U3DWind v2

Time-mean three-dimensional urban wind fields over five Chinese cities, computed with the CULBM lattice-Boltzmann large-eddy solver on a uniform 10 m grid. Each city has 144 cases: 16 inflow directions x 3 reference speeds x 3 inflow profiles, 720 cases in total.

Folder	Centre (N, E)	Domain, east x north	Horizontal cells
beijing	39.9050, 116.4040	50 km x 40 km	5000 x 4000
shanghai	31.2300, 121.4737	45 km x 35 km	4500 x 3500
guangzhou	23.1290, 113.2640	40 km x 35 km	4000 x 3500
shenzhen	22.5400, 114.0580	35 km x 25 km	3500 x 2500
hongkong	22.3320, 114.1700	30 km x 25 km	3000 x 2500

The grid has slightly more than 100 levels in the vertical; the exact number depends on the terrain relief of the city and is given by the z dimension.

Layout

<city>/
    u<SS>_<profile>/ang_<DDD.D>/
        mean_global.nc   fields on the Cartesian grid
        mean_agl.nc      the same fields on levels above the local terrain
        conf.luwpf       solver configuration of the case
        profile.dat      inflow speed profile, height above ground vs speed
    geometry/
        <city>_DEM.stl        terrain and building solid used by the solver
        buildings.gpkg        building footprints with their heights
        terrain.tif           terrain elevation, 25 m grid
        interpolated_dem.csv  terrain height above the lowest point
        summary.json          projection, extents and statistics
PROVENANCE.md            sources of every geometric input

• SS is the reference speed at 10 m above ground: 03, 06 or 09 m/s.
• profile is annual, summer (April-September) or winter (October-March); it selects the power-law exponent of the inflow profile.
• DDD.D is the direction the wind comes from, in degrees clockwise from north, in steps of 22.5.

A single case is 20 to 50 GB. To download one:

hf download Saxon0520/U3DWind --repo-type dataset \
    --include "hongkong/u06_annual/ang_090.0/*" --local-dir U3DWind

Variables

Name	Units	Meaning
u, v, w	m s-1	time-mean velocity towards east, north and up
tke	m2 s-2	resolved turbulent kinetic energy
rho	kg m-3	time-mean air density
fluid	-	1 in air, 0 inside terrain or buildings
ground_z, first_level_height	m	AGL file only: terrain height and the height of level 0 above it

Fields are 32-bit floats stored as NetCDF-4 with shuffle and deflate. The means cover the last 2000 of 50 000 time steps, sampled every 10 steps.

Coordinates

x points east and y north, in metres from the south-west corner of the domain, in a transverse Mercator projection centred on the city centre with scale factor 1; the projection string of each city is in <city>/geometry/summary.json. In mean_global.nc, z is the height above the lowest terrain point of the domain. In mean_agl.nc, level is the height above the first grid plane over the local terrain.

Inputs

Buildings and their heights come from 3D-GloBFP (Che et al., 2024), terrain from FABDEM V1-2 (Hawker et al., 2022). PROVENANCE.md lists the tiles, identifiers and processing steps. The inflow is a power-law profile with synthetic turbulence (intensity 5 %, length scale 100 m); the atmosphere is neutral and the Coriolis force is not included.

Limitations

• A 10 m grid represents building height in steps of 10 m. A building lower than about 12 m is either one cell high or absent from the solid, and taller ones can be up to one cell lower than in the source data.
• 3D-GloBFP heights are estimates from Earth-observation features. They are too low for tall buildings, and landmark towers are not reproduced; the tallest building per city is 157 m (Beijing) to 338 m (Shanghai).
• Buildings are flat-roofed prisms; vegetation is not represented.
• The power-law exponents come from reanalysis wind speeds at one point per city, and the inlet turbulence parameters are prescribed constants.

Licence

CC BY-NC-SA 4.0. The terrain derives from FABDEM, which is licensed for non-commercial use.