The Potential for Earthquake Early Warning in Southern California

See allHide authors and affiliations

Science  02 May 2003:
Vol. 300, Issue 5620, pp. 786-789
DOI: 10.1126/science.1080912


Earthquake mitigation efforts in the United States currently use long-term probabilistic hazard assessments and rapid post-earthquake notification to reduce the potential damage of earthquakes. Here we present the seismological design for and demonstrate the feasibility of a short-term hazard warning system. Using data from past earthquakes, we show that our Earthquake Alarm System (ElarmS) could, with current TriNet instrumentation, issue a warning a few to tens of seconds ahead of damaging ground motion. The system uses the frequency content of the P-wave arrival to determine earthquake magnitude, an approach that allows magnitude determination before any damaging ground motion occurs.

Current efforts to mitigate seismic hazard in the United States include long-term (50-year) hazard assessment and rapid post-event notification. Long-term hazard mitigation is facilitated by probabilistic ground-shaking maps (1), which estimate the probability of ground motion exceeding some threshold during the next 50 years. Such maps are used in the development of building codes intended to prevent the collapse of buildings during an earthquake. Rapid post-event notification in southern California is provided by TriNet (now part of the California Integrated Seismic Network), a network of about 155 station sites with both high dynamic-range broadband and strong-motion instrumentation (2, 3). Rapid notification of earthquake source parameters is issued through a pager system called CUBE (4) and over the Internet, within minutes of substantial earthquakes. Peak ground motion observations are also used to generate a map of ground motion distribution (“ShakeMap”) within 3 to 5 min of an earthquake (5).

Earthquake early warning systems (EWSs) provide a few seconds to tens of seconds of warning of oncoming ground motion, allowing for short-term mitigation. EWSs that estimate the severity of ground shaking and the time till that shaking will commence are in operation in Japan, Mexico, and Taiwan. The most basic system, offering no warning time, issues an alarm when ground shaking at the same location exceeds some threshold. When the earthquake source region is some distance from a populated area or city, seismometers can be deployed between the source and the city to detect any earthquake and transmit a warning electronically, ahead of the more slowly moving ground motion. Mexico City is protected by such a front-detection EWS: Seismometers along the coast detect earthquakes in the Guerrero Gap ∼300 km southwest of the city and an alarm is issued, providing ∼70 s of warning time (6, 7). The Central Weather Bureau of Taiwan also uses a front-detection EWS, which requires an average of 22 s to determine earthquake magnitude and location and thus provides warning for areas greater than ∼75 km from the epicenter (8, 9).

All these EWSs use observations of peak ground motion to estimate the magnitude of an earthquake, which is the most commonly applied and most accurate method of local magnitude determination. However, this approach does not provide the most rapid magnitude estimate. The first seismic arrival from an earthquake is the P wave, which is usually relatively low-amplitude and causes little damage (Fig. 1). It is followed by the S wave, which usually has a larger amplitude and includes the peak ground motion (Fig. 1), causing most of the damage to buildings in an earthquake. The warning time can be increased by use of the P-wave arrival to estimate the magnitude of an earthquake. This approach was introduced in Japan in the 1990s; the UrEDAS system uses the P-wave arrival to estimate both the magnitude and location of an earthquake (1012).

Fig. 1.

Typical seismic waveform for a local earthquake. This trace of horizontal ground motion was recorded 50 km from the epicenter of a magnitude 3.9 earthquake in southern California. Indicated are the P-wave and S-wave arrival times and the peak ground motion, which is required todetermine local magnitude.

The early warning problem in southern California is particularly challenging because the region is dissected by many active faults, including some beneath metropolitan areas (Fig. 2). A previously detailed EWS proposed instrumentation of known faults (13) but would have offered no protection against blind thrust faulting such as the Northridge earthquake (14). The utility of a front-detection EWS has also been argued (15) and applied for aftershocks. After the Loma Prieta earthquake, a front-detection system was used to warn workers of aftershocks as they cleared debris ∼100 km from the epicentral region (16). Front detection, however, cannot issue any warning in the epicentral region, impeding the effectiveness of such a system when active faults and population centers are in the same location.

Fig. 2.

Tectonic map of southern California showing topography (color scale), mapped fault distribution (thin black lines), locations of the TriNet stations with the necessary real-time capabilities toimplement ElarmS (gray squares), and epicenters of the 53 earthquakes used in this study (blue and red circles). The 28 earthquakes used to demonstrate the performance of ElarmS as a function of time are shown in red.

