Earthquake Early Warning 5 seconds before ‘killer pulse’

By David Jacobson and Ross S. Stein, Temblor

Check your hazard rank

Rescue workers walk outside a building which partially-collapsed in last week’s M=6.4 earthquake north of the city of Hualien. (AP Photo/Chiang Ying-ying)


Vital Warning

Stand-alone “P-wave detectors” provided about five seconds of warning before the ‘killer pulse’ of violent shaking launched by last week’s M=6.4 Taiwan quake reached Hualien, where seven buildings fell. Had these detectors been placed in the buildings, this would have been sufficient time for occupants to ‘drop, cover, and hold on,’ potentially saving lives. Instead, these Palert units are currently located in schools throughout Taiwan, and the earthquake occurred around midnight, when the children were not at school.


This figure shows how long the warning times were before the peak ground velocity occurred in last week’s M=6.4 Eastern Taiwan earthquake. These times were recorded by Palert devices, which are standalone P-wave alarm detectors invented by Prof. Yih-Min Wu at National Taiwan University. They work by detecting the faster but less violent P-waves, to warn against the slower but more violent S-waves. For scientists reading, see the end of this article for the peak ground acceleration and peak ground velocity lead times.


Taiwan is one of only a few countries that has already implemented an earthquake early warning system (three others are Japan, Turkey, and Mexico). Two systems exist in the country, a regional telemetered system run by the Taiwan Central Weather Bureau, and on-site stand-alone devices we report on here. While the regional system provided an 11-sec warning in Taipei that was broadcast on National TV, Taipei was far from the epicenter and so only perceptibly shaken. Hualien, instead, fell within the ‘blind zone’ (where no warning is received). This is mainly because the regional system took 16 seconds to detect and locate the mainshock and release the warning. However, on-site stand-alone devices do not have this processing period and thus provided warning. The regional detection could be made faster, which would shrink the blind zone. The weakness of standalone detectors is that there is no ‘economy of scale;’ one is needed in every building..


How it works

These stand-alone units (Palerts, manufactured by the Taiwanese company Sanlien to Prof. Wu’s specifications, can, within 4 seconds detect the P-wave and so forecast the S-waves.

Palert instrument and strobe light
This photo shows a standard Palert device. The cylinder on the left is a strobe (warning) light. Such a unit can also be connected to a siren, alerting those nearby of impending shaking. The car key is for scale.


Around Hualien, warning times to the peak ground velocity ranged from 5-8 seconds. Encouraged by these results, John Vidale, Professor at USC and Director of the Southern California Earthquake Center, told Temblor that, “even the simple system operating in Taiwan provided several seconds of advance warning. This time allows many quick actions to protect lives, property, computer and industrial facilities. This scenario of a moderately large earthquake nearby is the most common threat. Bigger earthquakes would take even longer to build to their strongest shaking, and thus allow even a wider range of actions.”

The figure above shows the peak ground acceleration (top) and peak ground velocities experienced in Hualien in last week’s M=6.4 earthquake. Both are measures of seismic shaking, with acceleration more important for homes and velocity more important for tall buildings. The peak shaking tended to occur approximately 2-3 seconds after it commenced. The very strong peaks at 6-7 s, often termed a ‘killer pulse’ are extremely dangerous for taller buildings. (Data from: Praveen K. Malhotra, Strong Motions, Inc.)


Killer Pulse

The figure above shows the seismogram (in the East component) at Hualien, with peak shaking occurring 2-3 seconds after shaking commenced. Such shaking is also when most damage occurs. Thomas Heaton, Professor of Engineering Seismology and Director of the Earthquake Engineering Research Laboratory at Caltech, said, “These peak ground accelerations and velocities in Hualien appear consistent with a near-source directivity pulse—sometimes called a killer pulse—which are dangerous for taller buildings. This also occurred in the 1979 M=6.5 Imperial Valley, California, earthquake.” Fortunately, there were no tall buildings in the Imperial Valley.

