camlab-ethz/NS-Sines

Short Description

Based on the incompressible Navier-Stokes equations, this dataset contains trajectories starting from sinusoidal initial conditions of the velocities, see https://arxiv.org/abs/2405.19101. It also contains a passive tracer that is carried by the flow.

Dimensions

The assembled NetCDF file has a single variable called velocity with dimensionality

• 20000 (number of trajectories)
• 21 (time steps)
• 3 (horizontal velocity, vertical velocity, passive tracer)
• 128 (x-dim)
• 128 (y-dim)

It was simulated on the unit square up to T=1 and saved as uniformly spaced in space and time.

Train/Val/Test-split

19640/120/240 trajectories

Download & Assembly

The dataset can be downloaded, e.g., via huggingface-cli download.

After download, the chunked data can be assembled into a single NetCDF file using the provided assemble_data.py script. Use it as follows:

python assemble_data.py --input_dir . --output_file NS-Sines.nc