Performance of a Low-Cost Earthquake Early Warning System (P-Alert) and Shake Map Production during the 2018 M w 6.4 Hualien, , Earthquake by Yih-Min Wu, Himanshu Mittal, Ting-Chung Huang, Benjamin M. Yang, Jyh-Cherng Jan, and Sean Kuanhsiang Chen

ABSTRACT

On 6 February 2018, an M w 6.4 earthquake struck near the city of Taiwan, the EP subducts eastward under the PSP (Tsai et al., of Hualien, in eastern Taiwan with a focal depth of 10.4 km. 1977; Wu et al., 2008). Tectonically, most of Taiwan is under The earthquake caused strong shaking and severe damage to northwest–southeast compression, with a measured conver- many buildings in Hualien. The maximum intensity during this gence rate of ∼8cm=yr (Yu et al., 1997). The damaging earth- earthquake reached VII (> 0:4g) in the epicentral region, which quakes happening in Taiwan can be divided into two is extreme in Taiwan and capable of causing damage in built categories, those associated with the plate boundary and the structures. About 17 people died and approximately 285 were others associated with active faults in western Taiwan (Wu injured. Taiwan was one of the first countries to implement an et al., 1999). earthquake early warning (EEW) system that is capable of issu- The Hualien earthquake is located in northeastern Tai- ing a warning prior to strong shaking. In addition to the official wan, where seismic activity is due to the of the EEW run by the (CWB), a low-cost PSP under the EP. In the neighborhood of the epicenter, there EEW system (P-alert) has been deployed by National Taiwan is the Milun fault (Fig. 1a). It caused an M 7.3 earthquake in University (NTU). The P-alert network is currently operational October 1951, although it has not been active for the past two and is capable of providing on-site EEW as well as a map of decades. A detailed study of in-land seismogenic features in expected ground shaking. In the present work, we demonstrate Taiwan by Shyu et al. (2016) suggests that the region is a the performance of the P-alert network during the 2018 curved reverse-fault system. The system consists of two reverse Hualien earthquake. The shake maps generated by the P-alert faults, the Milun fault and the other east–west-striking reverse network were available within 2 min and are in good agreement fault located offshore north of the Milun tableland (Fig. 1a). with the patterns of observed damage in the area. These shake The recent earthquake of 6 February 2018 may be associ- maps provide insights into rupture directivity that are crucial ated with the Milun fault. The peak ground acceleration (PGA) for earthquake engineering. During this earthquake, individual at some stations recorded by the P-alert network reached around P-alert stations acted as on-site EEWsystems and provided 2–8s 0:6g, giving rise to a maximum intensity of VII. This earthquake lead time in the blind zone around the epicenter. The coseismic was well located by the Central Weather Bureau (CWB) (offi- deformation (Cd) is estimated using the records of P-alert sta- cial agency in Taiwan). Based on the CWB rapid-reporting sys- 2 tions. The higher Cd values (Cd > ) in the epicentral region tem (Wu et al.,1997, 2000), the earthquake was located 18.3 km are very helpful for authorities for the purpose of responding to north-northeast of city of Hualien, with a focal depth 10.0 km. damage mitigation. The Broadband Array in Taiwan for Seismology (BATS) and the U.S. Geological Survey (USGS) reported focal mechanisms for this event (Fig. 1a), and both confirm oblique slip faulting INTRODUCTION with two nodal planes striking northeast–southwest and north- west–southeast. The notion that the Milun fault is responsible Located on the western circum-Pacific seismic belt, Taiwan is for this earthquake is based on the structure and location of the frequently struck by earthquakes. The collision between the Milun fault, which is mapped as a reverse fault with a strike-slip nearby (PSP) and Eurasian plates (EP) component. The obtained focal mechanism seems to support contributes to the complex geological features of Taiwan. this finding. This earthquake caused severe damage to many To the east of Taiwan, the PSP subducts northward under buildings in the city of Hualien. About 17 people were reported the EP along the Ryukyu trench (Fig. 1a). To the southwest dead and 285 injured (Wikipedia, 2018). Maximum causalities

doi: 10.1785/0220180170 Seismological Research Letters Volume 90, Number 1 January/February 2019 19

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/90/1/19/4603292/srl-2018170.1.pdf by National Taiwan Univ - Lib Serials Dept user on 23 April 2019 (a) (b)

24.2°24.2° 550 Damaged building 900 600

750 850 Okinawa trough USGS BATS 800

Eurasian plate 750 Ryukyu Trench 2018.02.06 M 6.4 2018.02.06 Mw 6.4 w 700

24.1°24.1° 650 Philippine 650 Sea plate

Manila Trench 700 600

550 600 450 550

500 Hualien city 500 30 (m/s) S

Latitude (N) 450 350 Milun Fault V 450 400 300 24.0°24.0° 65 Milun Tableland 500 400 Pacific Ocean 350

300

250 5 km 600 5 km 200 Damaged building 550 23.9°23.9° 150 121.5° 121.6°121.6° 121.7°121.7° 121.8°121.8° 121.5°121.5° 121.6°121.6° 121.7°121.7° 121.8°121.8° Longitude (E)

▴ Figure 1. (a) Tectonics map of the Taiwan region and Hualien area. The red line indicates the Milun fault. (b) V S30 map of the Hualien area. Stars show the epicenter of the 6 February 2018 Mw 6.4 Hualien earthquake, and squares show the buildings damaged by the Hualien earthquake.

