Group A, Poster #205, Earthquake Forecasting and Predictability (EFP)

Are Regionally Calibrated Seismicity Models more Informative than Global Models? Insights from California, New Zealand, and Italy

José A. Bayona, William H. Savran, Pablo C. Iturrieta, Matthew C. Gerstenberger, Warner Marzocchi, Danijel Schorlemmer, & Maximilian J. Werner
Poster Image: 

Poster Presentation

2022 SCEC Annual Meeting, Poster #205, SCEC Contribution #12344 VIEW PDF
Earthquake forecasting models express a wide range of hypotheses about seismogenesis that underpin regional and global seismic hazard assessments. Until recently, comparisons between global and regional models remained largely qualitative and retrospective, lacking a more quantitative analysis that could help more rigorously assess the performance of global models at regional scales, analyze whether regionals might be overfitting their training datasets, and better inform seismic hazard programs. Here, we prospectively assess the ability of the Global Earthquake Activity Rate (GEAR1) model and nineteen regionally calibrated time-invariant models to forecast M4.95+ seismicity in California, N...ew Zealand, and Italy from 2014 through 2021, using a suite of likelihood tests developed by the Collaboratory for the Study of Earthquake Predictability (CSEP). Our comparative test results show that, in California, a model that adaptively smooths the locations of small (M2+) earthquakes is the most informative, while GEAR1, based on global seismicity and geodesy datasets, outperforms all other regional models. In New Zealand, GEAR1 is the most informative model, obtaining an information gain score per earthquake of about 0.5 over the 2002 New Zealand Hazard Model, which uses seismicity and fault kinematic data. In Italy, a model based on relative intensities and two models using adaptive smoothed (M3+) seismicity can be considered statistically more informative than GEAR1, while the global model, ranking fourth, can be considered as informative as the 2004 Italian National Seismic Hazard Model. When analyzing the spatial dimension of the models, we find that those that provide forecasts that are too dispersed or too smooth perform poorly, while the models that apply “intermediate” smoothing procedures, such as adaptive smoothing, are the most consistent with the observations. In addition, we identify that the New Zealand models obtain higher spatial likelihood scores than the models for California and Italy, suggesting a possible link between the tectonic setting and the spatiotemporal variability of earthquakes. These results suggest that adaptively smoothing small earthquake locations and combining geodetic data with smoothed seismicity are useful methods for forecasting the occurrence of M~5 earthquakes over a period of less than a decade in these regions and provide preliminary support for using GEAR1 as a robust reference model of global moderate seismicity.
SHOW MORE