SCEC Award Number 20083 View PDF
Proposal Category Individual Proposal (Integration and Theory)
Proposal Title Statistical evaluation tools for earthquake simulation models.
Investigator(s)
Name Organization
Frederic Schoenberg University of California, Los Angeles
Other Participants 1 PhD and 1 MS student might potentially work on the project in an unfunded capacity.
SCEC Priorities 5c, 5d, 5a SCEC Groups EFP, Seismology, CXM
Report Due Date 03/15/2021 Date Report Submitted 03/01/2021
Project Abstract
 This research involved assessing the fit of spatial-temporal models for forecasting seismicity as part of the Collaboratory for the Study of Earthquake Predictability (CSEP). We used recent advances in residual analysis and model assessment for space-time point process models to compare the fit of competing models, including notably models such as ETAS that are commonly used for earthquake hazard assessment. Super-thinning and Voronoi deviance residuals were used to ensure that model components are correctly specified and properly incorporated into seismicity models. We developed algorithms for rapidly estimating the triggering functions nonparametrically and the efficient automatic assessment of such models and forecasts using Voronoi deviance residuals and superthinning. Our model fitting and residual analysis tools were found to be useful in assessing models for seismicity, both using real data as well as simulations. We find slightly superior fit of the Gordon (2017) model compared to the version of the ETAS model implemented by Zhuang et al. (2003), though both models fit the data overall very well. Our proposed nonparametrically estimated model performs well relative to the Helmstetter et al. (2007) ETAS model, and is
Hawkes models estimated using MISD, with spatial distributions computed relative to local strike angle estimates, are able adequately to summarize the spatial distribution of aftershocks and dependence upon region, distance from known faults, and mainshock magnitude. Simulation studies confirm the nonparametric estimation performs well, even under highly nonstandard conditions.
Intellectual Merit Our research contributed in important ways to the development of more accurate modeling, estimation, and evaluation of models for earthquake occurrences. We developed a new way of estimating the best-fitting models for earthquake occurrences, and showed that our method is not only faster but also more accurate and more robust, as it does not rely on particular parametric assumptions about how earthquake productivity varies with magnitude or spatial location. Our research also showed that super-thinning and Voronoi residuals are extremely useful and powerful tools for evaluating the fit of models for earthquake occurrences, both using real data and simulations. Moreover, we were able to compare in detail the fit to data of the best-fitting models for earthquake occurrences, in purely prospective analyses, and discover areas of potential improvement in these models.

Broader Impacts Our research on modeling seismicity and evaluating forecasts contributes to the global effort to improve earthquake forecasting which benefits society via better understanding earthquake hazard and preparedness. Statistics PhD student J. Gordon, MS student R. Jia, and undergraduate student C. Tamadon were trained by PI Schoenberg. Two of these students, Jia and Tamadon, are female, and Tamadon belongs to an underrepresented minority group.
Exemplary Figure Figure 2 of the project report.
Super-thinned residuals for Zhuang et al. (2003) model (left) and Gordon (2017) model (right). Red triangles indicate observed events after thinning and blue squares indicate superposed points.
From Jia et al. (2021).