Skip to content

Unsupervised Classification and Displacement Estimation of Very-Long-Period Seismic Signals at Mount Marapi

Tania Espinosa-Ortega1, Benoit Taisne 1,2, Jurgen W. Neuberg3, and Yasa Suparman4

  • Affiliations: 1Earth Observatory of Singapore, Nanyang Technological University, Singapore; 2Asian School of the Environment, Nanyang Technological University, Singapore; 3School of Earth and Environment, University of Leeds,UK; 4Center for Volcanology and Geological Hazards Mitigation, Geological Agency, Ministry of Energy and Mineral Resources, Bandung, Indonesia

  • Presentation type: Poster

  • Presentation time: Thursday 16:30 - 18:30, Room Poster Hall

  • Poster Board Number: 139

  • Programme No: 2.1.49

  • Theme 2 > Session 1

Abstract

Very long-period (VLP) seismic signals are commonly observed in volcanic settings, indicating significant ground deformation events associated with magma movements, gas emissions, and caldera collapses. However, these signals which typically have frequencies around 0.01 Hz, are often overlooked since they fall outside the flat frequency response of most seismometers. For these frequencies, standard signal restitution from raw data can lead to noise amplification, distorting the original seismic signal. One alternative to this issue is the use of forward modelling to generate synthetic VLP seismic signals corresponding to various ground deformation scenarios, which can then be compared to the uncorrected seismic traces for identification. In this study, we analysed VLP seismic signals recorded at Mount Merapi prior to the phreatic explosion on April 27, 2018. Using the Dynamic Time Warping (DTW) algorithm, we performed an unsupervised classification of these signals and compared the classified data with synthetic traces to identify the most likely ground deformation responsible for the observed seismic waveforms. Additionally, to assess the effectiveness of our method, we examined the correlation between synthetic VLP traces representing different ground deformation scenarios and explored how, under certain conditions, these signals could be misinterpreted due to their similarity. Our method provides an alternative for systematic analysis of VLP signals in raw seismic data, overcoming issues of noise distortion. This enhances volcano monitoring, by enabling better understanding of volcanic deformation.