Taiwan Borehole Seismometer Application in Earthquake Early Warning

Earthquake early warning (EEW) is an effective approach to mitigating earthquake damage. This is the first study evaluating borehole seismograph application to EEW in Taiwan. We selected inland and offshore earthquakes with ML larger than 4.0 occurring between 2012 and 2014 for this study. We investigated the Pd attenuation relationship as a function of the hypocentral distance (R) and magnitude (M). The new Pd attenuation relationship specific for the borehole records is expressed as: log(Pd) = 0.689ML 0.741log(R) 4.608 ± 0.248. Once the earthquake location is determined, this regression equation is used to quickly estimate Pd magnitude (MPd). According to the new regression equation formulated specifically for borehole observations in Taiwan, our result shows that the standard MPd deviation is about 0.21 relative to ML. This smaller standard deviation of 0.21 compared to that of the free-surface records might be attributed to the reduced influence of near-surface effects in the borehole records. We propose a new robust Pd regression equation for the Taiwan borehole seismic network.


InTroducTIon
Earthquake early warning (EEW) systems are proven to be effective tools for real-time seismic hazard mitigation, and are operated in many countries, for instance, Taiwan, Japan, Mexico, and Southern California (Nakamura 1988;Aranda et al. 1995;Kanamori et al. 1997;Wu et al. 1999;Allen and Kanamori 2003;Kamigaichi 2004;Horiuchi et al. 2005;Zollo et al. 2006;Allen et al. 2009;Hsiao et al. 2009; Lee and Wu 2011;Satriano et al. 2011).EEW provides alerts to urban areas near the epicenter of a forthcoming strong ground shaking caused by large earthquakes that may affect sensitive facilities such as public transportation systems.
There are two approaches adopted for EEW systems: regional warning and on-site warning.In regional warning, the seismic records of the closer-to earthquake seismic sensors or network are used to predict strong ground motions at more distant regions.In on-site warning, the initial P-wave motion is used to predict the ground motions of later arriving S and surface waves, which commonly have higher amplitudes and destructive energy than that of the initial P-wave.Generally, the regional approach is more compre-hensive and accurate, but it takes a longer time to produce results because it requires information from a number of stations.The on-site warning system has a shorter reporting time than the regional one, but the regional one has a higher accuracy than the on-site one because more stations are used, providing complete phase arrivals.
As an amplitude parameter, the peak amplitude of the initial 3-sec P-wave vertical displacement, Pd, reflects the attenuation relationship of the ground motion with distance (Wu and Zhao 2006;Wu et al. 2006), which leads to practical applications for EEW.This suggests that we can estimate the earthquake magnitude once the epicentral distance and Pd are available.When a large earthquake happens, its location can be quickly obtained from a few P-wave arrivals at nearby stations.Pd can then be used to determine the magnitude via the attenuation relationship.The relationships between the earthquake magnitude and several characteristic parameters obtained in the first few seconds from the P-wave have been developed for EEW applications (Allen and Kanamori 2003;Wu and Kanamori 2005a, b;Wu and Zhao 2006;Hsiao et al. 2011).
Taiwan is located on the western portion of the Circum-Pacific seismic belt with a plate convergence rate of 8 cm per year.Earthquakes are both frequent and serious disasters in Taiwan.We considered Pd from borehole seismic recording as an amplitude parameter for earthquake magnitude (M Pd ) determination for EEW.We defined a new robust Pd regression equation for the Taiwan borehole seismic network.

SEISmIc nETWork And dATA
We acquired borehole seismographs from the Central Weather Bureau (CWB) borehole seismic network.The vertical acceleration signals were numerically doubleintegrated to obtain the displacements and filtered using a 0.075 Hz high-pass recursive Butterworth filter to remove the low-frequency drift after the last integration.Pd is defined as the peak amplitude of filtered displacement during the initial 3 sec of P-waves.Since only the vertical motion component is used in this study, we assume that the vertical borehole seismometer component is well orientated and any deviation from this can be neglected (Krieger and Grigoli 2015;Wang et al. 2016;Zaldívar et al. 2016).The CWB has been installing the borehole seismometers since 2007.However, we selected events occurring after 2012 in this study to ensure stable network operation and data quality.Figure 1 shows the 29 stations used in this study.Each station has strong-motion seismographs installed at the free surface and bottom of the borehole with a depth range between 100 and 400 m.We used only the borehole data to reduce the receiver site near-surface effects and better extract the source signature registered in the initial P-wave arrival portion.
We selected in-land and offshore (distance to shoreline < 10 km) regional earthquakes from the CWB catalog that occurred between 2012 and 2014 with M L > 4.0 and focal depths shallower than 50 km (Table 1).Most inland and inshore earthquakes that cause ground shaking to a maximum peak ground acceleration > 80 gal (1 gal = 1 cm s -2 ) have focal depths shallower than 25 km in Taiwan.We further restricted the study to events that were recorded by a minimum of five stations (i.e., at least 5 Pd), and there are a total of 68 events (Fig. 1) and 465 Pd recordings.

