ALBERT

Description: A surprising discovery has been made that in some cases the complex, highly scattered Lg wave is found to be similar for clusters of events. We analyze in detail a subset of 28 out of 90 events from the 1999 Xiuyan sequence. Cross correlations provide highly accurate differential travel-time measurements. Their error estimated from the internal consistency is about 7 msec. These travel-time differences are then inverted by the double-difference technique to obtain epicenter estimates that have location precision on the order of 150 m. The locations are computed with waveform data from four to five regional stations 500 to 1000 km away. The epicenter estimates are not substantially affected by the sparseness of stations or large azimuthal gaps. Comparison with a surface trace a few kilometers away and location estimates based on much more dense networks led us to conclude that the absolute positions are accurate to the 5-km level. Regional event locations must often be based on a small number of phases and stations due to weak signal-to-noise ratios and sparse station coverage. This is especially true for monitoring work that seeks to locate smaller magnitude seismic events with a handful of regional stations. Two primary advantages of using Lg for detection and location are that it is commonly the largest amplitude regional wave (enabling detection of smaller events) and it propagates more slowly than P waves or Sn (resulting in smaller uncertainty in distance, for a given uncertainty in travel time.

Description: Earthquake location using relative arrival time measurements can lead to dramatically reduced location errors and a view of fault-zone processes with unprecedented detail. There are two principal reasons why this approach reduces location errors. The first is that the use of differenced arrival times to solve for the vector separation of earthquakes removes from the earthquake location problem much of the error due to unmodeled velocity structure. The second reason, on which we focus in this article, is that waveform cross correlation can substantially reduce measurement error. While cross correlation has long been used to determine relative arrival times with subsample precision, we extend correlation measurements to less similar waveforms, and we introduce a general quantitative means to assess when correlation data provide an improvement over catalog phase picks. We apply the technique to local earthquake data from the Calaveras Fault in northern California. Tests for an example streak of 243 earthquakes demonstrate that relative arrival times with normalized cross correlation coefficients as low as approximately 70%, interevent separation distances as large as to 2 km, and magnitudes up to 3.5 as recorded on the Northern California Seismic Network are more precise than relative arrival times determined from catalog phase data. Also discussed are improvements made to the correlation technique itself. We find that for large time offsets, our implementation of timedomain cross correlation is often more robust and that it recovers more observations than the cross spectral approach. Longer time windows give better results than shorter ones. Finally, we explain how thresholds and empirical weighting functions may be derived to optimize the location procedure for any given region of interest, taking advantage of the respective strengths of diverse correlation and catalog phase data on different length scales.

Description: We processed the complete digital seismogram database for northern California to measure accurate differential travel times for correlated earthquakes observed at common stations. Correlated earthquakes are earthquakes that occur within a few kilometers of one another and have similar focal mechanisms, thus generating similar waveforms, allowing measurements to be made via cross-correlation analysis. The waveform database was obtained from the Northern California Earthquake Data Center and includes about 15 million seismograms from 225,000 local earthquakes between 1984 and 2003. A total of 26 billion cross-correlation measurements were performed on a 32-node (64 processor) Linux cluster, using improved analysis tools. All event pairs with separation distances of 5 km or less were processed at all stations that recorded the pair. We computed a total of about 1.7 billion P-wave differential times from pairs of waveforms that had cross-correlation coefficients (CC) of 0.6 or larger. The P-wave differential times are often on the order of a factor of ten to a hundred times more accurate than those obtained from routinely picked phase onsets. 1.2 billion S-wave differential times were measured with CC〉 or =0.6, a phase not routinely picked at the Northern California Seismic Network because of the noise level of remaining P coda. We found that approximately 95% of the seismicity includes events that have cross-correlation coefficients of CC〉 or =0.7 with at least one other event recorded at four or more stations. At some stations more than 40% of the recorded events are similar at the CC〉 or =0.9 level, indicating the potential existence of large numbers of repeating earthquakes. Large numbers of correlated events occur in different tectonic regions, including the San Andreas Fault, Long Valley caldera, Geysers geothermal field and Mendocino triple junction. Future research using these data may substantially improve earthquake locations and add insight into the velocity structure in the crust.

