The global population is increasingly concentrated in urban areas, with projections indicating that up to 80% will reside in cities by 2050. This trend underscores the need for efficient and reliable tools to model the urban microclimate, which play a key role in supporting urban planners and policymakers in designing more sustainable and livable environments.

At the urban scale, pollutant dispersion is strongly influenced by transient atmospheric conditions, requiring high-fidelity computational fluid dynamics (CFD) simulations with fine spatial resolution and small time steps. These requirements lead to substantial computational and memory demands, often necessitating access to high-performance computing (HPC) resources to obtain results within practical time frames.

To address these challenges, surrogate modeling techniques offer a promising alternative, enabling rapid predictions with controlled accuracy loss. In this talk, I will present recent advances in numerical methods for real-time and many-query simulations of urban microclimate dynamics. The discussion will cover projection-based reduced-order models, physics-informed neural networks, graph neural networks, and scale-adaptive simulation strategies. This research is developed within the framework of the ERC Starting Grant DANTE.

Contatto:
nicola.parolini@polimi.it