This study explores wavefield reconstruction using machine learning methods for data compression and wavefield separation. We test various architectures to treat DAS data as two-dimensional arrays, including the Implicit Neural Representation (INR) models and the SHallow REcurrent Decoder (SHRED) model.
This repository provides independent notebook examples of model training and inference performed in the manuscript. All codes are implemented using PyTorch.
Notebooks of SHRED model inference and training on the Cook Inlet DAS data are available. The model training notebook is available. See below for instructions of getting the training data.
- Random Fourier Feature Network (RFFN, Tancik et al., 2020): notebooks/training_RFFN_KKFLS.ipynb
- Sinusoidal Representation Network (SIREN, Sitzmann et al., 2020): notebooks/training_SIREN_KKFLS.ipynb
The earthquake data from the Cook Inlet DAS experiment are hosted here. Earthquakes and daily data reports are updated daily.
Due to the size of the data used in this study (~260 GB per cable), they are not uploaded directly in this repository. However, a Python script is available to download these data from our archival server. Please refer to the script download.py and list of events event_list.csv.
Ni, Y., Denolle, M. A., Shi, Q., Lipovsky, B. P., Pan, S., & Kutz, J. N. (2024). Wavefield Reconstruction of Distributed Acoustic Sensing: Lossy Compression, Wavefield Separation, and Edge Computing. Journal of Geophysical Research: Machine Learning and Computation, 1(3), e2024JH000247. 10.1029/2024JH000247
@article{ni2024wavefield,
title={Wavefield reconstruction of distributed acoustic sensing: Lossy compression, wavefield separation, and edge computing},
author={Ni, Yiyu and Denolle, Marine A and Shi, Qibin and Lipovsky, Bradley P and Pan, Shaowu and Kutz, J Nathan},
journal={Journal of Geophysical Research: Machine Learning and Computation},
volume={1},
number={3},
pages={e2024JH000247},
year={2024},
publisher={Wiley Online Library},
doi={10.1029/2024JH000247}
}