Description: Statistical analyses were conducted on the capability of correlation detectors for similar events. Semiempirical synthetic runs took a 50-sec window on an Lg wave recorded at 750-km distance filtered from 1 to 3 Hz and embedded it 300,000 times in real continuous background seismic noise. The noise was selected for 36 days spread throughout the year to capture diurnal and seasonal variations. No screening for random, unknown signals in the noise was performed. A correlation detector has a 50% probability of detection with 1.5 false alarms per day for a signal-to-noise ratio (SNR) of 0.32, which corresponds to a full magnitude unit reduction in detection threshold over a standard short-term average/long-term average (STA/LTA) technique. A scaled cross-correlation coefficient performs slightly better with one false alarm per day and has fewer false triggers on unknown, random signals. Summing the cross-correlation traces together for all three components enhances the detection signal similar to beamforming. A correlation detector summing the correlation traces for the three components together has a 96% probability of detection with zero false alarms in 36 days for an SNR of 0.32. The significant result of this study is that a correlation detector has more than an order of magnitude improvement in detection threshold for similar events with acceptably low false alarm rates to be used in practice.

Description: There are basically three ways that have been demonstrated to reliably measure seismic velocity changes in the crust of the Earth--repeating events, controlled sources, and ambient noise. Each has its complementary benefits. Repeating events and controlled sources provide measurements at discrete points in time and space with high signal-to-noise ratios. But to achieve a sufficient signal-to-noise ratio for ambient noise it has been necessary to average over long periods of time and/or numbers of station pairs. Often 30 days has been the average, but researchers have experimented with one-day averages to capture short-term velocity variations at the expense of reduced measurement precision. Coseismic and postseismic velocity changes have been established for many earthquakes. But measuring a clear preseismic signal has continued to elude us. One of the main reasons is due to insufficient temporal sampling for repeating events and controlled sources making the findings inconclusive as to the existence of a preseismic signal because of a lack of data. The benefit of ambient noise monitoring is that it provides continuous measurements in the preseismic period and offers hope to alleviate this problem. This paper improves the measurement precision for one-day ambient noise monitoring by averaging over the three station components to construct the full nine-element Green's tensor. Doing so, I am able to measure a 95% confidence limit for an upper bound on preseismic velocity changes to be 0.0265%. I also compare my results constructing the Green's function by the correlation of the coda of the correlation of ambient noise.

Description: Waveform correlation is garnering attention as a method for detecting, locating, and characterizing similar seismic events. To explore the opportunities for using waveform correlation in broad regional monitoring, we applied the technique to a large region of central Asia over a three-year period, monitoring for events at regional distances using three high-quality stations. We discuss methods for choosing quality templates and introduce a method for choosing correlation detection thresholds, tailored for each template, for a desired false alarm rate. Our SeisCorr software found more than 10,000 detections during the three-year period using almost 2000 templates. We discuss and evaluate three methods of confirming detections: bulletin confirmation, high correlation with a template, and multistation validation. At each station, 65%-75% of our detections could be confirmed, most by multistation validation. We confirmed over 6500 unique detections. For monitoring applications, it is of interest that a significant portion of the Comprehensive Nuclear-Test-Ban Treaty Organization's Late Event Bulletin (LEB) catalog events was detected and that adding our confirmed detections for the LEB catalog would more than double the catalog size. Waveform correlation also allows for relative magnitude calculation, and we explore the magnitudes of detected events. The results of our study suggest that doing broad regional monitoring using historical and real-time-generated templates is feasible and will increase detection capabilities.

Description: From a local high-resolution base catalog at Parkfield, California, 5076 earthquakes (M 0.2 to 6) are used to study the comparative performance of a correlation detector and standard energy detector on the sparse regional network of continuously operating stations. Eighty-six percent of the events detected by a standard energy detector can also be detected by cross correlation. Correlation detection is able to find additional events by lowering the detection threshold by about 1 unit beyond what standard processing detects for Parkfield, a factor of 10 increase in number of events such as those predicted by Gutenberg-Richter. Most event separation distances for events that correlate at Parkfield are less than 1 km. The distribution of magnitude differences for events that correlate at Parkfield is not distinguishable from the input magnitude distribution. More robust measures to quantify reduction in detection threshold are introduced. Detection magnitude threshold reduction of about 1 unit holds for large-scale application to the 18,886 events in China and 5,076 events in Parkfield with false-alarm rates of a few percent. Large and small events are seen to correlate well enough for detection. Two examples are shown with magnitude differences as large as 2.3 and 3.3 units. The correlation detector also finds two cases of buried aftershocks in the coda of mainshocks that were previously unreported in the Annual Bulletin of Chinese Earthquakes (ABCE).