EARTHQUAKE EARLY WARNING SYSTEM FOR RAILWAYS AND ITS PERFORMANCE

Journal of JSCE, Vol. 1, 322-328, 2013 Special Topic - 2011 Great East Japan Earthquake (Invited Paper) EARTHQUAKE EARLY WARNING SYSTEM FOR RAILWAYS AND ITS PERFORMANCE Shunroku YAMAMOTO 1 and Masahiko TOMORI 2 1 Laboratory Head, Earthquake Disaster Prevention, Disaster Prevention Technology Division, Railway Technical Research Institute (2-8-38 Hikari-cho, Kokubunji-shi, Tokyo 185-8540, Japan) E-mail: syama@rtri.or.jp 2 Member of JSCE, Principal Chief Researcher, Disaster Prevention Research Laboratory, Research and Development Center of JR East Group, East Japan Railway Company (2-479 Nisshin-cho, Kita-ku, Saitama-shi, Saitama 331-8513, Japan) E-mail: tomori@jreast.co.jp During the 2011 off the Pacific coast of Tohoku Earthquake, the earthquake early warning system operated by East Japan Railway Company controlled the Shinkansen trains through information from its seismic stations before large shakings hit the line. By analyzing the event history of the system recorded in its monitoring PCs, it was confirmed that the system first detected the seismic motion at 14:46:38.9 (JST) based on the data of the Kinkazan seismic station located closest to the epicenter, and issued the first control signal to trains between Shiroishi-Zaoh station and Kitakami station of the Tohoku Shinkansen line at 14:47:02.9 through the excess of acceleration threshold of the Kinkazan seismic station. The signal was issued 12-22 seconds earlier than the time SI value exceeding 18 cm/s along the line, which is the required value to stop trains based on company rules. After issuing the first control signal, other seismic stations began issuing signals through the excess of acceleration threshold almost sequentially according to certain delays caused by wave propagation from the hypocenter. Eventually, 27 trains (19 of them were running) along the line were safely controlled. In this paper, after summarizing the configuration of the earthquake early warning system of East Japan Railway Company and the warning logics adopted for the system, the system performance during the 2011 off the Pacific coast of Tohoku Earthquake is evaluated and discussed in detail. Key Words : earthquake early warning system, railway, Tohoku Shinkansen, 2011 off the Pacific coast of Tohoku Earthquake 1. INTRODUCTION During the off the Pacific coast of Tohoku Earthquake (M w =9.0) that occurred on the 11th of March 2011, the earthquake early warning system operated by East Japan Railway Company (hereafter, the JR EEW system) safely controlled 27 trains, some of which were running at a speed of about 270km/h, in the Tohoku Shinkansen line. No passengers and crew in the Shinkansen trains were reported injured nor killed by the earthquake 1). Though it is obvious that the absence of critical accidents on the Shinkansen trains during the earthquake was mainly due to various advance countermeasures to protect railway structures and trains against earthquakes, the JR EEW system may have played an important role in minimizing earthquake damage. Thus, evaluation of the JR EEW system for this very large earthquake is significant for the improvement of earthquake disaster prevention for railways. The JR EEW system has been originally developed to output control signals to Shinkansen trains as rapidly as possible before a large shaking hits trains and railway facilities by utilizing P-wave and/or S-wave of seismic motions. Urgent Earthquake Detection and Alarm System (UrEDAS), the first system using P-wave information, had been in operation for the Shinkansen line since 1992 2). UrEDAS had the function of estimating seismic parameters (epicenter and magnitude) using P-wave data from a single seismic record and judging damaged areas through estimated seismic parameters 3). UrEDAS is known as the first practical EEW (Earthquake Early Warning) system in the world using P-wave information. 322