were reported in a 12-story residential building that tilted due to in the plains (Mittal et al.,2013, 2016). During the Hualien collapse of some lower floors. Correlation of this earthquake earthquake, the P-alert system performed very well and recorded with the Milun fault is also supported by the damage pattern good-quality data. A detailed shake map was provided by the P- of buildings that are within a few hundred meters of this fault. alert network within 2 min of the occurrence of the Hualien Although five of the collapsed structures are in the vicinity of the earthquake. The areas with maximum values of PGA and peak Milun fault, soil liquefaction and resonance may also play a role ground velocity (PGV) show some correlation with the damaged because the city of Hualien is built on deep sediment (Fig. 1b). areas. For example, the damaged buildings are located in the area This earthquake occurred 2 yrs after the 6 February 2016 Mei- where PGV is maximum. The network also has the capability to nong earthquake that killed 117 people (Wu et al.,2016)in image rupture directivity, a key parameter in damage assessment; southern Taiwan. Wu et al. (2016) reported the performance stations in the direction of propagation of the rupture front will of the P-alert network-based earthquake early warning record larger acceleration or velocity values, whereas stations in (EEW) system during this earthquake. the other directions will observe smaller values. This phenome- Taiwan was one of the first countries to implement an non was well documented in previous studies (Hsieh et al.,2014; EEW system that is capable of issuing a warning prior to strong Wu, 2015). For regional EEW, signals are transferred continu- shaking. The CWB only issues regional warnings for cities more ously from field stations to a central recording station situated at than 50 km away from the epicentral region. For cities within NTU, as well as at Institute of Earth Science, Academia Sinica 50 km, an on-site EEWsystem would be helpful. In addition to (IESAS) for research purposes. In addition to regional EEW, the official EEW of CWB, a low-cost P-alert system deployed by individual P-alert stations act as on-site EEW systems and pro- National Taiwan University (NTU) is in operation and is vide warnings when predefined thresholds are exceeded (Wu capable of providing on-site EEW, as well as rapid shake maps. et al.,2011, 2013; Hsieh et al.,2015). In this article, we examine This network consists of 636 P-alert stations. This network was the data obtained from the P-alert network and its performance initiated by a research group at NTU and is based on low-cost during the Hualien earthquake. microelectromechanical system (MEMS) accelerometers (Wu et al., 2013; Wu, 2015). The NTU group is also helping other P-ALERT NETWORK countries build their own EEW networks using P-alert. Most recently, a network of 100 P-alert instruments was installed The P-alert seismic network is a dense array of 636 stations in one portion of the Himalayan belt of India (Kumar et al., in Taiwan (Fig. 2). As mentioned earlier, these sensors are 2014), where large earthquakes pose great danger to people living low-cost MEMS-based acceleration sensors introduced in

20 Seismological Research Letters Volume 90, Number 1 January/February 2019

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/90/1/19/4603292/srl-2018170.1.pdf by National Taiwan Univ - Lib Serials Dept user on 23 April 2019 25.5° Each P-alert instrument provides three-component data Taipei city with a sampling frequency of 100 Hz. The system is designed Total : 699 stations 2 P-alert : 636 stations to have 16-bit resolution and a g clip level. P-alert has a low- 25.0° BATS : 27 stations pass filter that can be configured for 10 or 20 Hz. The data NCREE : 36 stations Okinawa trough received by each P-alert instrument in the field is processed to obtain the vertical peak displacement Pd for on-site EEW (Wu 24.5° Ilan County and Kanamori, 2005). The process of estimating Pd follows a two-stage integration process with a 0.075 Hz high-pass filter Hualien city (Shieh et al., 2008). Once the predefined thresholds, that is, 24.0° Eurasian plate 0 35 cm PGA 0 08 Pd > : or > : g (the CWB intensity V thresh- old), are exceeded, an on-site warning is generated (Wu et al., 23.5° 2011). The CWB intensity V is defined by PGA between 0.08 0 25 Ryukyu trench and : g, and can cause light damage. According to Wu and

Latitude (N) Kanamori (2005),aPd of 0.35 cm corresponds to a PGV of 23.0° 17 cm=s, and according to Wu et al. (2003) CWB intensity V corresponds approximately to a PGV between 17 and 49 cm=s. The data are transferred to the NTU office and are processed 22.5° for regional warning as well as the generation of real-time shake Phlippine maps. The regional warning generated by this network is cur- Sea plate rently used for research work; the CWB is the official agency in 22.0° Taiwan issuing regional warnings. For the Hualien event, both Manila trench 50 km the P-alert and CWB networks provided regional warnings 21.5° within 15 s after an earthquake occurrence. 119.5° 120.0° 120.5° 121.0° 121.5° 122.0° 122.5° Longitude (E) SHAKE MAP ▴ Figure 2. Station distribution of the P-alert network along with 0 0012 the Broadband Array in Taiwan for Seismology (BATS) and Na- Once 12 instruments are triggered with a PGA of : g, the tional Center for Research on Earthquake Engineering (NCREE) P-alert network starts plotting shake maps. These shake maps real-time strong-motion networks. are updated every minute as long as other instruments are trig- gering, and the maps are delivered automatically via e-mail to concerned persons, including the National Science and Tech- the early 1990s (Holland, 2003). Encouraged by its cost and nology Center for Disaster Reduction (NCDR), to provide efficiency, people at NTU started a pilot project in which 15 P- adequate time for relief work. These shake maps are automati- alert instruments were installed in the eastern part of Taiwan cally posted on the official P-alert webpage on Facebook. The (Wu and Lin, 2014). Coincidentally, Hualien was the area Hualien earthquake happened at 23:50:42.6 (local time), and where P-alerts stations were installed for the first time. Based the first PGA shake map was posted on Facebook at 23:50:48, on the performance of these instruments, this network was roughly 6 s later, based on data from 76 instruments (Fig. 3a). later extended to other parts of Taiwan as well (Wu et al., This shake map was updated every minute following the earthquake. The shake map was last updated 7 min later, at 2013, 2016; Wu, 2015). At present, P-alert stations are in- 23:57:08, based on data from 539 instruments. Figure 3 shows stalled densely (almost every 5 km) in every populated part the near-real-time shake maps generated by the P-alert network of Taiwan. Most of the P-alert stations are installed in two- during this event. At the time the first shake map was gener- or three-story buildings (preferably schools), where suitable ated, only the instruments closest to the epicenter had attained power and internet connectivity is available. Because P-alert maximum PGA values. Because the outermost instruments instruments are installed in the buildings, the site effects and – have not yet recorded their maximum accelerations, the outer- soil structure interaction also affect the PGA recorded by most contours are absurd. Figure 3b shows the final shake map these instruments. Real-time signals from these instruments delivered for the Hualien earthquake using all triggered P-alert are continuously transferred to a server station at NTU and instruments. In total, 539 instruments triggered and contrib- IESAS, where these data are stored and processed using Earth- uted to the shake map. worm software (Johnson et al., 1995; Chen et al., 2015). Data Figure 4 compares PGA and PGV shake maps using com- from the P-alert network is augmented by real-time strong- bined data from P-alert, BATS, and NCREE with the CWB motion data from a network of 36 stations of the National PGA shake map. We employ the distance inverse method to Center for Research on Earthquake Engineering (NCREE) interpolate the shake map, which is in accordance with the and another network of 27 stations from BATS, collectively previous study (Yang et al., 2018). The different interpolation forming a network of 699 stations. The station distribution schemes used by different versions of ShakeMap are well for all three kinds of instrumentation is shown in Figure 2. documented. Worden et al. (2010) describe the interpolation

