Tem 2020
Tem 2020
Tem 2020
Earthquake Spectra
Probabilistic seismic hazard 2020, Vol. 36(S1) 137–159
Ó The Author(s) 2020
assessment for Taiwan: TEM Article reuse guidelines:
sagepub.com/journals-permissions
Abstract
The Taiwan Earthquake Model (TEM) published the first version of the Taiwan probabil-
istic seismic hazard assessment (named TEM PSHA2015) 5 years ago. For updating to
the TEM PSHA2020, we considered an updated seismogenic structure database, includ-
ing the structures newly identified with 3D geometry, an earthquake catalog made cur-
rent to 2016, state-of-the-art seismic models, a new set of ground motion prediction
equations, and site amplification factors. In addition to earthquakes taking place on each
individual seismogenic structure, the updated seismic model included the possibility of
an earthquake occurring on multiple structures. To include fault memory for illustrating
activity on seismogenic structure sources, we incorporated the Brownian passage time
model. For the crustal seismicity that cannot be attributed to any specific structure, we
implemented both area source and smoothing kernel models. A new set of ground
motion prediction equations is incorporated. In addition to the calculation of hazard at
engineering bedrock, our assessment included site amplification factors that competent
authorities of governments and private companies could use to implement hazard pre-
vention and reduction strategies.
Keywords
Probabilistic seismic hazard assessment, Taiwan, the Taiwan Earthquake Model, multi-
ple-structure rupture, Brownian passage time model
Date received: 20 July 2020; accepted: 22 July 2020
1
Earthquake-Disaster & Risk Evaluation and Management (E-DREaM) Center, National Central University, Taoyuan City
2
Department of Earth Sciences, National Central University, Taoyuan City
3
Institute of Earth Sciences, Academia Sinica, Taipei City
4
Department of Geosciences, National Taiwan University, Taipei City
5
Department of Nuclear Backend Management, Taiwan Power Company, Taipei City
6
Institute of Applied Geology, National Central University, Taoyuan City
7
Sinotech Engineering Consultants, Inc., Taipei City
8
Department of Earth Sciences, National Cheng Kung University, Tainan City
Corresponding author:
Chung-Han Chan, Earthquake-Disaster & Risk Evaluation and Management (E-DREaM) Center, National Central
University, No. 300, Zhongda Rd., Zhongli Dist., Taoyuan City 32001.
Email: hantijun@googlemail.com
138 Earthquake Spectra 36(S1)
Introduction
In 2015, the Taiwan Earthquake Model (TEM) proposed its first version of the probabilis-
tic seismic hazard analysis (PSHA) for Taiwan, named TEM PSHA2015 (Wang et al.,
2016a). This model adopted 38 seismogenic structures identified by Shyu et al. (2016), 28
shallow-area sources, 4 subduction-interplate sources, and 12 subduction-intraplate
sources (Figure 3 of Wang et al., 2016a) to illustrate seismic activities in Taiwan and its
vicinity. This assessment considered seismic hazard only for engineering bedrock, that is,
Vs30 (i.e. averaged shear-wave velocity in the top 30 m) of 760 m/s, and therefore site
amplification was neglected. Although such an assumption is straightforward for engineer-
ing applications, its application requires incorporating site amplification that could be
overlooked by non-experts. The TEM PSHA2015 provided the first published national
seismic hazard map of Taiwan, and it provided the ground basis for understanding the
national seismic hazard potential. Several years after publishing this map, it was necessary
to conduct advanced work that incorporated recent developments in PSHA practice and
up-to-date data sets into a new PSHA model. Thus, this article details the official version
of this model by the TEM team, named TEM PSHA2020.
To upgrade seismic hazard assessment properly, an evaluation of the TEM PSHA2015
seismic model was required. Chan et al. (2019) examined the performance of the TEM
PSHA2015 through retrospectively forecasting the seismic behaviors of the 2018 Hualien,
Taiwan, sequence and seismicity during 2012 and 2016. For seismogenic structure sources,
the TEM PSHA2015 model forecasted a high earthquake probability of 53% in the com-
ing 50 years on the Milun fault, which ruptured during this sequence. Its probability could
be even higher by implementing a time-dependent Brownian passage time (BPT) model
(probability of 80% in the coming 50 years), raising the importance of the BPT model for
PSHA. Chan et al. (2019) also tested the shallow-background area sources of the TEM
PSHA2015 model. Their study compared the shallow-background area sources with the
Hualien sequence as well as the seismicity from 2012 to 2016 and confirmed their positive
correlation through a Molchan diagram (Molchan, 1990). This test implies the shallow-
background area model could forecast long-term seismic activity, applicable for subse-
quent PSHA models. Note that Chan et al. concluded similar forecasting reliability
between the area sources in the TEM PSHA2015 model and the smoothing model of Woo
(1996) although the spatial patterns of these two models are significantly different. Also,
after announcing the TEM PSHA2015, the TEM seismogenic structures database was
updated, including adding six newly identified structure sources (ID 39-44 in Table 1,
structure alignments shown in Figure 1a), and revising the geometries of some structures
from fixed dip angles (planar geometries) to depth-dependent dip angles (three-dimen-
sional geometries). In addition, based on the TEM historical earthquake database (http://
tec.earth.sinica.edu.tw/TEM/hisevent/hisdoc.php) and seismogenic structure alignments,
potential ruptures on multiple structures have been identified, and corresponding para-
meters have been summarized in Table 2.