With the increase in speed of Shinkansen trains and densification of Shinkansen diagrams, the JR EEW system has been upgraded to the present JR EEW system 4) to improve speed, accuracy and reliability. It began to be operated on the Shinkansen lines in 2004 and now all JR companies having Shinkansen lines operate the present JR EEW systems independently 5). In this paper, first we give a summary of the configuration of the present JR EEW system. Second, we briefly explain the warning logics adopted for the system. Third, we evaluate and discuss the system performance during the 2011 off the Pacific coast of Tohoku Earthquake according to the event history of the JR EEW system recorded in its monitoring PCs. 2. SUMMARY OF THE JR EEW SYSTEM Fig.2 shows the distribution of seismic stations operated by East Japan Railway Company. The system has 127 seismic stations as of June 2013. Note that open triangles show additional seismic stations constructed after the 2011 off the Pacific coast of Tohoku Earthquake. It can be seen that railside seismic stations are located along the Shinkansen lines with an interval of less than 15km, and coastline seismic stations are located along coastlines with an interval of roughly 0km. (2) Functions of the EEW seismometer The main functions of the EEW seismometer are P-wave warning output, S-wave warning output and maximum amplitude output. P-wave warning is issued when a Shinkansen line is expected to suffer seismic damage by analyzing (1) System configuration The present JR EEW system basically consists of seismic stations, central servers and monitoring PCs (Fig.1). Seismic stations are classified into two types: one is a railside seismic station located in a substation to detect mainly inland earthquakes; the other is a coastline seismic station located along a coastline to detect mainly subduction-zone earthquakes. Each seismic station has a conventional mechanical seismometer as well as an EEW seismometer for redundancy. Central servers and monitoring PCs, placed in a traffic control room of Shinkansen, are duplicated. Seismic stations and central servers are connected with broadband networks to communicate with each other. Monitoring PC Central server Tokyo Railside seismometer Coastline seismometer Fig.1 Configuration of the JR EEW system. Fig.2 Distribution of seismic stations operated by East Japan Railway Company as of June 2013. Solid squares and solid triangles show railside seismic stations and coastline seismic stations, respectively. Note that open triangles show additional stations constructed after the 2011 off the Pacific coast of Tohoku Earthquake. 323

P-wave data observed at a single seismic station. This warning also can be activated by using seismic information from other seismometers. In the latter case, seismic information is transmitted though network lines via central servers. By using P-wave, which travels faster than S-wave, the warning naturally produces a lead time depending on its hypocentral distance. S-wave warning is issued when the amplitude of bandpass-filtered acceleration exceeds a pre-defined threshold. This is a conventional but secure warning against earthquakes. It must be noted that the S-wave warning by coastline seismic stations still gives a certain lead time for a Shinkansen line in case of subduction-zone earthquakes. Both the signals of P-wave and S-wave warnings are directly transmitted to a substation from the EEW seismometers to allow shutting off the power supply to Shinkansen trains as rapidly as possible. Moreover, the value of the maximum amplitude recorded at each seismic station is output after an earthquake in order to provide shaking information to a traffic control center, where the necessity for a field inspection after an earthquake for Shinkansen is determined. The system is designed for redundancy by duplicating hardware and by using different warning logics from a number of seismic stations densely distributed in and around the Shinkansen lines. 