DeepLearning_Infrasound

Towards real-time assessment of infrasound event detection capability using deep learning-based transmission loss estimation
Associated publication: Cameijo, A., Le Pichon, A., Sklab, Y., Souhila, A., Brissaud, Q., Näsholm, S. P., ... & Aknine, S. (2025), Authorea Preprints.

---

Project Overview

DeepLearning_Infrasound provides a deep learning model that predicts ground-level infrasound transmission loss (TL) over distances up to 4,000 km based on realistic range-dependen atmospheric conditions (combining horizontal wind speed, temperature and small-scale disturbance fields).

It enables:

• Near real-time assessment of infrasound detection capabilities at a global scale.
• Faster studies of regional events such as volcanic eruptions or atmospheric explosions.

This repository contains the pre-trained model, data preprocessing scripts, and example codes to easily run predictions, even without deep AI expertise.

---

Repository Structure

DeepLearning_Infrasound/
├── cnn_gru/
│   ├── main.py               # Script to test the pre-trained model
│   ├── quick_test.ipynb      # Jupyter Notebook for a simple demonstration
│   ├── requirements.txt      # List of required Python packages
│   ├── to_load/              # Folder containing pre-trained model weights
├── Data/
│   ├── Preprocessing/
│   │   ├── inputs.py          # Script to prepare atmospheric input data
│   │   ├── outputs.py         # Script to prepare simulation target data
├── CITATION.cff               # Citation file
├── README.md                  # This documentation

---

Requirements

• Python 3.10.6
• TensorFlow 2.8.3
• Keras 2.8.0
• CUDA 11.7 and cuDNN 8.6.0 (optional for GPU acceleration)

Install the required packages:

pip install -r cnn_gru/requirements.txt

Tip: It is recommended to use a virtual environment (python -m venv env) to avoid conflicts.

---

Quick Start: Using the Model

1. Clone the repository

git clone https://github.com/your-repo/DeepLearning_Infrasound.git
cd DeepLearning_Infrasound/cnn_gru

2. Run a quick test with the pre-trained model

python main.py

This will generate a prediction using the provided demo atmospheric input.

3. Explore via Jupyter Notebook (optional)

jupyter notebook quick_test.ipynb

The notebook will guide you through loading the model, preparing the inputs, and visualizing a ground-level transmission loss map.

---

Using Your Own Data

If you want to use your own atmospheric profiles:

1. Prepare atmospheric input files

python Data/Preprocessing/inputs.py --input your_data_file.nc

2. Prepare simulated output targets

python Data/Preprocessing/outputs.py

Make sure your input data follows the expected format described in the preprocessing scripts.

---

Associated Publication

For scientific background, methodology, and validation:

Towards real-time assessment of infrasound event detection capability using deep learning-based transmission loss estimation
Cameijo, A., Le Pichon, A., Sklab, Y., Souhila, A., Brissaud, Q., Näsholm, S. P., ... & Aknine, S. (2025), Authorea Preprints.

and:

Predicting infrasound transmission loss using deep learning Brissaud, Q., Näsholm, S. P., Turquet, A., & Le Pichon, A. (2023), Geophysical Journal International, 232(1), 274-286.

---

External Data Sources

• Atmospheric Data: WACCM6 Model – NCAR/UCAR
• Parabolic Equation Simulation: ePape Model – NCPA, University of Mississippi

---

Citation

If you use this repository or any results from it, please cite as follows:

@article{cameijo2025,
  title={Towards real-time assessment of infrasound event detection capability using deep learning-based transmission loss estimation},
  author={Cameijo, A., Le Pichon, A., Sklab, Y., Souhila, A., Brissaud, Q., Näsholm, S. P., ... & Aknine, S. (2025)},
  journal={Authorea Preprints},
  year={2025}
}

---

Contact

For any questions or feedback, feel free to reach out:
📧 alice.cameijo@cea.fr

---

Enjoy exploring infrasound propagation with AI!