Robust Quantification of Earthquake Clustering: Overcoming the Artifacts of Catalog Errors

Friday, 19 December 2014
Ilya V Zaliapin, University of Nevada, Reno, Reno, NV, United States and Yehuda Ben-Zion, University of Southern California, Los Angeles, CA, United States
Quantitative characterization of earthquake clustering in space and time in relation to different event sizes and physical properties of the lithosphere are fundamental problems of statistical seismology. Recently, we approached these problems by taking advantage of the new statistical results, improved empirical constraints, and relatively uniform high-quality earthquake catalogs available for southern California and other selected regions. This led to identification and classification of statistically significant earthquake clusters in southern California, relating cluster characteristics to effective viscosity of the crust, and documenting some robust properties of observed earthquakes not simulated by the Epidemic Type Aftershock Sequence (ETAS) model. Extending these results to other seismically active areas and lower magnitude ranges, however, is impeded by inferior data quality. Most available catalogs are based on non-uniform recordings/analyses that lead to non-uniform (in space, time, magnitude) location errors, varying magnitude of completeness, and other problems. These non-uniformities may (and do) produce artificial patterns in the space-time-magnitude clusters of seismicity detected by our, as well as other, methods. In the present work we document the effects of catalog errors on inferred cluster properties, and report some striking patterns that emerge as artifacts of those errors. This includes apparent magnitude dependence, fluctuations in the proportion of singles (clusters consisting of individual events), space-dependent distance to the likely parents of events, and other effects. We also discuss additional differences between the ETAS model and observed seismicity. Finally, we propose a generalization of our method that involves assigning multiple possible parents to each event, and discuss some graph-theoretic techniques that may provide results that are more robust to location errors and other catalog deficiencies.