Rapid access to reliable information is critical for any emergency response effort. In the case of a major earthquake, the mobilization of local, state, and federal disaster operations can be greatly enhanced by dependable, near real-time estimates of location, magnitude, mechanism, and extent of strong-ground shaking. This information can be used to identify endangered communities, to evaluate the impact on lifelines, and to provide input for damage and loss estimation programs. Current applications of rapid earthquake information include the emergency services, transportation, utilities, telecommunications, and insurance industries.
Over the last 7 years, the BSL has invested in the development of the hardware and software necessary for a rapid earthquake notification system. The Rapid Earthquake Data Integration (REDI) project is a research program at the BSL for the rapid determination of earthquake parameters with three major objectives: to provide near real-time locations and magnitudes of northern and central California earthquakes; to provide estimates of the rupture characteristics oand the distribution of ground shaking following significant earthquakes, and to develop better tools for the rapid assessment of damage and estimation of loss. A long-term goal of the project is the development of a system to warn of imminent ground shaking in the seconds after an earthquake has initiated but before strong motions begin at sites that may be damaged.
On April 18, 1996, the BSL and the USGS announced the formation of a joint notification system for northern and central California earthquakes (Gee et al., 1996b). The new system merges the programs in Menlo Park and Berkeley into a single earthquake notification system, combining data from the NCSN and the BDSN (Figure 7.1). On the USGS side, incoming analog data from the NCSN are digitized, picked, and associated as part of the Earthworm system (Johnson et al., 1995). Preliminary locations, based on phase picks from the NCSN, are available within seconds, based on the association of a few arrivals, while final locations and preliminary coda magnitudes are available within 2-4 minutes. Earthworm reports events - both the "quick-look" 25 station hypocenters (without magnitudes) and the more final solutions (with magnitudes) to the Earlybird alarm module in Menlo Park. This system sends the Hypoinverse archive file to the BSL for additional processing, generates pages to USGS and UC Berkeley personnel, and updates the northern California earthquake WWW server.
On the UC Berkeley side, the Hypoinverse archive file is used to drive the Rapid Earthquake Data Integration (REDI) processing system. This is a modification of the original REDI design, which identified and located events using raw phase data from the BDSN and NCSN (Gee et al., 1996a). In the revised REDI processing, the magnitude of each hypocenter is assessed. If the coda magnitude is greater than or equal to 3.0, waveforms from the BDSN are analyzed to estimate local magnitude. Once an updated magnitude is obtained, the event information is paged to USGS and UC Berkeley personnel, notification is sent to emergency response agencies, and the revised magnitude is transmitted to the Earlybird system in Menlo Park. The earthquake is then evaluated for the next level of REDI processing. If the local magnitude is greater than 4.5, waveforms from the BDSN strong-motion instruments are analyzed to determine peak ground acceleration, velocity, and displacement and to estimate the duration of strong shaking. When the strong-ground motion processing is complete, the event is scheduled for moment tensor estimation.
At present, two Earthworm-Earlybird systems in Menlo Park feed two REDI processing systems at UC Berkeley (Figure 7.2). One of these systems is the production or paging system; the other is set up as a hot backup. The second system is frequently used to test new software developments before migrating them to the production environment. In addition, the BSL operates a third system, which uses BDSN picks to form an independent list of associated events. This third system provides redundancy in case the communication links with the USGS Menlo Park are disrupted. A fourth system is under construction in Sacamento in order to provide a redundant notification facility outside of the Bay Area.
This new structure has greatly expedited automatic earthquake processing in northern California (Figure 7.3). The dense network and Earthworm-Earlybird processing environment of the NCSN provides rapid and accurate earthquake locations, low magnitude detection thresholds, and first-motion mechanisms for smaller quakes. The high dynamic range dataloggers, digital telemetry, and broadband and strong-motion sensors of the BDSN and REDI analysis software provide reliable magnitude determination, moment tensor estimation, peak ground motions, and source rupture characteristics. Robust preliminary hypocenters are available about 25 seconds after the origin time, while preliminary code magnitudes follow within 2-4 minutes. Estimates of local magnitude are generally available 30-120 seconds later, and other parameters, such as the peak ground acceleration and moment magnitude, follow within 1-4 minutes.
