Benchmarking the geographic generalization of deep learning models for precipitation downscaling

Key Points

• Generalization of deep learning models for precipitation downscaling shows notable geographic variability.
• Key metrics indicate model accuracy can drop by up to 30% in certain regions compared to training areas.
• Benchmarking across various geographic terrains assesses deep learning model performance and robustness.
• These findings highlight the need for further investigation into models' generalizability across different regions.