Real-time Earthquake Detection and Hazard Assessment by ElarmS Across California

Holly Brown, Richard Allen, Margaret Hellweg, Oleg Khainovski, Peter Lombard, Douglas Neuhauser, and Adeline Souf


ElarmS is a network-based methodology for rapid earthquake detection, location, and hazard assessment in the form of magnitude estimation and peak ground motion prediction. The methodology is currently being tested as part of the real-time seismic system in California, leveraging the resources of the California Integrated Seismic Network (CISN) and the Advanced National Seismic System. ElarmS processing modules at three network processing centers reduce waveforms to a few parameters. These are then collected and processed at UC Berkeley to provide a single statewide prediction of future ground shaking that is updated every second.

The development of a statewide realtime warning system has presented several methodological and programming challenges. Here we focus on the system delays as data is passed between processing centers.


ElarmS uses the P-wave recorded on velocity and acceleration sensors to detect, locate, and estimate the magnitude of an earthquake. The initial estimate of location is beneath the first station to trigger, then between the first two stations based on the arrival times. Once three triggers are available the event is located using a grid search to minimize arrival time residuals. The depth for all events in California is set at 8 km.

The ElarmS magnitude estimate is based on the amplitude and frequency content of the P-wave arrival. For a given trigger, the maximum observed displacement, $P_{d}$, and predominant period, $t_{pmax}$, are converted to a magnitude estimate using empirical scaling relations [Allen and Kanamori, 2003; Tsang et al., 2007; Wurman et al., 2007]. Magnitude estimates for all triggered stations are averaged at each time step to provide a single estimate for the event.

The methodology naturally divides into a waveform processing module (WP) and an event monitoring module (EVM). WP operates on each data channel individually to reduce the seismic waveform to parameters including trigger times, $P_{d}$, $P_{v}$, $t_{pmax}$, peak ground velocity (PGV), peak ground acceleration (PGA) and signal-to-noise levels. The WP module can therefore be distributed. It currently runs at UC Berkeley processing waveforms from the Berkeley Digital Seismic Network (network code BK) at the USGS Menlo Park, processing the Northern California Seismic Network (NC) and some USGS Strong Motion Network (NP) data, and at Caltech/USGS Pasadena processing Southern California Seismic Network (CI), the Anza Network (AZ), and additional NP data. WP output parameters are telemetered to UC Berkeley where a single implementation of the EVM module integrates data from across the state to detect and analyze earthquake occurrence in real-time.

The EVM module currently processes waveform data from 383 stations (222 velocity instruments and 381 accelerometers) of the BK, NC, NP, CI, and AZ networks (Figure 2.33). The real-time implementation of ElarmS is now processing all stations in California that can be used by the system.

Figure 2.33: Map of stations currently used by realtime system.
\epsfig{file=brown09_2_1.eps, width=8cm}\end{center}\end{figure}


Every second required to telemeter data is a second reduction in warning time. Minimizing the time needed for data telemetry and processing is therefore an important aspect of a network-based early warning system. Figure 2.33 shows the current data latencies for ElarmS in California. The first cause of latency is the delay in getting data to the first shared memory region at its network processing site at UC Berkeley, Caltech/USGS Pasadena, or USGS Menlo Park (Figure 2.33a). Most of this delay is due to packetization of data. Data loggers at each station wait until a data packet is full before sending it across the telemetry system. Modifications to the configuration of many data loggers could reduce the size of these packets, reducing the overall latency. The actual telemetry (communication) delay at most sites most of the time is a fraction of a second, although packets can be delayed, resulting in the long tail to the distribution. The median waveform data latencies are 2.0, 6.5, 6.6, 6.6, and 11.5 second for NC, CI, BK, NP, and AZ, respectively. AZ has long latencies because AZ data is forwarded from Scripps to Caltech before it is processed by ElarmS. The median latency over all station channels is 6.5 sec.

Once the waveform data has arrived at the first shared memory region at a network center, WP processes the data to determine parameters. Parameters from Caltech/USGS Pasadena and USGS Menlo Park are then forwarded to UC Berkeley where they are incorporated into EVM as they arrive. Figure 2.33b shows the total latency in incorporating P-wave trigger times into the EVM output representing the current latency of the entire system. The median delay is 11.8 seconds and the distribution has a positive-skew, meaning that most frequent latency is in the 9-10 second window but there is a long tail at higher latencies. The actual processing of data by WP or EVM takes a fraction of a second. Most of the additional latency is therefore due to the process of moving data between the shared memory regions at the various stages of processing (including between network centers).

Figure 2.34: Histogram of latencies in the statewide system. (a)Waveform telemetry latency: delay between the absolute time of a P-wave trigger and when it arrives at the network processing center. Blue is BK, purple is NC, green is NC, orange is CI, red is AZ. (b)Total ElarmS latency: delay between the absolute time of a P-wave trigger and the timestamp of the EVM output file that incorporates that trigger.
\epsfig{, width=8cm}\end{center}\end{figure}


The state-wide, real-time test of ElarmS has only just begun (November 2008) so there will be additional methodology development necessary to handle the myriad technical issues that arise as a result of processing the approximately 15 billion observations per day that arrive with varying data latencies and various seismological and electronic sources of noise. In addition, further development is necessary to minimize the warning latency by improving the codes (currently 5 second delay) and upgrading station hardware (currently 6.5 second delay).


The real-time implementation and testing of ElarmS in California is part of a project by the California Integrated Seismic Network ( to test several early warning methodologies in the state. We have worked in collaboration with M. Böse, E. Hauksson, T. Heaton and K. Solanki at Caltech; T. Jordan and P. Maechling at USC and SCEC; D. Given and D. Oppenheimer at the USGS; M. Zeleznik of Saya Systems; and G. Cua at the Swiss Seismological Service. The project is funded by the USGS through cooperative agreement 06HQAG0147.


Allen, R. M., and H. Kanamori, The potential for earthquake early warning in southern California, Science 300, 786-789, 2003.

Allen, R. M., H. Brown, M. Hellweg, O. Khainovski, P. Lombard, and D. Neuhauser, Real-time earthquake detection and hazard assessment by ElarmS across California, Geophys. Res. Lett. 36, L00B08, 2009.

Böse, M., E. Hauksson, K. Solanki, H. Kanamori, and T. H. Heaton, Real-time testing of the on-site warning algorithm in southern California and its performance during the July 29 2008 $M_{w}$5.4 Chino Hills earthquake, Geophys. Res. Lett. 36, L00B03, 2009.

Tsang, L., R.M. Allen, and G. Wurman, Magnitude scaling relations from P-waves in southern California, Geophys. Res. Lett. 34, L19304, 2007.

Wurman, G., R.M. Allen and P. Lombard, Toward Earthquake Early Warning in Northern California, J. Geophys. Res. 112, B08311, 2007.

Berkeley Seismological Laboratory
215 McCone Hall, UC Berkeley, Berkeley, CA 94720-4760
Questions or comments? Send e-mail:
© 2007, The Regents of the University of California