rESulTS And dIScuSSIon
According to previous studies {Fowler 2004 [Eq.(4.13)]; Wu and Zhao 2006}, we assumed a linear regression model among the logarithmic Pd.The CWB reported local magnitude M L , and the logarithmic hypocentral distance R: where a, b, and c are constants to be determined from the regression analysis.
Equation ( 1) can be rewritten in matrix form as: where n and m are the numbers of recordings and events, respectively.Equation ( 2 Figure 2 shows the observed Pd values compared with the values predicted by Eq. (3) separately for magnitudes of 4.5, 5.5, and 6.5.We used the average of all available M Pd as the final magnitude for each event.For the real-time operation, the earthquake locations and M Pd can be determined from the first couple of P-wave arrivals to reduce the Pd collection time.In Fig. 3, we compared the CWB reported M L to the M Pd , and the resulting standard deviation between the reported M L and the average M Pd is of 0.21 (M Pd = M L ± 0.21).This standard deviation is fairly accept-able in the EEW system.
In order to verify the borehole Pd attenuation equation [Eq.( 2)], we used Pd values in our study to calculate M Pd using Eq. ( 4) and the free-surface Pd regression equation (Hsiao et al. 2011), respectively.Since the event magnitudes used in the study of Hsiao et al. (2011) are larger than 4.5, we only compared event magnitudes larger than 4.5.Figure 3 clearly suggests the necessity for a specific Pd regression equation for the borehole data.Figure 3 shows that for larger events of about M > 6.0, M Pd calculated by Hsiao et al. (2011) are closer to these using Eq. ( 4).This might imply that a larger magnitude event has longer period signal content (i.e., longer wavelength) than a smaller one in the early P-wave potion.This magnitude-dependent frequency characteristic is understood as the frequency-dependent source excitation (Hanks and Kanamori 1979).For a wave with a wavelength comparable to the borehole depth.The amplitude properties of the propagating wave between the borehole and free surface should be more similar than those of a propagating wave with a shorter wavelength.
In the study of Hsiao et al. (2011), M Pd estimated using the free-surface Pd regression equation has a 1:1 relationship with the CWB catalog M L with a standard deviation of 0.43.The borehole record application in M Pd in theory should be more robust due to the reduced near surface effect caused by the low shear-wave velocity, low quality factor (Q), or topographic effect.The smaller standard deviation of 0.21 presented in Eq. ( 4) than that of Hsiao et al. (2011) might be attributed to the reduced influence of the near-surface effect in the borehole records to some extent.
The site effect is one of the important factors that affects ground motion prediction.We determined the Pd site correction (S) by averaging the residuals between the observed and predicted values as follows (Wu et al. 2001 where n is amount of Pd for the corresponding station.The total amount of Pd obs. is 669 from the events that were recorded by at least three stations.Figure 4 presents the resulting S  Since the Taiwan borehole network recording period is very short compared to free-surface ground motion observations, only 12 stations out of 29 stations (41%) had quality Pd recordings greater than 20.The standard deviation of the free-surface Pd regression equation (Hsiao et al. 2011) is larger than that derived using the borehole Pd values in this study.Therefore, it might not be so encouraging to use the Pd regression equation for surface sensors incorporated with the borehole Pd site effect as another approach to bring the borehole data into the routine EEW operation.

Fig. 1 .M
Fig. 1.Station distribution of the Taiwan borehole network and the epicenter locations of the 68 events used in this study.Geological provinces are indicated by the solid lines and denoted by I, the Coastal Plain; II, the Western Foothills; III, Hsuehshan Range; IV, Backbone Range; V, Longitudinal Valley; VI, Coastal Range.The Central Range is composed of two ranges, the Backbone Range in the east and the Hsuehshan Range in the west.

Fig. 2 .
Fig. 2. Distribution of the 424 observed Pd measurements.The predicted curves by Eq. (3) are depicted by the solid lines for a variety of magnitudes from 4.0 -7.0.(Color online only)

Fig. 4 .
Fig. 4. Site correction (S) for the borehole Pd observations at each station.