Seismological Research Letters Volume 90, Number 1 January/February 2019 21

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/90/1/19/4603292/srl-2018170.1.pdf by National Taiwan Univ - Lib Serials Dept user on 23 April 2019 (a) 120° 121° 122° (b) 120° 121° 122°

UT Time : 20180206155000 UT Time : 20180206155100 To : 20180206155058 To : 20180206155158 Trigger sta. 76 Trigger sta. 538 25° 25° 25° 25°

24° 24° 24° 24°

23° 23° 23° 23°

PGA (Gal) PGA (Gal) 400.0 400.0 250.0 250.0 150.0 150.0 80.0 80.0 40.0 40.0 25.0 25.0 22° 15.0 22° 22° 15.0 22° 8.0 8.0 2.5 2.5

120° 121° 122° 120° 121° 122°

▴ Figure 3. Peak ground acceleration (PGA) shake map of the 6 February 2018 Mw 6.4 Hualien earthquake produced by the P-alert network. (a) The first shake map after 6 s of the earthquake occurrence with the triggering of 76 instruments. (b) The shake map after all available P-alert instruments have triggered.

procedure used in ShakeMap 3.5, but it was found to have seri- means that smooth contours are obtained. In the same way, ous issues so the methodology was overhauled in ShakeMap 4 the 0:25g contour for CWB shake map is based on data (Worden et al., 2018). One of the improvements in Worden received from two or three instruments, whereas the com- et al. (2018) was to eliminate the parameterization of their bined-data shake map generates the same contour using data smoothing/interpolation using a radius of influence from nearly 15 instruments. Toward city of Taipei in the (ROI ˆ 10 km); this eliminated islands near stations with north, the observed PGA values using combined data are found peak ground-motion outliers. between 0:008g and 0:025g, whereas for CWB network at the The results in our case demonstrate that, although the same places, observed PGA values are found to be less than shape of shaking contours is the same for the two networks, 0:008g. All these differences in PGA contours may be attrib- the combined networks’ PGA shake map is smoother than the uted to sparse instrumentation of the CWB. CWB network. The CWB network provides a shake map based on 110 real-time instruments, so the distribution of con- ON-SITE EEW tours is more scattered and unreasonable. P-alert instruments are less densely installed in the area near Hualien, as compared The CWB takes around 15 s to issue a regional warning (Chen to other parts of Taiwan, because of difficult logistics in moun- et al., 2015) and does so only for cities at least 50 km away from tain areas. However, the shake map generated by the combined the epicentral region. Lead time in Taiwan may range from a networks is more reasonable because much less interpolation is few seconds to 30 s, depending on the distance from the epi- needed. For example, there are four or five spurious 0:4g con- center. The lead time at a particular location is defined as the tours or islands in the CWB shake map. These contours could time gap between the issuing of warning and the arrival of not be smoothed because the interpolation is based on the strong shaking (maximum PGA or PGV). For cities located instruments placed within a 5 km range. On the other hand, within 50 km, no regional warning is possible; this is the blind the addition of a few P-alert instruments close to the epicenter zone. Because the P-alert system is capable of issuing both

22 Seismological Research Letters Volume 90, Number 1 January/February 2019

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/90/1/19/4603292/srl-2018170.1.pdf by National Taiwan Univ - Lib Serials Dept user on 23 April 2019 (a) (b) (c) PGA (Gal) PGV (cm/s) PGA (Gal) 25.5° PGA PGV CWB 585 All stations 585 All stations 102 All stations 25.0° 400 75 400

24.5° 250 49 250

24.0° 80 17 80

23.5° 25 5.7 25

Latitude (N) 23.0° 8.0 1.9 8.0

22.5° 2.5 0.65 2.5

Tainan city Tainan city Tainan city 22.0° 0.8 0.22 0.8

Kaohsiung city Kaohsiung city Kaohsiung city 21.5° 0 0 0 119.7°120.2° 120.7° 121.2° 121.7° 122.2° 119.7°120.2° 120.7° 121.2° 121.7° 122.2° 119.7°120.2° 120.7° 121.2° 121.7° 122.2° Longitude (E)

▴ Figure 4. Comparison of shake maps from the combined (P-alert, BATS, and NCREE) networks and Central Weather Bureau (CWB) stations alone. (a) PGA shake map from combined networks. (b) Peak ground velocity (PGV) shake map from combined networks. (c) PGA shake map from CWB stations alone.

