Backazimuth Estimates Research Articles

Estimating seismometer component orientation of the Brazilian seismographic network using teleseismic P-wave particle motion analysis and directional statistics

Seismometer component orientation is one of the most critical parameters for seismology, especially for methods that take wave field polarization effects into account, such as receiver functions, seismic anisotropy, surface wave dispersion and seismic tomography. Using P-wave particle motion combined with directional statistics, we estimated the orientation for 90 stations of the Brazilian Seismographic Network (RSBR), corresponding to 105 seismometers. The difference between the number of stations (90) and seismometers (105) is due to instrument replacement during station operation. We observed that 81 seismometers have an orientation of ±5° and 19 seismometers greater than ±5°. Only one seismometer has an orientation greater than 90° (BOAV station). ON and BR are the subnetworks of RSBR with the largest number of seismometers with orientation errors bigger than ±10° (6 for each one) and NB is the only subnetwork with no misoriented station. All the estimated orientations were close to the measured in the field, which is evidence that our approach is more accurate to detect misoriented stations than previous studies that analyzed the orientation of RSBR stations. The Watson's U2 statistical test indicated that the back azimuth deviations for Precambrian Basement and Phanerozoic Cover were significantly different, suggesting that the regional geology could affect back azimuth estimates.

Direction Estimates for Short-Period P-Waves on Three-Component Stations and Arrays

Abstract P-arrival backazimuth estimates can be crucial in locating poorly constrained seismic events. Correlating short windows of the vertical waveform with corresponding windows of the radial rotation for different backazimuths can provide estimates, but these are often uncertain and biased due to skewness in the Z–R correlation functions. Assessing how well cosine curves centered on different backazimuths match the Z–R correlation functions provides more reliable estimates that depend less upon the time-window used. Stacking best-fit-cosine curves from neighboring three-component stations improves stability further in a form of array-processing that does not require coherence between the waveforms themselves. We demonstrate for recordings of North Korean nuclear tests at the Pilbara 3C array in Australia that the biases in the Z–R correlation functions vary greatly between adjacent stations. This bias is reduced both by the cosine curve fitting and stacking operations. We advocate obtaining backazimuth estimates for all P arrivals at three-component stations globally. This could improve phase association and event location, identify sensor orientation problems, and provide baseline backazimuth corrections and uncertainty estimates. We propose two benchmark datasets for developing, documenting, and comparing backazimuth estimation algorithms and codes. All the data and code used to generate the results presented here are open.

Back-Azimuth Estimation of Air-to-Ground Coupled Infrasound from Transverse Coherence Minimization

AbstractWe present the transverse coherence minimization method (TCM)—an approach to estimate the back-azimuth of infrasound signals that are recorded on an infrasound microphone and a colocated three-component seismometer. Accurate back-azimuth information is important for a variety of monitoring efforts, but it is currently only available for infrasound arrays and for seismoacoustic sensor pairs separated by 10 s of meters. Our TCM method allows for the analysis of colocated sensor pairs, sensors located within a few meters of each other, which may extend the capabilities of existing seismoacoustic networks and supplement operating infrasound arrays. This approach minimizes the coherence of the transverse component of seismic displacement with the infrasound wave to estimate the infrasound back-azimuth. After developing an analytical model, we investigate seismoacoustic signals from the August 2012 Humming Roadrunner experiment and the 26 May 2021 eruption of Great Sitkin Volcano, Alaska, U.S.A., at the ranges of 6.5–185 km from the source. We discuss back-azimuth estimates and potential sources of deviation (1°–15°), such as local terrain effects or deviation from common analytical models. This practical method complements existing seismoacoustic tools and may be suitable for routine application to signals of interest.

ArrayNet: A Combined Seismic Phase Classification and Back-Azimuth Regression Neural Network for Array Processing Pipelines

