A Nonparametric Hawkes Model for Forecasting California Seismicity

Abstract

ABSTRACT A variety of nonparametric models have been proposed for estimating earthquake triggering. We investigate the ability of the model-independent stochastic declustering method developed by Marsan and Lengliné (2008) to estimate variable spatial triggering that can vary with direction, magnitude, and region. We develop an approach for local fault estimation and demonstrate forecasting methods that use the nonparametric estimates. Simulation studies are conducted to verify the effectiveness of the method, and the nonparametric estimates are applied to a California earthquake catalog. Model forecast performance is evaluated retrospectively by comparing our models with the long-term forecast of Helmstetter et al. (2007), using both deviance and Voronoi residuals. We show improved performance compared with Helmstetter et al. (2007) in various regions while using a full nonparametric estimation and forecasting approach.