Validation of the Prithvi WxC Model Through Atmospheric Process Evaluation

The advancement of Weather Foundation Models represents a notable progression in computational weather forecasting by incorporating data-driven methods to enhance both accuracy and efficiency. This study critically evaluates the Prithvi WxC, a foundation model designed for weather and climate, focusing on its compliance with the physical constraints dictated by atmospheric flow equations. While most AI model validations have traditionally emphasized error statistics related to atmospheric fields and phenomena, our research takes a different approach. We conducted a series of experiments to systematically assess the Prithvi WxC model’s performance by: 1) evaluating its adherence to core physical principles such as mass conservation; 2) analyzing the consistency of midlatitude flow fields with vorticity dynamics and quasi-geostrophic (QG) theory; and 3) verifying whether the AI-predicted evolution of flow fields aligns with the time tendencies dictated by physical processes. This presentation will outline our validation methodologies, present significant findings, and explore how these insights could contribute to refining the model’s application in both weather forecasting and climate research.