Near-field triggering of microearthquakes along the Longitudinal Valley fault in eastern Taiwan

1.   Introduction

The passage of seismic waves from large earthquakes has the potential to impose stress perturbations on a fault zone and prompt fault failure. Dynamic triggering of seismicity has been reported in many regions, such as triggered earthquakes in Greece following the 1999 Izmit, Turkey earthquake, low-frequency earthquakes within non-volcanic tremor in southern Taiwan triggered by the 2005 M_W8.6 Nias earthquake and so on (e.g., Brodsky et al., 2000; Tang et al., 2010; Kato et al., 2013; Walter et al., 2015; Cattania et al., 2017). Seismicity variation is usually related to the distribution of stresses required for fault failure under different tectonic settings and following seismicity (Brodsky and van der Elst, 2014). However, a significant portion of microearthquakes are missing from the existing catalogs due to waveform contamination by background noise or overlapping waveform arrivals from different earthquakes, which impedes our understanding of fault behavior (Kato et al., 2013).

Taiwan is located at the boundary between the Eurasian and the Philippine Sea Plates (Figure 1a), where the dynamic triggering of seismic events has been reported in previous studies (Chao et al., 2012; Tang et al., 2010, 2013). Triggered seismicity, induced by surface waves from large remote earthquakes, has mainly been detected in the north and south areas of Taiwan (Ma et al., 2001, 2005); however, there are few triggering records in existence for eastern Taiwan. Hence, the potential for triggered seismicity on the east coast of Taiwan, a boundary of compression and collision from two plates with relatively dense seismic activity, is worth exploring.

Figure  1.  (a) Tectonic map of eastern Taiwan showing the distribution of seismic stations (black triangles) along the Longitudinal Valley fault system (LVF; red lines). The 2019 February 15 M_L5.2 Hualien earthquake (yellow star) occurred at 121.748°E and 23.798°N with a focal depth of 45.6 km, as proposed by the Taiwan Weather Bureau. The focal mechanism is by the Broadband Array in Taiwan for Seismology Centroid Moment Tensor solutions. The long-term seismicity of M_L $\ge$ 1.5 from 2013 to 2017 are shown as gray circles and those exceeding M_L 5.0 are marked with black stars. Two physiographic features, the Central Range (CR) and the Coastal Range (CoR), are also shown. The inset shows the tectonic setting of Taiwan. (b) Distributions of earthquakes in the new catalog created by merging newly detected and original events. The dark gray rectangle indicates the selected region along the LVF, which is divided into two sub-areas, A and B. Events occurring before and after the Hualien earthquake in the LVF region are shown as red and blue circles, respectively, and events outside this window are shown as light gray circles

DownLoad: Full-Size Img PowerPoint

The Longitudinal Valley fault (LVF) in eastern Taiwan is an active strike-slip fault approximately 150 km long (Yu and Kuo, 2001) (Figure 1a). It represents a suitable site to investigate dynamic triggering from strong local earthquakes and their underlying mechanisms. The February 15, 2019, M_L 5.2 Hualien earthquake is used as the mainshock in this study, and we analyze seismicity rate changes as a consequence of this event on the LVF in detail. To improve the completeness of the seismic record, we employ a matched-filter technique (Peng and Zhao, 2009; Kato et al., 2012; Kato and Nakagawa, 2014; Tang et al., 2019, 2020) to detect missing microearthquakes. Then, we discuss the implications of our findings. This study will contribute to, and extend the study of dynamic stress transfer on earthquake triggering.

2.   Data and methods

Earthquakes listed in the Taiwan Weather Bureau (TWB) catalog that occurred from 2018-09-01 to 2019-08-31 and located on the east coast of Taiwan (120.5°E–122.25°E, 22.6°N–24.22°N) were used in this study. We used seismic waveform data recorded at five three-component broadband stations of the Broadband Array in Taiwan for Seismology (Figure 1a). We removed the mean from all seismic waveform data and applied a 1–8 Hz bandpass filter to eliminate long-period noise. Then, we selected those earthquakes with a signal-to-noise ratio above 5 for at least 12 seismic waveform components as templates (Liu et al., 2019). We visually identified the S wave arrivals of 1,404 local earthquakes and used the waveforms 2 s before and after the arrival time as template waveform windows.

