Manfred Joswig

Clustering and location of mining induced seismicity in the Ruhr basin by automated master event comparison based on dynamic waveform matching (DWM)

Abstract Most of the local seismicity in the Ruhr Basin can be separated into characteristic clusters of similar, mining induced earthquakes. Each cluster can be represented by a strong master event. Therefore, it is possible to associate weak events to the corresponding clusters by master event comparison. The seismic signal matching is performed by a nonlinear correlation termed DWM for the entire seismogram length. DWM permits stretchings and shortenings between the two signals and overcomes the ambiguities in phase correlation by a consistent matching path. The automatic cluster association searches for the best DWM-correlation between the actual event and all master events of the appropriate epicenter region. Knowing the P- and S-onsets of the master event, they can be transposed to the actual event by the correlation path with one sample accuracy. The method has been applied to all BUG small array recordings 1987–1990 of local events from the Hamm -region to investigate spatial and temporal clustering. Within the clusters, a high percentage of weak events could be located relative to its master event. The temporal clustering resolved seismic activities that typically last a few months per cluster, but single aftershocks occur in the following years.

Mapping Seismic Quiescence in California

The decrease of microearthquake activity is a controversial possible precursor for mainshock prediction. The calculation of seismic quiescence is a convenient tool to quantify this decrease. However, the actual size of an anomaly significantly depends on the calculation parameters. The necessary averaging over space and time acts as 3D lowpass filter, the sliding statistics windows introduce highpass behavior, and the magnitude threshold shall ensure catalog completeness but also biases the data selection. To argue on any claimed anomaly, the only conclusive solution would be an exhaustive search over all parameters, over an area significantly larger than just the epicenter region, and against a test of the null hypothesis. QMAP is a new mapping approach to process full catalogs, that is, some twenty years of data for areas of one million km 2 , with one fixed set of parameters. It utilizes fractil statistics to avoid errors otherwise induced by clusters of aftershock series. QMAP features two different scaling modes, one to display absolute decrease and a second for weighted anomalies by erosion filtering. Applied to California, one can recognize quiescence anomalies of regional scale prior to the Coalinga and Loma Prieta earthquakes. But there are more and greater anomalies distributed all along the San Andreas fault system. Without further criteria to understand these patterns, the additional anomalies degrade seismic quiescence to an observation with high false alarm rate for earthquake prediction.

Single-trace detection and array-wide coincidence association of local earthquakes and explosions

Abstract Local earthquakes and explosions can be recognized automatedly for the Bochum University Germany (BUG) small array by a sequence of knowledge-based approaches performed in the field and in the central hub. In single-trace detection, the recognition is based on sonogram patterns adapted for a wide variety of noise conditions on all array sites. The adaptation is performed by two steps: first each pattern is adjusted to the actual signal energy, second all those weaker phases that are below the new detection threshold are excluded. In the hub, a rule-based approach performs the coincidence evaluation. It is described by its 14 rules and the implicit assumptions. This scheme was tested on 1 month of data. The knowledge base consisted of 12 seismograms transformed automatically into the detectors internal knowledge representation of sonograms. The results show excellent performance for noise rejection and quarry blast recognition; for earthquakes clustering, a 85% success is achieved. The network success—usually below the best single performance—could be improved above any single-station optimum. Results of the rule-based approach are compared to the routine processing of the same data by Walsh-detection and the ‘2 of 4’ coincidence voting.

Automated reevaluation of local earthquake data by application of generic polarization patterns for P- and S-onsets

The particle motion of local earthquake seismograms is affected strongly by the fine structure details of upper crust. The angle of incidence gets frequency dependent, shear wave splitting occurs and strong P-SV conversions contaminate the P-coda. Contrary to teleseism, it is not possible any more to detect S-onsets by conformance tests between data and simple models. Instead, we must derive polarization images in the time-frequency plane that display particle motion without any assumptions. By suitable scaling, these images neutralize all high frequency effects and allow for onset recognition by simple patterns. The method was applied to the 800 events of 1989 evaluated by the Bochum University Germany (BUG) observatory. We determined a 67% success rate with 13% wrong and 20% rejected because of unstable phase energy. For two source regions, the automated results are shown to be more reliable than interactive routine evaluation by man.

