CityML: A high-fidelity LES urban flow dataset

This repository contains wall-modelled large-eddy simulations (WMLES) of a neutral atmospheric boundary layer over 50 real urban neighbourhoods located in 12 European cities of 6 countries: Barakaldo, Barcelona, Bilbao, Madrid, Sevilla, Valencia and Zaragoza (Spain), Berlin (Germany), Brussels (Belgium), London (United Kingdom), Paris (France) and Porto (Portugal). The neighbourhoods were selected from a pool of candidate 1 km x 1 km grid cells by clustering their urban-morphology descriptors, so that the database covers a wide range of building densities, heights, street layouts and orientations. For every neighbourhood, the three-dimensional building geometry covers a 1.5 km x 1.5 km region and was reconstructed with City4CFD from Global Building Atlas footprints and national airborne-LiDAR heights. The flow was solved with SOD2D, a GPU-enabled spectral element method (SEM) computational fluid dynamics (CFD) code developed at the Barcelona Supercomputing Center (BSC), on the MareNostrum 5 supercomputer. All cases share the same inflow conditions and a single wind direction. For each of the 50 cases the dataset provides: (i) the building geometry (STL); (ii) the time-averaged flow statistics (mean velocity, mean pressure, Reynolds stress tensor, mean-velocity products and turbulent kinetic energy) on the simulation mesh, clipped around the buildings; and (iii) 1001 instantaneous snapshots of velocity and pressure on a horizontal plane at pedestrian level (z = 1.75 m), covering 1800 time units (10 flow-through times). A table gives the location and the morphology descriptors of every case. The data are intended for the development and validation of data-driven urban flow models (reduced-order models, machine-learning surrogates) and for the validation of CFD codes over urban flows.