Machine Learning for Turbulence Modeling in Complex Terrain
Internship Nantes, France 13.10.2025Meteodyn & Sorbonne University
METEODYN is a very dynamic and well-known international High-Tech company created in 2003, leader in its sectors, namely Wind Engineering (wind energy & safety), Climatology (urban & climate change) and Meteorology (forecasting). Our headquarter is based in France, having 3 subsidiaries (2 in China, 1 in India). METEODYN is part of the CLS group, a company with global reach and expertise in Ocean Engineering and Satellite Data Processing. CLS is also a company with a mission to understand and protect the planet.
Sorbonne University is a public university, created in 2018, through the merger of Paris-Sorbonne and Pierre et Marie Curie Universities. As a multidisciplinary research university in the heart of Paris, Sorbonne University is resolutely international, research-intensive and world-class university. In cooperation with its partners, it creatively and innovatively implements its missions of education, research, innovation, mediation and technology transfer for the common good.
Mission
Meteodyn specializes in advanced wind flow simulation using Computational Fluid Dynamics (CFD) to model atmospheric behavior in environments with significant complexity—such as mountainous terrain, variable land cover, and urban structures. Our proprietary software, Meteodyn WT, is a CFD based tool designed to model wind flow and deliver high-accuracy wind resource assessments, particularly in complex terrains.
In the framework of the PostGenAI@Paris cluster in collaboration with Sorbonne University, this research leverages recent advances in machine learning-assisted turbulence modeling [1] to enhance the predictive capabilities of CFD simulations. Traditional turbulence models, such as Reynolds-Averaged Navier-Stokes (RANS), often face significant limitations when dealing with complex flow phenomena—particularly in scenarios involving flow separation, recirculation zones, and turbulence induced by irregular terrain.
The intern will explore and implement data-driven strategies to refine the turbulence closure, based on the framework such as Sparse Bayesian Learning [2] to generate corrections relying on high-fidelity data. These corrections aim to address limitations of standard RANS models by introducing more flexible and context-aware turbulence representations. The enhanced turbulence models will be evaluated through simulations over representative test cases (e.g., turbulent boundary layer flows, steep terrain with roughness variations) and validated using available observational or reference data. The goal is to achieve a more robust prediction of wind fields in operational conditions, thus improving wind resource assessments and sitting decisions in complex environments.
This project contributes to ongoing research into improving RANS closures using data-driven corrections, supporting the broader goal of more physically accurate turbulence modeling over complex terrain. The intern will collaborate closely with researchers from Sorbonne University and Meteodyn’s R&D team. Supervision will include regular meetings with academic and industry mentors.
Candidate Profile
This internship is intended for a final-year engineering and master student with a strong background in fluid dynamics, CFD, or data-driven modeling who would like to pursue a PhD Thesis via the CIFRE instrument.
Human & Technical skills
- Strong understanding of turbulence modeling and RANS-based CFD
- Interest or experience in data-driven modeling and machine learning for physical systems
- Programming skills in Python and Fortran (experience with scientific computing libraires)
- Good written and spoken scientific English
- Scientific rigor and autonomy
Technical information
Location: Meteodyn Headquater, Saint-Herblain (Nantes), France
Duration: 6 months
Starting Date: First quarter of 2026
Please send you resume and letter of motivation through the form below.
References
[1] M. Oulghelou, S. Cherroud, X. Merle and P. Cinnella, Machine-learning-assisted Blending of DataDriven Turbulence Models, Flow Turbulence and Combustion, 2025
[2] M. Schmelzer, R.P. Dwight, P.Cinnella, Discovery of algebraic Reynolds-stress models using sparse symbolic regression, Flow Turbulence and Combustion 104(2) 2020 579-603
"*" indicates required fields