ABSTRACT Array processing is an integral part of automatic seismic event detection pipelines for measuring apparent velocity and backazimuth of seismic arrivals. Both quantities are usually measured under the plane-wave assumption, and are essential to classify the phase type and to determine the direction toward the event epicenter. However, structural inhomogeneities can lead to deviations from the plane-wave model, which must be taken into account for phase classification and back-azimuth estimation. We suggest a combined classification and regression neural network, which we call ArrayNet, to determine the phase type and backazimuth directly from the arrival-time differences between all combinations of stations of a given seismic array without assuming a plane-wave model. ArrayNet is trained using regional P- and S-wave arrivals of over 30,000 seismic events from reviewed regional bulletins in northern Europe from the past three decades. ArrayNet models are generated and trained for each of the ARCES, FINES, and SPITS seismic arrays. We observe excellent performance for the seismic phase classification (up to 99% accuracy), and the derived back-azimuth residuals are significantly improved in comparison with traditional array processing results using the plane-wave assumption. The SPITS array in Svalbard exhibits particular issues when it comes to array processing in the form of high apparent seismic velocities and a multitude of frost quake signals inside the array, and we show how our new approach better handles these obstacles. Furthermore, we demonstrate the performance of ArrayNet on 20 months of continuous phase detections from the ARCES array and investigate the results for a selection of regional seismic events of interest. Our results demonstrate that automatic event detection at seismic arrays can be further enhanced using a machine learning approach that takes advantage of the unique array data recorded at these stations.

Locating the Largest Event Observed on Mars With Multi‐Orbit Surface Waves

AbstractPrior to the 2018 landing of the InSight mission, the InSight science team proposed locating Marsquakes using multiple orbit surface waves, independent of seismic velocity models, for events larger than MW4.6. The S1222a MW4.7 of 4 May 2022 is the largest Marsquake recorded and the first large enough for this method. Group arrivals of the first three orbits of Rayleigh waves are determined to derive the group velocity, epicentral distance, and origin time. The mean distance of 36.9 ± 0.3° agrees with the Marsquake Service (MQS) distance based on body wave measurements of 37.0 ± 1.6°. The origin time from surface waves is systematically later than the MQS origin time by 20 s. Backazimuth estimation is similar to body wave estimations from MQS although suggesting a shift to the south. Backazimuth estimates from R2 and R3 are more scattered, but do show clear elliptical motion.

Receiver orientation and event back-azimuth estimation for downhole microseismic monitoring using a probabilistic method based on P-wave polarization

Microseismic event back-azimuth is an indispensable parameter for source localization in downhole microseismic monitoring, and the accurate orientation of horizontal components of downhole seismic receivers is vital for reliably determining the event back-azimuth. Variation in the monitoring data quality may jeopardize the accuracy of receiver orientation which will further affect the event back-azimuth estimation. To mitigate this issue, we proposed a new probabilistic method based on P-wave polarization analysis for receiver orientation and event back-azimuth estimation. The algorithm constructs the von Mises distribution function using the polarization angle and corresponding rectilinearity of the P-wave, then determines the target angle using the maximum of the probability function. The receiver having the highest rectilinearity from the active-source event is used to quantify a reliable absolute orientation angle, and the relative orientation angles are calculated by the probability distributions based on the measurement angle differences and the associated averages of rectilinearity from all events. After receiver orientation, the P-wave polarization angles with different rectilinearity values are applied to construct the probability distribution functions to estimate the event back-azimuths. By using high-quality events and multi-receiver recordings, our methodology can greatly reduce the unintentional error in receiver orientation and increase event back-azimuth accuracy. We investigate the feasibility and reliability of the proposed method using both synthetic and field data. The synthetic data results demonstrate that, compared to the conventional methods, the proposed method can minimize the variance of the receiver orientation angle and back-azimuth estimation. The weighted standard deviation analysis demonstrates that the proposed method can reduce the orientation error and improve the event back-azimuth accuracy in the field dataset.

Master event based backazimuth estimation and its application to downhole microseismic monitoring

Microseismic monitoring provides a valuable tool for evaluating the effectiveness of hydraulic fracturing operations. However, robust detection and accurate location of microseismic events are challenging due to the low signal to noise ratio (SNR) of their signals on seismograms. In a downhole monitoring setting, P-wave polarization direction measured from 3-component records is usually considered as the backazimuth of the microseismic event, i.e., the direction of the event. The direction and arrival time difference between the P and S waves is used to locate the seismic event. When SNR is low, an accurate estimate of event backazimuth becomes very challenging with the traditional covariance matrix method. Here we propose to employ a master event and use a grid search method to find the backazimuth of a target event that maximizes the dot product of the two backazimuthal vectors of the master and target events. We compared the backazimuths measured with the proposed grid-search and the conventional covariance-matrix methods using a large synthetic dataset. We found that the grid-search method yields more accurate backazimuth estimates from low SNR records when measurements are made at single geophone level. When array data are combined, the proposed method also has some advantage over the covariance-matrix method, especially when the number of geophones is low. We also applied the method to a microseismic dataset acquired by a hydraulic fracturing project at a shale play site in southwestern China and found that the relocated microseismic events tend to align along existing faults more tightly than those in the original catalog.

