The understanding of climate change and its impacts has significantly advanced in recent decades, revealing the formidable challenges it poses to humanity. In particular, it is now well established that the continued anthropogenic emissions of greenhouse gases are not only driving an increase in global mean temperature but also amplifying the frequency and intensity of extreme weather events such as heatwaves and heavy precipitation. However, some aspects remain poorly understood, such as the influence of climate change on complex weather phenomena like convective storms, which can produce significant hazards, including floods, hail, lightning, and severe winds. This knowledge gap arises mainly from the multiscale, multivariate, and nonlinear nature of such phenomena, which involve intricate interactions among atmospheric variables. These processes are governed by turbulence, characterised by highly disordered fluid motions that present major challenges for both theoretical understanding and numerical modeling. In particular, turbulent flows require a very large number of degrees of freedom to be comprehensively accounted for in computer simulations, which is far beyond the capabilities of current computational resources in realistic environmental or industrial settings. As a result, numerical models use subgrid-scale modelling strategies with many adjustable parameters to represent unresolved processes, which introduces uncertainties.

This thesis addresses these challenges by focusing in particular on the analysis and modelling of coherent structures—organized, recurrent flow patterns such as eddies that emerge from the apparent disorder and provide a more tractable view of turbulence. Special emphasis is placed on convective storms in the context of climate change. First, we show how high-dimensional atmospheric data can be simplified through statistical approaches that extract recurring patterns. Specifically, a generative probabilistic model originally developed for topic modeling in text analysis is adapted to decompose maps of atmospheric variables into spatial patterns that can be identified as atmospheric coherent structures. Second, we examine the challenges in studying convective storms and their evolution with climate change, outlining the limitations of available observations and numerical models. In particular, an initial climatology of derechos, a severe type of convective storm, is constructed for France using multiple observational datasets. We also analyse historical changes in atmospheric environments associated with such events with a methodology that accounts for the conditioning role of the large-scale atmospheric configuration. Finally, a modelling approach for turbulence based on the dynamics of coherent structures is proposed. This approach accounts for the intermittent nature of energy dissipation and leverages the sparsity of coherent structures to limit computational costs. Such strategy could provide more efficient and accurate alternatives to current subgrid-scale models of turbulence, with potential application for atmospheric convection and climate models.

Lieu
Laboratoire des Sciences du Climat et de l’Environnement
LSCE-IPSL, UMR CEA-CNRS-UVSQ 8212 CEA Saclay
Bât 714, Site de l’Orme des Merisiers Chemin de Saint Aubin
RD 128 F-91191 Gif sur Yvette Cedex
Salle 1129

Visio
https://cnrs.zoom.us/j/95369036553?pwd=BvpEn7V8oX1VehhY5omlEmr7dHoXWD.1

• Sergey NAZARENKO, Directeur de recherche, Institut de Physique de Nice (UMR 7010) : Rapporteur
• Nikki VERCAUTEREN, Professeur, Université de Cologne – Rapporteur
• Corentin HERBERT, Chargé de recherche, Laboratoire de physique (UMR 5672) – Examinateur
• Juliette BLANCHET, Directrice de recherche, Institut des Géosciences de l’Environnement (UMR 5001), – Examinatrice