Earthworm : A Flexible Approach to Seismic Network Processing

Carl E. Johnson, University of Hawaii, Hilo
Alex Bittenbinder, Barbara Bogaert, Lynn Dietz and Will Kohler
U.S. Geological Survey, Menlo Park

Rapid technological evolution is both a boon and a bane to regional seismic network operators. On the one hand, smaller and faster computers and dramatic improvements in com-munications technology fuel the development of increasingly capable network data analysis systems. On the other hand, innovative systems are soon antiquated and the cost of 'keeping up' can be prohibitive. An adaptable, scalable seismic data processing system that can evolve with technological trends is the Holy Grail of regional network operators.

In June of 1991, regional seismic network operators gathered at the Alta II conference in Utah to address common interests. Network acquisition and processing systems were identified as being a source of common problems, including: obsolescent hardware, high maintenance costs, lack of commonality between institutions, and difficulties in integrating enhancements such as digital telemetry and rapid response functions.

What emerged was a clear need for a regional network data processing system that would offer a number of features:

  • Scalability- The system design should permit configurations to meet the requirements of the small as well as large networks. The cost of each such configuration should be proportional to the scale of its application.

  • Flexibility- It should be possible to tailor the system to specific network requirements, including various types of digital telemetry, processing algorithms, and interfaces to ancillary systems. It should also be possible to enhance the system after it has been put into operation. This implies the ability to readily incorporate preexisting computer codes, and to support development efforts without destabilizing other functions.

  • Longevity- The system should permit contemporary, cost-effective hardware to be incorporated as it becomes available.

  • Data Exchange- The system should support exchange of various data in near real-time to permit adjacent regional networks to improve their coverage and accuracy.

  • Support- Long term, committed support is required and should be tailored to complement the existing resources available at the various regional networks. Such support should range from supplying complete turn-key service to accepting suggestions and contributions.
  • The Seismology Branch of the USGS at Menlo Park undertook to address these issues. The Earthworm project was initiated shortly after the Alta II conference and has produced a system which attempts to meet the above requirements.

    Earthworm: A Modular Network Processing System

    The Earthworm system can be thought of as a toolkit of processing modules for integrating regional seismic network data processing. Individual modules provide such functionality as: acquisition of digital data, digitization of analog data, phase arrival identification, amplitude and duration measurements, event association, phase identification, hypocenter determination, event catalog generation, and earthquake alarm generation. Processing is fully automatic, proceeding from the collection and integration of digital and analog seismic data, through the generation of earthquake catalogs and near real-time event alarms, without human intervention. Of course, any or all automatically processed events can be reexamined by trained analysts as desired.

    The basic system architecture is an object-oriented, message-passing design. Each seismic function (e.g. data acquisition, P-phase picking, phase association) is encapsulated into a software module. Modules can generate messages (e.g. pick messages), and broadcast such messages onto a common media (e.g. a dedicated local network). Modules can also listen on such media and 'tune in' to messages of interest to them (e.g. trace data messages).

    Figure1. The basic Earthworm architecture is an object-oriented, massage-passing design. Each seismic function is encapsulated into a software module which can broadcast and receive messages on common communication media. The specific configuration of modules and media can be tailored for individual installation.

    Two implementations of this broadcast scheme are used in the system. One utilizes standard, dedicated local area networks and is used for communication between modules residing on separate computers. The second is based on interprocess communication mechanisms, and permits communication between modules residing within one computer. A pair of 'adapter' modules are used to move messages between the two media.

    System integrity is monitored via 'heartbeat' and error messages. Any module which ceases to function or detects an error (e.g. missing or garbled message) will cause the system to issue a suitable notification. The entire system, in turn, issues a heartbeat which can be monitored by a separate computer capable of generating alarms.

    This architecture offers several benefits:

  • A variety of computer systems can be used and freely intermixed to create a system of desired power and features. To date, a variety of configurations using Sun (SunOs and Solaris), Intel 486 (DOS and OS/2), and Pentium (OS/2) computers have been demonstrated. This permits the Earthworm system to be based on the most suitable and cost-effective platform for each installation.

  • For large networks, multiple copies of the same module may be run in parallel to achieve the requisite throughput. At Menlo Park, for example, the analog 'front end' consists of three digitizer modules feeding two P-phase picking modules.

  • Modules can be easily migrated individually or collectively to more capable or more cost effective hardware as computer technology continues its current rapid evolution.

  • Developmental modules, executing on their own computers, can be connected to an operational system without affecting other modules. For example, an experimental teleseismic trigger can be added without disrupting critical notification functions. New technologies, such as digital telemetry can be smoothly brought on line without disruption of existing data acquisition.

  • Various types of data can be exchanged with other networks in near real-time. The system includes "Import" and "Export" modules, capable of exchanging internal messages, such as trace data, picks, or hypocentral estimates, with other systems via internet.
  • Key Algorithms

    Currently, several existing algorithms have been integrated into Earthworm to perform the rapid notification functions: The Rex Allen P-picking scheme (Allen, 1978, 1982) is used to determine P-phase and coda characteristics. A variety of computer codes developed at the USGS in Menlo Park are used to verify and filter automatically located events. Hypo71 and Hypoinverse are currently used to compute final locations.