Earthquake information from the joint notification system is distributed by pager, e-mail, and the WWW. The first two mechanisms "push" the information to recipients, while the current Web interface requires interested parties to actively seek the information. Consequently, paging and, to a lesser extent, e-mail are the preferred methods for emergency response notification. The Northern California Web site has enjoyed enormous popularity since its introduction and provides a valuable resource for information whose bandwidth exceeds the limits of wireless systems and for access to information which is useful not only in the seconds immediately after an earthquake, but in the following hours and days as well.
As noted in previous reports, data from the BDSN are transmitted to UC Berkeley over continuous telemetry. Through funding provided by the CalREN Foundation, the older analog circuits have been upgraded over the last 24 months to 56 Kbits per sec (Kbps) frame relay connections and data from 24 stations are currently transmitted using this fast, packet-switching technology. The use of frame relay has increased the robustness and reliability of the BDSN telemetry and provides a significant increase in the available bandwidth. The switch to frame relay telemetry also offers new opportunities for expanding the capabilities of the REDI system.
Over the last year, we have updated our frame relay telecommunications hub to support for SLIP (Serial Line IP) in order to migrate existing stations to TCP/IP, and have installed new software on our remote frame relay access which will provide SLIP support to our dataloggers. We are testing SLIP software for our Quanterra dataloggers, which can support the multiple socket connections necessary for the development of an accelerated processing system. In addition, we are using the 56 Kbps bandwidth to reduce data transmission delays. The BDSN Quanterra dataloggers currently fill a 512 byte packet before sending it to the acquisition system. For the BH channels (20 sps), this corresponds to 20 sec of data. By decreasing the packet size, we can significantly reduce the delays associated with telemetry at the cost of increasing temporary data storage. We are also beginning to convert stations with triggered strong motion channels into continuous 80 or 100 sps data streams. This modification will allow us to run continuous processing algorithms for peak ground acceleration and other critical seismic parameters.
In early 1998, we completed the migration of the REDI processing software from the SunOS to the Solaris operating system. The Solaris operating system provides more features for real-time processing and this migration has led to the development of new features and capabilities in the REDI system. For example, new developments permit the exchange of waveform data using the Earthworm transport software. As a preliminary implementation of this capability, the BSL and the USGS are currently exchanging waveform data from Hayward Fault Network stations. This initial test will soon be expanded to include data from other Bay Area strong motion instruments. We also plan to begin exchanging waveform data with the University of Nevada at Reno in order to improve our coverage of events in eastern California. In addition, the new developments allow the transparent integration of waveform data from the NSN VSAT system. Data from 3 BDSN stations (CMB, SAO, and WDC) are transmitted via satellite as well as frame relay and the new software merges these data seamlessly with the data obtained through frame relay.
In addition to expanding capabilities of the REDI system, we have taken several steps toward improving the robustness and reliability of the processing at UC Berkeley. During 1997, we observed a degradation in the performance of the REDI processing due to overloading of the central data acquisition and processing computer. With support from the California Office of Emergency Services, we purchased a second computer and split the data acquisition between the two. This provides two fully independent processing computers, each one handling half of the BDSN acquisition as well as full REDI processing (Figure 7.2). The second computer alleviates the processing load previously borne by a single machine, as well as providing redundancy against failure. In addition to strengthening the processing capabilities, we are planning on replacing the current dedicated phone line from UC Berkeley to AirTouch paging with a radio link, reducing our dependency on this critical landline. The radio link will also provide a direct means of communication with most of the Bay Area, for those organizations which wish to receive the information directly. We are currently investigating different radio types.
