Yavor Kamer, Phd thesis presentation, 2.10.2015 Analysis of Large Seismicity Catalogs: The Californian Experience Examination Committee: Prof. Dr. Didier Sornette (ETH Zürich, Switzerland) Prof. Dr. Eduard Kissling (ETH Zürich, Switzerland) Prof. Dr. David Jackson (UCLA, United States of America) Dr. Guy Ouillon (Lithophyse, France) Thesis manuscript: https://goo.gl/CBeAso Abstract Recent years have seen a substantial increase in the amount of available seismic catalog data. Extracting information from these data, on a case-by-case basis is gradually becoming infeasible. In order to harness the influx of data, we need new methods that are objective and data-driven. These methods need to provide the community with models that are modest in their complexity and assumptions. This thesis undertakes the task of developing such methods that can be readily applied to large seismic catalogs. One of the main characteristics of earthquake phenomena is the power-law relation of moment release and frequency, i.e. the Gutenberg-Richter law. We developed a penalized likelihood-based method for spatial estimation of Gutenberg-Richter’s b-value. We used the obtained results to generate an earthquake forecast model for California. Our forecast shows a significant information gain with respect to the competing models. These results provide evidence for small amplitude, large-scale spatial b-value variations that can, in most cases, be attributed to inconsistencies in magnitude estimations. Another widely observed feature of seismicity is its multifractal spatial distribution. We developed an adaptive multifractal estimation method that reduces edge effects and improves precision. The method was successfully validated on synthetic fractals and used to assess the scaling properties of the widely studied diffusion-limited aggregation process. We developed also a condensation technique that allows accounting for the location uncertainty information in spatial analysis. We utilized these two approaches to investigate the spatial distribution of Southern Californian seismicity. Our analysis provided evidence for possible issues with the relocation procedure of the current catalog. This motivated us to re-locate the last three decades of Southern Californian seismicity by employing current state-of-the-art probabilistic non-linear methods. The outcome of these efforts is the new KaKiOS-15 catalog that contains the full expression of each event’s location uncertainty. Lastly, we developed a new fault network reconstruction method based on agglomerative clustering. We applied this method to the events in the condensed KaKiOS-15 catalog and obtained a fault network for Southern California. In retrospective tests, we were also able to demonstrate that the obtained fault network has significantly higher forecasting skills compared to commonly used methodologies.

• published: 22 Oct 2015
• views: 322