Our proposed Earthquake Alarm System (ElarmS) issues a warning based on information determined from the P-wave arrival only, providing the potential to issue a warning before peak ground motion at the epicenter. ElarmS initially determines the location, origin time, and magnitude of an event using the first detected P-wave arrivals. The spatial distribution of anticipated peak ground motion is then determined with attenuation relations. Finally, the warning time is estimated based on the origin time and travel time curves for S waves. Here, we focus on the determination of event magnitude using P-wave arrivals, as we use relatively standard procedures for the other components of ElarmS.

To develop and test the concept of ElarmS, waveform data were gathered for a large number of southern California earthquakes recorded by two or more broadband seismometers within 100 km of the epicenter. The event set consisted of 53 earthquakes. It included all earthquakes with magnitude ≥5.0 since 1995, plus the Northridge, Hector Mine, and Landers earthquakes with magnitudes 6.7, 7.1, and 7.3, respectively (table S1). In addition, two events were selected at random with magnitudes 3.0 to 4.9 (events were binned in 0.1-magnitude-unit bins).

We estimated the magnitude of an earthquake from the frequency content of the P-wave arrival, using an approach similar to that used in Japan (12). Small-magnitude events are the result of slip over a small patch of fault and result in the radiation of relatively high-frequency energy compared to larger magnitude events, which rupture large patches and radiate lower frequency energy. Therefore, a measurement of the predominant period of a seismic waveform in the few seconds after the P-wave arrival onset can constrain the magnitude. The predominant period is determined continually in real time from the vertical component of the velocity sensor at each station and is defined with the recursive relation Math1 where Math2 Math3 Math is the predominant period at time i, xi is the recorded ground velocity, Xi is the smoothed ground velocity squared, Di is the smoothed velocity derivative squared, and α is a smoothing constant (17). The higher frequency content of smaller magnitude earthquakes is measurable within a shorter time period after the P-wave arrival than the low-frequency energy of larger events. Correspondingly, the magnitude of smaller events can be determined more rapidly than that of larger events. This also means that the magnitude estimate after 1 s is a minimum estimate and that once 2, 3, and 4 s of data are available, the magnitude estimate may increase.

Two linear relations between Math and magnitude were used (Fig. 3). For smaller earthquakes (magnitudes 3.0 to 5.0), broadband data low-pass filtered at 10 Hz were used and a good magnitude estimate was possible given just 1 s of data. With 2 s of data the magnitude error reduced slightly, but additional data did not improve the estimate. Using our Math observations from the broadband waveforms after 2 s and minimizing the average absolute deviation, we determined the relation Math4 to estimate the magnitude of low-magnitude earthquakes (ml), with an average absolute deviation of 0.3 magnitude units. For larger magnitude events (magnitude >4.5), better estimates are possible with the application of a 3-Hz low-pass filter. The best estimates of magnitude require 4 s of data, although minimum-magnitude estimates can be made as soon as 1, 2, and 3 s after the P-wave arrival. The best-fit high-magnitude (mh) relation is Math5 with an average absolute deviation of 0.67 magnitude units.

Fig. 3.

Relation between predominant period and magnitude. The predominant period is measured on the vertical component of a broadband velocity sensor. (A) Maximum predominant period observed within2softhe P-wave arrival low-pass filtered at 10 Hz versus TriNet magnitude for individual stations (gray dots) and event averages (black dots) for all 53 events. The black line is the best fit (least absolute deviation of event averages) for data from magnitude 3.0 to 5.0 events. (B) The maximum predominant period within4softhe P-wave arrival on the same data stream low-pass filtered at 3 Hz. The black line is the best fit to event averages for earthquakes with magnitude between 5.0 and 7.3. Observations of the predominant period at single stations show noticeable scatter (gray dots); once data from several stations are averaged, the scatter reduces substantially (black dots).

ElarmS uses both ml and mh to produce the best estimate of magnitude. Initially, 1 s after a station triggers, ml is calculated from Math. The estimate is updated when 2 s of data are available. Station-magnitude estimates (one from each triggered station) are averaged to provide an event-magnitude estimate. If the event-magnitude estimate becomes greater than 4.0, then mh is also calculated and the event-magnitude estimate is the average of both ml and mh from each triggered station. The accuracy of the magnitude estimate increases with the number of stations reporting a Math observation (Fig. 3). With a single Math from the closest station, the average absolute error in the magnitude estimate is ±0.70, but when the closest 10 stations are used, the average error drops to ±0.35 (Fig. 4A and fig. S1).

Fig. 4.

