Goodness-of-fit assessment of earthquake forecasts using Voronoi and superthinned residuals

Ruoxuan Jia, Claire Tamadon, & Frederic P. Schoenberg

Under Review February 28, 2021, SCEC Contribution #10936

Many versions of the Epidemic-Type Aftershock Sequence (ETAS) model of Ogata (1988) have been proposed to model the occurrence times and locations of earthquakes. Variants proposed by Zhuang et al.\ (2003) and by Helmstetter et al.\ (2007) have been shown to fit best to data for prospective forecasting in the Collaboratory Study for Earthquake Predictability (CSEP), according to analyses by Schorlemmer et al.\ (2010) and Zechar (2013). However, Gordon (2017) showed retrospectively that his proposed version of the ETAS model,
which uses the Model Independent Stochastic Declustering (MISD) method of Marsan and Lenglin\'e (2008) to estimate a spatial triggering function varying direction, magnitude, and region,
would have outperformed prior versions of the ETAS model, using the Helmstetter et al.\ (2007) model as a benchmark and using data from 2005-2015. Here we evaluate the prospective fit of the nonparametric ETAS model of Gordon (2017) using earthquake data from 2017, comparing it to the Zhuang et al.\ (2003) model. The fit of both prospective model forecasts is evaluated using deviance and Voronoi residuals. We find slightly superior performance of the Gordon (2017) model compared to Zhuang et al.\ (2003) in various regions.

Key Words
CSEP, Epidemic Type Aftershock Sequence model, Hawkes process, Model Independent Stochastic Declustering, Point process, Seismology.

Citation
Jia, R., Tamadon, C., & Schoenberg, F. P. (2021). Goodness-of-fit assessment of earthquake forecasts using Voronoi and superthinned residuals. Seismological Research Letters, (under review). https://www.seismosoc.org/publications/srl