Method for Estimating the Tripartite Array Back-Azimuth Error Caused by the Far-Field Approximation

Abstract In recent decades, tripartite arrays (i.e., three-element array) have become an important tool for various seismoacoustic applications, mainly due to their superior back-azimuth estimation. However, the back azimuth is estimated assuming the far-field approximation. Lately, tripartite arrays have been used to monitor microseismicity and aftershocks at distances comparable with the size of the array in which the far-field assumption might not hold. In this work, we determined the validity of the far-field assumption by analyzing the plane-wave errors, that is, the errors of the back azimuth and slowness computations caused by the plane-wave assumption. Computational formulas for estimating the absolute errors, due to the plane-wave assumption, were developed. Several case studies demonstrated that the plane-wave errors are not theoretical issues only and taking them into account can improve the results of field measurements. Proposed practical methods to account for plane-wave errors can improve the performance of arrays aimed to measure low magnitude events such as in induced microseismicity monitoring or on-site inspection.

Low-Frequency Marsquakes and Where to Find Them: Back Azimuth Determination Using a Polarization Analysis Approach

ABSTRACT National Aeronautics and Space Administration’s Interior Exploration using Seismic Investigations, Geodesy, and Heat Transport (InSight) mission on Mars continues to record seismic data over 3 yr after landing, and over a thousand marsquakes have been identified, to date. With only a single seismic station, the determination of the epicentral location is far more challenging than on the Earth. The Marsquake Service (MQS) produces seismicity catalogs from data collected by InSight, and provides distance and back azimuth estimates when these can be reliably determined; when both are available, these are combined to provide a location. Currently, MQS does not assign a back azimuth to the vast majority of marsquakes. In this work we develop and apply a polarization analysis method to determine the back azimuth of seismic events from the polarization of observed P- and S-wave arrivals. The method is first applied to synthetic marsquakes and then calibrated using a set of well-located earthquakes that have been recorded in Tennant Creek, Australia. We find that the back azimuth is estimated reliably using our polarization method. The same approach is then used for a set of high-quality marsquakes recorded up to October 2021. We are able to estimate back azimuths for 24 marsquakes, 16 of these without MQS back azimuths. We locate most events to the east of InSight, in the general region of Cerberus Fossae.

Evaluating the location capabilities of a regional infrasonic network in Utah, US, using both ray tracing-derived and empirical-derived celerity-range and backazimuth models

SUMMARYMore realistic models for infrasound signal propagation across a region can be used to improve the precision and accuracy of spatial and temporal source localization estimates. Motivated by incomplete infrasound event bulletins in the Western US, the location capabilities of a regional infrasonic network of stations located between 84–458 km from the Utah Test and Training Range, Utah, USA, is assessed using a series of near-surface explosive events with complementary ground truth (GT) information. Signal arrival times and backazimuth estimates are determined with an automatic F-statistic based signal detector and manually refined by an analyst. This study represents the first application of three distinct celerity-range and backazimuth models to an extensive suite of realistic signal detections for event location purposes. A singular celerity and backazimuth deviation model was previously constructed using ray tracing analysis based on an extensive archive of historical atmospheric specifications and is applied within this study to test location capabilities. Similarly, a set of multivariate, season and location specific models for celerity and backazimuth are compared to an empirical model that depends on the observations across the infrasound network and the GT events, which accounts for atmospheric propagation variations from source to receiver. Discrepancies between observed and predicted signal celerities result in locations with poor accuracy. Application of the empirical model improves both spatial localization precision and accuracy; all but one location estimates retain the true GT location within the 90 per cent confidence bounds. Average mislocation of the events is 15.49 km and average 90 per cent error ellipse areas are 4141 km2. The empirical model additionally reduces origin time residuals; origin time residuals from the other location models are in excess of 160 s while residuals produced with the empirical model are within 30 s of the true origin time. We demonstrate that event location accuracy is driven by a combination of signal propagation model and the azimuthal gap of detecting stations. A direct relationship between mislocation, error ellipse area and increased station azimuthal gaps indicate that for sparse networks, detection backazimuths may drive location biases over traveltime estimates.