The understanding gained from the tests and use of the updated database could
improve the quality of PSHA for Taiwan. First, we implemented an updated seismogenic
structure database, including the consideration of earthquakes from multiple-segment rup-
tures. Considering the memory time-elapse of last rupture, we evaluated the time-
dependent rupture probability for seismogenic structures. For shallow-background
sources, we revised parameters for the area sources based on the updated earthquake cata-
log. In addition to the area source model, the background seismicity rate is also repre-
sented in the form of a smoothing model. To better model ground-shaking behaviors, we
Chan et al.
Table 1. The source parameters for the 44 seismogenic structures in the TEM database.
ID Seismogenic Type Length Depth 1 Dip (°) Depth 2 Dip2 (°) Depth 3 Dip3 (°) Width Area MW Displacement Recurrence
structure name (km) (km) between (km) between (km) between (km) (km2) (m) Slip rate interval
depth 0 depth 1–2 depth 2 (mm/year) (year)
and 1 and 3
1 Shanchiao fault N 53.10 7.0 60 10.0 45 14 30 19.84 1053.50 7.01 1.29 1.66 780
2 Shuanglienpo R 11.00 3.0 45 5.0 15 – – 11.97 131.67 6.24 0.72 0.13 5540
structure
3 Yangmei structure R 22.10 3.0 60 – – – – 3.46 76.47 6.03 0.60 0.18 3330
4 Hukou fault R 25.60 10.0 30 – – – – 20.00 512.00 6.77 1.16 0.46 2520
5 Fengshan river SS 30.60 13.8 85 – – – – 13.90 425.34 6.66 0.95 3.18 300
strike-slip structure
6 Hsinchu fault R 14.50 10.0 45 – – – – 14.14 205.03 6.41 0.83 0.66 1260 (2823)
7 Hsincheng fault R 28.50 12.9 30 – – – – 25.71 732.74 6.91 1.31 1.12 1170
8 Hsinchu frontal R 12.10 10.0 30 – – – – 20.00 242.00 6.48 0.90 1.44 620 (1351)
structure
9 Touhuanping SS 25.80 12.0 85 – – – – 12.05 310.89 6.52 0.80 0.13 6150 (15,166)
structure
10 Miaoli frontal R 30.90 10.0 30 – – – – 20.00 618.00 6.84 1.22 1.84 660 (1383)
structure
11 Tunglo structure R 15.70 3.5 30 – – – – 7.00 109.90 6.17 0.68 0.50 1360
12 East Miaoli R 14.40 4.0 30 – – – – 8.00 115.20 6.19 0.69 0.84 820
structure
13 Shihtan fault R 30.70 10.8 75 – – – – 11.18 343.23 6.61 0.99 1.38 720 (1590)
14 Sanyi fault R 29.80 9.0 15 – – – – 34.77 1036.15 7.04 1.45 0.85 1710
15 Tuntzuchiao fault SS 27.00 14.8 85 – – – – 14.85 400.95 6.64 0.94 0.50 1880 (4105)
16 Changhua fault R 82.20 3.0 45 5.0 30 12 10 48.55 3990.81 7.57 2.35 1.87 1260
17 Chelungpu fault R 91.90 12.0 15 – – – – 46.36 4260.48 7.60 2.45 6.94 350
18 Tamaopu— R 69.20 6.0 30 – – – – 12.00 830.40 6.96 1.38 1.06 1300
Shuangtung fault
19 Chiuchiungkeng R 33.60 12.0 30 – – – – 24.00 806.40 6.95 1.37 4.66 290
fault
20 Meishan fault SS 25.20 14.7 85 – – – – 14.75 371.70 6.60 0.89 2.51 350 (16,869)
(continued)
139
Table 1. Continued
140
ID Seismogenic Type Length Depth 1 Dip (°) Depth 2 Dip2 (°) Depth 3 Dip3 (°) Width Area MW Displacement Recurrence
structure name (km) (km) between (km) between (km) between (km) (km2) (m) Slip rate interval
depth 0 depth 1–2 depth 2 (mm/year) (year)
and 1 and 3
21 Chiayi frontal R 34.10 12.0 15 – – – – 46.36 1580.88 7.21 1.71 3.36 510 (3006)
structure
22 Muchiliao—Liuchia R 26.40 12.0 30 – – – – 24.00 633.60 6.85 1.23 5.75 210