Automated seismogram analysis for the tripartite BUG array: an introduction

Abstract The tasks for automated epicenter determination in the Bochum University Germany (BUG) small array are subdivided for different signal-processing modules that utilize knowledge-based approaches. The modules are designed for complementary advantages to yield best system performance in an interdependent architecture. This “bottom-up” solution proceeds from reliable waveform parameters to more simple interpretation rules than in seismic expert systems that must cope with traditional detectors as erratic front ends.

Pattern Recognition for Earthquake Detection

Automatic detection of earthquakes in environmental noise is subject to research for more than 25 years. The traditional approaches describe the stationary noise level by some kind of median and standard deviation and determine a detection threshold [1],[2]. Each temporary signal with energy above the threshold triggers a detection. Bandpass- or ARMA-filtering is used to improve the signal to noise ratio. The main disadvantage of these detectors is their high false alarm rate, since they are not able to distinguish between small earthquakes and noise from traffic or industry, which has the same or even higher amplitude. This is due to their decision logic of negative kind: Every signal above the threshold is assumed to be caused by an earthquake.

Automated Detection and Interpretation of Earthquake Seismograms by Adaptive Pattern Recognition

The measurement and interpretation of ground motion in seismograms is the primar source for our knowledge of the earth’s interior. As in other scientific areas, seismology has seen great improvements of its experimental facilities in the last decades. State-of-the-art observatories monitor earthquakes magnitudes below they are recognized by humans. Additionally, these observatories record explosions in quarries up to 1000 km apart and nuclear explosions all over the world. Over a year, these events sum up to some 10.000 data sets that must be found and processed in TeraBytes of raw data which are contaminated by local noise bursts of car traffic, sonic bangs and other sources. This task of routine preselection and processing is about to exceed the possibilities of human data analysis; any further progess will depend on automated methods. However, these methods must yield human-like performance not to degrade the achieved sensitivity and reliability.

Automated Event Location by Seismic Arrays and Recent Methods for Enhancement

This paper describes procedures to locate seismic events at regional distances by single seismic arrays. First it reviews the methods used for interactive analysis and compares their performance using array data to that of single three-component stations. Then the paper focuses on automated location. Ten days of continuous-operation ARCESS data are analyzed. They are evaluated by standard NORSAR routines of beamforming and ƒ-k analysis, and by a complementary, independent approach. This approach utilizes pattern recognition (PR) and rule-based system (RBS) techniques to derive an automated bulletin from a subset of four array stations. The content of both lists, i.e., the ARCESS log of NORSAR and the new, PR-based list, are combined to form a joint bulletin. Its enhanced quality is achieved by rules for consistency check or “pluralistic reasoning.”

Wissensbasierte Erdbebenerkennung mit Sonogrammen als mentalen Bildern

Hochempfindliche seismologische Observatorien identifizieren heute bis zu 20.000 Erdbeben pro Jahr, deren Vielfalt von lokalen uber regionale bis zu teleseismischen Ereignissen reicht (z.B. Schneider [1975]). Diese Unterteilung spiegelt qualitativ verschiedene Laufwegeffekte fur seismische Wellen in den Schichten des Erdkorpers (Kruste, Mantel, Kern) wieder. Besonders gros ist die Variationsbreite fur Fernbeben, hier treten bei Raumwellen noch Abschattungs- bzw. Bundelungseffekte des Erdkerns und bei Oberflachen wellen Unterschiede zwischen kontinentaler und ozeanischer Kruste hinzu. Eine weitere Differenzierung ergibt sich aus verschiedenen Herdtiefen und Herdmechanismen.