Ensemble Data Assimilation for High-Resolution Urban Simulations

Accurate characterisation of local urban wind flows is relevant for applications like pedestrian comfort, microclimate modelling, and pollutant dispersion. High-resolution computational fluid dynamics simulations provide detailed flow fields, but are highly uncertain due to model assumptions, boundary conditions, and turbulence parameterisations. By incorporating observations into these models, we can quantify this uncertainty, particularly in complex urban geometries, and make corrections to the model itself and its output.

This work is part of the UrbanAIR project, which aims to develop model representations of urban environments (Digital Twins), high-resolution physics-based computational models that mirror the urban atmosphere. By combining complex atmospheric models with city geometry, simulations of mesoscale climate, field measurements, and user input, the project aims to support decision-makers in designing climate-resilient cities.

In this context, we explore ensemble-based data assimilation strategies to improve the estimation of key drivers of urban flow, focusing on reduced-order formulations such as boundary condition and parameter estimation. These ensemble-based methods are considered, given the strong nonlinearities and high dimensionality of the system, while maintaining computational tractability. Synthetic experiments are used to investigate how effectively limited observations can constrain model states and parameters, and to assess the role of observation type, availability, and location.

In steady RANS urban simulations, preliminary results indicate that ensemble smoothing can recover the inflow wind angle distribution in synthetic twin experiments, but the performance of measurement locations is strongly dependent on the type of observable. In particular, velocity magnitude appears to carry directional information primarily in wake regions, leading to non-unique inversions elsewhere, whereas local velocity components provide more complementary information.

This work contributes to the connection between high-resolution urban modelling and observational data, supporting the development of integrated Digital Twin frameworks for urban climate and air quality applications.