The Virtual Seismologist in SeisComP3 (VS(SC3)) provides near instantaneous estimates of earthquake magnitude as soon as SeisComp3 origins are available. With a well-configured SeisComP3 system running on a dense network, magnitudes for local events can be available within 8-15 s of origin time. VS(SC3) can be a key component of an earthquake early warning system, and can be used to provide rapid earthquake notifications. With the capability to estimate magnitude (given a location estimate) with 3-seconds of P-wave information at a single station, VS(SC3) magnitude estimates are tens of seconds faster than conventional Ml calculations, which require waiting for the peak S-wave amplitudes. The VS magnitude estimation relationships consist of 1) a relationship between observed ground motion ratios (between vertical acceleration and vertical displacement) and magnitude, and 2) envelope attenuation relationships describing how various channels of envelope amplitudes vary as a function of magnitude and distance. These relationships were derived from a Southern California dataset with events in the magnitude range 2.5 <= M <= 7.6 and the Next Generation Attenuation (NGA) strong motion dataset. Once a SeisComp3 location estimate is available, VS magnitude estimates can be computed with as little as 3 seconds of envelope data at a single station (i.e., 3 seconds after trigger time at a single station). Typically, usable envelope data is available at numerous stations by the time the first SeisComp3 origin is available. The VS magnitude estimates are then updated every second for 30 seconds (configurable). The SeisComp3 implementation allows for use of broadband high-gain seismometers (with clipping value selected) as well as strong motion data. For co-located stations, VS magnitudes are calculated using the strong motion channels if the broadband channels saturate. VS magnitudes in SeisComp3 are called MVS.
The Virtual Seismologist method is a Bayesian approach to earthquake early warning (EEW) that estimates earthquake magnitude, location, and the distribution of peak ground shaking using observed picks and ground motion amplitudes, predefined prior information, and envelope attenuation relationships (Cua, 2005; Cua and Heaton, 2007; Cua et al., 2009). The application of Bayes' theorem in EEW (Cua, 2005) states that the most probable source estimate at any given time is a combination of contributions from prior information (candidate priors include network topology or station health status, regional hazard maps, earthquake forecasts, and the Gutenberg-Richter magnitude-frequency relationship) and constraints from the available real-time ground motion and arrival observations. VS is envisioned as an intelligent, automated system capable of mimicking how human seismologists can make quick, relatively accurate “back-of-the-envelope” interpretations of real-time (and at times, incomplete) earthquake information, using a mix of experience, background information, and real-time data. The formulation of the VS Bayesian methodology, including the development of the underlying relationships describing the dependence of various channels of ground motion envelopes on magnitude and distance, and how these pieces come together in EEW source estimation, was the result of the PhD research of Dr. Georgia Cua with Prof. Thomas Heaton at Caltech, from 1998 through 2004.
The first real-time VS prototype system was developed by Georgia Cua and Michael Fischer at ETH Zurich from 2006-2012. This prototype used location estimates generated by the Earthworm Binder module (Dietz, 2002) as inputs to the VS magnitude estimation. This architecture was used in continuous real-time testing in California (from 2008-2016) and Switzerland (2010-2016). During this period, VS was one of the three EEW algorithms that make up the CISN ShakeAlert EEW system (http://www.cisn.org/eew/). The other algorithms were the ElarmS algorithm from UC Berkeley and the TauC/Pd OnSite algorithm from Caltech. VS was retired from the ShakeAlert system in 2016.
In 2012/13, with funding from the EU projects NERA ("Network of European Research Infrastructures for Earthquake Risk Assessment and Mitigation") and REAKT ("Strategies and Tools for Real-Time EArthquake RisK ReducTion"), VS was integrated into SeisComP3 by the Seismic Network group at the SED in ETH Zurich and gempa GmbH.
Both real-time VS implementations (Binder- and SeisComp3-based) focus on real-time processing of available pick and envelope data. Prior information is not included.
Although the codes were effectively re-written, the basic architecture used in the original Earthworm-based implementation is used in SeisComp3. The SeisComp3 modules scautopick, scautoloc, and scevent replace the Earthworm Binder module for providing location estimates. Two VS-specific modules were developed to continuously calculate envelope amplitudes and to calculate and update VS magnitudes (MVS) once a SeisComP3 origin becomes available.
MVS is calculated and updated (with updates attached to the preferred origin) each second for 30 seconds (unless configured differently) after it is first invoked by the availability of a new SeisComp3 event. If configured, Ml can also be calculated for these events.
An additional module, scvsmaglog, creates log output and mails solutions once a new event is fully processed. It also provides an interface to send alerts in real-time.
In an effort to provide method-agnostic processing and logging modules (e.g. in order to also support FinDer with the same modules), scenvelope is being replaced by sceewamps and scvsmaglog is being replaced by sceewlog.
Configuring and optimizing VS(SC3) for EEW
The performance of VS(SC3) is strongly dependent on: 1) the quality and density of the seismic network; 2) the configuration of the general SeisComp3 system. scautoloc requires at least 6 triggers to create an origin. Given the network geometry, maps of when VS estimates would be first available (indicative of the size of the blind zone as a function of earthquake location relative to stations) can be generated for regions where EEW is of interest. VS(SC3) uses scautoloc, which was not built for EEW, so an additional delay of at most a few seconds is required for origin processing. VS magnitudes (MVS) can be expected within 1-2 seconds after a SeisComp3 origin is available. In the densest part of the Swiss network, SeisComp3 origins are available within 10-15 seconds after origin time; MVS is typically available 1-2 seconds later.
The VS magnitude estimation relationships in Cua (2005) were derived from a dataset consisting of Southern California waveforms and the NGA strong motion dataset. In theory, customizing VS to a specific region requires deriving a set of envelope attenuation relationships (168 coefficients) and relationships between ground motion ratios and magnitude (6 coefficients) from a regional dataset. In practice, the VS magnitude estimation relationships derived from Southern California have been shown to work reasonably well in Northern California and Switzerland (Behr et al, 2012). The envelope and ground motion ratio coefficients from Cua (2005) are hard-coded in scvsmag, and should not be modified without full understanding of the VS methodology and potential consequences of the modifications.
Although scautoloc can produce origins at any depth, the VS magnitude estimation relationships assume a depth of 3 km. For this reason, it is expected that MVS will systematically underestimate magnitudes for deep earthquakes. It may be most practical to simply add empirically derived offsets to MVS for deeper events, or for particular regions.
The VS system currently being offered is a test version. A tool for dissemination of results is not part of the core modules.
The rate of false alarms and missed events is determined by the output of the normal SeisComp3 origin chain (scautopick, scautoloc), and will be similar to the performance of the automatic setup for typical network operations (i.e. if you do not trust your automatic origins for the network, you will not trust them for VS either). A solution quality is independently estimated by VS, combining information on location quality and station quality. See scvsmag on how the VS specific solution quality is computed.
The SeisComp3 VS modules are free and open source, and are part of the SeisComp3 distribution from Seattle v2013.200. They are distributed under the 'SED Public License for SeisComP3 Contributions'.