It is likely that because the detectors are currently largely in schools, few if any received these warnings. Nonetheless, had the quake struck 12 hours earlier, the information could have saved lives. Prof. Richard Allen, Director of the Berkeley Seismology Laboratory and Chair of the U.C. Berkeley Department of Earth and Planetary Science, commented that “it is exciting to see that the schools in the Hualien region got a warning of a few seconds in the earthquake. Hopefully ShakeAlert will be providing warnings to schools in the U.S. later this year.”


ShakeAlert for the U.S.

ShakeAlert is the earthquake early warning system being developed by the USGS, U.C. Berkeley, Caltech, the University of Oregon, and the University of Washington for the western U.S. Rather than using stand-alone detectors, ShakeAlert will use networked seismometers and GPS receivers and cloud computing to send the warning to everyone with a cell phone. But ShakeAlert still requires long-term funding before it can be launched. Examining the Taiwan results, Dr. Lucy Jones, Founder of the Dr. Lucy Jones Center for Science and Society, said, “This is one more reminder that earthquake early warning systems work and can provide useful information for automatic systems or for people trained to use the warnings. This deepens my disappointment that the newly proposed President’s budget for 2019 once again tries to eliminate the development of early warning systems for the western United States.”



Prof. Yih-Min Wu for these preliminary results

Palert, Sanlien Co., Link

Praveen K. Malhotra, Strong Motions Knowledge and Clarity, Inc., Ground Motion Records from M 6.4 Taiwan Earthquake of February 6, 2018.

P-Alert research team, TEC, ASGC, and the TEC Data Center (Link)

Yih-Min Wu, Da-Yi Chen, Ting-Li Lin, Chih-Yih Hsieh, Tai-Lin Chin, Wen-Yen Chang, Wei-Sen Li, and Shaw-Hsung Ker (2013), A High-Density Seismic Network for Earthquake Early Warning in Taiwan Based on Low Cost Sensors, Seismological Research Letters 84, doi: 10.1785/0220130085

Yih‐Min Wu, Wen‐Tzong Liang, Himanshu Mittal, Wei‐An Chao, Cheng‐Horng Lin, Bor‐Shouh Huang, Che‐Min Lin (2016), Performance of a Low‐Cost Earthquake Early Warning System (P‐Alert) during the 2016 ML 6.4 Meinong (Taiwan) Earthquake, Seismological Research Letters, 87, 1050-1059, doi: 10.1785/0220160058

Central Weather Bureau


This figure shows lead times to peak ground acceleration (left) and peak ground velocity recorded by Palert detectors. (Figure from: Yih-Min Wu, Professor/Chair, Dept. Geosciences, National Taiwan University)


  • Eric Fielding

    Very interesting to see very high peak ground acceleration values near the marked epicenter of the M6.4 earthquake but low-to-moderate peak ground velocities there. The peak ground velocities are much higher in the area of Hualien, closer to where the large deformation is visible on InSAR south of Hualien. This could be consistent with a complex event with an early rupture near the epicenter but large moment release later further south near and south of Hualien.

  • Allen Hulse

    In the US I do not think that the system being developed will work to alert thru cell phones in a timely manor without the phone systems expending a great amount of money on the infrastructure of the phone systems.
    A stand alone system has more value to me. A raspberry pi seismograph might be another way to setup a stand alone alert system. The Raspberry shake is listed at $99 on the web page I just looked at with software at no cost. When I last looked at this it was just starting and they were talking about $300 and a license for the software. It sounds like this could be turned into a low cost alerting system that many people could use.

    When Cell phones are used for Amber alerts I can sit at a high school football game as the alerts come in on phones for over half an hour. and over 20 minutes for two phones on the same network and account. I have asked people who work on the cell phone network and they tell me that software changes as well as configuration changes will not be able to get the time under a minute without capital expense for infrastructure.

    I would like to see some real studies made by the cell phone companies to see just how hard it will be to get that type of response time. Till then I will look more at the cost of stand alone alert solutions.