on-site and regional warnings, P-alerts installed in the blind were recorded at instruments close to the epicenter, the corre- zone alleviate this problem. Each P-alert in the blind zone acts sponding PGVs at these stations are reduced, as compared to as an individual EEW system and issues a warning when the other stations southwest of the epicenter where buildings were 0 08 PGA or Pd is greater than : g or 0.35 cm, respectively (Wu damaged. This finding reinforces the notion that PGV is a bet- and Kanamori, 2005; Wu et al., 2011). The P-alert network ter indicator of building damage than PGA (Wu et al., 2004, worked very well for on-site warning during the Hualien earth- 2016). Fling effect characterized by a large amplitude velocity quake and provided useful lead time. pulse and a monotonic step in the displacement time history Figure 5 shows the lead time for PGA and PGV around (Kalkan and Kunnath, 2006) is clearly visible in velocity traces the epicenter. The warning time for PGA or PGV ranges from close to the damaged area. 1.5 to 8 s. The closest instruments to the epicenter are W011 The P-alert data can be used efficiently in many applica- and W029, where PGA lead time is 1.1 and 2.1 s, respectively. tions other than EEW and shake-map generation. Yin et al. In the same way, PGV lead time for these two stations is found (2016) and Hsu et al. (2018) used P-alert data for structural to be 1.7 and 3.0 s, respectively. The average PGA lead time for health monitoring (SHM), whereas Jan et al. (2017) used these instruments in the blind zone is 4.5 s, whereas for PGV it is data to estimate coseismic deformation (Cd) using a two-stage 4.2 s. The instruments placed closest to the damaged building baseline and integration approach (Wu and Wu, 2007). Jan that caused 17 fatalities reported relatively high lead times et al. (2017) computed Cd values using records from P-alert (6–8 s). stations and compared them with Cd values from Global Po- sitioning System data and records from the Taiwan Strong DATA QUALITY OF P-ALERT NETWORK Motion Instrumentation Program. The results match well. Fig- ure 7 shows the Cd value obtained from two stations, W002 The P-alert instruments recorded timely, good-quality data and W028, close to the damaged-building area. Cd reached during this earthquake. Almost all the instruments placed in 57:80  4:50 cm on the east–west (EW) component and the vicinity of the epicenter transferred data to the central sta- 14:05  2:53 cm on the Z component at W002. In same tion without gaps. This represents a significant improvement way, at W028, 61:45  5:01 cm was observed on the north– since the the Meinong earthquake of 2016, when some of the south (NS) component and 6:50  0:94 cm on the Z compo- 2cm stations provided only the initial few seconds of data and lost nent. High Cd estimates (Cd > ) in the epicentral region the rest of the data because of a break in internet connectivity based on P-alert data can help authorities respond appropri- (Wu et al., 2016). Figure 6 shows three-component time series ately for the purpose of emergency response and damage mit- from P-alert stations close to the epicenter. Though high PGAs igation. Figure 8 shows a section plot of traces from the P-alert

Seismological Research Letters Volume 90, Number 1 January/February 2019 23

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/90/1/19/4603292/srl-2018170.1.pdf by National Taiwan Univ - Lib Serials Dept user on 23 April 2019 (a) (b) 24.4° W031 W031 189.4 8.8 0.6 0.6

W012 Station code W012 Station code 99.5 Peak acceleration(gal) 5.7 Peak velocity(cm/s) 4.0 Leading time(s) 3.1 Leading time(s) W029 W029 294.8 16.6 W023 2.1 W023 3.0 24.2° W014 431.0 W014 23.3 604.5 2.7 30.2 1.4 3.9 2018/02/06 4.0 2018/02/06 W007 W007 W083 573.0 Mw 6.4 W083 28.1 Mw 6.4 148.2 3.5 4.0 3.7 0.4 W011 W00E 0.4 W011 W00E W00B 147.6 252.1 W00B 16.4 71.4 371.5 1.1 6.3 Hualien County 31.1 1.7 6.4 3.5 4.8 W002 W002 24.0° W013 251.9 W013 61.8 D005 240.6 5.7 D005 12.9 8.2 139.1 4.2 W008 Pacific Ocean 21.4 4.7 W008 Pacific Ocean 7.0 241.7 W00F 3.4 38.7 W00F Latitude (N) 4.7 W028 4.2 W028 336.9 317.0 125.5 104.4 W006 7.0 7.3 W006 6.4 7.6 144.3 W10F 18.4 W10F 4.5 252.3 3.5 27.9 6.3 4.2 W00C W00C 110.1 D004 12.0 D004 4.8 165.9 4.2 24.2 23.8° D001 6.2 D001 3.3 W00D 234.2 W00D 24.3 96.8 3.2 W022 5.5 3.3 W022 2.4 386.1 3.1 36.7 5.0 5.1 W003 W01D W003 W01D 299.3 201.3 14.9 19.5 4.6 8.1 10 km 4.8 5.3 10 km

Damaged building Damaged building 23.6° 121.2° 121.4° 121.6° 121.8° 122° 121.2° 121.4° 121.6° 121.8° 122° Longitude (E)

▴ Figure 5. Early warning lead time distribution for the P-alert network during the Hualien earthquake. (a) Lead time based on PGA. (b) Lead time based on PGV. Solid squares show the location of the damaged buildings.