While these changes will improve the current situation, they do not guard against the single point of failure provided by the single T1 connection between UC Berkeley and Pacific Bell's frame relay network. Nor does it assure that the REDI processing in McCone Hall will be operational after a major Bay Area earthquake. Consequently, we are working to install a backup REDI facility in the Sacramento headquarters of the California Office of Emergency Services. We have installed a 56-Kbps frame-relay connection at this facility along with a data acquisition and processing computer. We found it necessary to work closely with Pacific Bell in order to get the permanent virtual circuits from 6 BDSN stations (YBH, ORV, CMB, SAO, BKS, and HOPS) and the BSL up and running, as we experienced initial problems with these 8-Kbps circuits. At this point, all circuits are operational and we are beginning to establish data acquistion from the selected sites (Figure 7.1). We hope to have sufficient bandwidth to transmit 20 sps data from the broadband seismometers and 80 or 100 sps data from the strong motion accelerometers. These stations will provide a minumum network for determining the location, magnitude, mechanism, and rupture characteristics for large events in northern California. We plan to use this facility as a proof-of-concept in off-site acquisition, processing and distribution and to work with the USGS and Caltech toward establishing a California-wide backup system.
In the last year, REDI processing was expanded to include the estimation of the seismic moment tensor. Initially automated in 1994, the waveform modeling (Dreger and Romanowicz, 1994) and the surface wave inversion (Romanowicz et al., 1993) methods utilize data from the BDSN and are now fully incorporated in REDI processing. Efforts to streamline the programs and accelerate the computations have been the focus of recent work, with particular attention to the question of determining solution robustness (Pasyanos et al., 1996). As implemented in the REDI processing, both inversion methodologies are run for every qualifying event (earthquakes with ML greater than 3.5) and the results are compared. The estimate of moment magnitude is determined by computing a weighted average of the two solutions, and reliability of the mechanism is measured using the normalized root-mean-square of the moment tensor elements Pasyanos et al., 1996). The moment magnitude is considered "pageable" if the resulting event quality is less than or equal to 2 (on a scale from 0-4, where 0 indicates high quality), while more exacting criteria are required for broadcast of the mechanism (the mechanism correlation must be greater than 75%) and the solution with the best quality is paged.
A review of the REDI moment tensor solutions in the last year indicates the the automatic determination compares favorably to the final solutions obtained by BSL scientists Mike Pasyanos, Hrvojve Tkalcic, and Doug Dreger. In particular, the estimates of moment magnitude are quite robust. When the agreement between the two methods is quite high, the mechanisms also show excellent agreement with the final solutions. However, we have noted problems in the estimation of depth and are working to improve this aspect of the algorithms.
At the present time, the estimates of Mw and fault plan solutions are paged only to BSL staff. However, we have recently initiated a program where this information is transmitted to the strong-motion group at the USGS and integrated in the generation of intensity maps. This is part of a larger scale effort to develop the capability of identifying areas which may have experienced strong shaking for emergency response.
Over the last year, we have worked on the development of a new phase picking algorithm with the goal of replacing the Murdock, Hutt, and Halbert event detection code (Murdock and Hutt, 1983) currently installed on the Quanterra dataloggers. While the detection codes have been extremely effective for triggering the high sampling rate data streams, they have been less useful as phase pickers. Although they generally produce reliable P-picks, the retriggering algorithm creates "phase echoes" while failing to pick later arrivals. This has produced problems using these detections as part of the REDI processing.
The primary objective of this development is to design a reliable algorithm for phase picking of local, regional, and teleseismic events for use in the REDI system. Unlike the Murdock, Hutt, and Halbert detector, which is based on the vertical component channels, we want an algorithm which utilizes the full power of the broadband systems to detect later arrivals, identify phase type, and estimate arrival azimuth. Although our major focus is on the enhancement of our automatic systems, we anticipate that this will reduce the amount of time BSL analysts spend hand picking events.
We initiated this project with a review of existing algorithms. After testing a number of published algorithms, we decided that they did not meet the desired standards of picking P and S arrivals consistently.
We are currently experimenting with a characteristic function based on three terms. The first two terms of this characteristic function are those used in classical pickers such as those of Allen (1978; 1982), expanded for three-component data. The third term is a contribution for rapid changes in phase and is particularly sensitive to later arriving energy. This characteristic function is then processed with a recursive short term average/long term average detector.
We have applied this algorithm on filtered BDSN data. Although we had hoped to use the raw broadband data directly, the microseismic noise levels at the coastal sites severely limit the detection threshold. Instead, we introduced a high-frequency filter between 0.5 and 5.0 Hz and a low frequency filter between 0.01 and 0.1 Hz. These two bands generally provide phase detections for local/regional and teleseismic events respectively, although impulsive bodywave phases are often picked in the high-frequency filtered characteristic function.