Separation and location of microseism sources

[1] Microseisms are ground vibrations caused largely by ocean gravity waves. Multiple spatially separate noise sources may be coincidentally active. A method for source separation and individual wavefield retrieval of microseisms using a single pair of seismic stations is introduced, and a method of back azimuth estimation assuming Rayleigh-wave arrivals of microseisms is described. These methods are combined to separate and locate sources of microseisms in a synthetic model and then applied to field microseismic recordings from Ireland in the Northeast Atlantic. It is shown that source separation is an important step prior to location for both accurate microseism locations and microseisms wavefield studies.

Improvement of back-azimuth estimation in real-time by using a single station record

We propose a new approach to improve the accuracy of the back-azimuth estimation in real-time by using a single station record. Compared with the conventional approach in which the length of the time window for the analysis is fixed for all data, the accuracy and speed of the estimation are drastically improved by introducing a variable-length time window which is determined by the first half cycle of the wavelength of the initial P -wave, because this window tends to reduce the influences of trailing scattered waves. The analysis, using the K-NET dataset, shows that the estimation, using this new approach, is improved by 28% and 0.25 s in accuracy and speed, respectively.

<b>New Method for Estimating Earthquake Parameters for Earthquake Early Warning</b>

In 2004-2005 the Railway Technical Research Institute introduced a new earthquake early warning (EEW) for Shinkansen. This EEW system rapidly estimates earthquake parameters using data from a single station. Epicenter locations in this system are estimated utilizing the principal component analysis (PCA) and the B-Δ method. In order to improve the accuracy and speed of the single station method, this paper proposes two new methods to improve the performance of epicenter estimation. The first method introduces a variable-length time window instead of the conventional fixed-length time window to improve the PCA. A second method is proposed to improve the speed of the B- Δ method by using the time series characteristics of coefficients A and B. The proposed new methods improved the accuracy by 28% and the speed by 0.25 seconds of the back-azimuth estimation. Speed of estimation of the epicentral distance was improved by approximately 1.32 seconds, or 66% compared to the conventional method. These finding indicate that the new methods could produce faster and more robust earthquake warnings.

An Automated Short-Period Surface-Wave Detection Algorithm

We describe an automated short-period Rayleigh wave ( Rg ) detector designed to work on local (<2.5° epicentral distance) events recorded at three- component stations. The detector was modeled after an automatic 17- to 22-sec Rayleigh-wave detection method; however, we have modified the algorithms for local distance and short-period applications. We have tested the detector on a well-located cluster of mining events from central India and on a set of ground-truth events in the area. The Rg detector was also integrated into a semiautomatic event detection and location algorithm and applied on continuous data. Fourier and wavelet-based methods are evaluated for prefiltering. We observe that sample standard deviations of backazimuth estimates using the Rg detector, after wavelet prefiltering, are comparable to fk 3C P backazimuth estimates from event clusters. Our results indicate that using the Rg -phase backazimuths for event location is a promising alternative to using small signal-to-noise ratio first-arrival backazimuths. We recommend wavelet prefiltering versus Fourier prefiltering because it is more consistent for the detection of low signal-to-noise ratio events at local distances.

Backazimuth estimation reliability using surface wave polarization

Backazimuth estimates derived from surface wave polarization can be used to test association of an observed surface wave with a seismic event. However, commonly used techniques for estimating the backazimuth are subject to a number of errors and are often inaccurate. We evaluate the performance of a new algorithm for determining Rayleigh wave propagation direction from the cross‐correlation of the horizontal and Hilbert transformed vertical seismograms. This proves to be more accurate and much less error‐prone than current automatic processing procedures, and the cross‐correlation value provides a good measure of the reliability of the backazimuth estimate. Consequently, the new algorithm will permit more reliable event association at a lower magnitude threshold.

