Steven R. Taylor

Amplitude corrections for regional seismic discriminants

Abstract — A fundamental problem associated with event identification lies in deriving corrections that remove path and earthquake source effects on regional phase amplitudes used to construct discriminants. Our goal is to derive a set of physically based corrections that are independent of magnitude and distance, and amenable to multivariate discrimination by extending the technique described in Taylor and Hartse (1998). For a given station and source region, a number of well-recorded earthquakes is used to estimate source and path corrections. The source model assumes a simple Brune (1970) earthquake source that has been extended to handle non-constant stress drop. The discrimination power in using corrected amplitudes lies in the assumption that the earthquake model will provide a poor fit to the signals from an explosion. The propagation model consists of a frequency-independent geometrical spreading and frequency-dependent power law Q. A grid search is performed simultaneously at each station for all recorded regional phases over stress-drop, geometrical spreading, and frequency-dependent Q to find a suite of good-fitting models that remove the dependence on mb and distance. Seismic moments can either be set to pre-determined values or estimated through inversion and are tied to mb through two additional coefficients. We also solve for frequency-dependent site/phase excitation terms. Once a set of corrections is derived, effects of source scaling and distance as a function of frequency are applied to amplitudes from new events prior to forming discrimination ratios. Thus, all the corrections are tied to just mb (or M0) and distance and can be applied very rapidly in an operational setting. Moreover, phase amplitude residuals as a function of frequency can be spatially interpolated (e.g., using kriging) and used to construct a correction surface for each phase and frequency. The spatial corrections from the correction surfaces can then be applied to the corrected amplitudes based only on the event location. The correction parameters and correction surfaces can be developed offline and entered into an online database for pipeline processing providing multivariate-normal corrected amplitudes for event identification. Examples are shown using events from western China recorded at the station MAKZ.

1 Hz Lg Q tomography in central Asia

We have applied tomographic techniques to a data set of over 1700, regional distance, Lg amplitudes from 12 stations for paths inside a 30° by 40° region of Central Asia. Our purpose is to create high resolution path correction maps for use in regional distance source discrimination and magnitude determination, as well as to study Lg propagation characteristics in one of the most heterogeneous portions of the earth. For 1.0 Hz data, tomography reduced variance 33%, relative to a best fit, uniform Q model. LgQ varied with geologic region, with low values in Tibet, the Tien Shan and the Pamir range (Q=200 to 400), intermediate values in basins such as the Tarim (Q=500) and high values for platforms and older crust (Q>800). The relatively efficient propagation across the Tarim differs from the poor propagation that is generally observed across sedimentary basins.

A procedure for estimation of source and propagation amplitude corrections for regional seismic discriminants

We outline a procedure for the estimation of frequency-dependent source and propagation amplitude corrections for regional seismic discriminants (source path amplitude correction (SPAC)). For a given station and phase a number of well-recorded earthquakes are inverted for source and path corrections. The method assumes a simple Brune [1970] earthquake-source model and a simple propagation model consisting of a frequency-independent geometrical spreading and frequency-dependent power law Q. The inverted low-frequency levels are then regressed against mb to derive a set of corrections that are a function of mb and distance. Once a set of corrections is derived, effects of source scaling and distance as a function of frequency are applied to amplitudes from new events prior to forming discrimination ratios. The resulting discriminants are normally distributed and amenable to multivariate feature selection, classification, and outlier techniques. To date, most discrimination studies have removed distance corrections once a particular amplitude ratio is formed (distance corrected ratio (DCR)). DCR generally works well for phase ratios taken in a particular frequency band. However, when different frequency bands are combined (for phase spectral ratios or cross spectral ratios), significant source-scaling effects (e.g., corner-frequency scaling) can remain, causing the discriminants to vary as a function of event size and to be nonnormally distributed. It is then often necessary to construct nonphysical transformations in an attempt to make the discriminants multivariate normal. The SPAC technique can be used to construct discriminants that are multivariate normal by using simple physical seismic source and propagation models. Moreover, phase amplitude residuals as a function of frequency can be spatially averaged and used as additional path-specific corrections to correct for additional propagation effects such as phase blockages.

Path correction using interpolated amplitude residuals: An example from central China

To isolate effects of the laterally varying lithosphere on regional seismograms, we geographically smoothed source- and distance-corrected amplitudes from 853 events recorded at station LZH in central China. This procedure was applied to P n , P g , S n and L g phases and various coda windows for eight bands between 0.5 and 8 Hz. We assume smoothing reduces effects such as source radiation, leaving an estimate of the path effect that can be used to correct data. A cross-validation procedure showed that path corrections performed best with low-frequency L g data, reducing variance up to 39% for 1 Hz L g , and up to 57% for 0.75 Hz P/L g ratios, compared to correction for distance alone. These corrections reduce scatter in magnitude estimates and discriminant ratios without needing to know detailed geological structure and will be important for effective test-ban monitoring.

