Distinguished Seminar in Computational Science and Engineering

Distinguished Seminar in Computational Science and Engineering

 

September 10, 2026, 12-1PM

 

45-432 in Building 45 and Zoom Webinar

 

Toward Accurate and Efficient Large-Eddy Simulation of Urban Canopy Flows

 

Catherine Gorlé

Associate Professor of Civil and Environmental Engineering, Stanford University

 

Abstract: 

Computational fluid dynamics is increasingly being used to support the sustainable and resilient design of buildings and cities by predicting wind loading, natural ventilation, pedestrian wind comfort, and urban air quality. Large-eddy simulation (LES) has emerged as the most promising approach because it explicitly resolves the energetic turbulent motions that govern these processes. This presentation explores recent advances that are bringing LES closer to routine engineering practice.

A first step toward broader adoption of LES for urban and building design is demonstrating its predictive capability across different applications. For wind loading, validation studies have shown that a two-step LES methodology provides accurate predictions of peak pressures on buildings in realistic urban environments. For natural ventilation, coupled indoor-outdoor LES reproduces measured ventilation rates while capturing the strong influence of surrounding buildings on local flow patterns. These validation studies demonstrate that LES can accurately predict the governing flow physics.

The second challenge is reducing the computational cost of a design analysis, where tens to hundreds of simulations may be required to account for changing wind directions, atmospheric variability, and design alternatives. For wind loading, high-fidelity LES can be combined with lower-fidelity simulations through multi-fidelity machine learning frameworks, enabling accurate predictions across many wind directions while requiring only a limited number of expensive LES calculations. For applications such as urban airflow and natural ventilation, deep-learning surrogate models trained on LES databases can predict flow fields in seconds, providing orders-of-magnitude reductions in computational cost while maintaining the desired engineering accuracy. Together, these developments point toward a new generation of computational tools that combine physics-based simulation and machine learning to enable accurate and efficient prediction of complex turbulent flows.

 

Bio:

Catherine Gorlé is an Associate Professor of Civil and Environmental Engineering at Stanford University. Her research focuses on developing predictive computational methods for turbulent flows to support the design of sustainable and resilient buildings and cities. Her group develops computational frameworks that combine large-eddy simulation, uncertainty quantification, multi-fidelity and multiscale modeling, and machine-learning techniques to improve both the accuracy and computational efficiency of simulation methods.

Catherine received her BSc (2002) and MSc (2005) degrees in Aerospace Engineering from the Delft University of Technology, and her PhD (2010) from the von Karman Institute for Fluid Dynamics in cooperation with the University of Antwerp. She has been the recipient of a Stanford Center for Turbulence Research Postdoctoral Fellowship (2010), a Pegasus Marie Curie Fellowship (2012), and an NSF CAREER award (2018).