S53D-01
Time reversal imaging and cross-correlations techniques by normal mode theory
Time-reversal methods were successfully applied in the past to acoustic waves in many fields such as medical imaging, underwater acoustics, non destructive testing and recently to seismic waves in seismology for earthquake imaging. The increasing power of computers and numerical methods (such as spectral element methods) enables one to simulate more and more accurately the propagation of seismic waves in heterogeneous media and to develop new applications, in particular time reversal in the three-dimensional Earth. Generalizing the scalar approach of Draeger and Fink (1999), the theoretical understanding of time-reversal method can be addressed for the 3D- elastic Earth by using normal mode theory. It is shown how to relate time- reversal methods on one hand, with auto-correlation of seismograms for source imaging and on the other hand, with cross-correlation between receivers for structural imaging and retrieving Green function. The loss of information will be discussed. In the case of source imaging, automatic location in time and space of earthquakes and unknown sources is obtained by time reversal technique. In the case of big earthquakes such as the Sumatra-Andaman earthquake of december 2004, we were able to reconstruct the spatio-temporal history of the rupture. We present here some new applications at the global scale of these techniques on synthetic tests and on real data.
S53D-02
Time-Reversal to Estimate Focal Depth for Local, Shallow Earthquakes in Southern California
Current approaches for focal depth estimation are typically based on travel times and result in large uncertainties primarily due to poor data coverage and inaccurate travel time picks. We propose an alternative method based on an adaptation of time-reversed acoustics (TRA). In the context of TRA theory, the autocorrelation of an earthquake recording can be thought of as the convolution of the source autocorrelation function with the autocorrelation of the Green's function describing propagation between source and receiver. Furthermore, the signal to noise ratio (S/N) of stationary phases in the Green's function may be improved by stacking the autocorrelations from many receivers. In this study, we employ such an approach to estimate the focal depth of shallow earthquakes based on the time lag between the direct P phase and pP converted phase, which is assumed to be stationary across the receiver array. Focal depth estimates are easily obtained by multiplying half the pP time lag by the average velocity above the earthquake. We apply this methodology to estimate focal depths for several local earthquakes in Southern California. Earthquake recordings were obtained from the Southern California Earthquake Center (SCEC) for events with accurate, independent estimates of focal depth below about 15 km, and local magnitudes between 4.0 and 6.0. We observe pP in the stacked autocorrelations that correspond to the focal depths listed in the SCEC catalog for earthquakes located throughout Southern California. The predictive capability of the method is limited by S/N, defined as the pP amplitude divided by the background noise level of the stacked correlation. By considering subsets of the Southern California array, we explore the sensitivity of the S/N on station density and location (i.e. epicentral distance & azimuth). We find S/N is generally better for subsets of receivers within regions with relatively simple geologic structure. We are currently developing an extension of this methodology using the time- frequency correlation function, which may significantly reduce the station coverage required for accurate focal depth estimation.
S53D-03 INVITED
Application of the Time Reversed Acoustic Concept to Earthquake Location and Focal Depth Determination
Local and regional seismograms have long codas due to strong scattering of seismic waves in the crust. Accurate identification of individual phases (P, pP, PmP, S, sS, etc.) is difficult because the scattered arrivals complicate the local seismograms and introduce errors in picking phases and their arrival times. These, in turn, introduce errors in hypocenter parameters. Strong scattering, that is a detriment to picking individual phases, makes it possible to apply the Time Reversed Acoustic (TRA) concept to local earthquake location and focal depth determination. The basic idea in TRA is to time reverse the recorded signals (seismograms) and to inject them into the earth. If the earth structure is known, the back-propagated signals could focus at the source. We demonstrate this by synthetic (numerical) examples and with seismograms from earthquakes. Foci, determined by TRA and by traditional methods with arrival times from close-in stations, agree very well. Independently, we present a method based on the TRA concept for earthquake focal depth determination. In a highly scattering medium, the source time function determination and pP identification can be accomplished simply, by autocorrelation of the seismograms.
