P217 Automated P- and S-Wave Picking of Micro Earthquakes Re-Corded by a Vertical Array

Transcription

1 P217 Automated P- and S-Wave Picking of Micro arthquakes Re-Corded by a Vertical Array T. Fischer* (Geophysical Institute, Czech Academy of Science), A. Bouskova (Geophysical Institute, Czech Academy of Science), L. isner (Schlumberger Cambridge Research) & J. Le Calvez (Schlumberger DCS, Houston) SUMMARY Large number of induced events in microseismic datasets requires automated picking of P- and S-wave arrivals. Furthermore, the automated measurement of the wave arrival times en-ables, compared to manual processing, fast and repeatable results. However, automatic pick-ing usually suffers from lower accuracy and often only P-wave pickers are being used. We developed a new method for picking P- and S-waves at a linear receiver array, which employs polarization properties and array consistency of the detected phases. The automated method mimics manual picking procedure. We benchmark the automated method on a real micro-seismic dataset by comparison with manual picking. We show that the automated picker is robust as it has found more than 80% of the true arrivals and only 1.5% automatic picks were false alarms. The accuracy of the automated picker measured by standard error of the differ-ence between automatic and manual picks is 1.2 ms for P- and 1.0 ms for S-waves. AG 69 th Conference & xhibition London, UK, June 2007

2 Introduction: The growing use of seismic monitoring of hydraulic fracture treatments in hydrocarbon reservoirs brings the need for automation of the waveform processing. Besides faster measurement of the required parameters, the automated procedure provides also more objective results as it treats all waveforms in an equal and repeatable manner. In earthquake seismology, numerous techniques have been developed for automatic picking of wave arrival times. Some of these techniques are based on one-component P-wave pickers (e.g. Allen, 1978; Baer and Kradolfer, 1987). More advanced techniques use three-component P- and S-wave pickers that utilize the polarization properties of the seismic signal (e.g. Magotra et al., 1987; Jurkewicz, 1988; Roberts et al., 1989). Furthermore, seismic monitoring with arrays of receivers provides the opportunity to check consistency of signal detection across seismic network of receivers (Böðvarsson et al., 1999, Fischer, 2003; Kao and Shan, 2004; Drew et al., 2005 Gentili and Michelini, 2006). In this study we show an adjustment of the automated picking technique for P- and S-waves of Fischer (2003) for a linear array of receivers (in a vertical borehole) commonly used in hydrocarbon-reservoir microseismic monitoring. Automated measurement of arrival times and wave polarizations The proposed automated event and phase detection technique takes advantage of dominant character of direct S-waves in microseismic datasets and of array processing. Our approach attempts to mimic phase picking carried out by a human analyst who usually follows, among others, the following features of the three-component microseismic data: (i) waveforms recorded at local distances are dominated by direct S-waves, (ii) S-waves are polarized nearly orthogonally to P-waves, (iii) P- and S-wave arrival times must be mutually consistent on receivers of the array, and (iv) the S-P travel time difference is limited by a distance of induced events in a limited volume around the injection. 3 4 event 425.2; S-pick detail S-wave picker We are using polarization 6.9 analysis of threecomponent waveforms to 5 identify the peak of polarized energy that distin guishes the body-wave arrivals from seismic 22.9 noise, which tends to be uncorrelated between the 7 signal components. Thus, we calculate the signal covariance matrix S = σ ij in a moving time window and diagonalize it to get eigenvalues λ k and corresponding eigenvectors e k with k = The ratio of the three eigenvalues gives the shape of the polarization ellipsoid. One dominating time [ms] Fig. 1. xample of picking S-wave arrival of microseismic event recorded at a vertical array of receivers. The upper three traces of each receiver show the,, and components. The lower traces show the characteristic function λ H (magenta line), and the STA/LTA ratio (dash line). The magenta and black vertical abscissas indicate array-compatible maxima of λ H and the S-wave picks. The numbers indicate the maximum STA/LTA ratio at the pick. eigenvalue λ 1 indi- cates a linear polarization, whereas two dominating eigenvalues of comparable size point to elliptical polarization. In AG 69 th Conference & xhibition London, UK, June 2007