NS componentEW component Z component 24.4° PGV=16.6 cm/s, Dist=4 km PGV=8.0 cm/s, Dist=4 km PGV=3.9 cm/s, Dist=4 km W029

PGV=16.4 cm/s, Dist=4 km PGV=13.0 cm/s, Dist=4 km PGV=6.6 cm/s, Dist=4 km W011

PGV=23.3 cm/s, Dist=6 km PGV=17.4 cm/s, Dist=6 km PGV=8.8 cm/s, Dist=6 km W023 24.2° PGV=19.6 cm/s, Dist=7 km PGV=30.2 cm/s, Dist=7 km PGV=6.6 cm/s, Dist=7 km W014 Milun fault PGV=28.1 cm/s, Dist=9 km PGV=22.0 cm/s, Dist=9 km PGV=4.8 cm/s, Dist=9 km W007

PGV=24.2 cm/s, Dist=13 km PGV=31.1 cm/s, Dist=13 km PGV=9.4 cm/s, Dist=13 km Damaged building W00B 24.0° PGV=57.1 cm/s, Dist=16 km PGV=71.4 cm/s, Dist=16 km PGV=19.8 cm/s, Dist=16 km W00E

Latitude (N) PGV=56.9 cm/s, Dist=17 km PGV=61.8 cm/s, Dist=17 km PGV=26.3 cm/s, Dist=17 km W002

PGV=64.4 cm/s, Dist=19 km PGV=125.5 cm/s, Dist=19 km PGV=38.0 cm/s, Dist=19 km W028 23.8° PGV=75.7 cm/s, Dist=19 km PGV=104.4 cm/s, Dist=19 km PGV=20.5 cm/s, Dist=19 km W00F

PGV=38.7 cm/s, Dist=25 km PGV=19.8 cm/s, Dist=25 km PGV=17.1 cm/s, Dist=25 km W008 10 km PGV=26.3 cm/s, Dist=33 km PGV=36.7 cm/s, Dist=33 km PGV=16.1 cm/s, Dist=33 km W022 23.6° 121.2° 121.4° 121.6° 121.8° 0 1020304050600 1020304050600 102030405060 Longitude (E) Time (s) PGV (cm/s) 0.00 0.22 0.65 1.90 5.70 17.00 49.00 75.00

▴ Figure 6. A series of velocity traces recorded by P-alert during the Hualien earthquake. They represent the data quality of the network for north–south (NS), east–west (EW) and Z components.

network. After applying a Gaussian filter with a 0.2 Hz cutoff, DISCUSSION AND CONCLUSIONS a series of surface wave appears, with dispersion in time for traces away from the epicenter. This suggests that the records The first P-alert stations were installed in 2010 in the Hualien obtained from P-alert network can be used in surface-wave in- area (Wu and Lin, 2014). Encouraged by the network’s version as well. efficiency and robustness, it was extended to other areas of

24 Seismological Research Letters Volume 90, Number 1 January/February 2019

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/90/1/19/4603292/srl-2018170.1.pdf by National Taiwan Univ - Lib Serials Dept user on 23 April 2019 W002.NS component W028.NS component 200 200 150 150 100 100 50 50 0 0 −50 −50 −100 −100 −150 −150 −200 −200

60 100 50 80 40 60 30 20 40 10 20 0 0 −10 −20 −20 −30 −40 −40 −60 −50 −80 −60 −100 50 80 40 30 60 20 40 10 20 0 0 −10 −20 −20 −40 Disp.(cm)−30 Vel.(cm/s) Acc.(Gal) −40 Disp.=25.25cm 1 St. dev.=4.22cm −60 Disp.=61.45cm 1 St. dev.=5.19cm −50 −80 W002.EW component W028.EW component 250 200 280 150 210 100 140 50 70 0 0 −50 −70 −100 Acc.(Gal) −140 −150 −210 −200 −280 −250 80 150 60 120 90 40 60 20 30 0 0 −20 −30 −60

Vel.(cm/s) −40 −90 −60 −120 −80 −150 100 80 80 60 60 40 40 20 20 0 0 −20 −20 −40 −40 Disp.(cm) −60 −60 −80 Disp.=57.80cm 1 St. dev.=4.98cm −80 Disp.=10.34cm 1 St. dev.=9.56cm −100 W002.Z component W002.Z component 120 90 120 90 60 60 30 30 0 0 −30 −30 −60 −60 −90 −90 −120 −120 30 20 20 10 10 0 0 −10 −10 −20 −20 −30

30 20 20 10 10 0 0 −10 −10 Disp.(cm)−20 Vel.(cm/s) Acc.(Gal) −30 Disp.=14.05cm 1 St. dev.=2.53cm −20 Disp.=6.50cm 1 St. dev.=0.94cm

0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 Time (s)

▴ Figure 7. Coseismic deformation at stations W002 and W028 during the Hualien earthquake.

Taiwan. Since then, the network has recorded some moderate- recorded by the P-alert network. PGA shake maps were deliv- to-large earthquakes and many small earthquakes. Its perfor- ered in real time following this earthquake (Fig. 3). Later, these mance during previous moderate earthquakes has been de- data were augmented by the data from NCREE and BATS scribed earlier in other studies (Wu et al., 2013, 2016; instruments to produce PGA (Fig. 4a) and PGV (Fig. 4b) Hsieh et al., 2014; Wu, 2015). The number of instruments shake maps. Figure 4c shows the PGA shake map produced is increasing every year. Two years ago, at the time of the by the 110 instruments of CWB network. Although both net- 2016 Meinong earthquake, 581 P-alerts were installed in Tai- works show similar patterns in the contour map of PGA, that wan; to date, that number has increased to 636. It is the low produced by combined networks looks more reasonable, con- cost (around one-tenth of a traditional accelerometer) that al- sidering the smoothness of PGA contours. The area between lows us to expand this network. 0:025g and 0:25g contours in CWB shake map shrinks toward The 2018 Hualien earthquake was the first destructive south because of the low density of instruments in that area. earthquake after the Meinong earthquake of 2016 to be widely These maps provide insight about PGA and shaking intensity,

Seismological Research Letters Volume 90, Number 1 January/February 2019 25

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/90/1/19/4603292/srl-2018170.1.pdf by National Taiwan Univ - Lib Serials Dept user on 23 April 2019 340

320

300

280 Time (s) Time 260

240

220

200 50 75 100 125 150 175 200 225 250 Offset (km)

▴ Figure 8. Section plot of P-alert traces. The signature of the surface wave is evident, including dispersion.

