Comprehensive Validation of the Weather Foundation Model Through Atmospheric Process Analysis

Abstract

Weather Foundation Models in general represent a significant advancement in computational weather prediction by leveraging data-driven techniques to improve accuracy and speed. In this work we evaluate Prithvi WxC, a foundation model for weather and climate, from the perspective of adherence to physical constraints imposed by the equations governing atmospheric flow. The majority of AI model validation to date has focused on computing error statistics of modeled atmospheric fields and phenomena. We conducted experiments to systematically assess the performance of Prithvi WxC model to: 1) examine adherence to fundamental physical constraints such as conservation of mass; 2) consistency of midlatitude flow fields with respect to vorticity dynamics and QG theory and; 3) test if the AI modeled evolution of flow fields evolves in a way that is consistent with time tendencies implied by physical processes. This presentation will cover our validation methods, highlight key findings, and discuss how these insights can help improve the model's future applications in weather forecasting and climate research.