The matched filter technique (MFT; Peng and Zhao, 2009; Liu et al., 2019; Tang et al., 2019; 2020) was used to detect missing microearthquakes during February 2019. We shifted the 4 s template waveform window in increments of 0.05 s (one sample) through continuous waveforms, calculated the correlation coefficient between them for each seismic waveform component at every station, and took the average throughout. A positive detection occurred when the mean correlation peak exceeded six times the standard deviation (Tang et al. 2014). After removing multiple counts, we estimated the magnitude of new detections based on the amplitude ratio between the detection and the template by assuming that a ten-fold increase in amplitude corresponded with a one-unit increase in magnitude (Peng and Zhao, 2009). Figure 2 shows an example of a positive detection with an estimated magnitude of 1.5 by its template event, and a local magnitude of 2.8. The hypocenter of the detected event was the same as the corresponding template.

Figure  2.  Example of a detected earthquake. (a) Mean correlation coefficient (CC) function for the template 20190310094849.71 scanning of the continuous waveform data for 2019 February 24 shown on the left. The red dot indicates a positive detection above the threshold (black dashed line). A count histogram of mean CC values shown on the right. (b) Continuous waveforms shown in gray, and template waveforms in red. The red arrow represents the origin time of the detected event. Station components and corresponding correlation coefficients are shown on the left and right sides, respectively. The average CC value of all components is 0.52 and the estimated event magnitude is 1.5

DownLoad: Full-Size Img PowerPoint

We computed both the $z$ and $\,\beta$ values (Aron and Hardebeck, 2009) to examine seismicity rate changes in the LVF system (Figure 1b). We averaged the seismicity rate for the 15 days prior to the Hualien earthquake and used this value as a constant background rate. The $z$ value compares the difference between the mean rates of two independent groups from the same seismic sequence, given by:

where ${m}_{1}$ and ${m}_{2}$ are the mean rates of earthquake numbers in two independent time periods, ${\sigma }_{1}$ and ${\sigma }_{2}$ are the standard deviations for the two time periods, and ${n}_{1}$ and ${n}_{2}$ are the number of events in each time period. The $\beta$ statistic compares the difference between the number of events in the target time period and the expected number in that time period assuming a constant seismicity rate, given by:

where ${N}_{\rm{a}}$ and $N$ are the number of events in the target time period and the total number of events in the whole time period, and ${T}_{\rm{a}}$ and $T$ are the durations of the target time period and the entire catalog. An absolute value of $z$ or $\beta$ $\ge$ 1.64 represents a statistically significant change in the seismicity rate at a 90% confidence level (low threshold), $\ge$ 1.96 is significant at a 95% confidence level, and $\ge$ 2.57 is significant at a 99% confidence level (high threshold).

3.   Results

A total of 2,460 earthquakes were newly detected on the east coast of Taiwan within the 15 days prior to and 12 days following the M_L5.2 Hualien earthquake, twice the number listed in the standard TWB catalog. We calculated the magnitude of completeness (M_c) and $\beta$ value via the maximum curvature method as implemented in the ZMAP package (Wiemer, 2001). The M_c of the new catalog, which merged the detected and TWB catalog events, remains 1.6 due to a sharp increase in microearthquakes (Figures 3a–3b). We selected a seismic sequence prior to the Hualien earthquake along the LVF strike (Figure 1c) with a $\beta$ value of 1.39. The $\,\beta$ value in sub-region A of the LVF shows a slight decrease (Figure 3d) within three days following the mainshock.

