Benchmark DAS Datasets
DAS features are experimental
The SeisBench DAS API is currently experimental and some features, such as more datasets and models, will be added in the future.
SeisBench facilitates the downloading of a suite of publicly available DAS datasets for training machine learning models. An overview of the contents of each dataset is below, along with the corresponding citation.
RandomDASDataset
The RandomDASDataset contains, as the name might already suggest, random data.
This data is in no way physically meaningful. However, it has the correct format of a SeisBench DAS data set an contains
the keys and structures you’d expect, including (completely random) P and S pick labels. It is intended exclusively for
testing purposes.
MLSubDAS
The MLSubDAS contains 1236 DAS records from cables predominantly in Alaska, Chile,
and Japan. The records have been annotated semi-automatically: first, each individual trace has been picked using
classical three-component models. Afterwards, picks have been corrected manually. Nonetheless, the dataset has clear
imperfections: pick fronts are often incomplete, sometimes inconsistent, and can be missing completely, in particular,
for P arrivals. In addition, pick residuals in excess of 1 s occur. This should be kept in mind when using this dataset.
Warning
Dataset size: ~545Gb
Citation
Xiao, H., Tilmann, F., van den Ende, M., Rivet, D., Loureiro, A., Tsuji, T., … & Denolle, M. A. (2026). DeepSubDAS: an earthquake phase picker from submarine distributed acoustic sensing data. Geophysical Journal International, 245(2), ggag061.