Regional Climate Model Emulation Using Diffusion models

Abstract

Regional Climate Models (RCMs) are essential for producing high-resolution climate simulations tailored to specific regions. However, their reliance on discretized physical laws and boundary con ditions from Global Climate Models (GCMs) limits flexibility for assimilating new observations and makes long-term simulations computationally expensive, reducing the ability to explore multiple possible climate states from the same GCM. This thesis investigates the application of Score-Based Data Assimilation (SDA) ,reyling on diffusion models, in the context of the Modèle Atmosphérique Régional (MAR) over Belgium. The problem is framed from a stratistical perspective as an inverse problem.