3. WARNING LOGICS (1) P-wave warning P-wave warning is issued by the following steps: detection of P-wave, estimation of seismic parameters and decision to stop trains. To detect the P-wave of seismic motions, the EEW seismometer uses the standard STA/LTA trigger algorithm e.g. 6), which monitors the ratio of short-time average and long-time average of input signals. The trigger threshold ratio of each seismometer is determined independently taking into account the possibility of false alarm, which often disturbs the stable operation of railways. After detecting the P-wave, the EEW seismometer instantly searches the P-wave onset time by tracing back the waveform within a certain data length (1-2 s), then, conducts another process to distinguish a seismic signal from other signals, such as traffic vibrations, construction vibrations and electromagnetic noises by using the envelope of the signal. Next, the seismometer estimates the seismic parameters (epicentral distance, back-azimuth to epicenter and magnitude). To estimate the epicentral distance, the B-Δ method 7) is applied. This approach allows estimation from the empirical relationship between epicentral distance and an increasing ratio of acceleration of the initial P-wave (Fig.3(a)). The coefficient B is treated as an index to indicate the increasing ratio of the P-wave. This coefficient is obtained by fitting Eq. (1) to the envelope curve of amplitude for the observed P-wave s initial phase (Fig.3(b)). y(t) = B t exp(-a t), (1) where y(t) is the envelope curve of amplitude, A and B are coefficients, and t represents time. Since this method uses only the first 1-2 seconds of P-wave data at a single station, it allows the epicentral distance to be estimated immediately after P-wave detection. The seismometer also determines a back-azimuth Increasing rate 係数 B B(gal/s) (gal/sec) 00 0 1 0.1 Magnitude マグニチュード M7 M6 M5 M4 M3 0.01 0 epicentral 震央距離 distance (km) (Δ) (a) Epicentral distance (Δ) is then estimated by using the pre defined empirical relation between the coefficient B obtained in (a) and Δ. Logarithm of absolute amplitude Increasing rate B y(t)=btexp( At) P wave onset Time (b) The coefficient B is obtained by fitting Eq.(1) to the envelope curve of the observed P wave initial phase (1 2 sec). Fig.3 Explanation of estimating epicentral distance from a single station data in the JR EEW system by the B-Δ method. 324

to epicenter through the Principal Component Analysis for the displacement of the 1.0-1.1 second of the P-wave initial phase 8). Furthermore, magnitude is calculated using an empirical attenuation relationship of observed displacement, the coefficient B and magnitude shown below. M = 0.7387 log (D) - 1.02 log (B) + 7.07, (2) where M and D represent magnitude and maximum observed displacement. Magnitude estimated by the P-wave is calculated at a pre-defined interval after P-wave detection so as to follow rupture propagation on a seismic fault. Once epicentral distance, back-azimuth to epicenter and magnitude are determined, then, the potential damage area is estimated using the method, which adopts an empirical relation of magnitude, the epicentral distance and the potential damage area 9). Fig.4 shows the empirical relations that govern these factors. In the figure, damage data for railways (both Shinkansen lines and conventional lines) are compiled. Here, damage means structural or nonstructural damage that impedes safe operation of trains. It must be noted that damage to conventional lines is also included to define this damage area, because the damage data of Shinkansen lines are too few to define the empirical relation. During a large earthquake, the number of seismic stations may estimate seismic parameters with different values almost at the same time. In that case, the parameters that claim to stop trains are designed to take precedence for safety. (2) S-wave warning S-wave warning is a simple logic that monitors a threshold excess of bandpass-filtered acceleration as described in the previous chapter. The characteristics of the filter are shown in Fig.5. The pass band (0.5-5.0 Hz) was determined to standardize the characteristics of seismometers for railways considering structural responses of railway facilities in 1984 ). Threshold levels for a railside and a coastline seismometer are set to pre-defined values. For instance, the threshold level of a coastline seismometer is set at 120 gal. 4. OPERATION OF THE EEW SYSTEM DURING THE 2011 OFF THE PACIFIC COAST OF TOHOKU EARTHQUAKE (1) Event history of the system The event histories of the system during earthquakes