3 many datasets the S-wave polarization is mainly subhorizontal (e.g., Rutledge and Phillips, 2003). Thus, to detect an S-wave arrival it is advantageous to identify the maximum of the largest eigenvalue λ 1H in the horizontal plane (see also Fischer, 2003), which is rapidly evaluated by the equation (Magotra et al., 1987) 2 2 λ 1H = σ xx + σ yy + ( σ xx σ yy ) + 4( σ xy ) 2. In this study we use λ 1H (t) as a characteristic function λ(t) for S-wave detection. If one does not assume horizontal polarization of S-waves, the characteristic function for S-wave detection can be the maximum eigenvalue of the signal covariance matrix. The S-wave picking procedure illustrated in Fig. 1 consists of the following processing steps: (1) We calculate the characteristic function λ(t) for each receiver in a moving time window of the length corresponding to the S-wave duration, which acts as a low-pass filtering. (2) The vertical array geometry is employed to find the maxima of polarized energy arriving at consistent arrival time at each receiver. This is tested by varying the time shifts between the traces, similar to the beam-forming procedures (e.g. Blandford, 1974). Here, time shifts are derived from the traveltime differences calculated for a set of expected hypocenter depths and distances chosen in a grid spacing similar to that of the receiver offsets. S-wave groups are L t = λ t τ exceeds a fixed multiple of the identified if the maximum of the norm ( ) ( ) H median of L H, where τ j is the time shift of the j-th receiver. The dominant character of S- waves assures that the maximum of L H (t) in a given time interval of the length T SPmax identifies the S-wave arrival. We avoid misidentification of P-waves as S-waves of another event by setting T SPmax larger than the maximum expected S-P wave traveltime difference. (3) S-wave onsets t S are identified in a short-time window preceding the maxima of L H (t) (Swave group times) at the maxima of the short-time average over long-time average ratio (STA/LTA) applied to L H. For the sake of higher sensitivity nonoverlapping STA/LTA windows (Fischer, 2003) are used; to remove unreliable time picks a minimum required STA/LTA ratio is employed. (4) The onset time compatibility is further checked by fitting the hodochrone t S (z) by a parabolic curve. The outliers are repicked in the same way by identifying a more suitable local maximum of STA/LTA ratio that exceeds the required threshold. (5) The S-wave polarization vector e is measured in a short time window following the S- wave onset by diagonalizing the three-component signal covariance matrix and the S-wave slowness vector p is determined using the method of isner and Fischer (2006). P-wave picker In the second step P-waves are searched in the time interval T SPmax prior to each S-wave arrival. To detect P-waves we assume orthogonality of P and S-wave particle polarizations (i.e. colinea-rity of P- and S- rays), which is true in isotropic medium with constant ratio of P and S-wave velocities between source and receiver. This assumption is approximately satisfied for the microseismic datasets. P-wave groups are identified by search for the signal polarized in the S-ray direction, which is defined by the S-wave slowness vector p. We define the characteristic function c P for P-waves detection so that it increases with observed amplitude and maximizes when the signal s measured over a characteristic duration of the P-wave polarizes along the S-wave slowness vector p: c = s. p. s. s P The P-wave picking procedure illustrated in Fig. 2 consists of the following processing steps: (1) P-wave group times are measured at local maxima of c p detected within interval (t S T SPmax, t S ). j Hj j AG 69 th Conference & xhibition London, UK, June 2007