fault
23 Chungchou R 29.20 12.0 30 – – – – 24.00 700.80 6.89 1.28 12.20 100
structure
24 Hsinhua fault SS 14.80 15.0 85 – – – – 15.06 222.89 6.38 0.69 2.65 260
25 Houchiali fault R 12.20 5.0 45 – – – – 7.07 86.25 6.07 0.61 7.07 90 (99,655)
26 Chishan fault SS/R 40.00 10.8 75 – – – – 11.18 447.20 6.68 0.97 1.10 880
27 Hsiaokangshan fault R 11.10 7.0 30 – – – – 14.00 155.40 6.30 0.75 1.78 420
28 Kaoping River SS/R 33.40 12.3 75 – – – – 12.71 424.51 6.66 0.95 0.32 2970
structure
29 Chaochou fault SS/R 99.30 11.1 75 – – – – 11.50 1141.95 7.10 1.62 0.98 1650 (3481)
30 Hengchun fault SS/R 41.90 15.0 75 – – – – 15.53 650.71 6.85 1.20 6.15 200 (466)
31 Hengchun offshore R 19.70 4.0 30 – – – – 8.00 157.60 6.31 0.77 3.22 240
structure
32 Milun fault SS/R 32.60 10.0 75 – – – – 10.35 337.41 6.56 0.85 10.15 80
33 Longitudinal Valley R/SS 147.50 5.0 75 15.0 60 20 45 23.79 3509.02 7.52 2.25 11.35 200 (468)
fault
34 Central Range R 86.20 20.0 45 – – – – 28.28 2437.74 7.38 2.00 7.28 270
structure
35 Luyeh fault R 19.60 2.0 45 4.0 30 – – 6.83 133.87 6.24 0.71 5.28 130 (10,869)
36 Taimali coastline R/SS 43.00 10.6 75 – – – – 10.93 469.99 6.73 1.10 7.32 150
structure
37 Northern Ilan N 74.90 9.4 60 – – – – 10.87 814.16 6.90 1.14 3.29 350
structure
38 Southern Ilan N 21.90 11.3 60 – – – – 12.99 284.48 6.43 0.64 5.48 120
structure
(continued)
Earthquake Spectra 36(S1)
Table 1. Continued
ID Seismogenic Type Length Depth 1 Dip (°) Depth 2 Dip2 (°) Depth 3 Dip3 (°) Width Area MW Displacement Recurrence
Chan et al.
structure name (km) (km) between (km) between (km) between (km) (km2) (m) Slip rate interval
depth 0 depth 1–2 depth 2 (mm/year) (year)
and 1 and 3
39 Chushiang R/SS 19.80 3.0 55 – – – – 3.66 72.47 6.00 0.57 5.01 110
structure
40 Gukeng structure SS 9.20 12.0 85 – – – – 12.05 110.86 6.07 0.48 0.94 510
41 Tainan frontal R 32.90 3.0 30 12.0 15 – – 52.36 1722.64 7.24 1.74 0.92 1890 (9705)
structure
42 Longchuan R 23.10 12.0 60 – – – – 13.86 320.17 6.58 0.96 1.73 550
structure
43 Youchang sturcture R/SS 16.6 12.0 75 – – – – 12.42 206.17 6.41 0.83 1.64 510
44 Fengshan hills R 19.1 15.0 30 – – – – 30.00 573.00 6.81 1.19 0.92 1290
frontal structure
TEM: Taiwan Earthquake Model.
Based on this database, a seismogenic structure would fully rupture only during an earthquake with a characteristic magnitude shown in the table. That is, a partial rupture with a
smaller magnitude is not considered from the seismogenic structure source. Note that some of the structures obtain depth-dependent dip angles. The recurrence intervals in
parentheses were evaluated by considering slip partitioning for multiple-structure ruptures. The alignments of the seismogenic sources are presented in Figure 1a.
141
142 Earthquake Spectra 36(S1)
Figure 1. (a) Distribution of the 44 seismogenic structures in the TEM database and (b) TEM
PSHA2015 map of PGA (in g) for 10% probability in 50 years. Note that TEM PSHA2015 implemented
only 38 seismogenic structures (denoted as red alignments in (b)). Blue alignments represent newly
identified structures incorporated in this study.
