Unrest Detection Using Machine Learning Techniques on Seismic Data: A Case Study of Vulcano Island, Italy

Many volcanoes worldwide remain dormant, exhibiting mild to weak hydrothermal fuming activity. Detecting early signs of reactivation, particularly for those volcanoes near densely populated or touristic areas, presents a significant challenge in volcanology. Periods of unrest often involve complex interactions between magmatic and hydrothermal systems, obscuring clear eruption precursors.

 

In mid-September 2021, Vulcano Island, Italy, experienced significant degassing episodes at the La Fossa cone. Despite the unrest, no phreatic or phreatomagmatic eruptions occurred. This makes Volcano an ideal case study for applying advanced machine learning techniques to enhance the accuracy and timeliness of unrest detection and to understand the factors that prevented an eruption. This study aims to develop and evaluate machine learning models for detecting signs of volcanic unrest using seismic data. 

 

Our approach utilizes continuously recorded seismic data to capture the intricate precursors to volcanic eruptions. We explore various machine learning approaches, including supervised and unsupervised methods, to identify patterns and correlations indicative of volcanic unrest. These models undergo training and validation using historical data on Vulcano Island, ensuring their applicability in real-time monitoring scenarios. 

 

This study aims to improve the detection of volcanic unrest on Vulcano Island and our understanding of the precursors to eruptions, including the conditions that may inhibit eruption despite significant unrest.