Strong-Motion Data from the Two Pingtung, Taiwan, Earthquakes of 26 December 2006

1016 strong-motion records at 527 free-field stations and 131 records at 42 strong-motion arrays at buildings and bridges were obtained for the Pingtung earthquake doublet from the Taiwan Central Weather Bureau’s dense, digital strong-motion network. We carried out standard processing of these strong-motion records at free-field stations. A data set, including the originally recorded files, processed data files, and supporting software and information, is archived online http://tecdc .earth.sinica.edu.tw/data/EQ2006Pingtung/. We have not yet completed the processing of the strong-motion array data at buildings and bridges. However, some preliminary results and the strong-motion array data recorded at the second nearest instrumented building to the Pingtung earthquake doublet are shown. This paper is intended to document our data processing procedures and the online archived data files, so that researchers can efficiently use the data. We also include two preliminary analyses: (1) a comparison of ground motions recorded by multiple accelerographs at a common site, the TAP117 station in Taipei, and (2) attenuation of the horizontal ground motions (peak acceleration and response spectra at periods of 0.2, 1.0, and 3.0 s) with respect to distance. Our comparison study of multiple recordings at TAP117 indicates that waveform coherence among 20and 24-bit accelerograph records is much higher as compared to records from 16-bit or 12-bit accelerographs, suggesting that the former are of better quality. For the 20and 24-bit accelerographs, waveform coherence is nearly 1 over the frequency range 1 to 8 Hz for all components, and is greater than about 0.9 from 8 to 20 Hz for the horizontal component, but only from 8 to 12 Hz for the vertical component. Plots of pseudo-acceleration response spectra (PSA) as a function of distance, however, show no clear indication for a difference related to the performance level of the accelerographs. The ground-motions of the first event (MW = 7.0) are comparable, or even somewhat lower, than those from the smaller second event (MW = 6.9), consistent with the relative difference of the local magnitudes (ML = 6.96 and 6.99 for the first and second events, respectively). The ground motions from the first event are generally lower than those predicted from equations based on other in-slab subduction earthquakes, whereas the ground motions from the second event are closer to the predictions. Ground-motions for soil sites are generally larger than those from rock sites.


INTRODUCTION
One of the major tasks at the Central Weather Bureau (CWB) is to monitor earthquakes in the Taiwan region and to provide near realtime information about earthquake activities to the public.An extensive strong-motion instrumen-tation program was carried out by the CWB from 1992 to 1996 (Shin 1993;Liu et al. 1999).By the end of 2000, a total of 640 digital accelerographs and 56 accelerometer arrays had been deployed in free-field sites and in buildings and bridges, respectively (Shin et al. 2003).Since then, models of the new generation of 24-bit accelerographs have been purchased annually (from 50 to 100 units), mostly Terr. Atmos. Ocean. Sci., Vol. 19, No. 6, 595-639, December 2008doi: 10.3319/TAO.2008.19.6.595(PT) for replacing the older 16-bit accelerographs.Thus, several different types of digital accelerographs are used, and their general information is given in Table 1.Most of the digital For the first of the two Pingtung earthquakes, at 12:26 on 26 December 2006, a total of 484 strong-motion records at 457 free-field stations were obtained, and for the second event at 12:34, a total of 532 records at 502 free-field stations were obtained.The number of records exceeded the number of stations because some stations have more than one accelerograph, as some first generation 12-bit digital accelerographs are still in operation.Figures 1 and 2 show the locations of the triggered stations from the two Pingtung earthquakes, respectively.The origin time of two events was separated by about 8 minutes; and their magnitudes are about 7 [M W = 7.0, and M W = 6.9, respectively (http:// www.globalcmt.org/CMTsearch.html)].The first earthquake was clearly a normal-fault earthquake, whereas the second has a significant amount of strike-slip faulting (e.g., Wu et al. 2008).These two in-slab earthquakes were very well-recorded, and the recordings will greatly increase the worldwide sample of ground-motions from in-slab earthquakes, thus leading to an improvement of empirically based groundmotion prediction equations for in-slab, subduction-zone earthquakes.
Figure 3 shows the intensity maps of the two Pingtung earthquakes.Generally higher intensities were observed for the second event, which could be due to the somewhat closer distance of the event to Taiwan; we show later, however, that the ground motions seem to be somewhat higher for the second event than for the first event, at similar distances -accelerographs have a full scale of ± 2 g, 200 samples per second, 20 second pre-event data before the triggered time, and GPS timing.thus the second event may have had a somewhat larger stress drop.We will first describe our strong-motion processing procedures that are essentially the same as those for the 1999 Chi-Chi earthquake (Lee et al. 2001a, b).A data set, including the originally recorded files, processed data files in both binary and ASCII formats, and supporting software and information, is archived online for universal open access.
We also include in this paper two preliminary analyses: (1) a comparison of ground motions recorded by multiple accelerographs at the TAP117 site in Taipei, and (2) attenuation of the horizontal ground motion with respect to distance.We intend to publish our results from more detailed analyses in a future paper, which will also include some small amounts of recorded acceleration data that we did have time to process.

