CityFFD-UrbanMicroclimate

Time-averaged 3D wind velocity fields for 33 urban areas in 11 cities, solved with the CityFFD urban airflow solver, together with the gridded building geometry used in each simulation. The dataset was introduced in FLUME and is also used in GeoWind2Plan (demo, code). The data used in GeoWind2Plan, ready to use with its code, are in geowind2plan_data/.

License: CC BY-NC 4.0

Files

• metadata.csv: one row per area with id, city, the approximate longitude and latitude of the domain centre, and the grid size nz, ny, nx.
• data/<id>.nc: one NetCDF-4 file per area, holding u, v, w (velocity components along x, y, z, float32, m/s) and building (uint8, 1 = building, 0 = air) on a (z, y, x) grid, with cell-centre coordinates x, y, z in metres.

Grid

• Cells are 4 m × 4 m horizontally and 1.5 m vertically. x and y are measured from the domain centre. z is the height above the ground, starting at 0.75 m.
• Domains are 1.2 km × 1.2 km × 240 m, except osa01 and tyo01 at 3 km × 3 km × 180 m.
• Velocity is zero inside buildings.

Inflow

The inflow is along +x at 4 m/s, with a power-law profile of exponent 0.15 and reference height 10 m. Velocities are time averages taken after the flow statistics converged.

GeoWind2Plan data

geowind2plan_data/ is the input data of the GeoWind2Plan code. To run the code, place the contents of this folder in the data/ folder of the code.

It holds five of the urban areas above, hkg13, mtr01, mel02, syd02 and osa01, which are blocks A to E in GeoWind2Plan. Their building geometry and CFD wind are cropped to the UAV flight heights of GeoWind2Plan (30-120 m) and stored in the axis order of the code, which differs from data/<id>.nc. The folder also holds the wind predicted by neural operator for each block.

geowind2plan_data/
  profile_wind.npy      height-only wind profile, one value per layer (m/s)
  block_A/ ... block_E/
    buildings.npy       building occupancy
    sdf.npy             signed distance to buildings (m), positive in free space
    wind_cfd.npy        CFD wind, 4 m/s inflow
    wind_predicted.npy  wind predicted by neural operator, 4 m/s inflow
  block_A/
    wind_cfd_2ms.npy    CFD wind, 2 m/s inflow
    wind_cfd_8ms.npy    CFD wind, 8 m/s inflow

Citation

If you use this dataset, please cite:

@misc{qin2026fastlocalizedurbanmicroclimate,
  title={Fast localized urban microclimate emulator: wind and temperatures in unseen urban morphologies},
  author={Shaoxiang Qin and Theodore Potsis and Dongxue Zhan and Xue Liu and Ted Stahopoulos and Liangzhu Leon Wang},
  year={2026},
  eprint={2503.19708},
  archivePrefix={arXiv},
  primaryClass={physics.flu-dyn},
  url={https://arxiv.org/abs/2503.19708},
}

@misc{qin2026geowind2planmissiontime3durban,
  title={GeoWind2Plan: Mission-Time 3D Urban Wind Prediction for Energy-Efficient UAV Planning},
  author={Shaoxiang Qin and Yucheng Zhao and Fuyuan Lyu and Di Zhou and Jiachen Yao and Xue Liu and Anima Anandkumar and Liangzhu Leon Wang and Xiongye Xiao},
  year={2026},
  eprint={2609.36056},
  archivePrefix={arXiv},
  primaryClass={cs.AI},
  url={https://arxiv.org/abs/2609.36056},
}