Table 2. Potential multiple-structure ruptures and structure segmentations with rupture type,
corresponding rupture areas, magnitudes and recurrence intervals of potential earthquakes.
ID Seismogenic structure name Type Area MW Recurrence
(km2) interval
(year)
implemented a new set of ground motion prediction equations (GMPEs) for analysis. To
incorporate site amplification, our assessment applied an Vs30 map and new GMPEs to
describe site and path effects of strong ground motion attenuation. While we made the
effort of the TEM PSHA2020 model for the practical application, the challenges and sta-
tus of the TEM PSHA model for Taiwan continue required the long-term effort as
addressed in Gerstenberger et al. (2020), which review the state of the art and future chal-
lenges of PSHA at regional and national scale.
Multiple-structure rupture
Although the TEM PSHA2015 assumed an earthquake takes place on only one individual
seismogenic structure source, some historical earthquakes show that it is possible to have
simultaneous ruptures on multiple seismogenic structures. Implementing the multiple-
structure rupture factor could enable comprehension of some extreme cases in our hazard
144 Earthquake Spectra 36(S1)
(a) (b)
Figure 2. (a) The seismic hazard maps considering the new TEM seismogenic structure database,
including six newly identified structure sources (denoted as dark red alignments) and 3D structure
geometry and (b) the impact of this innovation. (b) The difference between Figures 2a and 1b. The unit
of the PSHA is PGA (in g) for 10% probability in 50 years (i.e. recurrence interval of 475 years).
where the a and b values are constants that can be obtained through regression of earth-
quake activity. We followed the procedure of Wang et al. (2016a), who determined a b
value for the Taiwan region through regression and obtained a b value of 1.10. Note that
Chan et al. 145
the acquisition of the earthquake catalog, data processing and analysis of the Gutenberg–
Richter law will be detailed later in the text.
Thus, the partitioned slip rates of the first and second structures (denoted as D_ 0L1 and
_D0L2 , respectively) can be represented as
D_ L1
D_ 0L1 = ð2Þ
AL1 + L2
AL1 3C1 + 1
and
D_ L2
D_ 0L2 = , ð3Þ
AL1 + L2
AL2 3C2 + 1
respectively, where D_ L1 and D_ L1 represent the original slip rates of the first and second
structures, respectively, available from Table 1; AL1 and AL2 represent the areas of the first
and second structures, respectively, available from Table 1; and AL1 + L2 represents the area
of the multiple-structure rupture. In addition, C1 represents the partitioned rate between
multiple- (D_ L1 _0
L1 + L2 ) and single-structure (DL1 ) ruptures from the first structure, represented
as
and
D_ L1 _0 ð5Þ
L1 + L2 = C1 3DL1 ,
where ML1 and ML1 + L2 represent the magnitudes of the first structure and multiple-
structure ruptures, respectively (available from Tables 1 and 2, respectively); DL1 + L2 repre-
sents the slip of the multiple-structure rupture, available from Table 2; and DL1 represents
the slip of the first structure, available from Table 1. C2 represents the partitioned rate
between multiple- (D_ L2 _0
L1 + L2 ) and single-structure (DL2 ) ruptures from the second structure,
represented as
and
D_ L2 _0 ð7Þ
L1 + L2 = C2 3DL2 ,
where ML2 represents the magnitude of the second structure, and DL2 represents the slip of
the second structure (both available from Table 1).
Thus, the slip rate for the multiple-structure rupture, D_ L1 + L2 , can be represented as
D_ L1 + L2 = D_ L1 _ L2
L1 + L2 + DL1 + L2 ð8Þ
146 Earthquake Spectra 36(S1)
DL1 + L2
RL1 + L2 = : ð9Þ
D_ L1 + L2
Through this approach, the recurrence intervals of the first and second structures (RL1
and RL2 , respectively) are estimated based on the partitioned slip rates, represented as
DL1
RL1 = ð10Þ
D_ 0 L1
and
DL2
RL2 = , respectively: ð11Þ
D_ 0 L2
For the case of the Hsinchu fault (ID 6) and Hsinchu frontal structure (ID 8) as an
example,
(a) (b)
Figure 3. (a) The seismic hazard maps considering the possibility of an earthquake on multiple
seismogenic structures and (b) the impact of this innovation. (b) The difference between Figures 3a and
2a. The unit of the PSHA is PGA (in g) for 2% probability in 50 years (i.e. recurrence interval of
2475 years). The locations of the Chaochou fault (ID 29) and Tainan frontal structure (ID 41) are
denoted.
