5:15 PM - 7:15 PM
[SVC32-P30] Systematic detection of pre-eruptive tilt changes worldwide and classification by time functions
Keywords:Tilt change, Volcanic eruption
1. Introduction
Pre-eruptive tilt changes are widely observed at volcanoes and are crucial for understanding the immediate pre-eruptive processes. The general nature of tilt changes is not well understood because most previous studies have focused on individual events or volcanoes. Maeda (2023, JVGR) systematically surveyed tilt changes before all eruptions in Japan with known times at a temporal resolution of 1 min from all publicly available waveform data and showed the presence of tilt changes in 3970/22958 (17%) analyzed waveforms before 3104/7890 (39%) eruptions. In this study, we performed a similar systematic survey at foreign volcanoes and classified them based on waveform similarity to examine their features.
2. Detections of pre-eruptive tilt changes at foreign volcanoes
From the International Federation of Digital Seismograph Networks (FDSN), we listed 48 networks with the name ``volcano'' for which continuous waveforms were downloadable from the EarthScope Data Center (as of May 31, 2024). By comparing the station information of each network with a list of Holocene volcanoes in the Global Volcanism Program (GVP), we identified 28 volcanoes that erupted after the beginning of broadband seismic monitoring at distances closer to 10 km. We surveyed the times of eruptions from the beginning of broadband seismic observation at each volcano to the end of 2023 using the websites of national monitoring agencies and reviewed papers; the results were 61 eruptions at 16 volcanoes at a temporal resolution of 1 min. We applied the same procedure as Maeda (2023) to identify the existence or absence and the time window of a tilt change before each eruption. Tilt changes were detected for 16/173 (9%) waveforms before 12/61 (20%) eruptions at 8/16 volcanoes.
3. Grouping by time functions
We grouped the 3976 waveforms of pre-eruptive tilt changes in Japan detected by Maeda (2023) and the additional 16 waveforms at foreign volcanoes based on the similarity of the waveforms as follows. We computed the misfit Mij={2∫01[si(t)-sj(t)]2dt/[∫01si(t)2dt+∫01sj(t)2dt]}1/2 between every pair of waveforms si(t) and sj(t) of tilt changes after the normalization of durations and amplitudes to [0, 1]. The waveform that had the largest number of pairs with Mij less than a threshold value Mthre was selected as the master waveform of the first group, and all waveforms that had Mij < Mthre with the master were classified into this group. The same procedure was repeated for the remainder of the waveforms to define the second, third, and later groups. The optimal Mthre was determined by AIC based on the residuals between the individual and averaged waveforms in each group with ≧ 2 waveforms. We tried Mthre from 0.01 to 0.30 at an interval of 0.01, and the optimal value was 0.10. A total of 73 groups with ≧ 4 waveforms were identified using this Mthre (Fig. 1).
The waveforms in some groups showed simple acceleration (Acc type) or declaration (Dcc type), but the others showed acceleration followed by deceleration (Acc-Dec type) or vice versa (Dec-Acc type). We quantified these four categories (Fig. 2) based on changes in the slopes of the fitting lines in three sections of each waveform. The optimal sections were determined to maximize the slope changes, keeping ≧ 20 % of the time in each section. Acc-type waveforms were mostly from Sakurajima. The Dcc type was relatively rare, and a large fraction of this type was from Suwanosejima (Fig. 3). Tilt changes before the 2014 eruption of Ontakesan and the 2018 eruption of Kirishimayama (Ioyama) were in the same group; however, this group consisted of only four waveforms and showed significantly different patterns from most others (Figs. 1 and 2). Future work includes a trial of creating a numerical model that explains the waveform characteristics.
4. Acknowledgments
We used continuous waveforms provided by the EarthScope Data Center. This study was supported by JSPS Kakenhi Grant Number JP21H05203.
Pre-eruptive tilt changes are widely observed at volcanoes and are crucial for understanding the immediate pre-eruptive processes. The general nature of tilt changes is not well understood because most previous studies have focused on individual events or volcanoes. Maeda (2023, JVGR) systematically surveyed tilt changes before all eruptions in Japan with known times at a temporal resolution of 1 min from all publicly available waveform data and showed the presence of tilt changes in 3970/22958 (17%) analyzed waveforms before 3104/7890 (39%) eruptions. In this study, we performed a similar systematic survey at foreign volcanoes and classified them based on waveform similarity to examine their features.
2. Detections of pre-eruptive tilt changes at foreign volcanoes
From the International Federation of Digital Seismograph Networks (FDSN), we listed 48 networks with the name ``volcano'' for which continuous waveforms were downloadable from the EarthScope Data Center (as of May 31, 2024). By comparing the station information of each network with a list of Holocene volcanoes in the Global Volcanism Program (GVP), we identified 28 volcanoes that erupted after the beginning of broadband seismic monitoring at distances closer to 10 km. We surveyed the times of eruptions from the beginning of broadband seismic observation at each volcano to the end of 2023 using the websites of national monitoring agencies and reviewed papers; the results were 61 eruptions at 16 volcanoes at a temporal resolution of 1 min. We applied the same procedure as Maeda (2023) to identify the existence or absence and the time window of a tilt change before each eruption. Tilt changes were detected for 16/173 (9%) waveforms before 12/61 (20%) eruptions at 8/16 volcanoes.
3. Grouping by time functions
We grouped the 3976 waveforms of pre-eruptive tilt changes in Japan detected by Maeda (2023) and the additional 16 waveforms at foreign volcanoes based on the similarity of the waveforms as follows. We computed the misfit Mij={2∫01[si(t)-sj(t)]2dt/[∫01si(t)2dt+∫01sj(t)2dt]}1/2 between every pair of waveforms si(t) and sj(t) of tilt changes after the normalization of durations and amplitudes to [0, 1]. The waveform that had the largest number of pairs with Mij less than a threshold value Mthre was selected as the master waveform of the first group, and all waveforms that had Mij < Mthre with the master were classified into this group. The same procedure was repeated for the remainder of the waveforms to define the second, third, and later groups. The optimal Mthre was determined by AIC based on the residuals between the individual and averaged waveforms in each group with ≧ 2 waveforms. We tried Mthre from 0.01 to 0.30 at an interval of 0.01, and the optimal value was 0.10. A total of 73 groups with ≧ 4 waveforms were identified using this Mthre (Fig. 1).
The waveforms in some groups showed simple acceleration (Acc type) or declaration (Dcc type), but the others showed acceleration followed by deceleration (Acc-Dec type) or vice versa (Dec-Acc type). We quantified these four categories (Fig. 2) based on changes in the slopes of the fitting lines in three sections of each waveform. The optimal sections were determined to maximize the slope changes, keeping ≧ 20 % of the time in each section. Acc-type waveforms were mostly from Sakurajima. The Dcc type was relatively rare, and a large fraction of this type was from Suwanosejima (Fig. 3). Tilt changes before the 2014 eruption of Ontakesan and the 2018 eruption of Kirishimayama (Ioyama) were in the same group; however, this group consisted of only four waveforms and showed significantly different patterns from most others (Figs. 1 and 2). Future work includes a trial of creating a numerical model that explains the waveform characteristics.
4. Acknowledgments
We used continuous waveforms provided by the EarthScope Data Center. This study was supported by JSPS Kakenhi Grant Number JP21H05203.