S23A-4481:
Retrospective Evaluation of Earthquake Forecasts during the 2010-12 Canterbury, New Zealand, Earthquake Sequence
Tuesday, 16 December 2014
Maximilian J Werner1,2, Warner Marzocchi3, Matteo Taroni3, Jeremy D Zechar4, Matt Gerstenberger5, Masha Liukis6, David A Rhoades5, Camilla Cattania7, Annemarie Christophersen5, Sebastian Hainzl7, Agnes Helmstetter8, Abigail Jimenez9, Sandy Steacy9 and Thomas H Jordan6, (1)Princeton University, Princeton, NJ, United States, (2)University of Bristol, School of Earth Sciences and Cabot Institute, Bristol, United Kingdom, (3)National Institute of Geophysics and Volcanology, Rome, Italy, (4)ETH Zurich, Zurich, Switzerland, (5)GNS Science-Institute of Geological and Nuclear Sciences Ltd, Lower Hutt, New Zealand, (6)Southern California Earthquake Center, Los Angeles, CA, United States, (7)Deutsches GeoForschungsZentrum GFZ, Potsdam, Germany, (8)ISTerre Institute of Earth Sciences, Saint Martin d'Hères, France, (9)University of Ulster, Coleraine, United Kingdom
Abstract:
The M7.1 Darfield, New Zealand (NZ), earthquake triggered a complex earthquake cascade that provides a wealth of new scientific data to study earthquake triggering and the predictive skill of statistical and physics-based forecasting models. To this end, the Collaboratory for the Study of Earthquake Predictability (CSEP) is conducting a retrospective evaluation of over a dozen short-term forecasting models that were developed by groups in New Zealand, Europe and the US. The statistical model group includes variants of the Epidemic-Type Aftershock Sequence (ETAS) model, non-parametric kernel smoothing models, and the Short-Term Earthquake Probabilities (STEP) model. The physics-based model group includes variants of the Coulomb stress triggering hypothesis, which are embedded either in Dieterich’s (1994) rate-state formulation or in statistical Omori-Utsu clustering formulations (hybrid models). The goals of the CSEP evaluation are to improve our understanding of the physical mechanisms governing earthquake triggering, to improve short-term earthquake forecasting models and time-dependent hazard assessment for the Canterbury area, and to understand the influence of poor-quality, real-time data on the skill of operational (real-time) forecasts. To assess the latter, we use the earthquake catalog data that the NZ CSEP Testing Center archived in near real-time during the earthquake sequence and compare the predictive skill of models using the archived data as input with the skill attained using the best available data today. We present results of the retrospective model comparison and discuss implications for operational earthquake forecasting.
