Automated Detection, Extraction, and Measurementof Regional Surface WavesA. L. LEVSHIN1and M. H. RITZWOLLER1Abstract Ð Our goal is to develop and test an eective method to detect, identify, extract, and quantifysurface wave signals for weak events observed at regional stations. We describe an automated surface wavedetector and extractor designed to work on weak surface wave signals across Eurasia at intermediateperiods (8 s±40 s). The method is based on phase-matched ®lters de®ned by the Rayleigh wave grouptravel-time predictions from the broadband group velocity maps presented by RITZWOLLER and LEVSHIN(1998) and RITZWOLLER et al. (1998) and proceeds in three steps: Signal compression, signal extraction orcleaning, and measurement. First, the dispersed surface wave signals are compressed in time by applying ananti-dispersion or phase-matched ®lter de®ned from the group velocity maps. We refer to this as the`compressed signal.' Second, the surface wave is then extracted by ®ltering `noise' temporally isolated fromthe time-compressed signal. This ®ltered signal is then redispersed by applying the inverse of the phase-matched ®lter. Finally, we adaptively estimate spectral amplitude as well as group and phase velocity onthe ®ltered signal. The method is naturally used as a detector by allowing origin time to slide along the timeaxis. We describe preliminary results of the application of this method to a set of nuclear explosions andearthquakes that occurred on or near the Chinese Lop Nor test site from 1992 through 1996 and oneexplosion on the Indian Rajasthan test site that occurred in May of 1998.Key words: Surface waves, Rayleigh waves, matched ®lters, group velocity, nuclear monitoring,Ms: mbdiscriminant.1. IntroductionThe Ms: mbdiscriminant and its regional variants are the most reliabletransportable means of discriminating earthquakes from explosions. To measuresurface wave amplitudes accurately in order to estimate Msis very challenging forsmall events in which surface waves may not be readily identi®able in rawseismograms. To provide these amplitude measurements, it is crucial to be able toreliably detect small amplitude surface wave-packets and extract all and only thedesired wave-packets reliably so that spectral amplitude measurements can beobtained.1Department of Physics, University of Colorado at Boulder, Boulder, CO, 80309-0390, USA.E-mail: [email protected] appl. geophys. 158 (2001) 1531±15450033 ± 4553/01/081531 ± 15 $ 1.50 + 0.20/0Ó BirkhaÈuser Verlag, Basel, 2001Pure and Applied GeophysicsWe describe a surface wave detector and extractor designed to work on weaksurface wave signals across Eurasia at intermediate periods (8 s±40 s). It is foundedon a long history of surface wave frequency-time analysis (e.g., DZIEWONSKI et al.,1969; LEVSHIN et al., 1972, 1989, 1992; CARA, 1973; RUSSELL et al., 1988). However,successful detection and wave-packet extraction are both dependent on the ability tomake accurate predictions of surface wave arrival times at intermediate periods. Ourmethod is based on the Rayleigh wave group travel-time predictions from the recentbroadband group velocity maps of RITZWOLLER and LEVSHIN (1998) and RITZWOL-LERet al. (1998) and proceeds in three steps: Signal compression, signal extraction orcleaning, and measurement.First, the dispersed surface wave signals are compressed in time by applying ananti-dispersion or phase-matched ®lter de®ned from our group velocity maps. Werefer to this as the `compressed signal.' Second, the surface wave is then extracted by®ltering `noise' temporally isolated from the time-compressed signal. This ®lteredsignal is then redispersed by applying the inverse of the phase-matched ®lter. Werefer to this wave-form as the `®ltered' or `cleaned signal.' Finally, we adaptablyestimate spectral amplitude as well as group and phase velocity on the ®ltered signal.After amplitudes are measured, Msis estimated using an empirical relation such asthat recently presented by REZAPOUR and PEARCE (1998).The general methodology of matched ®ltering was developed previously by anumber of other researchers (e.g., HERRIN and GOFORTH, 1977; STEVENS, 1986;RUSSELL et al., 1988; STEVENS and MCLAUGHLIN, 1997). We introduce threeinnovations here: (1) the use of recent group velocity maps to de®ne the matched-®lters, (2) automation of the procedure, and (3) the use of the method as adetector.We describe preliminary results of the application of this method to a set ofnuclear explosions and earthquakes that occurred on or near the Chinese Lop Nortest site from 1992 through 1996 and one explosion on the Indian Rajasthan test sitethat occurred on May 11, 1998.2. Group Velocity Maps and Correction SurfacesElsewhere we have described the construction of intermediate period groupvelocity maps across Eurasia (e.g., RITZWOLLER and LEVSHIN, 1998; RITZWOLLERet al., 1998), the Arctic (LEVSHIN et al., 2001), South America (e.g., VDOVIN et al.,1999), and Antarctica (e.g., VDOVIN, 1999). The method of tomography and theconstruction of group velocity correction surfaces is described by BARMIN et al.(2001) in this volume. In this paper we will use the somewhat dated group velocitymaps presented by RITZWOLLER et al. (1998).Figure 1 displays group velocity correction surfaces computed from the 20 sgroup velocity map of RITZWOLLER et al. (1998) for four stations: AAK (Ala-Archa,1532 A. L. Levshin and M. H. Ritzwoller Pure appl. geophys.,Kirghizstan), ABKT (Alibek, Turkmenistan), BRVK (Borovoye, Kazakhstan), andTLY (Talaya, Russia). For a given period, the value at each point on these mapsrepresents the group velocity perturbation that a surface wave that originated at thepoint would experience if recorded on the speci®ed station. The perturbations arerelative to the group velocity at the station. In this form, group velocity maps can beused eciently to predict dispersion curves for any event:station pair. BARMIN et al.(2001) present examples of group velocity correction surfaces for Rayleigh waves at40 s period. We note two circumstances regarding the correction surfaces in Figure 1and those shown by BARMIN et al. (2001). First, the corrections can be very large.For example, the 20 s Rayleigh wave from an event in the Caucasus 2500 km toAAK would experience a total group velocity perturbation of almost 400 m/s relativeto the group velocity at AAK or a perturbation in arrival time of more than2.5 minutes. Second, the