Examples of phase picks from two test datasets are shown in Figures 7.4 - 7.5. In the local and regional event dataset, 55 earthquakes of magnitude 4 and higher were processed with the picker. The left panel of Figure 7.4 illustrates the picker travel times as a function of distance. The arrival times of P and S are very well defined out to 2o or so, but show increasing scatter at greater distances. Some of this scatter is due to true variability in regional structure, while part is due to the influence of attenuation on the signal to noise ratio. The distance dependent effect can also be seen in the right panel, where the travel time residual of the automatic picks to hand picks of BSL analyts are plotted for P and S waves. The heavy line represents the total residual of the entire dataset, while the dashed and dotted lines indicate different distance bins. Both the histograms are well peaked around a mean of 0, but the distribution is far from Gaussian. In fact, both show a significant "tail", with the automatic picks being later than the hand picks. This tail is more pronounced for the S-waves than for the P-waves and is also a function of distance.
For the teleseismic test, we selected all events of magnitude 6.5 and higher recorded from 1993-1997, approximately 180 earthquakes. Figure 7.5 illustrates the results for events with source depths between 0 and 70 kms. The picker outlines the P and S branches of the travel-time curve clearly as well as marking the fundamental-mode surface wave arrivals. Portions of the branches for PP, PKP, SKS, PS, SS, and SSS are also apparent. Even subtle back branches are apparent, particularly when compared with Figure 7.6, which illustrates the hand picks made by BSL analysts on teleseismic events between 1993 and 1997. We are very encouraged by the performance of this picker on the small data set here and feel that this will be an important contribution to BSL data processing in the future.
A version of this algorithm has been implemented in a real-time module as part of a REDI development platform. At the present, we are using BH data from the BDSN for the local/regional picker, although we plan to use HH data when the 80 or 100 sps data streams are converted to continuous recording. The teleseismic detector uses LH data. Although initially plagued by some implementation bugs, the current version is operating smoothly and the picks are being fed into an association algorithm. The system is routinely processing earthquakes of magnitude 2.7 and higher in the San Francisco Bay Area and events of magnitude 3.3 and higher in northern and central California.
Although we have made good progress in development over the last year, several problems remain. One problem at the present is retriggering for local earthquakes. We find that the picker will detect the P-wave arrival and often retrigger within 1 or 2 seconds, due to the increasing power in the characteristic function. Rather than restricting the ability of the picker to retrigger (which is the typical solution), we are experimenting with penalizing these picks with lower qualities. By doing so, we hope to avoid the trap of missing a mainshock because of a small foreshock. We are also working to improve the capability of the algorithm to determine the incoming azimuth of the arriving wave. Finally, we are testing different time-domain filters, as the current long-period filter has a significant frequency dependent phase delay. When completed, this picker will form a critical component of both the REDI system and for the planned system for telesesimic event location.
On a daily basis, the BSL continues to locate and determine the magnitude of earthquakes in northern California and adjacent regions. As a general rule, events are analyzed if their magnitude is greater than 2.8 in the Central Coast ranges, greater than 3.0 in all of northern California, or greater than 3.8 in the bordering regions. Traditionally, these events were located using hand-picked arrival times from the BDSN stations in conjunction with P-arrival times from the NCSN using the algorithm st-relp. Over the past several years, the BSL has made a transition in the daily analsis to take advantage of the automatic processing system. As part of this transition, events which have been processed by the automatic system are not generally relocated, although phase arrivals are still hand-picked and the synthetic Wood-Anderson readings are checked. Instead, analysts are focusing on the determination of additional parameters, such as phase azimuth, measures of strong ground shaking. The BSL is also experimenting with the routine location of earthquakes using travel times and azimuth information in a new program developed by Bob Uhrhammer known as bw-relp.
In addition to the routine analysis of local and regional earthquakes, the BSL also processes teleseismic earthquakes. Taking advantage of the CNSS catalog (Malone et al., 1996), analysts review teleseisms of magnitude 5.8 and higher. All events of magnitude 6 and higher are read on the quietest BDSN station, while events of magnitude 6.5 and higher are read on the quietest station and BKS. Earthquakes of magnitude 7 and higher are read on all BDSN stations. Figure 7.6 displays the phase arrival catalog for local, regional, and teleseismic earthquakes for 1993-1997.