DATA PROCESSING AND ONLINE ARCHIVING
This large collection of data from hundreds of instruments requires extensive quality assurance.Data collection was also complicated by the use of 10 different models of accelerographs (see Table 1) with different data formats.

Assurance of Data Quality
The CWB strong-motion data were collected by four academic groups under the direction of Mr. Chun-Chi Liu and Dr. Kou-Cheng Chen of Academia Sinica, Prof. Gwo-Fig.2. Map showing the locations of the triggered stations from the second Pingtung earthquake.Free-field accelerograph stations are plotted as small black "diamonds", and strong-motion arrays in buildings or bridges are plotted as red "dots".The earthquake epicenter is shown as "star".Bin Ou of National Chung Cheng University, and Prof. Chien-Ying Wang of National Central University.The collected data were centralized at the CWB headquarters and collated for these two Pingtung earthquakes by the first author.
Several quality control tasks were performed on the recorded data.The principal tool we used was an interactive computer program called SMQC, written by Doug Dodge, for performing quality assurance tasks (see section C of the report by Teng and Lee 2000).Some of the accelerograms (mostly recorded by the A800, A900, and A900A accelerographs) have obvious spikes (mostly at the end of the record); these spikes were removed using the SMQC program, and the resulting processed data files have a slightly smaller total number of samples than the corresponding original records.

Header Information
Although information about the unit serial number, station name, coordinates, timing device, and other parameters are encoded in the header of each strong-motion data file, we found many errors.Because this header information must be entered manually by a technician, it is easy to make some mistakes.We corrected many errors using a master file of station information maintained at the strong-motion group at the CWB headquarters.We devoted a considerable amount of time building a reliable master file of station information, especially with respect to station coordinates and timing.In the data processing, we correct all header information using a computer program called "sudsfix" and our master station file (Lee and Dodge 2007).

DC-Offset Corrections, Filtering, and Peak Ground Acceleration and Peak Ground Velocity Values
The DC-offset in a given data waveform was corrected by removing the mean of the entire waveform.We then determined the peak ground acceleration (PGA) values in units of cm s -2 from the processed strong-motion records.We also filtered the accelerations using a low-cut 0.1 Hz acausal filter, and then we obtained peak ground velocities (PGV) and response spectral amplitudes from the filtered data, using the TSPP software of Boore (2008).The PGA and PGV values are summarized in Table 2 and Table 3 for