are recorded as system logs in the monitoring PCs. These include times of P-wave detection, estimated results by P-wave, P-wave warning, S-wave warning, and so on. It is possible to confirm the operating history by analyzing these logs. As a reference for comparing the timing and seismic information estimated by the system, earthquake information on the 2011 off the Pacific coast of Tohoku Earthquake published by Japan Meteorological Agency (JMA) is shown in Table 1. Epicentral distance (km) 00 0 1 East off Chiba (1987) Off Kushiro (1993) Southwest off Hokkaido (1993) East off Hokkaido (1994) Far off Sanriku (1994) Southern Hyogo (1995) Western Tottori (2000) Geiyo (2001) Off Miyagi (2003) Northern Miyagi (2003) Off Tokachi (2003) Mid Niigata (2004) Off Mid Niigata (2007) 5.5 log Δ=0.51M-1.5 Damage area 5 6 7 8 Magnitude 400km Fig.4 Empirical relations of epicentral distance and magnitude to cover damage area for railway. The hatched area indicates damage area. Plots represent actual railway damage. Sensitivity 1 0.1 0.01 0.01 0.1 1 Frequency (Hz) Fig.5 Characteristics of the band-pass filter for acceleration data used for S-wave warning. 325

The origin time of this earthquake is 14:46:18.1 (JST) and the epicenter is located about 130km east-southeast of Kinkazan seismic station. The locations of the epicenter and nearby seismic stations of East Japan Railway Company are shown in Fig.6. Table 1 Seismic information published by Japan Meteorological Agency. Date March 11, 2013(JST) Origin time 14:46:18.1 (JST) Latitude 38.35 N degree Longitude 142.86 E degree Depth 24 km M j 8.4 M w 9.0 A B C D E F G H I J K L Kinkazan Seismic station Estimated damage area By P wave analysis M=6.9 (46:48) M=5.8 M=6.1 (46:41) (46:43) M=5.5 (46:40) Epicenter by JMA Epicenter estimated by the system Fig.6 Locations of the epicenter (star) and Kinkazan seismic station. The epicenter estimated by the system (cross) and damage area (circles) are also shown. Squares and triangles are seismic stations. Black squares are seismic stations that issued the first control signal to Shinkansen. Seismic stations with capital letters correspond to the stations in Fig.7. Circles show estimated damage areas for P-wave warning that were issued by Kinkazan seismic station at 14:46:40.0, 14:46:41.0, 14:46:43.0 and 14:46:48.0. Estimated magnitudes for the warnings are also described beside the circles. (2) P-wave warning The system first detected the seismic motion at 14:46:38.9, 20.8 seconds after the earthquake occurred, through the EEW seismometer at Kinkazan seismic station located closest to the epicenter. On P-wave detection, the seismometer at Kinkazan seismic station immediately traced back the seismic data and determined the P-wave onset time as 14:46:38.0. The first estimation of seismic parameters was then conducted at 14:46:40.0. Though the estimated epicenter was almost allowable considering the average errors of the methods using single station data 11), the estimated magnitude (M=5.5) was too small to meet the requirement for issuing a P-wave warning. Subsequently, additional magnitude estimations were done at 14:46:41.0, 14:46:43.0 and 14:46:48.0 at Kinkazan seismic station; however, these did not trigger any P-wave warning due to the still smaller magnitude estimation (M=5.8, 6.1 and 6.9, respectively). Locations of the estimated epicenter, magnitude and estimated damage area at each timing by Kinkazan seismic station during the earthquake are also shown in Fig.6. Other stations near the epicenter began detecting seismic signals after the P-wave detection of Kinkazan, but they failed to determine P-wave onset times within a certain data length. The reason for the failure will be mentioned in the next chapter. (3) S-wave warning Meanwhile, the seismometer at Kinkazan seismic station detected the threshold excess of filtered acceleration at 14:47:02.9 and provided the control signal to trains between Shiroishi-Zaoh station and Kitakami station of the Tohoku Shinkansen line. Subsequently, other seismic stations began issuing signals through the excess of acceleration threshold almost sequentially according to certain delays caused by wave propagation from the hypocenter. Finally, all the lines of Tohoku Shinkansen (from Tokyo station to Shin-aomori station) were controlled. The major events of the system related to Kinkazan seismic station are summarized in Table 2. 