Small-Aperture Seismo-Acoustic Arrays: Design, Implementation, and Utilization

A four-element, 1-km-aperture seismo-acoustic array has been designed and installed northeast of Seoul, Korea. Each element of the array consists of a GS-13 vertical seismometer (1 Hz) in a shallow borehole (∼10 m) and a low-frequency acoustic gauge connected to an 11-element hose array (7.6-m hoses) at the surface. The array is being used to assess the importance of colocated seismic and acoustic sensors for the purposes of (1) quantifying wind as a source of seismic and acoustic noise, (2) constraining propagation path effects in the atmosphere and solid earth, (3) locating the sources of the waves, and (4) characterizing the source type. Seismic noise estimates illustrate a level that is only slightly above the low-noise model on average. Acoustic noise levels resolve the microbaroms during low-noise times but document a nearly 50-dB increase in noise during the windiest periods. Infrasonic noise in the 0.01- to 5-Hz band increases rapidly with wind velocity. The seismic noise shows little or no dependence on wind velocity. Analysis of the data from a 2-month time period suggests that there are many more acoustic signals than seismic (4–10 times as many). Approximately 1/4 of all seismic signals are associated with an acoustic arrival. The vast majority of seismo-acoustic observations come from sources in the 30- to 200-km range and occur during working hours, local time. The 30- to 200-km observation distance is surprising in that average atmospheric velocity models predict no acoustic returns in this range. Average atmospheric models modified by meteorological data for the troposphere indicate the possibility of ducting in the troposphere as an explanation for these arrivals. Event location is based upon regional seismic phase identification ( P n, P g, P m P, L g, R g) using the array and backazimuth estimates from both the seismic and acoustic data. Many of the infrasound signals have good signal-to-noise ratios from 1 to beyond 4 Hz. Despite the small size of the array, event clusters are identified at regional distances. Events associated with acoustic signals are presumed to be from mining regions. The existence of Rg arrivals and dominance of P energy at high frequency are consistent with this interpretation. Ground truth in the form of in-mine observations has validated that two of the clusters come from construction and mine blasts.

Contribution of Dense Array Analysis to the Identification and Quantification of Basin-Edge-Induced Waves, Part I: Methodology

Recent earthquakes have shown that edge-generated surface waves can significantly contribute to increased damages. Most observations of edge-generated surface waves are concerning long-period surface waves propagating in large-size valleys. Since travel times of such waves between valley edges can reach several tens of seconds, they are quite easy to isolate. In small-size structures, reverberating wave trains are mixed and very dense array analysis is required for the identification of basin-induced surface wave trains. The city of Grenoble (French Alps) is located in a small-size deep alluvial valley and faces important site effects (Lebrun et al. , 2001). In order to identify and quantify multidimensional site effects in this basin, a very dense array of 29 three-component seismometers over a 1-km aperture was installed within the city. The wave-field complexity as well as the in situ noise characteristics (colored/correlated noise and low signal-to-noise ratio) led us to develop a procedure based on time-frequency coherence of signal waveforms and the multiple signal classification (MUSIC) (Schmidt, 1981) algorithm to identify the main energetic contributions crossing the array. Next, the nature and energy of waves were estimated using some properties of the analytical three-component covariance matrix. Careful methodological investigations were performed in order to better understand and quantify the effects of site constraints on the estimation of wave parameters with the MUSIC technique. Simulations outline the ability of array antennas first to handle difficult scenarios involving multiple, nonstationnary, and correlated propagating phases and second to estimate the polarization and energy of waves. The velocity estimation is shown to be much more unstable than backazimuth estimation, and a low signal-to-noise ratio introduces some variation in estimates. Finally, considering the very large number of identified waves, a statistical view of final estimates is suggested for improving the reliability of analysis. In an accompanying article (Cornou et al. , 2003), we use this method to investigate the entire wave field of seismic events recorded by the array in order to isolate basin-induced waves. Manuscript received 21 June 2002.

Elastic neural net for the earthquake epicenter search

Automatic processing of seismic data is today a key element in the efforts to achieve high quality seismic systems. Automated procedures for locating seismic events with a network including arrays and single element seismometers usually incorporate back‐azimuth estimates, arrival‐time data, and associated uncertainties into a least‐squares‐inverse location algorithm. Such an algorithm is quite cumbersome and requires expanding a set of non‐linear equations in a Taylor series. Second‐order terms usually not included in the algorithm can be important if the initial estimate is far from the solution.We propose to use elastic neural nets (ENN) to find the initial estimation in automated procedures of locating seismic events and discuss the results for simulated seismic events. The advantages of ENN are the simplicity of the algorithm, the fast convergence and the high efficiency.