allowing authorities, including NCDR, to conduct rapid loss Shake maps are also helpful in identifying rupture direc- assessments following a destructive earthquake. tion (Wu, 2015; Wu et al., 2016; Legendre et al., 2017). The The maximum intensities were reported in central Hua- focal mechanism for this event was released by BATS and lien County and the nearby Yilan County, where the shaking USGS a few minutes after the occurrence of the earthquake. intensity reached VII (> 0:4g), as reported by the CWB. Five According to the focal mechanism provided by BATS and buildings (shown as squares in Fig. 9) were structurally dam- USGS, the two nodal planes are striking in the northeast– aged in the Hualien area. This earthquake caused 17 fatalities southwest and the northwest–southeast directions. The maxi- and several injured. A 12-story building tilted to a dangerous mum PGV is found at stations southwest of the epicenter. angle because of this event, claiming many lives. The damaged Also, the aftershock distribution is concentrated toward the buildings are located in an area of high PGV (> 49 cm=s, see southwest (Fig. 9a). Comparing PGA and PGV shake maps with the focal mechanism and the aftershock sequence, it is Fig. 9). This seems quite different from the 2016 Meinong – earthquake (Wu et al., 2016), where buildings collapsed in clear that the more likely fault plane is northeast southwest, the areas of PGV around 17 cm=s. One possible explanation where the aftershock sequence fits the shape of the shaking contours. This finding confirms the destruction in the city for this is that the structural quality in the Hualien area is of Hualien, which is situated in southwest of epicenter. It is much better than in the Tainan area. Only one or two build- also confirmed by rupture slip inversion from waveform mod- ings falling in the zone of PGV > 17 cm=s suffered structural eling (Lee et al., 2018). damage during the Hualien earthquake. Though the magni- The data for research work from the P-alert network can be tudes of both the 2016 Meinong earthquake and the 2018 downloaded from an open website (shown in Data and Resour- Hualien earthquake were similar, less destruction was reported ces). The data are in Seismic Analysis Code format in units of during the Hualien earthquake. This difference may be attrib- cm=s2 and polarity positive in vertical, north, and west. Most uted to the deeper sediment accumulation in the Tainan area as of the P-alert stations are installed in schools, so warnings were compared to Hualien. A strong energy pulse is observed at confined to schools. Because this earthquake occurred during the 2.5 s, at which PGA almost reached around 0:4g and PGV night, the maximum potential benefit of the network was not has a pulse-like phase of about 100 cm=s or more at the period achieved. If the earthquake had happened during daytime, of 1–2 s (Fig. 6). Assuming a rupture velocity of 3:5km=s and alarms would have gone off in the schools, and children are well a strong energy pulse at 2.5 s, the asperity would be about 8 km trained to respond to such panic situations in Taiwan because (Ruiz et al., 2011) from the epicenter and could be just under- earthquake drills are conducted frequently. If P-alert stations neath the damaged buildings. were installed more densely, for example, if it they were required

26 Seismological Research Letters Volume 90, Number 1 January/February 2019

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/90/1/19/4603292/srl-2018170.1.pdf by National Taiwan Univ - Lib Serials Dept user on 23 April 2019 (a) (b) PGA (Gal) PGV (cm/s) 24.6° 24.6° FS AS Ilan County M6 M5 USGS M W 6.4 M4 400.0 Ilan County 75.00 M3 M2 24.4° M1 24.4° 250.0 49.00

BATS M W 6.3 Pacific Ocean 80.0 17.00 24.2° 24.2° 2018/02/06 M W 6.4 25.0 5.70 Hualien Latitude (N) 24.0° 24.0° County 8.0 1.90 Damaged building

Pacific Ocean 2.5 Hualien 0.65 23.8° 23.8° County 0.8 0.22 10 km 10 km

23.6° 0.0 23.6° 0.00 121.2° 121.4° 121.6° 121.8° 122.0° 122.2° 121.2° 121.4° 121.6° 121.8° 122.0° 122.2° Longitude (E) Longitude (E)

▴ Figure 9. (a) PGA shake map with focal mechanisms, foreshock, and aftershock sequence. (b) PGV shake map.

in all high-occupancy commercial buildings, on-site EEW in the P-alert network can be downloaded at http://palert.earth blind zones could save many lives. .sinica.edu.tw/db/ (last accessed August 2018). In conclusion, we report the performance of the P-alert network during the 2018 Hualien earthquake, an event that ACKNOWLEDGMENTS caused severe shaking in northeastern Taiwan. Overall, the re- sults are quite satisfactory. The automatically generated shake authors are profusely thankful to the Ministry of Science and map faithfully depicted the shaking resulting from the earth- Technology (MOST) in Taiwan for funding the project quake. The first PGA shake map was posted on Facebook, (MOST 106-2116-M-002-019-MY3), under which this study roughly 6 s later after the earthquake, with the triggering of was carried out. This work was also financially supported 76 instruments. The final shake map got updated 7 min later, by the National Taiwan University (NTU) Research Center with the triggering of 539 instruments. These timely shake for Future Earth from The Featured Areas Research Center maps provided adequate time to NCDR for relief work. – Program within the framework of the Higher Education The on-site EEW has around a 2 8 s lead time, considering Sprout Project by the Ministry of Education (MOE) in the relative lower density deployed in eastern side. In terms of Taiwan. The constructive comments from Editor-in-Chief data availability, no outages were suffered during this event. In Zhigang Peng and one anonymous reviewer helped in improv- fact, the data quality is so high that it can be used for other ing the article. Authors are especially thankful to Nick Acker- seismological studies, including SHM and estimation of Cd. ley, Canadian Hazard Information Service, whose constructive The P-alert seismic network is robust and functions as needed comments and thorough checking of the article helped in im- during severe shaking. proving the overall shape of the article.