    A key component of the Earthworm automated processing system is the rapid phase associator (Johnson, 1994). The problem is to consolidate phases from a variety of sources (possibly many earthquakes intermingled in an arbitrary time order) into a catalog of discrete events, each associated with a list of supporting arrival time data. To obtain the efficiency necessary to process data during a vigorous aftershock series or swarm, a robust stacking algorithm was developed. Back-projections for all phases arriving within a given time window are stacked over a four dimensional grid of latitude, longitude, depth, and time. With a spatial grid size of 3 to 5 kilometers, the algorithm is capable of correctly resolving co-located events differing in origin time by a few seconds, or simultaneous events spaced at least 10 km apart. Both conditions are frequently encountered during vigorous seismic sequences.

    Figure 2. Earthworm configuration at USGS Menlo Park. The system offers acquisition of analog and digital data, rapid hypocenter determination, pager notification, catalog generation, and an interactive graphic quick-look facility.

    Current Applications

    Several Earthworm systems and derivatives are currently in operation: At USGS, Menlo Park, a 512-channel, analog telemetry Earthworm system is in production, as shown in Figure 2. Its performance has been certified by observing its performance on 38 selected segments of historic digital waveform data representing various seismic crises over the last 11 years. Figure 3 shows the seismicity reported by Earthworm in its first 20 days of real-time operation. In addition, integration of 24-bit digitally telemetered data has been demonstrated and is awaiting the completion of field stations. Exchange of phase data in near real-time has been demonstrated in a cooperative effort with the California Institute of Technology, and the USGS, Pasadena. Through the combined efforts of personnel at the Geophysical Institute, University of Alaska, Fairbanks and USGS, Menlo Park, the DataScope package developed at the IRIS JSP Center has been integrated and is scheduled to serve as the rapid-response event verification system at Menlo Park by the first quarter 1996.

    Figure 3. Map of California showing the epicenters reported automatically by the Menlo Park Earthqirm system in its first 20 days of operation on live CALNET data (July 19, 1995 to August 7, 1995) prior to system tuning. Red dots show hypocenters having at least 8 arrivals, rms<0.3 sec, horrizonal error <2.5km and vertical error <5 km. Pink dots show all other reported epicenters.

    At University of Alaska at Fairbanks, an Earthworm system capable of processing 256 channels is in operation. The DataScope integration mentioned above is in operation, and the system is being actively enhanced. Several valuable enhancements produced at Fairbanks have also been incorporated into the Earthworm system. Another 256-channel system is being assembled for the University of Utah at Salt Lake City, and is scheduled for installation by January1996. The overall objective of this effort is to install similar systems at various networks in the Intermountain region and to enable peer-to-peer data exchange in near real-time. In addition, derivative systems are in operation at University of California at Berkeley, University of Hawak‚ at Hilo, and University of Washington at Seattle.

    Future Directions

    Development of the Earthworm system is ongoing, and a number of enhancements are in various stages of planning and completion.

    We plan to enhance the facilities at Menlo Park to provide rapid urban damage assessment in case of catastrophic earthquakes. The initial phase will consist of integrating existing ground motion prediction codes into the Earthworm system, thus permitting maps of predicted ground shaking and damage patterns to be computed within minutes after an event. Subsequent phases will integrate telemetered strong-motion data to improve the reliability of such maps.

    The phase association module could be adapted to sparse regional networks of the sort used for nuclear test ban monitoring. The primary modification would be the stacking of multiple phases, and modifying the phase associator to operate with a spherical rather than a planar geometry, thus addressing the problem of teleseismic phase association.

    A real-time spectral processing module and associated display have been demonstrated. With the addition of a tremor amplitude detector, the basic Earthworm configuration could serve the needs of the volcano monitoring community. The scalable nature of the design facilitates its use in relatively small networks such as those used in volcano monitoring, where cost is a primary consideration. In a general sense, though, it would seem that the most important feature of the Earthworm system with respect to future applications is its ability to assimilate rapidly evolving technology so that network processing capabilities can evolve smoothly, without major breaks and periods of retraining. It is no minor concern in these times of tightening budgets that the survival of dense regional seismic networks depends upon efficient and economic processing capabilities.

    References
    Allen, R.V., 1978. Automatic earthquake recognition and timing from single traces. Bull. Seism. Soc. Am., vol. 68, no. 5, p.1521-1532.
    Allen, R.V., 1982. Automatic phase pickers: Their present use and future prospects. Bull. Seism. Soc. Am., vol.72, no.6, p.S225-S242.
    Johnson, C.E., Lindh A.G., and Hirshorn, B. 1994. Robust regional phase association. USGS Open File Report 94-621.


    Go to next Article
    Return to: Newsletter Title Page
    Return to: Information on IRIS