The locations and magnitude determined by the BSL are cataloged on the NCEDC. The phase and amplitude data are provided to the NEIC, along with the locations and magnitudes, as contributions to the global catalogs.
The BSL is currently collaboration with the USGS Menlo Park on the production of a Northern California earthquake Bulletin for the period 1993-1997. This publication will document the operations of both the BSL and the USGS in the joint earthquake monitoring activities. We plan to publish this bulletin in Seismological Research Letters.
The Berkeley Seismological Laboratory has been working to apply the available real-time broadband data and experience in automated processing to identification and characterization of teleseisms. The analysis of teleseisms is complicated by the limited azimuth sampled by the BDSN. Nonetheless, we have demonstrated viable methods to accurately identify and locate distant earthquakes, based on analysis of all wavefronts crossing the BDSN (e.g. Uhrhammer et al., 1996; Fulton et al., 1996)
For large teleseisms (M>6.5), we routinely perform Centroid-moment tensor analysis (Dziewonski et al., 1981) to determine the source mechanism and moment, a more accurate measure of size. The CMT analysis if performed automatically given the event location (currently taken from the NEIC preliminary event notification) and a reviewed solution is often available within one to two hours. Following a testing period, this year we began distribution of the BDSN CMT solutions to the community using both e-mail and our web page (www.seismo.berkeley.edu/seismo/tele_mt).
Two independent moment tensor inversion methodologies are currently being used in routine earthquake analysis at the Berkeley Seismological Laboratory: a) a regional surface wave inversion method, which is a two-step frequency domain inversion that uses 15 to 50 second surface waves and is adapted from a method originally developed for a global data (Romanowicz, 1992) and b) a regional time-domain inversion method using complete waveforms at frequencies between 0.01 and 0.10 Hz (Dreger and Helmberger, 1993). Both methods have been fully automated (e.g. Pasyanos et al., 1996) and integrated into REDI (Gee et al., 1996). Human reviewed solutions are still sent out to the community via an e-mail distribution list and also the website: www.seismo.berkeley.edu/ dreger/mtindex.html We estimated 68 moment tensor solutions for events which occurred in and around northern and central California between July 1997 and July 1998. They were determined using one of the methods mentioned above.
In the last year, the Mammoth Lakes region significantly increased activity, starting from November 1997, and continues to the present day. Figure 7.7 shows beach ball diagrams of 18 events in this region with moment magnitude larger than 4.0. The focus of current research is to evaluate the contribution of structural versus source effects in moment tensor inversions and to explain the existance of non-double-couple components in some of these mechanisms. Solutions for all events in the last year are plotted in Figure 7.8 and listed in Table 4..
In the current earthquake notification system, earthquake magnitude is determined using several different methodologies. For example, the preliminary magnitude is based on the decay of the P-wave amplitude with time and typically requires 120 seconds of data for analysis. A revised estimate of size, based on determination of local magnitude, requires analysis of the maximum amplitude of the S-wave packet and imposes delays of 15 to 150 seconds depending on the location of the event. Finally, the determination of the seismic moment tensor, which provides an estimate of moment magnitude, utilizes the complete waveform, a restriction that typically imposes delays of 5-7 minutes. In contrast, a preliminary location is generally available with 15-20 seconds after an earthquake occurs, based on the current technology and the dense network of the NCSN.
