Continuous CWB GPS Array in Taiwan and Applications to Monitoring Seismic Activity

GPS observations have revealed important information for studying active tectonics and plate motion and are a useful tool for monitoring crustal deformation. The CWB continuous GPS array consists of approximately 150 stations with dense spatial coverage throughout Taiwan and can be used not only to monitor crustal deformation and seismic activity, but also to analyze the earthquake precursors in Taiwan.


INTRODuCTION
The Central Weather Bureau (CWB) initiated its permanent Global Positioning System (GPS) network in 1993 to investigate the relationships between seismic activity and crustal deformation in Taiwan.Utilizing satellite positioning techniques, each station can provide precise global coordinates for its antenna position which can be used to monitor the horizontal and vertical crustal movement at the site (Altamimi et al. 2002).This approach provides important information about crustal deformation caused by plate motion in the region.Using these data allows further studies to be carried out on related questions about seismic activity, stress accumulation or release, and mechanisms of earthquake formation (Dixon 1991;Segall and Davis 1997).

CONTINuOuS CWB GPS ARRAy IN TAIWAN
After the 1999 Chi-Chi earthquake, CWB enhanced data collection in its network to scrupulously monitor crustal deformation in Taiwan.The total number of the GPS stations was increased to 150 between 2001 and 2006.Pres-ently, the CWB GPS array consists of about 150 stations (Table 1 and Fig. 1), with dense spatial coverage in Taiwan, and operates continuously.
GPS observations have provided important information about tectonic deformation (Seno et al. 1993;Yu et al. 1997Yu et al. , 1999Yu et al. , 2001;;Hu et al. 2001;Rau et al. 2008).The continuous CWB GPS array is used to monitor site displacement, site velocity, and crustal deformation and can be used to study earthquake precursors in Taiwan.In order to analyze earthquake precursors and improve efforts to predict future crustal activity and strong earthquakes, just as in the case of weather forecasting, GPS observations must be mathematically connected with the physical processes of earthquake generation.Crustal deformation data are important for this purpose.Previous research has shown that dislocation theory works well for describing co-seismic, post-seismic and inter-seismic deformation (Wdowinski et al. 1997;Yu et al. 2001Yu et al. , 2003;;Hu et al. 2007;Cheng et al. 2009;Hsu et al. 2009b).The dense continuous GPS array provides daily GPS site coordinates, and the spatial coverage in Taiwan is now satisfactory.

ROuTINE GPS DATA PROCESSING AND ANAL-ySIS
All GPS stations in the CWB GPS network are equipped with dual-frequency geodetic receivers.The receiver instruments and antenna types include: (1) Leica RS500 and LEIAT504 SCIT (20 stations), (2) TRIMBLE 5700 and TRM41249.00(113 stations), (3) TRIMBLE NETRS and TRM41249.00(17 stations).The sampling interval for data collection is either 30 or 1 sec.Raw data from each station are converted to Receiver Independent Exchange Format (RINEX) and the continuous GPS data are processed independently using both Gamit/Globk and Bernese software to obtain the precise coordinates (Hugentobler et al. 2001;Herring et al. 2008).
GPS phase data acquired at continuous stations are transferred to the central control system at the CWB main office.All the station coordinates are used to obtain the final solution using the IGS (International GNSS Service) precise orbit after a few weeks.Data screening, editing and double differencing of phase observations between sites are performed in order to estimate their relative positions.Regional and global adjustment for daily solutions can improve the accuracy, and an example of the site position time series is plotted in Fig. 2. For some special cases, such as the occurrence of a strong earthquake, data are processed at intervals of 30 or 1 sec.and the baseline components are calculated using broadcast orbits in a quasi-real time manner to monitor crustal activity.This approach can quickly provide rough information about co-seismic displacement.The database system stores daily coordinate solutions, their covariance matrices, and raw phase data for further analyses.Time series plots of station coordinates and velocity vector plots can be used for routine monitoring of crustal deformation.
GPS time series data are analyzed using the following formula (Nikolaidis 2002 where a, b represent the intercept and slope of linear trend of inter-seismic crustal movement, c, d, e, f are coefficients for annual and semi-annual periodic deformation, g j is the co-seismic displacement of jth earthquake, h j is the change of trend after the jth earthquake, k j , τ j are the constant and relaxation time for post-seismic deformation of jth earthquake, H(τ) is a step function where H(τ < 0) = 0 and H(τ ≥ 0) = 1, and v i is the random error term.
In Fig. 2, the blue circle indicates the daily coordinate variation and the magenta line shows the regression curve for the above formula.Furthermore, the average inter-seis- mic site velocity can be estimated from the time series of each station as shown in Fig. 3 and Table 2 (for horizontal velocity).The horizontal strain field can be derived from horizontal site velocities and is also plotted in Fig. 3.The coseismic site slip for the 26 December 2006 strong earthquake in Taiwan is shown in Fig. 4.
GPS data in Taiwan have demonstrated related information and time series of coordinate variation for each site   in the network.Maps of the site velocity field and horizontal strain field in Taiwan derived from time series data are useful for future analysis.The general public can also learn about crustal deformation and plate motion in Taiwan.GPS-derived velocity/strain field reveal several features of tectonic deformation, which consistently indicate the northwestward movement in the direction of plate convergence (Yu and Chen 1994;Yu et al. 1997;Yu and Kuo 2001;Lin et al. 2006;Hsu et al. 2009a).

