HDAS Data from La Palma - DigiVolCan Project

This repository contains various co-eruptive VT datasets collected during the 2021 eruption at La Palma, Spain, by an underwater High-fidelity Distributed Acoustic Sensing (HDAS). These datasets have been used to train and test the RNN-DAS model, a deep learning framework designed for volcano-seismic event detection using DAS data.

This data was collected as part of the DigiVolCan project, which is a collaboration between the University of Granada, the Canary Islands Volcanological Institute (INVOLCAN), the Institute of Technological and Renewable Energies (ITER), the University of La Laguna, and Aragón Photonics. It is funded by the Ministry of Science, Innovation, and Universities / State Research Agency (MICIU/AEI) of Spain and the European Union through the Recovery, Transformation, and Resilience Plan, Next Generation EU Funds. The project reference is PLEC2022-009271, funded by MICIU/AEI /10.13039/501100011033 and by the European Union Next GenerationEU/PRTR.

Dataset Description

The shared dataset contains HDAS data recorded over several periods, with one file per minute. Each file is in .h5 format and follows the structure:

file_path
│
└───"data" (dataset)
    │
    ├───data (2D matrix of strain rate)
    │    └───[channels x time_samples]
    │
    ├───attrs
         │
         ├───"dt_s" (temporal sampling in seconds)
         ├───"dx_m" (spatial sampling in meters)
         └───"begin_time" (start date in 'YYYY-MM-DDTHH:MM:SS.SSS' format)

Five datasets are provided as separate compressed .zip archives due to their size. Each archive contains DAS waveform data in the HDF5 (.h5) format described above, organized in one-minute files. These datasets correspond to figures presented in the RNN-DAS model article and are intended to facilitate reproducibility and further analysis.

Dataset 1 – Main event and aftershocks (Figure 5)

This dataset contains a one-hour DAS recording from November 30, between 07:00 and 08:00 UTC, featuring a main seismic event with magnitude Ml = 3.22 along with several aftershocks.

Dataset 2 – Continuous Test Segment (Figure 6)

This dataset contains one hour of continuous DAS recordings from October 29, between 04:00 and 05:00 UTC. 

Dataset 3 – Events with Varying SNR and Magnitude (Figure 4)
This dataset includes three separate 3-minute DAS recordings, each corresponding to a different seismic event with distinct characteristics. The selected events represent a range of conditions, including attenuated signals, low signal-to-noise ratio (SNR), and nearby high-SNR events. 

Dataset 4 – High-Magnitude Event Example (Figure 3)
This dataset contains a 3-minute DAS recording corresponding to a seismic event with magnitude Ml = 4.23. This example demonstrates the model’s response to a clear, high-magnitude event.

Dataset 5 – Moderate Events and Noise-Only Sample (Figure 7)
This dataset includes four separate DAS recordings: three corresponding to moderate seismic events and another containing only seismic noise. 

All .zip archives can be easily decompressed and used directly.

Note: The full HDAS dataset from La Palma used for model training and evaluation is not included due to its large size. It is available upon request from the corresponding author.

RNN-DAS Model

The RNN-DAS model is an innovative Deep Learning model based on Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM) cells, developed for real-time Volcano-seismic Signal Recognition (VSR) using Distributed Acoustic Sensing (DAS) measurements. The model was trained on a comprehensive dataset of Volcano-Tectonic (VT) events from the 2021 La Palma eruption, recorded by a High-fidelity submarine Distributed Acoustic Sensing array (HDAS) located near the eruption site.

RNN-DAS can detect VT events, track their temporal evolution, and classify their waveforms with approximately 97% accuracy when tested on a database of over 2 million unique strain waveforms, enabling real-time continuous data predictions. The model has demonstrated excellent generalization capabilities for different time intervals and volcanoes, facilitating continuous, real-time seismic monitoring with minimal computational resources and retraining requirements.

The model is available in the RNN-DAS GitHub repository:

https://github.com/Javier-FernandezCarabantes/RNN-DAS 

Fernández-Carabantes, J., Titos, M., D'Auria, L., García, J., García, L., & Benítez, C. (2025). Javier-FernandezCarabantes/RNN-DAS: RNN-DAS v1.1.1 (v1.1.1). Zenodo. https://doi.org/10.5281/zenodo.15858492

A copy of the repository is also provided here as the RNN-DAS_main.zip file. This archive mirrors the contents of the GitHub repository at the time of submission (v1.0.0). For correct usage, it is recommended to read the included README file. Users are encouraged to refer to the GitHub repository for future updates or changes.

Usage

This dataset is provided as a sample data for the RNN-DAS model. It can be used to test and validate our model, as well as for the development of other machine learning approaches.

Citation

If you use this dataset in your research or if you use the RNN-DAS model, proper citation of the related article and this dataset are needed (Fernández-Carabantes et al., 2025)

Fernández-Carabantes, J., Titos, M., D'Auria, L., García, J., García, L., & Benítez, C. (2025). RNN-DAS: A new deep learning approach for detection and real-time monitoring of volcano-tectonic events using distributed acoustic sensing. Journal of Geophysical Research: Solid Earth, 130, e2025JB031756. https://doi.org/10.1029/2025JB031756

For further details, please refer to the project documentation or contact the research team (corresponding author email: javierfyc@ugr.es).