Finite-Fault Rupture Detector (FinDer): Going Real-Time in Californian ShakeAlert Warning System

Rapid detection of local and regional earthquakes and issuance of fast alerts for impending shaking is considered beneficial to save lives, reduce losses, and shorten recovery times after destructive events (Allen et al., 2009). Over the last two decades, several countries have built operational earthquake early warning (EEW) systems, including Japan (Hoshiba et al., 2008), Mexico (Espinosa-Aranda et al., 1995), Romania (Mărmureanu et al., 2011), Turkey (Erdik et al., 2003), Taiwan (Hsiao et al., 2011), and China (Peng et al., 2011). Other countries, such as the United States (Böse, Allen, et al., 2013), Italy (Satriano et al., 2011), and Switzerland (Behr et al., 2015), are currently developing systems or evaluating algorithms in their seismic real-time networks. Over the past eight years, scientists at the California Institute of Technology (Caltech), the University of California–Berkeley, the University of Southern California, the University of Washington, the U.S. Geological Survey (USGS), and the Swiss Federal Institute of Technology (ETHZurich, Switzerland) developed the U.S. west-coast-wide ShakeAlert EEW demonstration system for California, Oregon, and Washington (Böse, Allen, et al., 2013). Real-time alerts are currently being shared with around 300 individuals, companies, and emergency response organizations to gather feedback about system performance, to educate potential end users about EEW, and to identify needs and applications of a future operational warning system. To quickly determine earthquake magnitudes and (pointsource) hypocenter locations, the Californian ShakeAlert system processes and interprets real-time waveform data streams from several hundred seismic broadband velocity and strong-motion acceleration sensors as part of the California Integrated Seismic Network (CISN). Designed as a hybrid system, ShakeAlert combines the estimates from one single-sensor and two network-based EEWalgorithms that run in parallel, including τc Pd Onsite (Wu et al., 2007; Böse, Hauksson, Solanki, Kanamori,Wu, et al., 2009; Böse, Hauksson, Solanki, Kanamori, and Heaton, 2009), ElarmS (Allen, 2007; Kuyuk et al., 2014), and theVirtual Seismologist (Cua et al., 2009; Behr et al., 2015). Once an earthquake has been detected, each algorithm starts independently sending reports to the ShakeAlertDecision Module, which—based on these reports—determines the most probable event parameters (location, magnitude, and origin time) and uncertainties, which are then continuously updated with progressing time as new data arrive. Using these estimated source parameters, the ShakeAlert UserDisplay software, which is installed at the end user’s site and subscribes to the ShakeAlert system, predicts and displays the arrival and intensity of expected peak shaking (Böse, Allen, et al., 2013). Encouraged by its promising real-time performance during the recent moderate-size 2014M 5.1 La Habra andM 6.0 South Napa earthquakes in southern and northern California, the ShakeAlert Science team decided in 2014 to include the Finite-fault rupture Detector algorithm, FinDer (Böse et al., 2012), as a fourth seismic algorithm in the ShakeAlert system to enable the use of rupture-to-site distances in ground-motion predictions and hence improve EEW performance during large-magnitude (M ≥6:0) earthquakes. Based on a rapid nearor far-source classification and comparison with precalculated templates, FinDer provides rapid estimates of the centroid position (latcentroid, loncentroid), length L, and strike Θ of an ongoing fault rupture, assuming a line source (Fig. 1). The integration of FinDer into ShakeAlert was finalized in April 2015. Currently, FinDer is operated as a MathWorks MATLAB (www.mathworks.com/products/ matlab, last accessed September 2015) stand-alone code with a C++ waveform-processing module installed at three CISN datacenters at Caltech/USGS-Pasadena, UC Berkeley, and USGSMenlo Park. Translation of the FinDer algorithm to C++ by USGS-Menlo Park, ETH Zurich, and Caltech is currently under way. We will start this article with a brief review of the FinDer algorithm developed by Böse et al. (2012), followed by a summary and demonstration of recent improvements to the algorithm, including error estimates, usage of generic and faultspecific templates, and extension to subduction-zone earthquakes. We will demonstrate and evaluate the real-time and off-line performance of FinDer during the recent 2014M 5.1 La Habra and M 6.0 South Napa (California) earthquakes, as well as the 2010 M 7.2 El Mayor–Cucapah (Mexico) and 2011 M 9.0 TohokuOki (Japan) earthquakes.