Flow-Alert

📢 Welcome to Flow-Alert

If you're interested in leveraging machine learning and seismic signals for channelized flow early warning,
including, but not limited to, debris flows, glacial lake outburst floods (GLOFs), and lahars,
you've come to the right place!
Check out our repository to get started.

🛠️ 0. Major Changes for v1.3

Compared with previous versions, including:
version 1.0 (https://doi.org/10.5281/zenodo.15020368),
version 1.1 (https://doi.org/10.5281/zenodo.16811121),
version 1.2 (https://doi.org/10.5281/zenodo.16893616),
the latest version 1.3 (https://doi.org/10.5281/zenodo.18324322) includes the following major changes:

(1) Data: The debris flow events on 2019-10-09 and 2019-10-15, recorded at the ILL12 station, are now included in the training dataset.
These events were not used in previous versions because the ILL18 station was unavailable.
Earlier versions relied on a network of stations (ILL12, ILL13, and ILL18) for warning,
whereas the latest version focuses on single-station detection and classification.

(2) Labels: Previous versions used manually labeled event timestamps,
while the latest version employs STA/LTA-based event times,
which are theoretically more objective.
Please check here for details.

(3) Features: Previous versions used all 70+ available seismic features,
whereas the latest version selects 12 seismic features to train the model.
Please check feature_type_H for details.

(4) Model Structure: This version integrates an attention mechanism layer after the LSTM,
which is expected to better capture temporal dependencies in the seismic signals.

📁 1. Repository Structure

Flow-Alert
├── calculate_features   # Convert raw seismic data into features
├── config               # Configuration files
├── data                 # Seismic data and extracted features
├── demo                 # Examples of running the case
├── docs                 # Documentation for users
├── functions            # Core functions and scripts
├── pipeline             # Train and test code
├── trained_model        # Pre-trained models

🚀 2. How to Use Our Pre-trained Models on Your Data?

To get started:
2.1 Check the prepare_env.md for setting up the Python environment in your local PC
2.2 Check the tutorial for usage
2.3 Run the tutorial on your local PC or on

🐛 3. Found a Bug?

Feel free to open a Pull request, or reach out to us via email.

❓️4. Have Questions?

4.1 Start by reading our related paper
Qi Zhou, Hui Tang, Clément Hibert, Małgorzata Chmiel, Fabian Walter, Michael Dietze, and Jens M Turowski.
"Enhancing debris flow warning via machine learning feature reduction and model selection."
Journal of Geophysical Research: Earth Surface, 129, e2024JF008094.
Click here for the manuscript

Qi Zhou, Hui Tang, Michael Dietze, Fabian Walter, Dongri Song, Yan Yan, Shuai Li, and Jens M. Turowski.
"Similarity of Debris Flows in Seismic Records."
Click here for the preprint

If you still have questions, feel free to contact the project contributors.

4.2 Or reach out to our research groups
Hazards and Surface Processes Research Group and Digital Earth Lab
Led by Dr. Hui Tang
Physical Earth Surface Modelling Lab
Led by Dr. Jens Turowski.

💪 5. Contributors

Qi Zhou
qi.zhou@gfz.de or qi.zhou.geo@gmail.com

Kshitij Kar
kshitij.kar@gfz-potsdam.de