A Recurrent Neural Network Approach for Real-Time Detection and Monitoring of Volcano-Tectonic Events Using DAS

This work introduces a deep learning framework based on recurrent neural networks (RNNs) developed for real-time recognition of volcano-seismic signals from distributed acoustic sensing (DAS) data. The model was developed using a large dataset of volcano-tectonic events associated with the 2021 La Palma eruption, captured by a high-resolution submarine DAS array deployed close to the volcanic source. For training phase, we employed features derived from the signal energy across different frequency bands and spatial points, allowing the model to effectively exploit both spatial and temporal patterns inherent in seismo-volcanic signals. The proposed approach is capable not only of detecting volcano-tectonic events but also of characterizing their temporal behavior, identifying and classifying complete waveforms with an accuracy close to 97%. In addition, the model exhibits strong generalization capabilities across different time periods and volcanic settings. The results showed fast and automatic analysis with relatively low computational cost and limited retraining, enabling continuous real-time seismic monitoring and supporting the automatic generation of labeled seismic catalogs directly from DAS data, representing a significant step forward in the application of DAS technology for studying active volcanoes and their seismic activity.