Global versus local clustering of seismicity. Implications with earthquake prediction

The estimation of the maximum expected magnitude is crucial for seismic hazard assessment. It is usually inferred via Bayesian analysis; alternatively, the size of the largest possible event can be roughly obtained from the extent of the seismogenic source and the depth of the brittle–ductile transition. However, the effectiveness of the first approach is strongly limited by catalog completeness and the intensity of recorded seismicity, so that it can be of practical use only for aftershocks, while the second is affected by extremely large uncertainties. In this article, we investigate whether it may be possible to assess the magnitude of the largest event using some statistical properties of seismic activity. Our analysis shows that, while local features are not appropriate for modeling the emergence of peaks of seismicity, some global properties (e.g., the global coefficient of variation of interevent times and the fractal dimension of epicenters) seem correlated with the largest magnitude. Unlike several scientific articles suggest, the b-value of the Gutenberg–Richter law is not observed to have a predictive power in this case, which can be explained in the light of heterogeneous tectonic settings hosting fault systems with different extension.