National Seismic Hazard Maps for Japan (Fujiwara, 2014). The density function (DF) of
the BPT model can be expressed as follows:
!
m 1=2 ðt mÞ2
DF = exp , ð12Þ
pa2 t3 2a2 mt
where m represents the mean recurrence interval, t represents the time elapsed since the last
rupture of the event, and a represents the aperiodicity, usually between 0.3 and 0.7. Chan
et al. (2017) applied this model to the seismogenic structures in Taiwan and tested the
deviations of rupture probability and seismic hazard contributed by a using its two end
members. Their results show that the rupture probability differences for most cases are less
than 20% (Table 6 of Chan et al., 2017) and less than 2% of the entire study area with a
hazard level difference of more than 0.1 g (Figure 3 of Chan et al., 2017). This test sug-
gested an insignificant variance in hazard levels within a reasonable a. Thus, we assumed
a median value of 0.5 in our study.
148 Earthquake Spectra 36(S1)
Table 3. Rupture probabilities and relative rate differences determined by different models at the time
snapshots of 2018 based on corresponding recurrence intervals and time elapsed since the last event.
Fault name ID Recurrence Time- Time Time- Rate
interval independent elapsed dependent change
(years) rate (%) rate (%) (%)
(a) (b)
Figure 4. (a) The seismic hazard maps considering the BPT model and (b) the impact of this innovation.
(b) The difference between Figures 4a and 3a. The unit of the PSHA is PGA (in g) for 10% probability in
50 years (i.e. recurrence interval of 475 years). The structures incorporating the BPT model are
denoted as dark red alignments. The location of the Chelungpu (ID 17), Muchiliao-Liuchia (ID 22), and
Milun (ID 32) faults is denoted.
Figure 5. Distribution of the 28 area sources and seismicity implemented in this study. The
earthquakes’ parameters were obtained from the CWBSN catalog. We considered only the earthquakes
with M ø 4.0 from 1973 to 1993 and M ø 3.0 from 1993 to 2016. Foreshocks and aftershocks in the
catalog were removed based on the declustering approach proposed by Gardner and Knopoff (1974).
our study did not consider epistemic uncertainty resulting from various declustering
approaches. The analyzed earthquake catalog is shown in Figure 5 and was implemented
for constructing shallow-background source models, including area source and smoothing
models.
Table 4. a and b values, size (in km), and earthquake density for M ø 4.0 for 28 area sources.
Area a value b value Size (km) Density (M ø 4 per km2)
Smoothing model
Besides the area source model mentioned above, we implemented another seismogenic
model for the background seismicity based on the smoothing approach proposed by
Woo (1996). Unlike an area source model, it does not require a subjective definition of
area source geometry, which might result in epistemic uncertainty (Chan et al., 2019).
This approach summarizes seismic activity in an earthquake catalog based on the smooth-
ing kernel, which can be represented as a function of the magnitude M and the
distance between the site of interest and the epicenter of the ith earthquake, shown as
follows:
!PL
PL 1 x xi 2
K ðM, x xi Þ = 1+ , ð13Þ
pH 2 ð M Þ H ðM Þ
where PL denotes the power law index. The bandwidth function H(M) is defined as the
mean distance between each event with magnitude M and its nearest neighbor, represented
as follows:
152 Earthquake Spectra 36(S1)
(a) (b)
H ð M Þ = c edM , ð14Þ
where c and d are constants that can be obtained by regression. Through analyzing the cat-
alog mentioned in the previous section, we determined that the c and d values of the
Chan et al. 153
bandwidth function are 0.0146 and 1.1893, respectively. Thus, a seismic density model can
be proposed accordingly (Figure 6b), and the seismicity hazard can be subsequently
assessed (Figure 6d). Compared to the area source model (Figure 6c), implementing the
smoothing model obtained a higher hazard level in central-northern Taiwan and a lower
hazard at the western edge of S06 (Figure 6e). Such discrepancies could be attributed to
the spatial heterogeneity of the seismicity (Figure 5), which can be presented by a smooth-
ing model but would be averaged based on the area source model. Since Chan et al. (2019)
concluded similar levels of earthquake forecasting ability between the area source and
smoothing models, the TEM PSHA2020 model considered both models and assumed
equal weightings (50% for each) in the logic tree for the assessment.