Data Archived Online
All the recorded strong-motion data are archived online at the Taiwan Earthquake Research Center http://tecdc.earth.sinica.edu.tw/data/EQ2006Pingtung/.For documentation purposes, we include (1) original data files as recorded by the accelerographs, (2) software from the manufacturers that will read (and often display) the original recorded data; software is also provided by the manufacturers to convert an original recorded data file to a binary file in PC-SUDS format (see Banfill 1994 for format specifications), (3) software and associated master files that were used to correct the header information and produce the processed data in PC-SUDS format, (4) software that can be used to convert the PC-SUDS formatted files into ASCII-formatted files, and (5) the converted ASCII data files.
For completeness, we will archive all the strong-motion data files recorded for the two Pingtung earthquakes, including those that have defects, such as spikes, bad data channel(s), late triggering, Users can select the data files that are appropriate for their needs.For example, if a user wishes to use the data for picking P-wave arrivals in an earthquake location study, then they should use only the data files that have absolute timing.
In this paper, we classify the recorded data files into three quality groups: good, fair, and bad.For the archived data set, the recorded data files are classified into four quality classes.Since the recorded strongmotion data are based on a triggering algorithm, a main concern is whether the record has pre-event data and whether the record is long enough to cover adequately the duration of the strong ground shaking at that station.In addition, we are concerned whether the record has defects (e.g., spikes, or components that were not recorded), and whether the record has absolute timing.In general, QA-class records are the best and can be used for any studies.The QB-class records are the next best because they do not have absolute timing.The QC-class records include the principal strong motions but have less than one second of pre-event data.The QD-class records have significant defects; they are included here for completeness and should not be used for most studies.

A COMPARISON OF GROUND MOTIONS RECORDED BY MULTIPLE ACCELEOGRAPHS
Because of the Taiwan government open-procurement requirements, any accelerograph manufacturer can submit their instruments to CWB for technical evaluation to see if they meet the CWB technical specifications (which are included in the procurement announcement).If qualified, they can bid their instruments, and the order is then awarded to the lowest bidder.Consequently, CWB has purchased many different types of accelerographs, because the lowest bidder was often different from year to year.It is, therefore, desirable to compare the performance of the various instruments at one site.In collaboration with the National Taiwan University (NTU), CWB established a test site (TAP117) in September 2006 using the concrete foundation of an aban-doned weather station at NTU.Several different types of accelerographs were installed, and records were obtained from the two PingTung earthquakes.This experiment is similar to that conducted at the Hualien station as reported by Lee et al. (2005).
For a preliminary analysis, we used the records from the second Pingtung earthquake following the procedure described in Lee et al. (2005).Table 4 shows the results from six co-located accelerographs at TAP 117.Note that the bit-designation in the A/D column is not precise.Kinemetrics (http://www.kmi.com)states that their Etna and K2 accelerographs (basically the same, but the K2 has more features) have 19-to 20-bit resolution from DC to 50 Hz, although these accelerographs use 24-bit A/D chips similar to those in the other 24-bit accelerographs.Also, data from Etna's and K2's are recorded as 24-bit integers.In the following discussion, we will consider Etna and K2 accelerographs to belong to the 24-bit class of accelerographs.In Table 4, the PGA values from the six different accelerographs are comparable, but these are "nominal" values because corrections for the calibration data of each accelerograph have not been applied.
In order to compare the waveforms, we use the record from the CV-575 accelerograph as the standard, and compute coherence between it and the other four accelerographs.It is reasonable to assume that the newer 24-bit accelerograph (such as CV-575) is superior to the older generations of 16-bit and 12-bit accelerographs.The waveform coherence results are shown in Fig. 8.The 12-bit A800 accelerograph (due to very few digital counts in the waveform data) does not compare well: the coherence with the CV-575 data is less than 0.8, except for frequencies between 2 to 6 Hz.The 16-bit A900 accelerograph appears to perform slightly better: the coherence is near 1 for frequencies up to 8 Hz.Waveforms from the Etna accelerograph and the SMART-24A accelerograph agree well with that of the CV-575 accelerograph: coherence is nearly 1 from 1 to 8 Hz for all components.The coherence starts to fall below 0.9 for frequencies above about 20 and 10 Hz on the horizontal and vertical components, respectively.We did not show the K2 results because they are essentially the same as the Etna re-Fig.6. Displacement time series for both events at station KAU082, one of the closest stations.Fig. 7. Fourier acceleration spectra for both events at station KAU082, one of the closest stations.To make it easier to compare the spectra, they have been smoothed over log-spaced frequencies, with a Konno and Ohmachi (1998) smoothing operator with a width of 0.4 of a decade.
sults.We note that this is a particularly severe test of the lower-bit systems because TAP117 is very far from the source (356 km) and thus the long-period signal will be smaller than if the recording had been made at closer distances.
These results are similar to the unpublished results obtained by Lee et al. (2005), where they reported only the coherence comparisons between the 24-bit class of accelerographs (Kinemetrics' Etna and K2, Reftek's 130-SMA/01, Tokyo-Sokushin's CV-575 and G3) for records from 16 earthquakes that occurred in the spring of 2004.During that same time period, data were also recorded by 3 other colocated accelerographs: one A800 (12-bit) and two A900 (16-bit), and coherence between these data and the 24-bit accelerographs's data were also computed and showed results similar to those reported here.
The present experiment at TAP117 site is intended as a long-term test of co-located accelerographs to assess, for example, whether or not there are any aging effects of the accelerographs.We will perform calibrations of these accelerographs periodically and report the results in the future, along with calibrated data from more earthquakes.