EARThquAkE PRECuRSOR ANALySIS
Taiwan is located at the active plate boundary between the Eurasia Plate and Philippine Sea Plate.The continuous collision results in active seismicity and large earthquakes (Tsai 1986;Wang and Shin 1998).Because earthquakes occur frequently in Taiwan, research on earthquake prediction is vital.Earthquake prediction presents serious challenges, whether the effort is focused on the time of occurrence, epicenter location, magnitude, or intensity.Studies of seismic precursor are actively underway in Taiwan.Research on earthquake precursors is mainly focused on monitoring and analysis of pre-seismic anomalies, collecting data on precursor phenomenon before strong earthquakes occur.GPS data can show steady deformation processes in eastern Taiwan and the seismic deformations associated with some significant events (Yu et al. 2001;Hu et al. 2007;Cheng et al. 2009;Hsu et al. 2009b).The pre-seismic anomalies from GPS time series including the 19 December 2009 earthquake (M 6.9) in eastern Taiwan are shown in Fig. 5.We hope to use the GPS time series analysis to identify changes in earthquake precursor signals from background signals.Although GPS data can provide insights on the ongoing deformation only, continuing these kinds of monitoring efforts will lead to much more precise deformation fields and more details on their temporal variations.The typical accuracy as root mean squares error for GPS site coordinates is a few mm in the horizontal and 10 -20 mm in the vertical.Even though daily site coordinate solutions are not adequate for monitoring the crustal deformation at active plate boundaries, monthly mean solutions might be adequate.Seasonal variation signals still remain in the monthly time series, which presents some obstacles for geodetic or geophysical interpretations.However, daily or hourly analysis is required for monitoring precursory changes for large earthquakes.In this case, we will need to process much noisier data to detect small changes.

Fig. 2 .
Fig. 2. Time series showing the coordinate component variation of ILAN GPS station from 2004 to 2009.The magenta curves are the best-fitting lines of each component using the Nikolaidis (2002) regression model.

Fig. 3 .
Fig. 3. GPS site deformation map: horizontal velocity field (2004 ~ 2009, left) and crustal strain rate (right).The magnitude and direction of principal strain rates are shown by arrows.Color scale shows magnitudes of dilatation rates.Red color denotes contraction and blue color represents extension.

Fig. 4 .
Fig. 4. The coseismic displacement of GPS stations associated with the 26 December 2006 earthquake in southern Taiwan.

Table 1 .
Site locations for the CWB continuous GPS array in Taiwan.

Stations Longitude (deg) Latitude (deg) Antenna type Receiver type Starting date
Fig. 1.Site location in the CWB GPS Network.Table 1. (Continued)

Table 2 .
Velocity field for CWB GPS stations with respect to Penghu station.