Site effect
The TEM PSHA2015 did not include site effect, that is, its outcomes represented hazard
levels for engineering bedrock. Thus, application of the TEM PSHA2015 hazard map
required accessing site amplification for the sites of interest. Such a requirement could con-
fuse non-experts who do not have either an engineering or scientific background. For eas-
ier application for end users, our assessment included built-in site amplification that
requires an Vs30 map. We obtained the Vs30 at the 816 sites of the Taiwan Strong Motion
Instrumentation Program (TSMIP) stations from the Engineering Geological Database for
TSMIP (http://egdt.ncree.org.tw/) (solid circles with colors denoting Vs30 ). In this database,
the Vs30 at most sites are obtained from measurements (Kuo et al., 2012), whereas some are
acquired by estimation. Detailed parameters for each site can be accessed through http://
egdt.ncree.org.tw/DataList_eng.htm. After interpolation based on these data, we obtained
an Vs30 map with a resolution of 500 by 500 m2 (Figure 7c). This map shows low Vs30 at
some plains and basins, for example, Taipei Basin, Ilan Plain, and Chianan Plain. Based
on the implemented GMPEs by Lin (2009), the low Vs30 results in significant site amplifica-
tion and elevated seismic hazard (Figure 7d).
GMPE
For a seismic hazard assessment, illustrating ground-shaking behaviors requires GMPEs
representing the ground-shaking level as a function of magnitude, distance, and other fac-
tors. In the TEM PSHA2015, the implemented GMPEs for the crustal events are those by
Lin (2009) that were obtained based on the observations of the Taiwan events. This
GMPE set, however, has been criticized for overestimation in the near-field cases
(Reevaluation of Probabilistic Seismic Hazard of Nuclear Facilities in Taiwan Using
SSHAC Level 3 Methodology, http://sshac.ncree.org.tw). Our current study implemented
the GMPEs proposed by Lin et al. (2011) instead, which are also obtained based on
Taiwan observations, but have corrected near-field bias. We are aware that after the
framework of TEM PSHA2020 was determined, some references were published describ-
ing new sets of GMPEs based on the observations of the crustal events and earthquakes in
Taiwan (e.g. Chao et al., 2020; Phung et al., 2020). These new GMPEs will be incorpo-
rated in the next generation of the TEM PSHA.
By incorporating the GMPEs of Lin et al. (2011), the seismic hazard (Figure 8) shows
lower levels in the region close to some of the seismogenic structures in the plain regions
with low Vs30 (Figure 7c).
154 Earthquake Spectra 36(S1)
(a) (b)
(c) (d)
Figure 7. The seismic hazard maps considering (a) engineering bedrock and (b) site amplification using
the factor of Vs30 , (c) distribution of Vs30 , interpolated from the Taiwan region from the Engineering
Geological Database for TSMIP (solid circles with colors denoting Vs30 ), and (d) impact of the site
amplification. (d) The difference between Figure (a) and (b). The unit of the PSHA is PGA (in g) for 10%
probability in 50 years (i.e. recurrence interval of 475 years). The locations of the Taipei Basin, Ilan Plain,
and Chianan Plain are denoted.
Chan et al. 155
(a) (b)
Figure 8. (a) The seismic hazard maps considering the ground motion prediction equations proposed
by Lin et al. (2011) and (b) the impact of this innovation. (b) The difference between Figures 8a and 7b.
The unit of the PSHA is PGA (in g) for 10% probability in 50 years (i.e. recurrence interval of 475 years).
Discussion
We, as the team of the TEM, have proposed a new PSHA for Taiwan, the TEM
PSHA2020. Departing from the previous assessment (TEM PSHA2015), our assessment
included several innovations: (1) an updated seismogenic structure database, (2) rupture
probability on multiple structures, (3) a time-dependent rupture model for some seismo-
genic structures, (4) a revised area source model through an updated earthquake catalog,
(5) a smoothing model to illustrate background seismicity, and (6) a new set of GMPEs to
simulate strong ground motion behaviors. We accessed an updated seismogenic structure
database, incorporating six newly identified structure sources (ID 39–44 in Figure 1a) and
three-dimensional geometries for some structures (Table 1). And we obtained higher
hazard levels in the vicinity of these newly identified structures, whereas some faults with
longer rupture return periods (Figure 2) had a lower hazard along them. Considering
multiple-structure ruptures did not result in a significant difference in hazard levels for a
short return period (e.g. 475 years, corresponding to 10% probability in 50 years), but this
factor became crucial when the PSHA return period is assumed to be longer, especially
156 Earthquake Spectra 36(S1)
near the Chaochou fault (ID 29) and the Tainan frontal structure (ID 41), shown in Figure
3. To introduce fault memory of previous events, we adopted the BPT model to evaluate
time-dependent earthquake probability. It suggests a higher hazard near the seismogenic
structures with longer time elapsed since the last rupture and/or shorter recurrence inter-
vals, such as the Muchiliao-Liuchia (ID 22) and Milun (ID 32) faults. In contrast, earth-
quake probabilities for the structures with recent events, for example, the Chelungpu fault
(ID 17) after the 1999 earthquake, become lower, resulting in a lower hazard level in their
vicinities (Figure 4). The treatment of the background seismicity activity is based on the
updated earthquake catalog (Figure 5) for implementing both area source (Figure 6a) and
smoothing (Figure 6b) models. To be easily applied by various end users, the TEM
PSHA2020 included two versions: one based on engineering bedrock (Figure 7a) and the
other on built-in site amplification, considering an Vs30 map (Figure 7b).
