Thesis by Mathis Pasquier : « Quantification d’incertitudes pour la dispersion turbulente de polluants à l’échelle micro-urbaine » (Uncertainty quantification for the turbulent dispersion of traffic-related pollutants on a micro-urban scale).
In urban zones, road traffic is responsible for a high proportion of pollutant emissions, with a significant impact on air quality, which represents a major public health issue. Atmospheric dispersion can be comprehensively taken into account using CFD1. However, random uncertainties, of both natural and anthropogenic origin, can affect its predictive capacity.
Concerning this prediction, the PhD research was aimed at quantitatively evaluating the impact of uncertain parameters: firstly, meteorological and, secondly, those related to road traffic. To do this, “high-fidelity”2 simulations of pollution dispersion on a micro-urban scale (neighborhood) were conducted.
Initially, a complete modeling chain was constructed in order to produce two-dimensional spatial representations of pollutant concentrations in real urban geometries. This combined a CFD code based on the Lattice Boltzmann method (LBM), a microscopic traffic simulator (vehicle scale) and a road traffic emissions model. This LBM code was developed for highly turbulent flows and validated on academic test cases before being applied to realistic road geometry and emissions, in this case NOx emissions (Figure 1).
The subsequent calculations demonstrated that consideration of non-uniform emissions data, associated with the acceleration/deceleration phases of vehicles using urban roads, modified the estimation of inhabitant pollution exposure [1].
A sensitivity analysis was then conducted using this calculation chain, in order to identify which input uncertainty sources have the greatest impact on outputs of interest. To this end, substitution models3, which are inexpensive to evaluate, were constructed, in order to reduce CFD simulation time.
The outputs examined were time-averaged maps related to pollutant concentration and to the probability of threshold exceedance. The uncertainty input variables were: wind direction and speed, traffic volume and composition (proportion of diesel and gasoline vehicles), as well as the road network speed limit. In this way, 2D spatial sensitivity maps (Figure 2) and global sensitivity indices for the entire domain were obtained [2].
Conducted on a local scale and with data reflecting the most realistic conditions possible (geometry, weather, emissions), this study concluded that wind direction and the proportion of diesel engines were the most influential factors.
The research also highlighted the interest of the type of approach employed, combining CFD with statistical methods, to understand the influence of multiple parameters in complex scenarios. In terms of application, this makes it possible to objectively assess the impact of a given planning or regulatory decision.
et de la proportion de moteurs diesel/essence (à droite).
1- Computational Fluid Dynamics.
2- Enabling consideration of turbulence.
3- Proper Orthogonal Decomposition combined with Gaussian Processes Regression (POD-GPR).
4- Sensitivity index of an output variable with respect to an input variable (based on variance decomposition).
References:
-
Mathis Pasquier, Stéphane Jay, Jérôme Jacob, Pierre Sagaut. A Lattice-Boltzmann-Based Modelling Chain for Traffic-Related Atmospheric Pollutant, Dispersion at the Local Urban Scale. Building and Environment, 2023, 242, pp.110562.
>> https://doi.org/10.1016/j.buildenv.2023.110562
-
Noé Fellmann, Mathis Pasquier, Céline Helbert, Adrien Spagnol, Delphine Sinoquet, Christophette Blanchet-Scalliet. Sensitivity analysis for sets : application to pollutant concentration maps. Quality and Reliability Engineering International, Special issue, 2024.
>> https://doi.org/10.1002/qre.3638
Scientific contact: Stéphane Jay