DATA AND RESOURCES REFERENCES

The strong-motion waveform records used in this study were Chen, D. Y., N. C. Hsiao, and Y. M. Wu (2015). The Earthworm based obtained from the National Taiwan University (NTU), the earthquake alarm reporting system in Taiwan, Bull. Seismol. Soc. Am. 105, 568–579, doi: 10.1785/0120140147. Institute of Earth Sciences (IES) of Academia Sinica, and the Holland, A. (2003). Earthquake data recorded by the MEMS accelerom- National Center for Research on Earthquake Engineering eter: Field testing in Idaho, Seismol. Res. Lett. 74, 20–26, doi: (NCREE) of Taiwan. The Generic Mapping Tools (GMT) 10.1785/gssrl.74.1.20. software from Wessel and Smith (1998) was used in plotting Hsieh, C. Y., W. A. Chao, and Y. M. Wu (2015). An examination of the part of the figures and is gratefully acknowledged. The official threshold-based earthquake early warning approach using a low cost seismic network, Seismol. Res. Lett. 86, 1664–1667, doi: 10.1785/ page of P-alert shake map on Facebook is at https://www 0220150073. .facebook.com/Palert. Shakemap/ (last accessed August 2018). Hsieh, C. Y.,Y. M. Wu, T. L. Chin, K. H. Kuo, D. Y. Chen, K. S. Wang,Y. The strong-motion waveform records for research work from T. Chan, W. Y. Chang, W. S. Li, and S. H. Ker (2014). Low cost