ATTENUATION OF THE HORIZONTAL GROUND MOTION WITH RESPECT TO DISTANCE
The 5%-damped pseudo-acceleration response spectra (PSA) were computed for the horizontal components of the two earthquakes.The periods at which the PSA were computed were 0.0 (for which PSA = PGA), 0.2, 1.0, and 3.0 s.As mentioned earlier, a 0.1 Hz low-cut filter was applied, but this has an inconsequential impact on the PSA values (the filter was necessary in order to obtain PGV).Figures 9 and 10 show the ground motions for the two earthquakes, using different symbols for sites with Vs30 less than and greater than 360 m s -1 (these correspond in a rough way to soil and rock sites).The Vs30 estimates were taken from Lee and Tsai (2008).The distance measure is the closest distance from the station to the rupture surface, where the rupture surface was estimated from the aftershock locations in Wu et al. (2008), with guidance from the empirical relations between magnitude and rupture length, rupture width, and rupture area given by Wells and Coppersmith (1994).Figure 11  shows a direct comparison of the ground motions for the two earthquakes for "soil" sites.The ground motions for the two events are very similar to one another, with the motions from the second event being somewhat higher than from the first event for some periods and distances.
Comparisons were also made of the data with groundmotion prediction equations (GMPEs) for subduction earthquakes.The comparisons are given in the next four figures .Only data from stations with a value of Vs30 provided by an Excel file accompanying the paper of Lee and Tsai (2008) are included in the plot.The types of system (16-and 24-bit) are indicated by the colors.The close and far recordings are on 16-and 24-bit systems, respectively, but for distances with both types of recordings the motions seem comparable.The GMPEs used include Atkinson and Boore (2003;abbreviated AB03), Kanno et al. (2006;abbreviated Kea06), Youngs et al. (1997;abbreviated Yea97), and Zhao et al. (2006;abbreviated Zea06).In general, the motions at close and far distances are in reasonable agreement with the predicted motions.In general, however, motions for the second earthquake are better predicted for all distances than those from the first earthquake, for which the motions are somewhat lower than the predictions.

STRONG-MOTION ARRAY DATA RECORDED AT BUILDINGS AND BRIDGES
Over 60 strong-motion arrays have been installed and operated by the Central Weather bureau since the 1990s, and a brief description of these arrays is given in Lee and Shin (1997).Sixty five records at 41 strong-motion arrays were obtained from the first Pingtung earthquake as summarized in Table 5, and 66 records at 42 strong-motion arrays were obtained from the second Pingtung earthquake as summarized in Table 6.The dataloggers for these strong-motion arrays are either (1) the 16-bit, 32-or 64-channel PC-based systems (designed by W. H. K. Lee), or (2) the more recent 24-bit, 32-channel Tokyo-Sokushin systems (Model SAMTAC 700).Since the PC-based systems have both high-gain and low-gain units, two records are often obtained for a given strong-motion array.
As an example, we will select the building array at the National Taitung Senior High Commercial and Vocational School (TTNBA0).Accelerometers are installed at different levels of the building as shown in Fig. 16.Waveforms recorded in this strong-motion array from the first Pingtung earthquake is shown in Fig. 17, and that for the second   Kanno et al. 2006;Yea97 = Youngs et al. 1997;Zea06 = Zhao et al. 2006).The curves are plotted for a "stiff soil" site class (NEHRP C).The gray bands show the plus and minus one standard deviations for the Zea06 GMPEs.Pingtung earthquake is shown in Fig. 18.Please note that the waveform amplitudes are not shown in true scale (they are normalized to fit within the plotting space).