5. DISCUSSION (1) Lead time The East Japan Railway Company decides to stop the Shinkansen trains when the SI value, observed along a line, exceeds 18 cm/s for an inspection of earthquake damage. Thus, lead time up to 18 cm/s in excess of the warning time is considered to be a rough index for evaluating system performance. We used the acceleration data recorded at the railside seismic stations in the Tohoku Shinkansen 326

line and calculated the lead time in excess of the 18 cm/s of the first warning at 14:47:02.9. Fig.7 shows the lead times at the railside seismic stations related to the first control signal from Kinkazan seismic station. The figure shows that those stations had lead times of 12-22 seconds and this demonstrates that the system was able to issue a warning signal earlier than the strong shakings, although lead times show some variations mainly due to hypocentral distance and amplification characteristics at the stations. It is reported that deceleration by emergency brake of a Shinkansen train is roughly 2.6 km/h/s 12). This indicates that the lead time shown above corresponds to reducing speed of about 30-60 km/h. Though the reducing speed is limited, the speed reduction is considered very significant for the safety of Shinkansen trains, which run very fast. (2) P-wave warning It is thought that the system generally worked well during the 2011 off the Pacific coast of Tohoku Earthquake as shown in the previous section. However, the system is also deemed to function better if the following two points on P-wave warning will be improved. One point to be improved is the determination of P-wave onset time. During this earthquake, only one seismic station (Kinkazan seismic station) of the system was able to determine seismic parameters from the P-wave; other stations could not determine P-wave onset time. The major reason for this was that, compared to the final magnitude size of the earthquake, the initial P-wave amplitude of this earthquake was too weak to trace back to search the onset time within 2 seconds. Table 2 Major events of the system related to Kinkazan seismic station. time(jst) Events at Kinkazan seismic station 14:46:18.1 Origin time of the earthquake. (JMA) 14:46:38.9 P-wave detection. 14:46:40.0 1st estimation of seismic parameters (estimated magnitude is 5.5). No warning. 14:46:41.0 2nd estimation of seismic parameters (estimated magnitude is 5.8). No warning. 14:46:43.0 3rd estimation of seismic parameters (estimated magnitude is 6.1) No warning. 14:46:48.0 4th estimation of seismic parameters (estimated magnitude is 6.9). No warning Issue of S-wave warning to the line between 14:47:02.9 Shiroishi-Zaoh station and Kitakami station Fig.8 shows a comparison of amplitude growth of this earthquake with other smaller events recorded at almost the same hypocentral distances. This figure clearly shows the slow growth of the initial phase of this earthquake considering its final magnitude size. On the small amplitude of the initial phase, Chu et al. 13) indicated that a small earthquake (M w =4.9) occurred just a few second before the main shock in the hypocentral region. This likely caused a slow-growing appearance of the initial phase. Lead time (sec) acceleration (gal) 25 20 15 5 0 A B C D E F G H I J K L Seismic stations Fig.7 Lead time at railside seismic stations related to the first control signal from Kinkazan seismic station. Capital letters in the horizontal axis correspond to the stations shown in Fig.6. 2.0 1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 東北地方太平洋沖地震の地震波 2011 off the Pacific coast of Tohoku Earthquake 0 2 4 6 8 12 14 Time(sec) P wave onset Fig.8 Comparison of acceleration growth of vertical component for various earthquakes. The heavy line shows the 2011 off the Pacific coast of Tohoku Earthquake. The thin broken lines show 14 smaller events (6<M<8) that occurred around the Tohoku subduction zone. 327