Figure  3.  Frequency-magnitude relationships (FMR) for (a) the earthquakes listed in the original catalog within the study area, (b) the earthquakes listed in the new catalog within the study area, (c) events in the LVF region, and (d) events in sub-region A (refer to Figure 1b). The cumulative number of events listed in the original TWB catalog and the newly merged catalog are shown in blue and red, respectively. The M_c values (open triangle) and the FMR fitting slopes (i.e. $\beta$ value; solid line) were calculated via the maximum curvature method as implemented in the ZMAP package (Wiemer, 2001). The FMR fitting of the seismic sequence removed the M_L5.2 Hualien earthquake in subfigure (a) and a M_L4.4 local event in subfigure (c)

DownLoad: Full-Size Img PowerPoint

The spatial distributions of the seismic sequence in the LVF region during different time intervals and the corresponding seismicity rates along fault-striking are shown in Figure 4. Events are mainly distributed on the northern segment of the LVF, and a M_L 4.4 local event occurring approximately three days before the Hualien earthquake is the main driver of the high seismicity rate (Figure 4a). We used the average seismicity rate before the mainshock as a reference rate. The seismicity rate after the mainshock had an apparent increase in the north of the LVF (Figure 4b) which then gradually reduced (Figures 4c–4d). Conversely, seismicity in the south of the LVF only showed a slight increase.

Figure  4.  Epicenters of the seismic sequence in the LVF region for time intervals (a) −15 to 0 days; (b) 0 to 12 hours; (c) 0 to 20 hours and (d) 0 to 52 hours, and the corresponding histograms of seismicity rate versus along-strike distance. Note that the origin time of the Hualien earthquake (yellow star) is day 0. The average rate before the Hualien earthquake is approximately 2.7 events per day (vertical black line). The units of the seismicity rate within 24 hours following the mainshock were enlarged to the events per day. The M_L 4.4 earthquake occurring before the mainshock is marked as a black star in subfigure (a)

DownLoad: Full-Size Img PowerPoint

The $z$ and $\beta$ values continuously increased over time, exceeding the low threshold of statistically significant change in seismicity 20 hours following the mainshock and remaining significant for approximately 32 hours (Figure 5a). Values above the high significance threshold occurred at approximately 36 hours. The time-varying distributions of $z$ and $\beta$ values for the seismic sequence in sub-region A showed a similar pattern (Figure 5b).

Figure  5.  The $z$ and $\beta$ value changes for the seismic sequence up to 72 hours following the Hualien earthquake for (a) all events in the LVF region and (b) events in sub-region A. The three dashed lines show the thresholds of 2.57, 1.96, and 1.64, which represent statistically significant changes in the seismicity rate with confidence levels of 99%, 95%, and 90%, respectively

DownLoad: Full-Size Img PowerPoint

Figure 6 shows the distribution of the seismic sequence up to three days following the mainshock in two orthogonal profiles, including v_P/v_S structures from the local tomographic result (Wu et al., 2007). Most events are located in those regions where the v_P/v_S ratio exceeds 1.73. In profile B–B’, the events occurred along a northwest-dipping plane and were mainly concentrated in the shallow crust.

Figure  6.  Velocity ratio profiles (Wu et al., 2007) for the seismic sequence up to three days following the Hualien earthquake. The black star represents the location of the Hualien earthquake. The selected events in the LVF region (refer to Figure 1b) and the events outside the region are shown as colored and gray circles in the map and as black and gray circles in the profiles, respectively

DownLoad: Full-Size Img PowerPoint

4.   Discussion and conclusions

We applied the MFT to a continuous seismogram and recovered many missing microearthquakes on the east coast of Taiwan from the standard TWB catalog. The improved catalog provided an opportunity to conduct a detailed analysis of seismicity rate changes in the LVF following the M_L5.2 Hualien earthquake. Laboratory experiments have shown that shear stress increases on faults can decrease the b value of an earthquake during the seismic cycle (Rivière et al., 2018). Sensitivity of the b value to temporal stress variations in the tectonic setting of Taiwan has also been reported by Wu et al. (2018). Compared with the b value of the LVF before the mainshock, the b value of following events showed a slight decrease in the northern segment of the LVF, which may indicate that external stress from the Hualien earthquake loaded the fault and prompted microfracture failures.