Seismological Research Letters Volume 90, Number 1 January/February 2019 27

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/90/1/19/4603292/srl-2018170.1.pdf by National Taiwan Univ - Lib Serials Dept user on 23 April 2019 seismic network practical applications for producing quick shaking Wu, Y. M., and T. L. Lin (2014). A test of earthquake early warning maps in Taiwan, Terr. Atmos. Ocean. Sci. 25, 617–624, doi: system using low cost accelerometer in Hualien, Taiwan, in Early 10.3319/TAO.2014.03.27.01(T). Warning for Geological Disasters—Scientific Methods and Current Hsu, T. Y., R. C. Yin, and Y. M. Wu (2018). Evaluating post-earthquake Practice, F. Wenzel and J. Zschau (Editors), Springer-Berlin- building safety using economical MEMS seismometers, Sensors 18, Heidelberg, New York, New York, 253–261, ISBN: 978-3-642- 1437, doi: 10.3390/s18051437. 12232-3, doi: 10.1007/978-3-642-12233-0_13. Jan, J. C., W. A. Chao, Y. M. Wu, C. C. Chen, and C. H. Lin (2017). Wu, Y. M., and C. F. Wu (2007). Approximate recovery of coseismic How well can we extract the permanent displacement from low- deformation from Taiwan strong-motion records, J. Seismol. 11, cost MEMS accelerometers?, Sensors 17, 2643, doi: 10.3390/ 159–170. s17112643. Wu, Y. M., C. H. Chang, L. Zhao, T. L. Teng, and M. Nakamura (2008). Johnson, C. E., A. Bittenbinder, B. Bogaert, L. Dietz, and W. Kohler A comprehensive relocation of earthquakes in Taiwan from 1991 (1995). Earthworm: A flexible approach to seismic network process- to 2005, Bull. Seismol. Soc. Am. 98, 1471–1481, doi: 10.1785/ ing, IRIS Newsletter 14, 4. 0120070166. Kalkan, E., and S. K. Kunnath (2006). Effects of fling step and forward Wu, Y. M., C. C. Chen, T. C. Shin, Y. B. Tsai, W. H. K. Lee, and T. L. directivity on seismic response of buildings, Earthq. Spectra 22, Teng (1997). Taiwan rapid earthquake information release system, 367–390, doi: 10.1193/1.2192560. Seismol. Res. Lett. 68, 931–943. Kumar, A., H. Mittal, B. P. Chamoli, A. Gairola, R. S. Jakka, and A. Wu, Y. M., D. Y. Chen, T. L. Lin, C. Y. Hsieh, T. L. Chin, W. Y. Chang, Srivastava (2014). Earthquake early warning system for W. S. Li, and S. H. Ker (2013). A high density seismic network northern India, 15th Symposium on Earthquake Engineering, Indian for earthquake early warning in Taiwan based on low cost sensors, Institute of Technology, Roorkee, 11–13 December 2014, Seismol. Res. Lett. 84, 1048–1054, doi: 10.1785/0220130085. 231–238. Wu, Y. M., J. K. Chung, T. C. Shin, N. C. Hsiao, Y. B. Tsai,W. H. K. Lee, Lee, S. J., T. C. Lin, T. Y. Liu, and T. P. Wong (2018). Fault-to-fault and T. L. Teng (1999). Development of an integrated earthquake jumping rupture of the 2018 M w 6.4 Hualien earthquake in eastern early warning system in Taiwan—Case for Hualien earthquakes, Taiwan, Seismol. Res. Lett. (in this issue). Terr. Atmos. Ocean. Sci. 10, 719–736. Legendre, C. P., T. L. Tseng, H. Mittal, C. H. Hsu, A. Karakhanyan, and Wu, Y. M., N. C. Hsiao, and T. L. Teng (2004). Relationships B. S. Huang (2017). Complex wave propagation revealed by peak between strong ground motion peak values and seismic loss during ground velocity maps in the Caucasus area, Seismol. Res. Lett. 88, the 1999 Chi-Chi, Taiwan earthquake, Nat. Hazards 32, 357–373. 812–821, doi: 10.1785/0220160178. Wu, Y. M., W. H. K. Lee, C. C. Chen, T. C. Shin, T. L. Teng, and Y. B. Mittal, H., A. Kumar, and Kamal(2013). Ground motion estimation in Tsai (2000). Performance of the Taiwan rapid earthquake informa- Delhi from postulated regional and local earthquakes, J. Seismol. 17, tion release system (RTD) during the 1999 Chi-Chi (Taiwan) earth- 593–605, doi: 10.1007/s10950-012-9340-5. quake, Seismol. Res. Lett. 71, 338–343. Mittal, H., A. Kumar, Y. M. Wu, Kamal, and A. Kumar (2016). Source Wu, Y. M., W. T. Liang, H. Mittal, W. A. Chao, C. H. Lin, B. S. Huang, study of M w 5.4 April 4, 2011 India–Nepal border earthquake and and C. M. Lin (2016). Performance of a low-cost earthquake early scenario events in the Kumaon-Garhwal region, Arabian J. Geosci. warning system (P-alert) during the 2016 M L 6.4 Meinong 9, 348, doi: 10.1007/s12517-016-2330-0. (Taiwan) earthquake, Seismol. Res. Lett. 87, 1050–1059, doi: Ruiz, S., E. Kausel, J. Campos, G. R. Saragoni, and R. Madariaga (2011). 10.1785/0220160058. Identification of high frequency pulses from earthquake asperities Wu, Y. M., T. L. Lin,W. A. Chao, H. H. Huang, N. C. Hsiao, and C. H. along Chilean subduction zone using strong motion, Pure Appl. Chang (2011). Faster short-distance earthquake early warning using Geophys. 168, 125–139. continued monitoring of filtered vertical displacement—A case Shieh, J. T., Y. M. Wu, and R. M. Allen (2008). A comparison of and for study for the 2010 Jiasian earthquake, Taiwan, Bull. Seismol. Soc. magnitude estimation in earthquake early warning, Geophys. Res. Am. 101, 701–709, doi: 10.1785/0120100153. Lett. 35, L20301, doi: 10.1029/2008GL035611. Wu, Y. M., T. L. Teng, T. C. Shin, and N. C. Hsiao (2003). Relationship Shyu, J. B. H., Y. R. Chuang, Y. L. Chen, Y. R. Lee, and C. T. Cheng between peak ground acceleration, peak ground velocity, and inten- (2016). A new on-land seismogenic structure source database from sity in Taiwan, Seismol. Res. Lett. 93, 386–396. the Taiwan earthquake model (TEM) project for seismic hazard Yang, B. M., T. C. Huang, and Y. M. Wu (2018). ShakingAlarm: A analysis of Taiwan, Terr. Atmos. Ocean. Sci. 27, 311–323, doi: nontraditional regional earthquake early warning system based 10.3319/TAO.2015.11.27.02(TEM). on time-dependent anisotropic peak ground-motion attenuation Tsai, Y. B., T. L. Teng, J. M. Chiu, and H. L. Liu (1977). Tectonic im- relationships, Bull. Seismol. Soc. Am. 108, 1219–1230, doi: plications of the seismicity in the Taiwan region, Mem. Geol. Soc. 10.1785/0120170105. 2, 13–41. Yin, R. C., Y. M. Wu, and T. Y. Hsu (2016). Application of the low-cost Wessel, P., and W. H. F. Smith (1998). New, improved version of Generic MEMS-type seismometer for structural health monitoring: A pre- Mapping Tools released, Eos Trans. AGU 79, 579. study, 2016 IEEE International Instrumentation and Measurement Wikipedia (2018). 2018 Hualien Earthquake, available at https://en Technology Conf. Proceedings (I2MTC),1–5, doi: 10.1109/ .wikipedia.org/wiki/2018_Hualien_earthquake (last accessed July I2MTC.2016.7520389. 2018). Yu, S. B., H. Y. Chen, and L. C. Kuo (1997). Velocity field of GPS Worden, C. B., E. M. Thompson, J. W. Baker, B. A. Bradley, N. Luco, stations in the Taiwan area, Tectonophysics 274, 41–59. and D. J. Wald (2018). Spatial and spectral interpolation of ground- motion intensity measure observations, Bull. Seismol. Soc. Am. 108, 866–875. Yih-Min Wu1 Worden, C. B., D. J. Wald, T. I. Allen, K. Lin, D. Garcia, and G. Cua Himanshu Mittal (2010). A revised ground-motion and intensity interpolation Ting-Chung Huang scheme for ShakeMap, Bull. Seismol. Soc. Am. 100, 3083–3096. Wu, Y. M. (2015). Progress on development of an earthquake early Benjamin M. Yang warning system using low cost sensors, Pure Appl. Geophys. 172, Sean Kuanhsiang Chen 2343–2351, doi: 10.1007/s00024-014-0933-5. Department of Geosciences Wu, Y. M., and H. Kanamori (2005). Rapid assessment of damaging po- National Taiwan University tential of earthquakes in Taiwan from the beginning of P waves, Number 1, Section 4, Roosevelt Road Bull. Seismol. Soc. Am. 95, 1181–1185, doi: 10.1785/0120040193.

28 Seismological Research Letters Volume 90, Number 1 January/February 2019

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/90/1/19/4603292/srl-2018170.1.pdf by National Taiwan Univ - Lib Serials Dept user on 23 April 2019 Taipei 10617, Taiwan Taipei 11529, Taiwan [email protected] [email protected] [email protected] [email protected] Published Online 5 September 2018 [email protected] [email protected] 1 Also at NTU Research Center for Future Earth, National Taiwan Uni- versity, Taipei 10617, Taiwan; [email protected]; and Also at Insti- Jyh-Cherng Jan2 tute of Earth Sciences, Academia Sinica, Taipei 11529, Taiwan. 2 Institute of Earth Sciences Also at Department of Earth Sciences, National Central University, Academia Sinica Jhongli 320, Taiwan. Number 128, Section 2, Academia Road, Nangang

Seismological Research Letters Volume 90, Number 1 January/February 2019 29

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/90/1/19/4603292/srl-2018170.1.pdf by National Taiwan Univ - Lib Serials Dept user on 23 April 2019