UrbanFlow-oneObstacle-NoTrip

3D and 2D velocity field data from a Nek5000 spectral-element CFD simulation of urban flow past a single obstacle, without a turbulence trip. Volumes/planes were interpolated from the simulation's spectral-element mesh onto a uniform Cartesian grid in post-processing (postprocess-urbanflow).

Files

3d/ — full 3D velocity fields (turbulent fluctuations, mean subtracted)

The original Train (T=033-233, 40000 snapshots) and Test (T=233-283, 10000 snapshots) files share identical means/x/y/z and are time-contiguous, so they are combined into one continuous 50000-snapshot sequence and split into 8 shards by snapshot index (Hugging Face caps individual files at 500GB; the combined data is ~2.65TB). Each shard is self-contained.

File	Snapshots	Grid (x, y, z)	Time range	Size
combined-3d-ds1-fluctuations-snap-00000-06249.h5	6250	288 x 88 x 88	T = 33.00-64.24	~335 GB
combined-3d-ds1-fluctuations-snap-06250-12499.h5	6250	288 x 88 x 88	T = 64.25-95.49	~335 GB
combined-3d-ds1-fluctuations-snap-12500-18749.h5	6250	288 x 88 x 88	T = 95.50-126.74	~335 GB
combined-3d-ds1-fluctuations-snap-18750-24999.h5	6250	288 x 88 x 88	T = 126.75-157.99	~335 GB
combined-3d-ds1-fluctuations-snap-25000-31249.h5	6250	288 x 88 x 88	T = 158.00-189.24	~335 GB
combined-3d-ds1-fluctuations-snap-31250-37499.h5	6250	288 x 88 x 88	T = 189.25-220.49	~335 GB
combined-3d-ds1-fluctuations-snap-37500-43749.h5	6250	288 x 88 x 88	T = 220.50-251.74	~335 GB
combined-3d-ds1-fluctuations-snap-43750-49999.h5	6250	288 x 88 x 88	T = 251.75-272.99	~335 GB

Each shard contains:

• U, V, W: velocity fluctuation components, shape (6250, 288, 88, 88), float64
• means/U, means/V, means/W: mean velocity fields subtracted to produce the fluctuations, shape (288, 88, 88) (identical across all shards)
• x, y, z: grid coordinates (identical across all shards)
• t: snapshot timestamps for this shard's range
• fluctuations: boolean flag (always True) indicating the mean has been removed

To reconstruct the full sequence, concatenate U/V/W/t across shards 0-7 in order.

2d/ — 2D velocity midplane (z-index 44) extracted from the same simulation

File	Snapshots	Grid (x, y)	Time range
2d_plane_z44_T_033_233.h5	40000	288 x 88	T = 033-233
2d_plane_z44_T_233_283.h5	10000	288 x 88	T = 233-283

Each file contains U, V, W velocity components for the extracted midplane (raw, not fluctuations).

Loading the data

See load_example.py for a minimal script that inspects an HDF5 file's structure and reads a slice. Note each 3D shard is ~335 GB — always slice (f["U"][0]), never load a full array (f["U"][:]) into memory.

python load_example.py 2d/2d_plane_z44_T_033_233.h5

Visualization

Side-view animation of the flow field, T = 033-233 (downsampled, ds_ratio=1):

Source

Generated from Nek5000 simulations in the UrbanFlow_oneObstacleNOTRIP project (data-gen-obstacleNOTRIP: single obstacle, no turbulence trip) and post-processed with postprocess-urbanflow.

Code

Simulation and post-processing code, plus visualizations, will be released in a companion GitHub repository (coming soon — link will be added here).