Bayesian Lg Attenuation Tomography Applied to Eastern Asia

Bayesian attenuation tomography (Tarantola, 1987) is being used to refine existing Lg attenuation models in eastern Asia. The advantages of a Bayesian approach to tomography are that large-scale and high-resolution tomographic models available from other well-accepted studies can be used as prior background models. The resulting refined tomographic model will blend into these prior background models. Moreover, the uncertainties are well established in a Bayesian framework. We assume a general linear Gaussian (least squares) model, where the covariance matrix is partitioned into data and prior model components. Uncertain data are naturally down weighted and model components with small errors will be subject to little change. Amplitude tomography provides only an approximation to the propagation effects a seismic wave may experience. For example, phase blockage can occur over relatively short distances in the absence of any anelastic effects. Station-centric kriged amplitude correction surfaces on top of the tomographic models may assist in identifying blocked paths. Because a signal-to-noise criterion is used to select amplitudes for the inversion, the data are left-censored and the resulting Q 0 models will be biased high. We examine the utility of a maximum-likelihood data augmentation method to the left-censored data problem (Schafer, 1997). In this case, we can only measure an upper bound to measured amplitudes on the basis of prephase noise. Data augmentation is used to impute the missing data values with their conditional expectation based on the relationship of amplitudes with other completely observed variables. We have initially chosen a bivariate Gaussian model for filling in missing amplitudes based on the relationship between amplitude and prephase noise. We apply the technique to eastern Asia for Lg signals at 1 Hz using data from 1651 earthquakes recorded at 12 stations. Tomographic patterns correlate well with those expected from geophysical considerations (e.g., high attenuation in Tibet and low attenuation up into the stable regions of Kazakhstan). Many of the large cratonic basins surrounding the Tibet Plateau (e.g., Tarim, Junggar) show reduced attenuation consistent with the hypothesis that they represent regions of stronger lithosphere. Data augmentation tends to increase attenuation in Tibet (particularly western Tibet) and the cratonic basins to the North of but not to the East of Tibet. In addition, the resulting models are much smoother than models that do not account for censoring.

Forensic seismology and the sinking of the Kursk

On August 10, 2000, Russias Northern Fleet began its largest naval exercise in more than a decade. Among the vessels taking part was the heavily-armed Kursk, an Oscar class submarine that was the most modern cruise-missile sub in the fleet. Beginning on August 14, a series of reports in the press indicated that the Kursk had been severely damaged during the exercise and that the crew were likely dead. By August 17, news agencies were reporting that seismic networks in the Baltic area had detected two seismic events which appeared to correspond to the Kursk disaster in time and space (Figure 1). Specifically the seismic events were consistent with reports from the British Broadcasting Corporation on the location of ongoing rescue efforts. The fact that this section of the Barents Sea is essentially aseismic added credence to the assertion that the seismic events were directly related to the sinking of the Kursk.

An evaluation of generalized likelihood Ratio Outlier Detection to identification of seismic events in Western China

The Generalized Likelihood Ratio Outlier Detection Technique for seismic event identification is evaluated using synthetic test data and frequency-dependent P{sub g}/L{sub g} measurements from western China. For most seismic stations that are to be part of the proposed International Monitoring System for the Comprehensive Test Ban Treaty, there will be few or no nuclear explosions in the magnitude range of interest (e.g. M{sub b} < 4) on which to base an event-identification system using traditional classification techniques. Outlier detection is a reasonable alternative approach to the seismic discrimination problem when no calibration explosions are available. Distance-corrected P{sub g}/L{sub g} data in seven different frequency bands ranging from 0.5 to 8 Hz from the Chinese Digital Seismic Station WMQ are used to evaluate the technique. The data are collected from 157 known earthquakes, 215 unknown events (presumed earthquakes and possibly some industrial explosions), and 18 known nuclear explosions (1 from the Chinese Lop Nor test site and 17 from the East Kazakh test site). A feature selection technique is used to find the best combination of discriminants to use for outlier detection. Good discrimination performance is found by combining a low-frequency (0.5 to 1 Hz) P{sub g}/L{sub g} ratio with high-frequency ratios (e.g. 2 to 4 and 4 to 8 Hz). Although the low-frequency ratio does not discriminate between earthquakes and nuclear explosions well by itself, it can be effectively combined with the high-frequency discriminants. Based on the tests with real and synthetic data, the outlier detection technique appears to be an effective approach to seismic monitoring in uncalibrated regions.

