Adaptation of the hybrid Eulerian/Lagrangian stochastic model of the CFD code Code-Saturne to pollutant atmospheric dispersion at the micro-meteorological scale and comparison with the Eulerian method

This Ph.D. thesis is part of a project that aims at modeling pollutant atmospheric dispersion with the Computational Fluid Dynamics code Code-Saturne. The objective is to simulate atmospheric dispersion of pollutants in a complex environment, that is to say around power plants, industrial sites or in urban areas. In this context, the focus is on modeling the dispersion at micro-scale, that is for distances of the order of a few meters to a few kilometers and corresponding to time scales of the order of a few tens of seconds to a few tens of minutes: this is also called the near field area. The approach followed in this thesis follows a hybrid Eulerian/Lagrangian formulation, where the mean dynamical fields relative to the carrier fluid (pressure, velocity, temperature, turbulence) are calculated through an Eulerian approach and are then provided to the Lagrangian solver. This type of formulation is commonly used in the atmospheric literature for its numerical efficiency. The Lagrangian stochastic model considered in our work is the Simplified Langevin Model (SLM), developed by Pope (1985, 2000). This model belongs to the methods commonly referred to as PDF (Probability Density Function) methods, and, to our knowledge, has not been used before in the context of atmospheric dispersion. First, we show that the SLM meets the so-called well-mixed criterion (Thomson, 1987). This criterion, essential for any Lagrangian stochastic model to be regarded as acceptable, corresponds to the fact that if particles are initially uniformly distributed in an incompressible fluid, then they must remain so. We check the good respect of the well-mixed criterion for three cases of inhomogeneous turbulence representative of a wide range of practical applications: a mixing layer, an infinite plane channel, and an atmospheric-like case involving an obstacle within a neutral boundary layer. We show that the good respect of the well-mixed criterion lies simply in the good introduction of the pressure gradient term as the mean drift term in the Langevin model (Pope, 1987; Minier et al., 2014; Bahlali et al., 2018c). Also, we discuss the importance of consistency between Eulerian and Lagrangian fields in the framework of such Eulerian/Lagrangian hybrid formulations. Then, we validate the model in the case of continuous point source pollutant dispersion, under uniform wind and homogeneous turbulence. In these conditions, there is an analytical solution allowing a precise verification. We observe that in this case, the Lagrangian model discriminates well the two different near- and far-field diffusion regimes, which is not the case for an Eulerian model based on the eddy-viscosity hypothesis (Bahlali et al., 2018b). Finally, we work on the validation of the model on several experimental campaigns in real atmosphere, taking into account atmospheric thermal stratification and the presence of buildings. The first experimental program considered in our work has been conducted on the 'SIRTA' site (Site Instrumental de Recherche par Teledetection Atmospherique), in the southern suburb of Paris, and involves a stably stratified surface layer. The second campaign studied is the MUST (Mock Urban Setting Test) experiment. Conducted in the United States, in Utah's desert, this experiment aims at representing an idealized city, through several ranges of containers. Two cases are simulated and analyzed, respectively corresponding to neutral and stable atmospheric stratifications (Bahlali et al., 2018a). (author)