Appel à candidatures | Recherche, Emploi

PhD - Offre de thèse à l'IMFT : Data-driven reduced-order modelling of urban canopy flow

Du 1 octobre 2023 au 30 septembre 2026

Toulouse

Contact : Contacts : romain.mathis@imft.fr

The student will take part of the joint ANR research program MUFDD (Modelling Urban Flows using Data-Driven methods) between IMFT, LHEEA (Ecole centrale of Nantes) and PPRIME (Poitiers), to support the experimental activities carried out at IMFT. The work will focus on isothermal street canyon urban canopy configuration. Experiments will be performed in the large, 26 m long and 1.1 m wide, hydraulic flume of IMFT with optical access for laser optical measurements (LDV, PLIF, TomoPIV).

For health, safety and environmental reasons, good prediction of air ventilation in urban areas are increasingly important to guide policy makers, urban planners and environmental engineers. Urban areas are critical places where critical physical processes such as heat absorption and production, pollution emission, alteration of the water cycle, among others, take place. All these processes are driven by chemical and dynamical interactions with the atmosphere above. In most cases, it is indeed the wind field that drives the exchanges between the urban canopy (i.e. the region ranging between the ground and the top of the buildings) and the atmosphere, enhancing the dissemination of scalars (such as chemicals, particles or heat) and by acting aerodynamically on urban structures. Our ability to develop well-adapted management strategies, modelling and forecasting tools is therefore of crucial importance. Difficulties arise due to the fact that the urban canopy is a highly heterogeneous and multi-scale terrain. As a consequence, the wind flow in the lower part of the atmosphere over such a surface is highly turbulent and is composed of various eddies of different scales in space and time. Modelling and prediction of such flows are challenging for this type of configuration. Classical numerical simulation approaches, such as Large-Eddy Simulation (LES) or Direct Numerical Simulation (DNS), still have a prohibitive computational cost as they involve solving nonlinear partial differential equations with a huge number of degrees of freedom. In accidental hazards with short-range dispersion, it is essential to have precise and rapid models to reproduce the intermittent and unsteady dynamics of the flow responsible for the dispersion processes to minimise the impacts in terms of health and pollution. To produce such rapid models, the present project aims at developing reduced-order models (ROMs) of the urban canopy flow both derived from and driven by data. To achieve this goal, data-driven model identification techniques and data-assimilation methods will be combined. Development and validation tests will be performed on canonical and realistic urban canopy configurations, ranging from small-scale laboratory experiments, up to full-scale street canyons. This project will address challenges posed by this data revolution with the aim to obtain a simplified dynamical model of the turbulent flow within the urban canopy, at the scale of the street or neighborhood including local buoyancy effects and scalar dispersion