Centrale Lyon

Summary of the PhD research project - ** Context and objectives** Accelerated urbanization and the urban heat island effect are increasing thermal stress on populations, a phenomenon exacerbated by climate change (IPCC, 2022). In Southern Europe, where demographic aging heightens vulnerabilities, urban adaptation is becoming a priority. By 2070, cities in southern France could experience more than 30 days of heatwaves annually, with temperatures reaching up to 50 °C. **2. Scientific Challenges** Current urban atmospheric simulations rely on various modeling approaches based on the resolution of fluid dynamics equations (Maronga et al., 2020). To accurately simulate microclimatic conditions, these models must incorporate multiphysics modules that account for radiative exchanges between urban surfaces as well as the evapotranspirative and radiative exchanges of vegetation, which presents a major challenge. Simulating the effects of trees in these models is particularly complex: wind resistance of leaves must be represented by a drag force (Salim et al., 2015), while effects on radiative exchanges are modeled through an attenuation factor in specific algorithms (Krc et al., 2021). Due to their computational cost, these high-fidelity numerical simulations are limited to small domains and specific meteorological conditions. They are therefore not suitable for real-time simulations or for evaluating greening scenarios at the metropolitan scale. To overcome this limitation, operational models are needed. Designed to be accessible to non-experts, these models must (i) require far fewer computational resources and (ii) provide hourly values for variables such as air temperature, humidity, wind speed, and radiation at street level, capturing the main spatial variations (Di Sabatino et al., 2013). These variables are essential for calculating thermal comfort indices, such as the Universal Thermal Climate Index (UTCI) (Blazejczyk et al., 2012). Achieving this requires simplifying urban geometry and parameterizing urban dynamics, potentially by integrating field data through data assimilation (Nguyen and Soulhac, 2021) or machine learning (Buizza et al., 2022). The parameterization will depend on the model’s intended use—whether for real-time simulations (health risk management), scenario analysis (epidemiological studies), or urban planning optimization. While several operational models exist for air quality (Soulhac et al., 2011) and noise (Gabillet, 1990), very few are dedicated to simulating urban climate at the street scale. To address this, we propose in this PhD project to integrate new Artificial Intelligence (AI) algorithms into the operational model for predicting urban thermal stress, with scientific challenges notably including the integration of modalities of different types and scales as well as the need for data efficiency in model training. **3. Founding of the PhD thesis : Bouquet de thèses 2025** The doctoral thesis described above is part of a series of theses designed to build a multidisciplinary scientific approach to the societal challenge of a "responsible digital society", and more specifically, the specific theme of "Data and AI in a sustainable and responsible approach", identified as a priority issue by the 4 institutions of the Lyon Saint-Etienne Engineering College (Centrale Lyon, ENTPE, INSA Lyon, Mines Saint-Étienne) and by the Université Jean Monnet Saint-Étienne, which are providing financial support for the theses making up this 2025 package. The 2025 theses package includes 6 theses covering different facets of data science and artificial intelligence, addressing the following questions: - Monitoring crystallization processes using AI-assisted acoustic emission - AI-assisted design of biodegradable and/or biosourced biopolymers for the sustainable protection of agricultural crops - Machine learning methods for urban microclimate prediction - Data-driven non-linear structure identification - Inference and explicability in confidential mode: towards self-diagnosis via images - Towards certification of vibration monitoring with explanatory AI These theses involve a total of 16 supervisors from 11 laboratories on the Lyon Saint-Etienne site (Centre d'Innovation en Télécommunications et Intégration, Centre SPIN - Génie des Procédés, Ingénierie des Matériaux Polymères, Biologie Fonctionnelle, Insectes et Interactions, Institut Camille Jordan, Laboratoire de Mécanique des Fluides et d'Acoustique, Laboratoire de Tribologie et Dynamique des Systèmes, Laboratoire d'InfoRmatique en Image et Systèmes d'information, Laboratoire Hubert Curien, Laboratoire Vibrations Acoustique, Matériaux : Ingénierie & Science) of which the 5 funding institutions are supervisors. The 6 PhD students recruited under this package will be enrolled in 3 Doctoral Schools on the site: MEGA, EDML, SIS. The teams (doctoral students and their supervisors) involved in these 6 theses form a multi-disciplinary scientif