The static Coulomb stress hypothesis is a widely known physical mechanism for earthquake triggering, and thus a prime candidate for physics-based Operational Earthquake Forecasting (OEF). However, the forecast skill of Coulomb-based seismicity models remains controversial, especially in comparison to empirical statistical models. A previous evaluation by the Collaboratory for the Study of Earthquake Predictability (CSEP) concluded that a suite of Coulomb-based seismicity models were less informative than empirical models during the aftershock sequence of the 1992 M w 7.3 Landers, California, earthquake. Recently, a new generation of Coulomb-based and Coulomb/statistical hybrid models were developed that account better for uncertainties and secondary stress sources. Here, we report on the performance of this new suite of models in comparison to empirical Epidemic Type Aftershock Sequences (ETAS) models during the 2010-2012 Canterbury, New Zealand, earthquake sequence. Comprising the 2010 M 7.1 Darfield earthquake and three subsequent M ≥ 5.9 shocks (including the February 2011 Christchurch earthquake), this sequence provides a wealth of data (394 M ≥ 3.95 shocks). We assessed models over multiple forecast horizons (1-day, 1-month and 1-year, updated after M ≥ 5.9 shocks). The results demonstrate substantial improvements in the Coulomb-based models. Purely physics-based models have a performance comparable to the ETAS model, and the two Coulomb/statistical hybrids perform better or as well as the corresponding statistical model. On the other hand, an ETAS model with anisotropic (fault-based) aftershock zones is just as informative. These results provide encouraging evidence for the predictive power of Coulomb-based models. To assist with model development, we identify discrepancies between forecasts and observations.

The 2019 Ridgecrest sequence provides the first opportunity to evaluate Uniform California Earthquake Rupture Forecast v.3 with epidemic-type aftershock sequences (UCERF3-ETAS) in a pseudoprospective sense. For comparison, we include a version of the model without explicit faults more closely mimicking traditional ETAS models (UCERF3-NoFaults). We evaluate the forecasts with new metrics developed within the Collaboratory for the Study of Earthquake Predictability (CSEP). The metrics consider synthetic catalogs simulated by the models rather than synoptic probability maps, thereby relaxing the Poisson assumption of previous CSEP tests. Our approach compares statistics from the synthetic catalogs directly against observations, providing a flexible approach that can account for dependencies and uncertainties encoded in the models. We find that, to the first order, both UCERF3-ETAS and UCERF3-NoFaults approximately capture the spatiotemporal evolution of the Ridgecrest sequence, adding to the growing body of evidence that ETAS models can be informative forecasting tools. However, we also find that both models mildly overpredict the seismicity rate, on average, aggregated over the evaluation period. More severe testing indicates the overpredictions occur too often for observations to be statistically indistinguishable from the model. Magnitude tests indicate that the models do not include enough variability in forecasted magnitude-number distributions to match the data. Spatial tests highlight discrepancies between the forecasts and observations, but the greatest differences between the two models appear when aftershocks occur on modeled UCERF3-ETAS faults. Therefore, any predictability associated with embedding earthquake triggering on the (modeled) fault network may only crystalize during the presumably rare sequences with aftershocks on these faults. Accounting for uncertainty in the model parameters could improve test results during future experiments.

No abstract