Processing automatic seismic event detections: an iterative sorting algorithm improving earthquake hypocentres using interevent cross-correlation

Ó Gudmundsson,F Wagner,R Roberts,A Tryggvason

doi:10.1093/gji/ggz362

Abstract

SUMMARYWe present an iterative classification scheme using interevent cross-correlation to update an existing earthquake catalogue with similar events from a list of automatic seismic event detections. The algorithm automatically produces catalogue quality events, with improved hypocentres and reliable P- and S-arrival time information.Detected events are classified into four event categories with the purpose of using the top category, with the highest assessed event quality and highest true-to-false ratio, directly for local earthquake tomography without additional manual analysis. The remaining categories have varying proportions of lower quality events, quality being defined primarily by the number of observed phase onsets, and can be viewed as different priority groups for manual inspection to reduce the time spent by a seismic analyst.A list of 3348 event detections from the geothermally active volcanic region around Hengill, southwest Iceland, produced by our migration and stack detector (Wagner et al. 2017), was processed using a reference catalogue of 1108 manually picked events from the same area.P- and S-phase onset times were automatically determined for the detected events using correlation time lags with respect to manually picked phase arrivals from different multiple reference events at the same station. A significant improvement of the initial hypocentre estimates was achieved after relocating the detected events using the computed phase onset times. The differential time data set resulting from the correlation was successfully used for a double-difference relocation of the final updated catalogue.The routine can potentially be implemented in real-time seismic monitoring environments in combination with a variety of seismic event/phase detectors.

Highlights

In times of vast seismic networks and rapidly growing data storage capacity, the need for automatic analysis of seismic data becomes increasingly important
In a previous study (Wagner et al 2017), we investigated the benefits of back-propagating and stacking different trace attributes, such as, for example, the ratio of a short-term average over a longterm average (STA/LTA) or the Kurtosis function to detect and locate seismic events without explicitly detecting phase arrivals
We checked whether the estimated phase onsets were misidentified, that is, an event was assessed to be a true event at the location indicated, and potentially useful for tomographic analyses

Summary

Introduction

In times of vast seismic networks and rapidly growing data storage capacity, the need for automatic analysis of seismic data becomes increasingly important. Langet et al 2014; Pesicek et al 2014; Grigoli et al 2016) This type of seismic migration-based detection (MBD) uses an a priori velocity model to combine information from phase arrivals, which is the chosen trace attribute, at different observation locations. The stacked attribute parameter of an event (sum of attribute traces at the detected event hypocentre) can indicate event signal-to-noise ratio (SNR), depending on the attribute that is being stacked, and is often the intuitive choice for a preliminary quality assessment of detections. It can be difficult, and to some extent arbitrary, to determine an

Methods

Results

Discussion

Conclusion