potential of AI in offering scalable, precise solutions for forecasting, climate science, and disaster management.