Date of Award
2021
Document Type
Open Access Master's Thesis
Degree Name
Master of Science in Geology (MS)
Administrative Home Department
Department of Geological and Mining Engineering and Sciences
Advisor 1
Gregory Waite
Committee Member 1
Simon Carn
Committee Member 2
Roohollah Askari
Abstract
Volcan de Fuego is an active stratovolcano located in the Central Guatemalan segment of the 1100 m long Central America Volcanic Arc System (CAVAS). Fuego-Acatenango massif consists of at least four major vents of which the Fuego summit vent is the most active and the youngest member. The volcano exhibits primarily Strombolian and Vulcanian behavior along with occasional paroxysms and pyroclastic flows. Historically, Fuego has produced basaltic-andesitic rocks with more recent eruptions progressively trending towards maficity. Several studies have used short-term deployments of broadband seismometers, infrasound, and long-term remote sensing techniques to characterize the mechanism of Fuego. In our study, we analyze the tilt derived from transient broadband seismometers and tiltmeter stationed over several days during 2009, 2012, and 2015 near the summit crater using unsupervised learning.
Unsupervised learning has the potential to play a significant role in monitoring volcanoes dominated by large, unlabeled datasets. In our study, we make use of dynamic time warping distance measure along with unsupervised classification methods to identify precursory tilt signals. The unsupervised classification revealed two types of tilt signals with opposite polarity, one of which confirms features identified in previous studies while the other signal has been previously unknown. Template matching implemented with the known signal identified 268 events between October 1, 2015, and January 13, 2016, the duration of which varied between 7 and 39 minutes. The temporal distribution of these events as well as the maximum amplitude of inflation showed clustering activity accompanied by intra-cluster waxing and waning. We created subsets of temporal clusters and calculated repose times between successive events. Auto-correlation functions were calculated for each subset and probability density functions were fitted which support survival/failure processes. The long-term tilt records provided a useful tool to characterize the activity and revealed a near-continuous cyclicity.
Recommended Citation
Sivaraj, Kay, "Volcan de Fuego: A Machine Learning Approach in Understanding the Eruptive Cycles Using Precursory Tilt Signals", Open Access Master's Thesis, Michigan Technological University, 2021.