Datasets:

tmarkmann
/

dataset-rbc-fno

Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/hdf5/hdf5.py", line 49, in _split_generators
                  import h5py
              ModuleNotFoundError: No module named 'h5py'
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 66, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ~~~~~~~~~~~~~~~~~~~~~~~^
                      path=dataset,
                      ^^^^^^^^^^^^^
                      config_name=config,
                      ^^^^^^^^^^^^^^^^^^^
                      token=hf_token,
                      ^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                      path,
                  ...<6 lines>...
                      **config_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Rayleigh-Benard Convection (RBC) Dataset

This dataset contains Direct Numerical Simulation (DNS) data of Rayleigh-Benard convection in two and three dimensions, generated using Oceananigans.jl. It accompanies the paper Fourier neural operators as data-driven surrogates for two- and three-dimensional Rayleigh-Benard convection, published in Neurocomputing (2026).

Dataset Description

Rayleigh-Benard Convection describes convection in a fluid layer heated from below and cooled from above. The dataset provides time-series trajectories of the full flow state (velocity, temperature, pressure) at different Rayleigh numbers, suitable for training data-driven surrogate models.

Dataset Structure

2D/
  train/   # 50 episodes per Ra (7 Ra values)
  val/     # 10 episodes per Ra
  test/    # 20 episodes per Ra
3D/
  train/   # 50 episodes at Ra=2500
  val/     # 20 episodes
  test/    # 20 episodes

Each file is named ra{value}.h5 and stored in HDF5 format with the following datasets per episode i:

Key	2D Shape	3D Shape	Description
`states{i}`	`(1000, 5, 64, 96)`	`(1000, 6, 32, 48, 48)`	Flow state: velocity + temperature + pressure
`actions{i}`	`(1000, 12)`	`(1000, 8, 8)`	Boundary heating actions
`nusselts{i}`	`(1000,)`	`(1000,)`	Nusselt number (convective heat flux)

State channels

2D (5 channels): u, v, T, p_hyd, p_non (horizontal velocity, vertical velocity, temperature, hydrostatic pressure, non-hydrostatic pressure)
3D (6 channels): u, v, w, T, p_hyd, p_non (3 velocity components, temperature, hydrostatic pressure, non-hydrostatic pressure)

The last two channels (pressure) are typically excluded during surrogate model training.

Simulation Parameters

Parameter	2D	3D
Domain D	2pi x 2	4pi x 4pi x 2
Grid N	96 x 64	48 x 48 x 32
Ra	{1e4, 3e4, 1e5, 3e5, 1e6, 3e6, 1e7}	2500
Pr	0.7	0.7
(T_C, T_H)	(1, 2)	(0, 1)
dt	0.03	0.01
Timesteps per episode	1000	1000

Usage

import h5py

with h5py.File("2D/train/ra300000.h5", "r") as f:
    # Load episode 0 states: (1000, 5, 64, 96)
    states = f["states0"][:]

    # Velocity and temperature (exclude pressure)
    uvT = states[:, :3]  # (1000, 3, 64, 96)

For use with the training code, see the RBC-FNO-Surrogate repository.

Citation

@article{MARKMANN2026133201,
  title = {Fourier neural operators as data-driven surrogates
  for two- and three-dimensional Rayleigh-Benard convection},
  journal = {Neurocomputing},
  volume = {679},
  pages = {133201},
  year = {2026},
  issn = {0925-2312},
  doi = {https://doi.org/10.1016/j.neucom.2026.133201},
  author = {Thorben Markmann and Michiel Straat and Sebastian
  Peitz and Barbara Hammer},
}

Downloads last month: 7

Total file size:

207 GB