Azimuth estimation capabilities of the ARCESS regional seismic array

Abstract The effect of local geology and noise conditions on the performance of a small regional array is investigated by comparing the regional Pn backazimuth estimation capabilities of the ARCESS array in northern Norway to the NORESS array. A broadband frequency-wavenumber estimator was used to calculate backazimuths from the Pn arrival for each of 203 regional events recorded at ARCESS while varying element spacing, frequency band, and time window. Most of the errors in backazimuth are less than 20° when appropriate parameter combinations are used, and mean backazimuth errors are close to zero. The best results are obtained using a 13-element configuration that has a 1.4 km aperture and a maximum station spacing of about 600 m. With the 13-element configuration and the data filtered to include frequencies between 3 and 10 Hz, the mean errors for the 203 event data set are less than 0.9°, and S.D. are as small as 16.9°. There are differences seen in the backazimuth estimation capabilities of ARCESS and NORESS with specific parameter combinations. The larger aperture configurations (10- and 17-elements) have smaller means at ARCESS, although the precision is about the same. The estimates using unfiltered data at ARCESS are poor, because of local noise conditions that increase the level of background noise at low frequencies. Overall the precision is better at NORESS, but both regional arrays have the best results using the 13-element configuration and filtering the data in the middle frequency range (3 to 10 Hz). Other factors investigated include SNR and source region. Backazimuth estimation statistics improve if only events with 5 dB of SNR are included in the data set at both ARCESS and NORESS. The mean errors move closer to zero and standard deviations decrease. The differences between the two arrays are not as pronounced. There are some path effects from different source regions around the ARCESS array. However, combinations of small aperture configurations and middle (3 to 10 Hz) frequency bands work well for events over the entire distance range of 30 to 1200 km. ARCESS and NORESS have similar backazimuth estimation capabilities even though there are differences in the local geology and noise conditions. Because a 13-element configuration produces reliable results for both arrays, it would be reasonable to reduce the number of elements in a regional array. This in turn will reduce the costs associated with building and deploying small regional arrays.

Analysis of three-component data from IRIS/IDA stations in the USSR

Abstract This paper presents results of backazimuth, incidence angle, and phase identification studies based on the three-component polarization of 0- to 5-Hz seismic signals recorded at the IRIS/IDA stations within the USSR. A data set was built from the analysis of three-component (3-C) recordings from seismic stations at Arti (ARU), Garm (GAR), Kislovodsk (KIV), and Obninsk (OBN) for the month of March 1989. Approximately 260 events were studied. At stations ARU, KIV, and OBN, teleseisms dominated the sample of recordings. Station GAR, however, recorded a larger set of events that included a significant number of local and regional events. Except for station OBN, all of the IRIS/IDA stations are located near major geologic structural boundaries that are characterized by significant lateral variation in the deep velocity structure. These stations near boundaries show system-atic deviations in estimates of backazimuth and angle of incidence. At station GAR, both of these measurements were influenced by the structure. Polarization characteristics of seismic phases were studied individually for each station. At stations OBN and KIV, the initial teleseismic P wave was characterized, for the most part, by prograde elliptical motion. At ARU and GAR, the motion was rectilinear. Regional S phases from Hindu-Kush earthquakes recorded at GAR were dominated by transverse horizontally polarized motion ( SH ). At KIV and ARU, Lg motion was predominantly in a near-horizontal plane. A phase identification study was performed using discriminant analysis based on polarization parameters and frequency content. Two discriminants, linear combinations of polarization and frequency measurements, were able to separate arrival data into one of three groups of phases: teleseismic P , regional P , and regional S . At GAR, the first discriminant is correlated with polarization and separates arrivals with P -wave characteristics from arrivals with S -wave characteristics. The second discriminant accentuates the differences in the frequency content of the arrivals and provides a good separation between regional and teleseismic phases. Backazimuth estimates were made for local, regional, and teleseismic events at GAR and for teleseismic events at ARU, KIV, and OBN. The unique structural environment of each station required that custom processing parameters be developed for each station. The assumed particle motion (rectilinear or elliptical) used in the polarization computation, the window length, and the frequency band (characteristic for each station) were determinant factors. Backazimuth estimates within ±25° of the theoretical value were obtained for about 70% of the events at most of the stations. At GAR, systematic backazimuth errors caused by large lateral velocity discontinuities degraded the results. Corrections could be made to reduce these errors.