Application of regional phase amplitude tomography to seismic verification

Abstract — We have applied tomographic techniques to amplitude data to quantify regional phase path effects for use in source discrimination studies. Tomography complements interpolation (kriging) methods by extending our ability to apply path corrections into regions devoid of calibration events, as well as raising levels of confidence in the corrections because of their more physical basis. Our tomography technique solves for resolvable combinations of attenuation, source-generation, site and spreading terms. First difference regularization is used to remove singularities and reduce noise effects.¶In initial tests the technique was applied to a data set of 1488, 1.0 Hz, Pg/Lg amplitude ratios from 13 stations for paths inside a 30° by 40° box covering western China and surrounding regions. Tomography reduced variance 60%, relative to the power-law distance correction traditionally applied to amplitude ratios. Relative Pg/Lg attenuation varied with geologic region, with low values in Tibet, intermediate values in basins and high values for platforms and older crust. Spatial patterns were consistent with previous path effect studies in Asia, especially local earthquake coda-Q. Relative spreading was consistent with expected values for Pg and Lg. Relative site terms were similar to one another, yet some tradeoff with attenuation was evident.¶Tomography residuals followed systematic trends with distance, which may result from the evolution from direct to coda phases, focusing, model tradeoff or data windowing effects. Examination of the residuals using a kriging interpolator showed coherent geographical variations, indicating unmodeled path effects. The residual patterns often follow geological boundaries, which could result from attenuating zones or minor blockages that are too thin to be resolved, or that have anisotropic effect on regional phases. These results will guide efforts to reparameterize tomography models to more effectively represent regional wave attenuation and blockage. The interpolated residuals also can be combined with predictions of the tomographic model to account for path effects in discrimination studies on a station by station basis.

Shallow Velocity Structure at the Shagan River Test Site in Kazakhstan

Abstract — During 1997 and 1998, twelve chemical explosions were detonated in boreholes at the former Soviet nuclear test site near the Shagan River (STS) in Kazakhstan. The depths of these explosions ranged from 2.5 to 550 m, while the explosive yield varied from 2 to 25 tons. The purpose of these explosions was for closure of the unused boreholes at STS, and each explosion was recorded at local distances by a network of seismometers operated by Los Alamos National Laboratory and the Institute of Geophysics for the National Nuclear Center (NNC). Short-period, fundamental-mode Rayleigh waves (Rg) were generated by these explosions and recorded at the local stations, resultingly the waves exhibited normal dispersion between 0.2 and 3 seconds. Dispersion curves were generated for each propagation path using the Multiple Filter Analysis and Phase Match Filtering techniques. Tomographic maps of Rg group velocity were constructed and show a zone of relatively high velocities for the southwestern (SW) region of the test site and slow propagation for the northeastern (NE) region. For 0.5 sec Rg, the regions are separated by the 2.1 km/sec contour, as propagation in the SW is greater than 2.1 km/sec and less in the NE region. At 1.0 sec period, the 2.3 km/sec contour separates the two regions. Finally, for 1.5 and 2.0 sec, the separation between the two regions is less distinct as velocities in the NE section begin to approach the SW except for a low velocity region (<2.1 km/sec) near the center of the test site. Local geologic structure may explain the different regions as the SW region is composed predominantly of crystalline intrusive rocks, while the NE region consists of alluvium, tuff deposits, and Paleozoic sedimentary rocks. Low velocities are also observed along the Shagan River as it passes through the SW region of the test site for shorter period Rg (0.5–1.0 sec). Iterative, least-squares inversions of the Rg group velocity dispersion curves show shear-wave velocities for the southwestern section that are on average 0.4 km/sec higher than the NE region. At depths greater than 1.5 km the statistical difference between the models is no longer significant. The observed group velocities and different velocity structures correlate with P-wave complexity and with spatial patterns of magnitude residuals observed from nuclear explosions at STS, and may help to evaluate the mechanisms behind those observations.

Multivariate acoustic detection of small explosions using Fisher's combined probability test

A methodology for the combined acoustic detection and discrimination of explosions, which uses three discriminants, is developed for the purpose of identifying weak explosion signals embedded in complex background noise. By utilizing physical models for simple explosions that are formulated as statistical hypothesis tests, the detection/discrimination approach does not require a model for the background noise, which can be highly complex and variable in practice. Fishers Combined Probability Test is used to combine the p-values from all multivariate discriminants. This framework is applied to acoustic data from a 400 g explosion conducted at Los Alamos National Laboratory.