A Deep Learning Framework for the Automatic Classification of Microseisms at Cotopaxi Volcano

This paper presents a comprehensive framework for an automated classification system of Long Period (LP) and Volcano Tectonic (VT) seismic events from the Cotopaxi volcano, employing two prominent Deep Learning models: Stacked Autoencoders and Deep Neural Networks. The dataset was provided by the Instituto Geofísico de la Escuela Politécnica Nacional (IGEPN), which consists of 1,044 LP events and 101 VT events. To enhance the model’s performance, data augmentation techniques were applied, generating synthetic microseismic data using a Conditional Generative Adversarial Network (CGAN), ultimately expanding the dataset to 10,000 microseisms, comprising both real and synthetic samples. In the feature extraction stage, we calculated the Power Spectral Density (PSD) using Welch’s method and a 512-point Fast Fourier Transform (FFT), resulting in 257 frequency-based features. Additionally, wavelet coefficients were obtained via Discrete Wavelet Transform (DWT) using the Symlet wavelet with six levels of decomposition. PSD was also computed on these wavelet coefficients, producing 257 additional frequency-scale features. The enriched feature set was used to train the Deep Learning models, with performance evaluated using standard classification metrics. The Stacked Autoencoder model outperformed others, achieving an accuracy of 99.73% and a balanced error rate (BER) of 0.0027, corresponding to just 3 errors per 1,000 classified microseisms. These results substantially exceed the geophysical accuracy requirement of a BER below 0.01. This paper demonstrates the efficacy of Deep Learning models in distinguishing between LP and VT events, offering new opportunities for enhancing volcanic monitoring and hazard assessment strategies.