As a step toward the development of an early warning system, we are investigating methods for the rapid estimation of earthquake size with the goal of combining these methods with the rapid earthquake locations. We have developed an algorithm based on the growth rate of near-field ramps in the broadband displacement records (Uhrhammer, 1993). This method relates the seismic scalar moment to the ramp growth rate and provides an approach for the rapid estimation of moment magnitude. We tested this algorithm on over 50 earthquakes in central California (Figure 7.10) and observations confirm the results of Uhrhammer (1993) that only a few seconds of data following the P-wave is required for estimation of the relevant parameters (Gee et al., 1997). Figure 7.10 compares the magnitude determined from this method with the final magnitude (either Mw or Ml). For stations within 40 km of the event, the ramp-based magnitude estimates are generally within .5 units of the final magnitude. Beyond 40 km, however, the error in the magnitude estimate increases systematically with distance. This error can be attributed to problems with the attenuation correction and we plan to calibrate the code accordingly. These results are particularly exciting, since this estimate of moment magnitude can be generated within a few seconds after a station has recorded an event. In particular, we found that moment magnitude may be estimated with less than 2 seconds of data following the P-wave arrivals at stations within 20 km of the earthquake. Greater scatter is apparent for distances beyond 20 km. This reflects, in part, the influence of the source mechanism as the geometry of this study is such that the larger source-receiver distances correlate with nodes in P-wave excitation. The results of this analysis are sufficiently promising that we are working to implement this program in conjunction with the rapid event notifications produced by the USGS which will significantly advance rapid earthquake notification. When completed, we anticipate that preliminary locations and magnitudes will be available within 1 minute after the event, in contrast to the current 4-6 minutes. For some applications, preliminary information, even if unreliable, is of great value in the initial assessment of an earthquake's impact (e.g., Savage et al., 1998).
Allen, R., Automatic phase pickers: Their present use and future prospects, Bull. Seis. Soc. Am., 72, S225-S242, 1982.
Dreger, D., and B. Romanowicz, Source characteristics of events in the San Francisco Bay region, USGS Open-File-Report 94-176, 301-309, 1994.
Dziewonski, A. M., T.-A. Chou and J. H. Woodhouse, Determination of earthquake source parameters from waveform data for studies of global and regional seismicity, J. Geophys. Res., 86, 2825-2952, 1981.
Fulton, S., J.J. Durek, L.S. Gee, D.S. Neuhauser, and B. Uhrhammer, Teleseismic earthquake monitoring with a regional network (abstract), EOS Trans. AGU, 77, 451, 1996.
Gee, L. S., D. S. Neuhauser, D. S. Dreger, M. E. Pasyanos, B. Romanowicz, and R. A. Uhrhammer, The Rapid Earthquake Data Integration System, Bull. Seis. Soc. Am, 86, 936-945,1996a.
Gee, L. S., A. Bittenbinder, B. Bogaert, L. Dietz, D. Dreger, B. Julian, W. Kohler, A. Michael, D. Neuhauser, D. Oppenheimer, M. Pasyanos, B. Romanowicz, and R. Uhrhammer, Collaborative Earthquake Notification in Northern California (abstract), EOS Trans. AGU, 77, F451, 1996b.
Gee, L. S., D. S. Neuhauser, R. A. Uhrhammer, S. Fulton, and B. Romanowicz, Getting REDI for Early Warning (abstract), EOS Trans. AGU, 78, F45, 1997.
Johnson, C., A. Bittenbinder, B. Bogaert, L. Dietz, and W. Kohler, Earthworm: A flexible approach to seismic network processing, IRIS Newsletter, XIV (2), 1-4, 1995.
Murdock, J., and C. Hutt, A new event detector designed for the Seismic Research Observatories, USGS Open-File-Report 83-0785, 39 pp., 1983.
Neuhauser, D. S., B. Romanowicz, T. McEvilly, F. Morrison, W. Karavas, J. Friday, D. Rapkin, R. Clymer, R. Uhrhammer, L. Gee, and M. Murray, Digital telemetry and recording enhancements for the Berkeley Digital Seismic Network (abstract), Eos Trans. AGU, 76, F394, 1995.
Romanowicz, B., D. Dreger, M. Pasyanos, and R. Uhrhammer, Monitoring of strain release in central and northern California using broadband data, Geophys. Res. Lett., 20, 1643-1646, 1993.
Pasyanos, M., D. Dreger, and B. Romanowicz, Toward real-time estimation of regional moment tensors, Bull. Seis. Soc. Am, 86, 1255-1269, 1996.
Savage, W., N. Abrahamson, and M. Maclaren, Useful products for electric utilities from integrated regional seismic networks (abstract), Seis. Res. Lett., 69, 166, 1998.
Uhrhammer, R., Rapid seismic moment tensor estimation using near-field waveforms recorded by broadband seismographs (abstract), Eos Trans. AGU, 74, F400, 1993.