Our analyses including site effect have a significant impact on hazard levels and warned
of high hazard potential in the Taipei Basin, Ilan Basin, and Chianan Plain (Figure 7).
Furthermore, the Central Geological Survey has identified most of these regions having
high soil liquefaction potential (https://www.liquid.net.tw/CGS/Web/Map.aspx), which
might be triggered by strong ground shaking during an earthquake, resulting in a second-
ary hazard. Thus, a reliable Vs30 map is crucial for hazard assessment. The site condition
utilized in this study was based on the on-site surveys on the TSMIP station sites (Kuo
et al., 2012, and references therein), that is, the sites of interest that are far from a TSMIP
site obtain a less precise Vs30 . In addition, the Vs30 map we obtained (Figure 7c) is simply
based on interpolation, that is, geological and topographic information was not incorpo-
rated. Anyone wanting to obtain a more reliable Vs30 map could follow the model proposed
by Kwok et al. (2018) that is based on geological classification and topographic gradient.
Alternatively, it is desired to implement an Vs30 obtained by an on-site survey, such as a
microtremor survey and/or standard penetration test.
Conclusion
Since the implemented innovations are based on an up-to-date database and our work
could provide a more reliable hazard assessment for Taiwan that could be widely applied
for various aspects. Our hazard assessment includes site amplification (Figure 9b and d),
which could be widely applied by non-experts without science or engineering backgrounds,
such some as in government agencies (e.g. the National Science and Technology Center
for Disaster Reduction). In addition, we are aware that some applications do not require
this factor (e.g. building code legislation, engineering purposes). Thus, we also provide the
hazard maps without site effects (Figure 9a and c); end users could incorporate site ampli-
fications on their own.
While we have just proposed a new version of PSHA for Taiwan, we already expect to
make some updates in the next generation of the TEM PSHA. For example, the current
PSHA followed the TEM PSHA2015 for the subduction source model. An update could
include outcomes of the Reevaluation of Probabilistic Seismic Hazard of Nuclear Facilities
in Taiwan Using SSHAC Level 3 Methodology. Although we have updated the seismo-
genic structure database based on state-of-the-art geomorphology and geology studies
(Shyu et al., 2020), we will incorporate seismogenic parameters obtained from geodetic
research. In respect to the GMPEs, some new sets of GMPEs, for example, Chao et al.
(2020) and Phung et al. (2020), will be discussed, and their epistemic uncertainties (Phung
et al., 2019) will be evaluated and considered in a logic tree.
Chan et al. 157
Figure 9. The hazard maps of the TEM PSHA2020, considering (a, c) engineering bedrock and (b, d)
site amplification for (a, b) 2% and (c, d) 10% probability in 50 years in PGA (in g).
158 Earthquake Spectra 36(S1)
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/
or publication of this article: This study was supported by the TEM, and funded by the Ministry of
Science and Technology—under the grants MOST 107-2119-M-006-011 and MOST 108-2119-M-
006-004—and by the National Research Foundation Singapore and the Singapore Ministry of
Education under the Research Centers of Excellence initiative. The seismic hazard calculations were
conducted by the open source software OpenQuake, developed by the Global Earthquake Model.
ORCID iD
Yin-Tung Yen https://orcid.org/0000-0003-1338-3592
References
Chan CH, Ma KF, Lee YT and Wang YJ (2019) Rethinking seismic source model of probabilistic
hazard assessment in Taiwan after the 2018 Hualien, Taiwan, earthquake sequence. Seismological
Research Letters 90(1): 88–96.
Chan CH, Wang Y, Wang YJ and Lee YT (2017) Seismic-hazard assessment over time: Modeling
earthquakes in Taiwan. Bulletin of the Seismological Society of America 107(5): 2342–2352.
Chao SH, Chiou B, Hsu CC and Lin PS (2020) A horizontal ground-motion model for crustal and
subduction earthquakes in Taiwan. Earthquake Spectra 36(2): 463–506.
