Rapid Characterization of the July 2019 Ridgecrest, California, Earthquake Sequence From Raw Seismic Data Using Machine‐Learning Phase Picker

Abstract

The two principle earthquakes of the July 2019 Ridgecrest, California, earthquake sequence, MW 6.4 and 7.1, and their immediate foreshocks and thousands of aftershocks present a challenging environment for rapid analysis and characterization of this sequence as it unfolded. In this study, we analyze the first 6 days of the sequence using continuous data from available seismic networks to detect and locate earthquakes associated with the earthquake sequence. We build a high‐precision earthquake catalog using a deep‐neural‐network‐based picker—PhaseNet and a sequential earthquake association and location workflow. Without prior information, we automatically detect and locate more than twice as many earthquakes as the routine catalog. Our high‐precision earthquake catalog reveals detailed spatiotemporal evolution of the earthquake sequence and clearly defines multiple faults activated during the sequence. Our study demonstrates that it is possible to characterize earthquake sequences from raw seismic data using a well‐trained machine‐learning picker and our workflow.

Sequential Fault Reactivation and Secondary Triggering in the March 2019 Red Deer Induced Earthquake Swarm

“…Among these STA/LTA picks, we picked 964 P arrivals and 1,062 S arrivals from the raw waveforms using a machine‐learning phase picker (Zhu & Beroza, 2019). Events were associated and preliminarily located using a grid search‐based earthquake association and location method (Liu et al, 2020; M. Zhang, Ellsworth, et al, 2019). The search space is discretized into uniform grids with a constant size of 0.01° × 0.01° × 2 km, and a minimum of 5 P picks and 10 P + S picks is required for each event.…”

Hydraulic Fracturing‐Induced Seismicity

“…Rapid progress in the application of machine learning methods to event detection, timing, association, and location is transforming the analysis of both natural and induced seismicity (Bergen et al, 2019; Mousavi et al, 2019). New machine learning algorithms for automatic detection and picking of P and S wave arrivals in continuous data are beginning to achieve results comparable to template matching and superior, in some cases, to traditional methods (Liu et al, 2020). For example, Park et al (2020) applied the PhaseNet picker (Zhu & Beroza, 2018) to the entire Guy‐Greenbrier, Arkansas, sequence to detect over 280,000 events (single station) and located almost 90,000 events, compared to fewer than 2000 reported by Horton (2012) using standard network procedures or 13,000 found by FAST.…”

Seismotectonics and Fault Geometries of the 2019 Ridgecrest Sequence: Insight From Aftershock Moment Tensor Catalog Using 3‐D Green's Functions

“…Based upon the observed surface ruptures and the satellite images, there are very complex surface ruptures during the 2019 Ridgecrest earthquakes, including the splay ruptures at the SE tip, conjugate ruptures at the NW tip, and the along‐strike variation of surface ruptures near the M w 7.1 epicenter (Figures S1 and S2 in the supporting information) (Barnhart et al, 2019; Brandenberg et al, 2019; Ross et al, 2019). The locations of aftershocks, which provide insights into the subsurface faulting, in general correlate with the mapped surface breaks (Lin, 2020; Liu et al, 2020; Ross et al, 2019; Shelly, 2020), yet apparent differences exist. For example, at the location where the NW rupture of M w 6.4 stopped and the M w 7.1 started, the aftershocks mostly lie along a narrow compact straight strip, while the surface breaks show along‐strike variations with significant offsets to the seismicity (Figure 1; see Figure S1 for a zoom‐in figure).…”

“…Both the foreshock (M w 6.4) and mainshock (M w 7.1) were followed by notable aftershocks, with more than 111,000 M L 0.5+ aftershocks within the first 21 days (Ross et al, 2019). The spatial distribution of aftershocks of the M w 6.4 event illuminates that it ruptured a conjugated fault system that forming an “L” shape (Ross et al, 2019; Lin, 2020; Liu et al, 2020; Shelly, 2020). The kinematic subevent and finite fault inversions provide additional evidence that the M w 6.4 event ruptured both the NW‐SE and NE‐SW trending faults (Chen et al, 2020; Feng et al, 2020; Jia et al, 2019; Liu et al, 2019; Ross et al, 2019), although only the NE‐SW trending rupture is captured by the InSAR optical image (Barnhart et al, 2019).…”

“…The kinematic subevent and finite fault inversions provide additional evidence that the M w 6.4 event ruptured both the NW‐SE and NE‐SW trending faults (Chen et al, 2020; Feng et al, 2020; Jia et al, 2019; Liu et al, 2019; Ross et al, 2019), although only the NE‐SW trending rupture is captured by the InSAR optical image (Barnhart et al, 2019). About 34 hr later, a Mw 7.1 event initialized near the NW end of the aftershock zone of the M w 6.4 event and ruptured bilaterally on a ~50 km NW‐SE striking fault (Barnhart et al, 2019; Chen et al, 2020; Goldberg et al, 2020; Lin, 2020; Liu et al, 2019; Liu et al, 2020; Ross et al, 2019; Shelly, 2020), with clear surface breaks observed by geological surveys and satellite observations (Brandenberg et al, 2019; Fielding et al, 2020; Ross et al, 2019). To the south, the aftershock zone splits into several subparallel strands that disappear near the NE‐SW striking Garlock fault.…”