Real-Time Data-Driven Modeling of Road Traffic Pollutant Emissions and Atmospheric Concentrations at an Urban Scale

The importance of air quality cannot be underestimated as it represents a major public health issue: today, nearly 91% of the global population is exposed to air pollution levels exceeding the exposure thresholds set by the WHO. This thesis lies at the intersection of three scientific domains: numerical modeling, air quality, and deep learning. Its central objective is to enable precise assessment of population exposure to air pollutants, leveraging recent advances in artificial intelligence. In this context, the thesis aims to improve the description of emissions dispersion resulting from road traffic in urban areas. To achieve this, we will employ numerical modeling techniques integrating state-of-the-art simulation tools for atmospheric dispersion calculations such as the SIRANE code. Used in practice for operational needs, this tool can be combined with sensitivity analysis tools to identify impact and exposure factors by considering numerous parameters determining concentrations (meteorology, traffic composition and density, urban layout...) and requiring the exploitation of numerous simulations, which can represent significant computational time. To address this, we plan to introduce deep learning algorithms to develop accelerated models for determining pollutant concentrations in the atmosphere and to be used on lightweight computing platforms. A first step will involve conducting a comprehensive analysis of currently available tools. This analysis will provide a better understanding of their strengths and limitations and identify their potential complementarities. Subsequently, the use of deep learning will be deployed to accelerate atmospheric dispersion calculation times. This innovative approach will yield results more quickly while maintaining a high level of precision compared to more traditional approaches. In summary, this thesis aims to push the boundaries of simulation and air quality characterization by combining cutting-edge methods in numerical modeling, air quality estimation, and deep learning.

Keywords: Air quality, Numerical simulation, Deep Learning, Finite elements

• Academic supervisor    Prof Lionel SOULHAC, LMFA, ORCID : 0358-3486
• Doctoral School    ED 162 MEGA (https://edmega.universite-lyon.fr)
• IFPEN supervisor    Dr Guillaume SABIRON, ORCID : 6670-7322
• PhD location    IFPEN, Lyon, France
• Duration and start date    3 years, starting in the fourth quarter 2024 
• Employer    IFPEN
• Academic requirements    Master’s degree in data science, Mathematics, Fluid mechanics
• Language requirements    English level B2 (CEFR), Willingness to learn French    
• Other requirements    Python.

To apply, please send your cover letter and CV to the IFPEN supervisor indicated below.