CSEP Evaluations of 24-Hour Earthquake Forecasting Models for California: New Results and Ensemble Models

Maximilian J. Werner, Maria Liukis, Warner Marzocchi, David A. Rhoades, Matteo Taroni, Zechar D. Zechar, & Thomas H. Jordan

Submitted August 15, 2016, SCEC Contribution #6873, 2016 SCEC Annual Meeting Poster #314

Operational Earthquake Forecasting requires credible earthquake probability estimates at short time scales. The objective of the Collaboratory for the Study of Earthquake Predictability (CSEP) is to evaluate earthquake forecasting models and hypotheses in a blind, automated and prospective manner. CSEP supports OEF efforts by independently and rigorously evaluating the strengths and weaknesses of candidate OEF models and ensemble OEF models. CSEP has been evaluating over a dozen 24-hour forecasting models in California since 2009. Models include the STEP model, various ETAS model flavors, non-parametric models and other statistical clustered seismicity models. Here, we report on new results from CSEP’s 24-hour earthquake forecasting experiment in California. The data set consists of 132 earthquakes greater than magnitude 3.95. Relative probability gains indicate that the predictive skills of the recent models are improving. In addition, we explore protocols for constructing ensemble models. Ensembles are powerful forecasting tools when several models are available and their predictive skills not yet established. In addition, combinations of models may outperform individual models.

Key Words
earthquake forecasting, predictability, clustering, seismic hazard

Citation
Werner, M. J., Liukis, M., Marzocchi, W., Rhoades, D. A., Taroni, M., Zechar, Z. D., & Jordan, T. H. (2016, 08). CSEP Evaluations of 24-Hour Earthquake Forecasting Models for California: New Results and Ensemble Models. Poster Presentation at 2016 SCEC Annual Meeting.


Related Projects & Working Groups
Collaboratory for the Study of Earthquake Predictability (CSEP)