Access

Abstract

Microscale pollutant dispersion is a critical aspect of air quality assessment with significant implications for the environment and public health. At this scale, of the order of a hundred meters, the aim is to characterize pollutant concentration levels in detail, which is particularly relevant in urban environments due to their great heterogeneity. Designing accurate microscale dispersion models is of paramount importance for predicting air pollution exposure and assessing risks, for example in case of industrial accidents. However, this is a challenging task, as the structure and trajectory of pollutant plumes are strongly influenced by atmospheric flow, which is inherently multi-scale and turbulent and interacts in complex ways with the built environment.

Figure 1. Large-scale eddy simulation of atmospheric flow and pollutant dispersion in a simplified urban canopy.

Computational Fluid Dynamics (CFD) has emerged as a powerful tool to address this issue by providing obstacle-resolving flow and dispersion predictions. However, CFD models are very costly, which hinders their use in operational and emergency contexts. In addition, their accuracy remains limited because of the significant uncertainties involved, in particular those arising from a lack of knowledge about the large-scale atmospheric forcing and from the internal variability of the atmospheric boundary layer. For risk assessment applications, controlling and quantifying these uncertainties is essential, but made difficult by the cost of CFD models.

To address these dual issues, we design and validate a large-eddy simulation (LES) modeling system that includes a reduced-order model and a data assimilation algorithm based on an ensemble Kalman filter. This thesis provides a proof-of-concept of the system’s ability to improve LES pollutant concentration field predictions in a neutral trial of the MUST field experiment. In particular, we demonstrate that local pollutant concentration measurements can be used to reduce meteorological parametric uncertainties and correct bias in the model boundary conditions. The use of a reduced-order model enables generating ensemble predictions that accurately account for the strong nonlinearities of the LES model, in just a few tens of seconds.

Particular attention is paid to the uncertainty associated with the internal variability of the atmospheric boundary layer. We adapt a bootstrap approach to quantify its effect on microscale dispersion and demonstrate that internal variability significantly affects not only LES model predictions but also field observations. By propagating the associated uncertainties to the standard statistical metrics used for air quality model evaluation, we show that the resulting variability in the validation metrics is significant and cannot be ignored when evaluating LES model accuracy. We then go a step further by accounting for this internal variability in the construction of the reduced-order model and of the data assimilation system. In particular, we show that the analysis of internal variability is of great interest to make an informed choice on the number of reduced-basis modes to avoid introducing noise into the reduced-order model. Finally, we take into account the microscale internal variability in the data assimilation process, making it much more robust and realistic. These additions to the data assimilation framework are made elegantly and without generating implementation or computational heaviness.

From a broader point of view, this thesis shows some ways to adopt a probabilistic modeling approach for complex atmospheric phenomena based on LES, which are nowadays recognized as references, but remain subject to uncertainties, some of which are inherently irreducible.

Figure 2: Overview of the manuscript structure.

How to cite?

Lumet, E. (2024) Assessing and reducing uncertainty in large-eddy simulation for microscale atmospheric dispersion. PhD thesis, Université Toulouse III - Paul Sabatier. URL: https://theses.fr/2024TLSES007 

@phdthesis{lumet2024phd,
    title={Assessing and reducing uncertainty in large-eddy simulation for microscale atmospheric dispersion},
    author={Lumet, Eliott},
    year={2024},
    school={Universit{\'e} Toulouse III - Paul Sabatier},
    url={https://theses.fr/2024TLSES003}}