We propose a deep learning method, which only requires the synthetic data and a few real noise to train the designed network, can denoise DAS data sets from the FORGE and SAFOD projects. This work can uncover very weak signals from the highly strong noise, and thus enhance earthquake detection using the STA/LTA method.
We are in the process of organizing the data sets and trained models and are ready to upload them to Github. Then, the results in the paper can be easily reproduced.
Uploading data to GitHub is tough — I have to rely on mobile data (hotspot), and it’s pretty costly.
The raw FORGE DAS data are freely available in the GitHub repository: https://github.com/chenyk1990/dasdenoising-dataonly
The catalogs for earthquakes in FORGE data are freely available in the GitHub repository: https://github.com/ariellellouch/FORGE
The raw SAFOD DAS data are freely available in the GitHub repository: https://github.com/ariellellouch/DASDetection and https://github.com/chenyk1990/dasmrrcoh-dataonly