Seismicity rates along the LVF showed marked increases immediately following the Hualien earthquake (Figure 4), which was found to be a characteristic feature for earthquake triggering in previous studies (Kato et al., 2013; Walter et al., 2015; Cattania et al., 2017). The stress perturbation induced by the mainshock rupture can impact the seismicity of adjacent regions. We calculated the dynamic stress ($\mathrm{\sigma }$) from the mainshock at each seismic station using $\sigma =\left({\rm{PGV}}\right)\left(\mu \right)/{v}_{\rm{S}}$ by assuming a shear rigidity $\mu$ of 30 GPa (Aiken and Peng, 2014) and deriving the corresponding S wave velocity under each station from the local tomographic result (Wu et al., 2007). The results showed that the dynamic stresses of seismic stations in sub-region A were above the typical value of 5–10 kPa for triggering microseismicity (Aiken and Peng, 2014), while those in sub region B were approximately 1 kPa (Table 1). We suggest that this dynamic stress change was the dominant factor triggering increased seismicity on the LVF (Kato et al., 2013).

Table  1.  Dynamic stress from the Hualien mainshock recorded at each seismic station

Station	Lat (°N)	Lon (°E)	Dynamic stress (kPa)
LXIB	24.021	121.413	10.3
WARB	23.718	121.386	11.1
HGSD	23.492	121.424	6.5
YULB	23.393	121.297	4.1
TWGB	22.818	121.080	0.9

 | Show Table

DownLoad: CSV

High values of $z$ and $\beta$ also indicated that significant seismicity rate changes occurred 20 hours after the Hualien earthquake (Figure 5). Delayed triggered seismicity has also been shown in previous studies (Peng et al., 2015; Zigone et al., 2012). Possible attributions for delayed triggering include passing seismic waves from the mainshock causing fault nucleation (Peng et al., 2015), and postseismic stress transfer (e.g., poroelastic rebound and viscous relaxation; Freed, 2005). Furthermore, the time-varying distributions of $z$ and $\beta$ values showed that seismicity rate changes lasted for approximately two days, during which no moderate local events or teleseismic large earthquakes occurred. One possible explanation for this is that the triggered events following the mainshock may be caused by other fault motions, such as aseismic slip and creeping, because some events were located close to regions with a relatively high v_P/v_S ratio (> 1.75; Figure 6; Chao et al., 2012; Tang et al., 2010, 2013).

The number of triggered events and increased seismicity in the northern segment of the LVF was higher compared to the southern segment (Figures 3 and 4). One direct reason for this is that the Hualien earthquake was located closer to the northern segment; hence, the larger amplitude seismic waves could produce higher stress perturbations, consistent with the dynamic stress distribution observed by the seismic stations (Table 1). Another potential attribution is differential stress state variations within segments of the LVF. For example, seismicity in the north of the fault is always denser compared to the south in the long-term TWB record; however, the crustal motion velocities of the two regions remain close (Yu and Kuo, 2001). This suggests that the stresses required for failure on the northern fault are lower, where some asperities and patches may always be near critical (Brodsky and van der Elst, 2014).

In this study, we focused on seismicity rate changes in the LVF immediately following the Hualien earthquake that occurred offshore in eastern Taiwan. The newly detected events found by this study will provide valuable assistance in identifying dynamic triggering due to strong regional earthquakes. Future research will focus on obtaining the spatial variations in triggered seismicity on active faults in Taiwan from more cases, which will help in monitoring fault stress states prior to the next earthquake rupture.

Acknowledgments

This research is supported by the Fundamental Research Founds for National University, China University of Geosciences (Wuhan) (No. 1910491T09) and the National Natural Science Foundation of China (No. 42074061). We would like to thank the Taiwan Weather Bureau and Broadband Array in Taiwan for Seismology for providing the catalog and the waveform data, respectively.