DeepStrain

This repository implements a three-step pipeline for the detection and clustering of strain transients using borehole strainmeter data, wavelet transforms, and deep learning (AutoencoderZ).
The algorithm is developed for the detection of Slow Slip Events (SSEs).

---

Method Overview

The workflow consists of three main steps, with corresponding code organized in the scr folder:

1. Preprocessing (folder: 1-Preprocessing)

   • Data download and conversion to strain using EarthScope tools
   • Corrections: trend, tides, and atmospheric pressure

2. Input Creation (folder: 2-Wavelet_transform)

   • Wavelet transform of corrected regional strain components
   • Downsampling and normalization for machine learning input

3. Deep Clustering (folder: 3-Clustering_autoencoderZ)

   • Feature extraction using AutoencoderZ
   • K-means clustering applied to latent features

---

Workflow Figure

(The figure above illustrates the three main steps of the pipeline.)

---

Requirements

• Python ≥ 3.8
• NumPy
• Pandas
• Matplotlib
• PyWavelets (pywt)
• TensorFlow (for AutoencoderZ)
• earthscopestraintools

(See individual subfolders and repositories for more details.)

---

Contributions

Contributions are welcome. Please open an issue or submit a pull request if you would like to extend or improve the functionality.

---

Contact

Developer: Zahra Zali
Email: zali@gfz.de

---

Citations and Links

When using this model, please cite the following works:

1. Zali, Z., Martínez-Garzón, P., Mencin, D., Beroza, G.
   Slow slip modulates low-frequency seismicity on the San Andreas Fault
   https://doi.org/10.21203/rs.3.rs-7317566/v1

2. Zali, Z., Martínez-Garzón, P., Kwiatek, G., Núñez-Jara, S., Beroza, G., Cotton, F., Bohnhoff, M.
   Low-Frequency Tremor-Like Episodes Before the 2023 MW 7.8 Türkiye Earthquake Linked to Cement Quarrying.
   Scientific Reports 15, 6354 (2025).
   https://doi.org/10.1038/s41598-025-88381-x

3. Zali, Z., Mousavi, S. M., Ohrnberger, M., Eibl, E. P., & Cotton, F.
   Tremor clustering reveals pre-eruptive signals and evolution of the 2021 Geldingadalir eruption of the Fagradalsfjall Fires, Iceland.
   Communications Earth & Environment 5, 1 (2024).
   https://doi.org/10.1038/s43247-023-01166-w

---

Notes

• The clustering step uses AutoencoderZ (improved deep autoencoder).
• For the end-to-end clustering pipeline, also see ClusTremor.

---

License

This project is licensed under the MIT License.