S53D-04 INVITED
Regional Seismic Depth-Phase Identification: Strengths and Weaknesses of Cepstral Signal Processing Techniques
One of the primary discriminants between man-made versus natural seismicity is the depth of an event. The identification of seismic depth phases (surface reflections) is the most straightforward way to establish event focal depth. However, at regional distances the accurate detection and identification of seismic depth phases such as pPn and sPn remains an elusive goal. In this distance range the first arriving phase (Pn) is commonly believed to be a "whispering gallery mode" produced by multiple reflections from velocity gradients at the bottom of the Moho. These multiple reflections and other crustal reflections and reverberations, arriving soon after the direct wave, create a complex coda. The depth phases may be buried within this coda, but are often difficult to differentiate from other wave phenomena. We have extensively tested an array-based signal processing technique known as the Cepstral F-Statistic Method to detect regional depth phases in seismic coda. Cepstral methods exploit the periodicity that occurs in the power spectrum when echoes are present between primary and coda arrivals, such as those expected between the direct (P) and depth (pP and/or sP) phases. However, the original cepstral method does not distinguish between peaks (i.e. detections) due to depth phases and those due to other crustal and scattering wave phenomena. To address this weakness we assign significance to cepstral peaks through the calculation of an F statistic. In addition we estimate the slowness and back azimuth of the phases corresponding to peaks in the cepstral F statistic to help assess their potential as depth phases. Even with these and other improvements, the application of cepstral methods to data at regional distances from small-to-intermediate-sized events has produced inconsistent results, resulting in a high number of false detections. In this work we present a case study of depth-phase identification in the region surrounding the Korean Peninsula. This region represents the ideal proving ground for cepstral techniques at regional distances, because of its relatively simple tectonic structure, importance in nuclear monitoring, and availability of high quality short-period array data (KSAR/KSRS seismic array, Wonju, South Korea). Our results indicate that cepstral-based identification of the depth phases is problematic in more than 50% of the cases. However, we have found that while cepstral methods cannot independently and accurately identify regional depth phases, they can be effectively combined with other methods to produce good regional depth estimates.
S53D-05
Temporal distribution of strong high frequency energy radiation inferred from far-field broadband records
Recently, we developed a new method to determine earthquake magnitudes using durations of high frequency energy radiation and the maximum displacement amplitudes measured from processing of first arriving P-waves (Hara, 2007a. Earth Planets Space, 59, 227-231; Hara, 2007b. Earth Planets Space, 59, 561-565). In this technique, the difference between P arrival time and the time when the amplitude of high bandpass (2-4 Hz) filtered seismogram becomes the largest is used to set width of moving window for smoothing time series. In this study, we measured these differences for large shallow earthquakes that occurred between 1995-2007. We normalized them by centroid time shifts to find the low frequency in the range of 0-20 % and the high frequency around 50 %. This result supports the effectiveness of the measurement procedure of duration of the high frequency energy radiation of Hara (2007a, b). We also applied our measurement procedure to the January 15, 1993 Kushiro-Oki, July 12, 1993 Hokkaido-Nansei-Oki, and December 28, 1994 far east off Sanriku earthquakes. We compared the measured time differences to the results of inversion of strong motion records to find the qualitative agreement between them.
S53D-06
The 7-sec Microseism Origin Investigated From Amplitude Variations and Polarization of Noise.
In the absence of earthquakes, broadband seismic signal is dominated by a peak of noise around 7 sec of period, called microseisms. Variations of this peak amplitude are investigated using the 30 stations of the GEOSCOPE global network. For each station, power spectrum estimates of the seismic noise are averaged over sequences of 24 hours. The noise level of 10 years of continuous data has been computed. Variations of the seismic noise amplitudes are similar every year. We observe seasonal variations of the 7-sec microseismic peak that are correlated with the station latitude. For stations close to the equator the peak has constant low amplitude over the year. The maximum amplitude of the 7-sec peak is observed for stations in the southern hemisphere between June and September that is during local winter. Similarly, high noise level is observed for stations in the Northern hemisphere between November and February but with smaller amplitude than in the Southern hemisphere. These noise level variations are well correlated with the location and intensity of storms in the oceans that are stronger and more frequent in winter. In order to locate the source of the microseisms, we systematically computed the degree of polarization of noise sequences and when the noise is polarized, the source back azimuth is estimated using the method of Schimmel and Gallart (2004). The computation is performed using EGEE computing grid. The variation of the microseim sources can then be followed over the year and compared to the storm locations.
S53D-07
Acoustic and Elastodynamic Redatuming for VSP Salt Dome Flank Imaging
We apply an extension of the concept of Time Reversed Acoustics (TRA) for imaging salt dome flanks using Vertical Seismic Profile (VSP) data. We demonstrate its performance and capabilities on both synthetic acoustic and elastic seismic data from a Gulf of Mexico (GOM) model. This target-oriented strategy eliminates the need for the traditional complex process of velocity estimation, model building, and iterative depth migration to remove the effects of the salt canopy and surrounding overburden. In this study, we use data from surface shots recorded in a well from a walkaway VSP survey. The method, called redatuming, creates a geometry as if the source and receiver pairs had been located in the borehole at the positions of the receivers. This process generates effective downhole shot gathers without any knowledge of the overburden velocity structure. The resulting shot gathers are less complex since the VSP ray paths from the surface source are shortened and moved to be as if they started in the borehole, then reflected off the salt flank region and captured in the borehole. After redatuming, we apply multiple passes of prestack migration from the reference datum of the borehole. In our example, the first pass migration, using only simple vertical velocity gradient model, reveals the outline of the salt edge. A second pass of reverse-time prestack depth migration using the full, two-way wave equation, is performed with an updated velocity model that now consists of the velocity gradient and the salt dome. The second pass migration brings out the dipping sediments abutting the salt flank because these reflectors were illuminated by energy that bounced off the salt flank forming prismatic reflections.