(A) Error in ElarmS magnitude estimate as a function of the number of stations reporting. Each histogram shows the frequency distribution of the magnitude error given the number of reporting stations. The total number of event-magnitude estimates included and the average absolute error are shown for each histogram. (B) Error in ElarmS magnitude estimate as a function of time with respect to the S-wave arrival at the epicenter (–1 s is 1 s before the S-wave arrival). Magnitude estimates are available for 2, 12, and 19 events (of the 28-event data set) at –1, 0, and 1 s with average errors of 0.61, 0.52, and 0.56 magnitude units, respectively (figs. S1 and S2).

Another key element of an early warning system is its ability to process the necessary information in the shortest possible time, in order to issue an initial warning as soon as possible, and then to update that warning as more data become available. Our approach to magnitude determination provides the first magnitude estimate within 1 s of the first P-wave trigger. Depending on the S-minus-P time, which will vary according to the depth of the event and the proximity of the closest station, this estimate may be determined before the S-wave arrival at the epicenter. Thirty km from the epicenter, the magnitude would be available ∼8 s before the S-wave arrival, and at 60 km, the magnitude would be available ∼16 s before the S-wave arrival.

To demonstrate the time history of the ElarmS magnitude estimate and to show how the system could operate today given the current TriNet station density in the more populated regions of southern California, we used a subset of 28 of our 53 earthquakes (Fig. 2 and table S1). We assessed the timeliness with which the magnitude estimate becomes available by determining the error in the ElarmS magnitude estimate as a function of time (Fig. 4B). The zero time was the S-wave arrival at the epicenter, because it represents the earliest time of damaging ground motion. Of the 28 events, a magnitude estimate was available for 12 at the time of the S-wave arrival at the epicenter. The magnitude estimates were variable during the time window –1 to +2 s, but then became stable (Fig. 4B and fig. S2). The average absolute error in the magnitude estimate steadily dropped from 0.41 to 0.26 magnitude units in the period from 2 to 5 s after the S-wave arrival at the epicenter (Fig. 4B).

This subset of events did not include the three largest magnitude events (Landers, Northridge, and Hector Mine), because none of these were sampled by the dense station network now available in populated regions of southern California. Although these events require an additional 2 s for the best magnitude estimates, the peak ground motion for such large-magnitude events can occur tens of seconds after the S-wave arrival. In the case of Northridge, peak ground motion at the closest station (32 km from the epicenter) occurred 5 s after the S-wave arrival; the two broadband waveforms for Landers (at 62 and 65 km) showed a 10- to 15-s delay between the S-wave arrival and peak ground motion.

The tests of ElarmS presented here use data from past earthquakes. In a real-time implementation of ElarmS, the total processing time would be increased by data transmission times. Current TriNet stations are able to perform waveform processing on site. Because only parameter information is transmitted to central processing, transit time is reduced, allowing the processing of station information at the central site 1 s behind real time. The delay in transmitting the warning would be dependent on the technology used but could be reduced to less than 1 s. It is therefore conceivable to issue a ground-motion warning across southern California within 2 s of the times shown in Fig. 4B.

In conclusion, the implementation of ElarmS could provide a few to tens of seconds of warning to areas that may suffer structural damage in an earthquake. Buildings up to 60 km from the epicenter were red-tagged for demolition after the magnitude 6.7 Northridge earthquake. In a repeat event, occupants of buildings ∼60 km from the epicenter could receive ∼20 s of warning before peak ground motion. In a larger magnitude earthquake, the area damaged could be larger and even more warning time would be available to those further from the epicenter. For example, in the 1999 magnitude 7.6 Chi-Chi earthquake in Taiwan, many buildings were moderately damaged in the capital city of Taipei 145 km from the epicenter (18). With the ElarmS approach, there could be ∼40 s of warning at a distance of 145 km.

The potential uses of a few to tens of seconds of warning span both personal and institutional preservation. Personal protective measures that could be undertaken at home and in the workplace include getting under desks and moving away from dangerous chemicals and machinery. During the response to a major earthquake, ElarmS could provide warning to rescue and clean-up personnel as they work on unstable debris. Institutional uses of short-term warnings include automated mass-transportation systems that can use a few seconds to slow and stop trains, abort airplane landings, and prevent additional cars from entering the freeway. Industry can shut down, or initiate the shutdown process of, sensitive equipment before peak ground motion arrives, preventing cascading failures. In addition to these immediate uses, the development of an early warning system will lead to the development of infrastructure that can use the information. For example, engineering companies in Japan are developing buildings with active response systems: The buildings can change their mechanical properties within a few seconds to better withstand ground motion (19).

Supporting Online Material

Materials and Methods

Figs. S1 and S2

Table S1

References and Notes

View Abstract

Navigate This Article