This quake does not exist

Generative modelling of seismic waveforms using denoising diffusion.

About

This repository implements Generative Waveform Models (GMWs), i.e., generative models that can generate synthetic seismic waveforms. In particular, the repository implements HighFEM, the GWM introduced in High Resolution Seismic Waveform Generation using Denoising Diffusion.

The repository can be also be used to replicate the results from the manuscript using the experimental code in experiments, and for training custom GWMs from scratch.

Quick start

If you are only interested in generating synthetic waveforms, install tqdne using

pip install git+https://github.com/highfem/tqdne@<RELEASE>

where RELEASE should be replaced with the latest release version to be found here. You should then be able to simulate waveforms by calling

generate-waveforms

from the (Unix) command line.

Installation

If you are interested in reproducing the results from the paper, please refer to experiments/README.md.

You can install tqdne by following the steps below.

1. a) Recommended: Download the latest release if you do not require commit history. Releases have been tested and reproduced by us and partners.

   b) Alternatively, clone the repository using:

   git clone (--depth 1) https://github.com/highfem/tqdne.git

2. There are two ways to install the package: you can either install the package with uv which will use the versions of all dependencies that we are using for development, or you install the package using the pip installer which will install the latest versions of all dependencies.

   a) Recommended: Install all dependencies and the package with uv, via

   uv sync

   b) Install using pip install, for instance, within a conda or virtualenv environment, via:

   pip install -e .

Experiments

To reproduce the experiments from the manuscript, including data preprocessing, training, and evaluation, navigate to the experiments folder. Refer to the corresponding README files for step-by-step guidance.

Sampling waveforms

You can generate your own waveforms by using the scripts in scripts. See the corresponding README files for more information.

Contributing

Contributions in the form of pull requests are more than welcome. In order to contribute:

1. Clone tqdne and install uv from here.
2. Create a new branch locally git checkout -b feature/my-new-feature or git checkout -b issue/fixes-bug.
3. Install all dependencies using: uv sync --all-extras.
4. Activate the virtual environment: source .venv/bin/activate.
5. Install pre-commit (which we use for auto-formatting and checking code) using: pre-commit install.
6. Implement your contribution.
7. Run uv run ruff check . to check the code for issues.
8. Push your changes and submit a PR 🙂.

Acknowledgements

Some Python code has been adapted from the following repositories:

• EDM
• Consistency Models