Create synthetic seismic volumes with labeled meandering channels, tributary channel networks, and submarine canyons.
See our ESSD paper for details:
cigChannel: a large-scale 3D seismic dataset with labeled paleochannels for advancing deep learning in seismic interpretation
The cigChannel dataset, containing 1600 seismic volumes, is available here
- Clone this repository to your device:
git clone https://github.com/wanggy-1/cigChannel.git - Install all the required packages by running the following command:
conda env create -f env.yml
or prepare an environment with python >= 3.10, and then run the following command:
pip install -r requirements.txt
We have provided demonstration codes for your refenrence to create your own seismic volumes.
Run python meandering_channel.py, and you will get a seismic impedance model, a seismic volume, and a channel label volume.
You can also create multiple seismic volumes in parallel following these steps:
- Download the zip file of channel distance maps from here, which contains thousands of prefabricated meandering channels.
- Unzip it to
./Distmap. - Run
python meandering_channel_parallel.py. It accelerates the process of data generation by creating meandering channels from those distance maps.
Download the zip file of channel topography maps from here, which contains thousands of prefabricated tributary channel networks.
Unzip it to ./Topography.
Run python tributary_channel.py. It creates tributary channel networks from the topography maps.
You will get a seismic impedance model, a seismic volume, and a channel label volume.
You can also create multiple seismic volumes in parallel by running python tributary_channel_parallel.py
Run python submarine_canyon.py, and you will get a sedimentary facies model, a seismic impedance model, a seismic volume, and a channel label volume.
In the sedimentary facies model, 2 represents point-bars (yellow), 3 represents natural levees (dark gold), 4 represents abandoned meanders (saddle brown), and 0 represents the background (white).
You can also create multiple seismic volumes in parallel by running python submarine_channel_parallel.py
Download the zip file of channel distance maps and channel topography maps.
Unzip them to ./Distmap and ./Topography.
Run python assorted_channel.py, and you will get a sedimentary facies model, a seismic impedance model, a seismic volume, and a channel label volume.
In the sedimentary facies model, 1 represents channel lag deposits (green), 2 represents point-bars (yellow), 3 represents natural levees (dark gold), 4 represents abandoned meanders (saddle brown), and 0 represents the background (white).
You can also create multiple seismic volumes in parallel by running python assorted_channel_parallel.py
- Install the environment by running the following command:
conda env create -f unet/torchenv.yml - Download the trained UNet models: UNet_meander.pth, UNet_tributary.pth, and UNet_submarine.pth, save them to
./checkpoints. - Download field seismic volumes to
./fieldseismic. - Run
python unet/apply.py
The 3D volumes are visualized by cigvis, an outstanding open-source toolbox developed by the Computational Interpretation Group at USTC.
If you find this work useful in your research and want to cite it, please consider use this:
@article{wang2025cigchannel,
title={cigChannel: a large-scale 3D seismic dataset with labeled paleochannels for advancing deep learning in seismic interpretation},
author={Wang, Guangyu and Wu, Xinming and Zhang, Wen},
journal={Earth System Science Data},
volume={17},
number={7},
pages={3447--3471},
year={2025},
publisher={Copernicus Publications G{\"o}ttingen, Germany}
}
