Themistoklis (Themis) Vargiemezis

I am a Postdoctoral researcher at the Wind Engineering Lab, part of the Civil & Environmental Engineering Department of Stanford University. Prior to that, I finished my PhD from the same lab, supervised by Prof. Catherine Gorle.

My research focuses on advancing the understanding and predictive modeling of wind flow in urban areas through an interdisciplinary approach that combines Computational Wind Engineering and Deep Learning. I aim to establish simulation and data-driven frameworks to improve how we design, construct and operate buildings and urban spaces, supporting the development of more sustainable and resilient cities.

I'm always open to new research collaborations. If you're working on something aligned with these topics, I'd love to hear from you.

Email  /  Google Scholar  /  Github  /  LinkedIn

News

• 2025-10 Presented my poster "WindMiL: Equivariant Graph Learning for Wind Loading Prediction" at the Stanford Graph Learning Workshop 2025, part of the Stanford Data Science Affiliates Program, Oct. 14, 2025
• 2025-05 I will be presenting at EMI 2025 conference in Anaheim, California, May 27-30, 2025
• 2025-01 Started my postdoc at the Wind Engineering Lab, Stanford University
• 2024-06 Successfully defended my PhD from Stanford University
• 2023-08 ICWE16 conference presentation selected for the Special Issue of Elsevier's JWEIA

Research

WindMiL: Equivariant Graph Learning for Wind Loading Prediction Themistoklis Vargiemezis, Charilaos Kanatsoulis, Catherine Gorle Stanford Graph Learning Workshop 2025 Poster This research introduces WindMiL, a symmetry-aware graph neural network surrogate for predicting wind pressures on low-rise buildings. We build a systematic LES dataset (462 cases) by interpolating roof geometries with signed-distance function, then train a reflection-equivariant GNN that guarantees consistent predictions on mirrored building geometries. WindMiL accurately estimates the surface pressure coefficients and remains robust in extrapolation, while running in seconds rather than hours required for LES workflows.

This research proposes and validates a Large-Eddy Simulations (LES) framework for predicting wind pressures on low-rise buildings. By comparing LES results with two wind tunnel datasets for a realistic building model, we demonstrate that LES accuracy is comparable to the variability between experiments, boosting confidence in its reliability for wind load estimation.

This study validates Large Eddy Simulation (LES) for predicting wind-induced pressures on low-rise buildings in urban settings. Using wind tunnel data, the LES results showed great agreement across pressure statistics (mean, RMS, peak, skewness, kurtosis). The simulations also showed how surrounding buildings increase negative peak pressures by affecting local flow patterns. The work demonstrates LES as a reliable tool for evaluating wind loads in complex urban environments.

We investigated the pressure loads on a realistic low-rise building configuration within an urban area to study wind-induced pressure loads due to the surrounding buildings. The test case is the Y2E2 building in Stanford Engineering Quad. Experiments were performed at two wind tunnels; the NHERI Wall of Wind (WOW) wind tunnel at the Florida International University facility, and the NHERI terraformer Boundary Layer Wind Tunnel at the University of Florida (UF). The dataset is available for download below.

Teaching

Teaching Assistant, CEE 260G: Imaging with Incomplete Information, Spring 2024 Teaching Assistant, CEE261D: Data assimilation, Winter 2024 Teaching Assistant, ENGR203: Public Speaking for Engineers, Autumn 2023 Teaching Assistant, CEE261A: Physics of Wind , Winter 2021