Chen C-H, Wang J-P, Wu Y-M, Chan C-H and Chang C-H (2013) A study of earthquake inter-
occurrence times distribution models in Taiwan. Natural Hazards 69(3): 1335–1350.
Ellsworth WL, Matthews MV, Nadeau RM, Nishenko SP, Reasenberg PA and Simpson RW (1999)
A Physically Based Earthquake Recurrence Model for Estimation of Long-Term Earthquake
Probabilities (U.S. Geological Survey Open-File Report: 99–522). Reston, VA: U.S. Geological
Survey.
Fujiwara H (2014) Seismic hazard maps for Japan. In: Meyers RA (ed.) Encyclopedia of Complexity
and Systems Science. Berlin: Springer, pp. 1–28.
Gardner JK and Knopoff L (1974) Is the sequence of earthquakes in Southern California, with
aftershocks removed, Poissonian? Bulletin of the Seismological Society of America 64(5):
1363–1367.
Gerstenberger MC, Marzocchi W, Allen T, Pagani M, Adams J, Danciu L, Field EH, Fujiwara H,
Luco N, Ma K-F, Meletti C and Petersen MD (2020) Probabilistic seismic hazard analysis at
regional and national scale: State of the art and future challenges. Reviews of Geophysics, AGU
Centennial Special Issue 58: e2019RG000653.
Gutenberg B and Richter CF (1944) Frequency of earthquakes in California. Bulletin of the
Seismological Society of America 34(4): 185–188.
Kuo C-H, Wen K-L, Hsieh H-H, Lin C-M, Chang T-M and Kuo K-W (2012) Site classification and
VS30 estimation of free-field TSMIP stations using the logging data of EGDT. Engineering
Geology 129–130: 68–75.
Kwok OLA, Stewart JP, Kwak DY and Sun PL (2018) Taiwan-specific model for Vs30 prediction
considering between-proxy correlations. Earthquake Spectra 34(4): 1973–1993.
Lin PS (2009) Ground-motion attenuation relationship and path-effect study using Taiwan data set.
PhD Dissertation, Institute of Geophysics, National Central University, Taoyuan City, Taiwan
(in Chinese).
Lin PS, Lee CT, Cheng CT and Sung CH (2011) Response spectral attenuation relations for shallow
crustal earthquakes in Taiwan. Engineering Geology 121(3–4): 150–164.
Chan et al. 159
Molchan GM (1990) Strategies in strong earthquake prediction. Physics of the Earth and Planetary
Interiors 61: 84–98.
Phung VB, Loh CH, Chao SH and Abrahamson NA (2019) Analysis of epistemic uncertainty
associated with GMPEs and their weight within the logic tree for PSHA: Application to Taiwan.
Terrestrial, Atmospheric & Oceanic Sciences 29: 611–633.
Phung VB, Loh CH, Chao SH and Abrahamson NA (2020) Ground motion prediction equation for
Taiwan subduction zone earthquakes. Earthquake Spectra 36(S1): 1331–1358.
Shyu JBH, Chuang YR, Chen YL, Lee YR and Cheng CT (2016) A new on-land seismogenic
structure source database from the Taiwan Earthquake Model (TEM) project for seismic hazard
analysis of Taiwan. Terrestrial, Atmospheric & Oceanic Sciences 27(3): 311–323.
Shyu JBH, Yin YH, Chen CH, Chuang YR and Liu SC (2020) Updates to the on-land seismogenic
structure source database by the Taiwan Earthquake Model (TEM) project for seismic hazard
analysis of Taiwan. Terrestrial, Atmospheric & Oceanic Sciences. DOI: 10.3319/
TAO.2020.06.08.01. Available at: http://tao.cgu.org.tw/index.php/articles/archive/geophysics/
item/1723-2020060801t
Wang YJ, Chan CH, Lee YT, Ma KF, Shyu JBH and Rau RJ (2016a) Probabilistic seismic hazard
assessments for Taiwan. Terrestrial, Atmospheric and Oceanic Sciences 27(3): 325–340.
Wang YJ, Lee YT, Chan CH and Ma KF (2016b) An investigation of the reliability of the Taiwan
Earthquake Model PSHA2015. Seismological Research Letters 87: 1287–1298.
Woo G (1996) Kernel estimation methods for seismic hazard area source modeling. Bulletin of the
Seismological Society of America 86: 353–362.
Wu YM, Chang CH, Zhao L, Teng TL and Nakamura M (2008). A comprehensive relocation of
earthquakes in Taiwan from 1991 to 2005. Bulletin of the Seismological Society of America 98(3):
1471–1481.