RNN-DAS: A New Deep Learning Approach for Detection and Real-Time Monitoring of Volcano-Tectonic Events Using Distributed Acoustic Sensing

Abstract

In this article, we present a novel Deep Learning model based on Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM) cells, designed as a real-time Volcano-seismic Signal Recognition (VSR) system for Distributed Acoustic Sensing (DAS) measurements. The model was trained on an extensive database of Volcano-Tectonic (VT) events derived from the co-eruptive seismicity of the 2021 La Palma eruption, recorded by a High-fidelity submarine Distributed Acoustic Sensing array (HDAS) near the eruption site. The features used for supervised model training, based on signal energy average in frequency bands, effectively enable the model to leverage spatial contextual information and the temporal evolution of volcano-seismic signals provided by the DAS technique. The proposed model not only detects the presence of VT events but also analyzes their temporal evolution, selecting and classifying their complete waveforms with an accuracy of approximately 97% for correctly detected and classified VT events. Furthermore, the model has demonstrated robust performance in generalizing to other time intervals and volcanoes, enabling continuous real-time monitoring of seismicity. Such results highlight the potential of using RNN-based approaches with LSTM cells for application to other active volcanoes, enabling fast, automatic analysis with low computational requirements and the need of minimal retraining, for the creation of labeled seismic catalogs directly from DAS measurements. This represents a significant advancement in the use of DAS technology as a viable tool to study active volcanoes and their seismic activity.