5:15 PM - 6:45 PM
[SGD02-P04] Automatic detection of inflection points in high-rate GNSS data: A case study of slow slip events off the Boso Peninsula
Keywords:High-rate GNSS, Slow slip event
In the midst of advancing research on the diverse temporal scales of slow earthquakes, events lasting from approximately 100 seconds to one day have been less frequently detected (e.g., Ide and Beroza, 2023). Moreover, there is active discussion on whether events with longer durations might simply be aggregates of shorter-duration events (e.g., Frank and Brodsky, 2019). Therefore, detecting slow slip events (SSEs) using geodetic observation data with sampling intervals shorter than one day is considered crucial for deepening our understanding of slow earthquakes.
This study investigates a comprehensive event detection method using high-rate GNSS data with sampling intervals shorter than one day, utilizing the 5-minute interval precise point positioning dataset provided by the Nevada Geodetic Laboratory (e.g., Mitsui & Arai, 2023). Preliminary research indicated that due to the significant noise in high-rate GNSS data, signal detection using template matching as performed with daily sampling data (e.g., Rousset et al., 2017) was challenging. Therefore, we adopted a method for automatically detecting inflection points in the time-series data at the start and end times of events (Taylor and Letham, 2017). Specifically, we pre-set a large number of inflection point candidates and used piecewise regression based on maximum posterior estimation, assuming a Laplace distribution as the prior distribution, to extract significant inflection points.
To evaluate the performance of the method, we present cases where signals detected in previous studies based on low-noise daily sampling GNSS data were re-detected using high-rate GNSS data. Specifically, we focused on SSEs that occurred at the plate boundaries off the Boso Peninsula along the Sagami Trough and the Japan Trench subduction zones, in 2011, 2016, and 2018 (e.g., Hirose et al., 2012; Nishimura, 2021). The detection tests showed that for the events in 2011 and 2018, the direction of displacement, start times, and end times largely matched those identified in previous studies, indicating successful automatic signal detection. However, the 2016 event did not yield a clear signal, likely because, unlike the events in 2011 and 2018, which had displacements of more than 2 cm, the displacement in the 2016 event did not exceed 1 cm, rendering the signal insufficient against the noise.
The short sampling intervals of high-rate GNSS data are expected to capture the temporal evolution of events in detail. For example, we assessed whether the start or end times of SSEs were more clearly captured using the change rate of the slope of the regressed piecewise lines, indicated by delta values. The results showed that the 2011 SSE had a higher delta value at the start time, while the 2018 SSE had a higher delta value at the end time, suggesting variability in whether the onset or cessation of an event is more abrupt. The method introduced in this study, which does not require complex preprocessing, is capable of facilitating the comparative examination of the diversity and universality of SSE temporal developments worldwide.
This study investigates a comprehensive event detection method using high-rate GNSS data with sampling intervals shorter than one day, utilizing the 5-minute interval precise point positioning dataset provided by the Nevada Geodetic Laboratory (e.g., Mitsui & Arai, 2023). Preliminary research indicated that due to the significant noise in high-rate GNSS data, signal detection using template matching as performed with daily sampling data (e.g., Rousset et al., 2017) was challenging. Therefore, we adopted a method for automatically detecting inflection points in the time-series data at the start and end times of events (Taylor and Letham, 2017). Specifically, we pre-set a large number of inflection point candidates and used piecewise regression based on maximum posterior estimation, assuming a Laplace distribution as the prior distribution, to extract significant inflection points.
To evaluate the performance of the method, we present cases where signals detected in previous studies based on low-noise daily sampling GNSS data were re-detected using high-rate GNSS data. Specifically, we focused on SSEs that occurred at the plate boundaries off the Boso Peninsula along the Sagami Trough and the Japan Trench subduction zones, in 2011, 2016, and 2018 (e.g., Hirose et al., 2012; Nishimura, 2021). The detection tests showed that for the events in 2011 and 2018, the direction of displacement, start times, and end times largely matched those identified in previous studies, indicating successful automatic signal detection. However, the 2016 event did not yield a clear signal, likely because, unlike the events in 2011 and 2018, which had displacements of more than 2 cm, the displacement in the 2016 event did not exceed 1 cm, rendering the signal insufficient against the noise.
The short sampling intervals of high-rate GNSS data are expected to capture the temporal evolution of events in detail. For example, we assessed whether the start or end times of SSEs were more clearly captured using the change rate of the slope of the regressed piecewise lines, indicated by delta values. The results showed that the 2011 SSE had a higher delta value at the start time, while the 2018 SSE had a higher delta value at the end time, suggesting variability in whether the onset or cessation of an event is more abrupt. The method introduced in this study, which does not require complex preprocessing, is capable of facilitating the comparative examination of the diversity and universality of SSE temporal developments worldwide.