4 (2) The P-wave onsets t P are identified in a short-time window preceding P-wave group times by the STA/LTA algorithm, similarly to determining S-wave onsets. (3) The Wadati s relation t S tp = ts ( α β ) β, where α and β are the average P- and S-wave velocities (in the vicinity of the array), is used to remove outliers of the P-wave arrival time measurements. (4) The P-wave backazimuth is measured as the azimuth of the eigenvector corresponding to the largest eigenvalue of the signal covariance matrix in a short-time window following the P-wave onset time [ms] Fig. 2. xample of picking of P-wave arrival of microseismic event. The lower traces show the characteristic function c F (magenta line), and the STA/LTA ratio used for arrivaltime picking (dashed line). The P-wave picks are denoted by a full vertical abscissa; S-wave picks are also indicated. The rectangles show the P-wave search window defined by T SPmax = 60 ms. See Fig 1 for explanation of other symbols. Test application We have tested the automatic picker using several microseismic datasets of hydraulic fracture monitoring. Here we show the comparison of automatic and manual arrival- time measurements of waveform data recorded by a vertical array of eight threecomponent geophones during the hydraulic fracture treatment of a gas reservoir in the Canyonsand Formation in West Texas (isner et.al., 2006) with sampling rate of 1 khz. For this dataset we measured manually arrival times of 296 largest events amounting to 1753 P- and 1747 S-onset times. To evaluate the efficiency of the automated picker we measure the time difference between automatic and manual picks. We classify the automatic picks to the correct picks and false alarms by the time difference from the manual pick of 10 ms. From the total 3123 automatic picks only 1.7% were false alarms. The ratio of the number of correct automatic picks and the number of manual picks was 82.4% for P- and 93.2% for S-waves with values at individual receivers ranging from 63% to 91% for P- and from 90% to 96% for S-waves. The arrival-time differences show narrow Gaussian-like distribution with the mean values and standard errors of 1.0 ±1.2 ms for P- and 1.9 ±1.0 ms for S-waves. The negative time difference results from different picking method, whereas the automatic picker attempts to measure the time of wave onset, the human interpreter has measured the time of the first extreme. Conclusions We have developed and applied a robust automated algorithm for measurement of P- and S- wave arrival times and polarizations using waveforms recorded by a linear array of receivers. This technique employs the dominating character of direct S-waves in microseismic datasets. S-waves are detected by searching for local maxima of polarized energy arriving at a consistent vertical slowness to the receiver array. P-waves are identified prior to S-waves at the maxima of energy which is polarized in the S-ray direction. The algorithm has been successfully tested on a real dataset by comparison with manually picked arrival times. AG 69 th Conference & xhibition London, UK, June 2007

5 Acknowledgements We thank Dominion xploration Company for releasing the dataset used in this study. We also thank the uropean Union for funding the IMAGS Transfer of Knowledge project (MTKI-CT ). Fig. 3. Distribution of time differences between automatically and manually obtained arrival times of test dataset. Vertical axis shows the number of time measurements in bins of 1 ms width. References: Allen, R. [1978] Automatic earthquake recognition and timing from single traces. Bulletin of the Seismological Society of America 68, Baer M. and Kradolfer U. [1987] An automatic phase picker for local and teleseismic events. Bulletin of the Seismological Society of America 77, Blanford, R.R. [1974] An automatic event detector at the Tonto Forest Seismic Observatory. Geophysics 39 (5), Böðvarsson, R., S.Th. Rögnvaldsson, R. Slunga and. Kjartansson [1999] The SIL data acquisition system - at present and beyond year Physics of arth and Planetary Interior 113, Drew, J., Leslie, D., Armstrong, P. and G. Michaud [2005] Automated microseismic event detection and location by continuous spatial mapping. SP isner L. and T. Fischer [2007] Method of monitoring microseismic events, UK patent, application number GB isner L., Fischer T., and J. Le Calvez [2006] Detection of repeated hydraulic fracturing (out-of-zone growth) by microseismic monitoring, The Leading dge 25 (5), Fischer T. [2003] Automatic location of swarm earthquakes from local network data. Studia geophysica et geodaetica 47, Gentili, S., and A. Michelini [2006] Automatic picking of P and S phases using a neural tree. Journal of Seismology 10 (1), Jurkewicz, A. [1988] Polarization analysis of three-component array data. Bulletin of the Seismological Society of America 78 (5), Kao, H., and S. Shan [2004] The Source-Scanning Algorithm: mapping the distribution of seismic sources in time and space. Geophysical Journal International 157, Magotra., Ahmed. and. Chael [1987] Seismic event detection and source location using singlestation (three-component) data. Bulletin of Seismological Society of America 77, Roberts, R.G., Christoffersson, A., and F. Cassidy [1989] Real time event detection, phase identification and source location estimation using single-station three-component seismic data, Geophysical Journal International 97, Rutledge, J.T. and W.S. Phillips [2003] Hydraulic stimulation of natural fractures as revealed by induced microearthquakes, Carthage Cotton Valley gas field, east Texas. Geophysics, 68, AG 69 th Conference & xhibition London, UK, June 2007