The other point is magnitude estimation timing. The JR EEW system is designed to estimate magnitude until seconds after the P-wave onset time. No additional estimation by P-wave is done after seconds until S-wave comes. However, in the case of a very large earthquake with rather long hypocentral distance like the 2011 off the Pacific coast of Tohoku Earthquake additional magnitude estimation after seconds may give more effective performance for the system because it enhances the possibility of issuing early P-wave warning by following the growth of magnitude size. To make the JR EEW system more effective and reliable, improvements on the system, including those two points mentioned above, are now in progress. 6. CONCLUSIONS The evaluation of the earthquake early warning system of East Japan Railway Company for the 2011 off the Pacific coast of Tohoku Earthquake is carried out in detail by analyzing the event history recorded in the monitoring PCs and so on. As a result, the following conclusions are obtained: (1) Operation of the system The system first detected the seismic motion at 14:46:38.9 (JST) through the Kinkazan seismic station located closest to the epicenter. It then issued the first control signal to trains between Shiroishi-Zaoh station and Kitakami station of the Tohoku Shinkansen line at 14:47:02.9 through the excess of acceleration threshold of Kinkazan seismic station. The signal was issued 12-22 seconds earlier than the time SI value exceeding 18 cm/s along the line, which is the required value to stop trains based on company rules. (2) Improvements for a more effective system It is indicated that improving the warning logic for seismic motion with weak initial P-wave and additional magnitude estimation after seconds from P-wave onset time will make the system more effective and reliable. REFERENCES 1) Reconstruction Planning Dept., Facilities Dept. and Construction Dept. of East Japan Railway Company : Damaging to JR East from great east Japan earthquake and current situation, Japan Railway & Transport Review, No.60, pp.6-15, 2012. 2) Nakamura, Y. : On-site alarm the effective earthquake early warning, Proc. of 5th International Conf. on Urban Earthquake Engineering, pp. 45-53, 2008. 3) Nakamura, Y. : On the urgent earthquake detection and alarm system (UrEDAS), Proc. of 9th World Conf. on Earthquake Engineering, pp. VII673-VII678, 1998. 4) Iwahashi, H., Iwata, N., Sato, S. and Ashiya, K. : Practical use of earthquake quick alarm system, RTRI Report, Vol.18, No.9, pp.23-28, 2004 (in Japanese). 5) Yamamoto, S. and Sato, S. : History of earthquake early warning system for railway, RRR, Vol.67, No.3, pp. 16-19, 20 (in Japanese). 6) Allen, R. V. : Automatic earthquake recognition and timing from single traces, Bull. Seismo. Soc. Am., Vol.68, No.5, pp. 1521-1532, 1978. 7) Odaka, T., Ashiya, K., Tsukada, S., Sato, S., Ohtake, K. and Nozaka, D. : A new method of quickly estimating epicentral distance and magnitude from a single seismic records, Bull. Seismo. Soc. Am., Vol.93, No.1, pp. 526-532, 2003. 8) Yokota, T. : Study on earthquake prediction by geophysical method, Technical Reports of the Meteorological Research Institute, No.16, pp. 56-0, 1985 (in Japanese). 9) Nakamura, H., Iwata, N. and Ashiya, K. : Statistical relationships between earthquake disaster and seismic parameters used for train operation control after earthquake, RTRI Report, Vol. 19, No., pp. 11-16, 2005 (in Japanese). ) Bito, Y., Nakamura, Y. and Tomota, K. : On a method of reopening the train service after an earthquake for the Tokaido-Sanyo Shinkansen, Railway Technical Research Report, No.1294, pp. 1-38, 1985. 11) Yamamoto, S., Noda, S. and Korenaga, H. : An estimation method of epicentral distance based on characteristics of P-wave initial envelope, RTRI Report, Vol.26, No.9, pp. 5-, 2012 (in Japanese). 12) Arai, H., Kanno, S., Fujino, K., Kato, H. and Asano, K. : Development of a brale system for Shinkansen speed increase, JR East Technical Review, No.16, pp. 17-21, 20. 13) Chu, R., Wei, S., Helmberger, D. V., Zhan, Z., Zhu, L. and Kanamori, H. : Initial of the great Mw 9.0 Tohoku-Oki earthquake, Earth and Planetary Science Letters, No. 308, pp. 277-283, 2011. (Received July 1, 2013) 328