INTEGRATING MACHINE LEARNING WITH BAROCLINIC INSTABILITY MODELS FOR ADVANCED MESOSCALE ENERGY CASCADE ANALYSIS

Abstract

This study combines traditional methods, like normal mode decomposition, with cutting-edge machine learning (ML) techniques to enhance the analysis of energy transfer across scales in baroclinic instability models. By leveraging high-resolution, non-hydrostatic simulations, we explore the energy distribution between geostrophic and ageostrophic modes, uncovering distinctive spectral slopes of −3.1 and −2.7, respectively, which underscore the role of inertia-gravity waves at the mesoscale. Employing Convolutional Neural Networks (CNNs), we automate the identification and classification of these modes, streamlining spectral analysis and improving accuracy, even in highly turbulent environments. This approach not only advances our understanding of mesoscale energy cascades but also highlights the transformative potential of machine learning in atmospheric dynamics, paving the way for more precise weather and climate predictions.