Conclusions
Over 1000 strong-motion records were obtained by CWB for the two Pingtung earthquakes.Since large (M ~7) normal-faulting earthquakes are rare, these data are signifi-cant for studying the subduction and collision processes near Taiwan.Our main effort thus far has been to document and process this data set following a standard procedure so that it can be archived online to provide universal access for research purposes.We have done some preliminary data analysis and presented some tentative results.We plan to perform additional data analyses for a future paper intended to be published in about one year.The computed pseudo-acceleration response spectra (PSA) of these two earthquakes are similar, with the motions from the second event being somewhat larger on average.Motions from soil sites are clearly larger than for rock sites for most distances.Comparisons were also made of the data with ground-motion prediction equations (GMPEs) for subduction earthquakes.The motions from the second event are in better agreement with the predictions than for the first event, which are generally lower than the predicted motions.A preliminary coherence analy-sis of the recorded waveforms between several co-located accelerographs at the TAP117 site was also made.The results indicate that data from 24-bit accelerographs are similar, and are superior to that from 12-and 16-bit accelerographs, but in the period range 0 -3 s, there are no obvious systematic differences between PSA values computed from these two classes of instruments.

Fig. 1 .
Fig.1.Map showing the locations of the triggered stations from the first Pingtung earthquake.Free-field accelerograph stations are plotted as small black "diamonds", and strong-motion arrays in buildings or bridges are plotted as red "dots".The earthquake epicenter is shown as "star".
the first and second of the Pingtung earthquakes, respectively, along with other details.Sample acceleration, velocity, and displacement time series at one of the closest stations (KAU082) are shown in Figs. 4, 5, and 6, respectively.Smoothed Fourier spectra are shown in Fig. 7.Although the detailed ground motions differ for the two events, overall they are similar.

Fig. 4 .
Fig. 4. Acceleration time series for both events at station KAU082, one of the closest stations.

Fig. 8 .
Fig. 8. Waveform coherence at Station TAP117 between the record of the second Pingtung earthquake from the CV-575 accelerograph and the records from the A800, A900, Etna, and SMART-24A accelerographs.

Fig. 10 .
Fig. 10.Ground-motions vs. distance for the second earthquake (P2), separated by Vs30 values at each station.Shown are the geometric means of the two horizontal components.

Fig. 11 .
Fig. 11.Comparison of ground motions for the two events for "soil" sites.Shown are the geometric means of the two horizontal components.

Fig. 16 .
Fig. 16.Cross section view of the National Taitung Senior High Commercial and Vocational School where a strong-motion array is installed.The numbers in the left-hand side is in cm units, i.e., 4 m in height between different floors.The numbers (1 to 20) indicate where accelerometers are installed and correspond to the channel numbers in the Figs.17 and 18.

Fig. 17 .
Fig. 17.Waveforms recorded by the strong-motion array (TTNBA0) at the National Taitung Senior High Commercial and Vocational School from the first Pingtung earthquake.The channel number corresponds to the accelerometer number shown in Fig. 16.Please note that the waveform amplitudes are not shown in true scale (they are normalized to fit within the plotting space).

Fig. 18 .
Fig. 18.Waveforms recorded by the strong-motion array (TTNBA0) at the National Taitung Senior High Commercial and Vocational School from the second Pingtung earthquake.The channel number corresponds to the accelerometer number shown in Fig. 16.Please note that the waveform amplitudes are not shown in true scale (they are normalized to fit within the plotting space).