In the last few years, deep learning has solved seemingly intractable problems, boosting the hope to find approximate solutions to problems that now are considered unsolvable. Earthquake prediction, the Grail of Seismology, is, in this context of continuous exciting discoveries, an obvious choice for deep learning exploration. We reviewed the literature of artificial neural network (ANN) applications for earthquake prediction (77 articles, 1994–2019 period) and found two emerging trends: an increasing interest in this domain over time and a complexification of ANN models toward deep learning. Despite the relatively positive results claimed in those studies, we verified that far simpler (and traditional) models seem to offer similar predictive powers, if not better ones. Those include an exponential law for magnitude prediction and a power law (approximated by a logistic regression or one artificial neuron) for aftershock prediction in space. Because of the structured, tabulated nature of earthquake catalogs, and the limited number of features so far considered, simpler and more transparent machine-learning models than ANNs seem preferable at the present stage of research. Those baseline models follow first physical principles and are consistent with the known empirical laws of statistical seismology (e.g., the Gutenberg–Richter law), which are already known to have minimal abilities to predict large earthquakes.

The hypothesis that earthquake foreshocks have a prognostic value is challenged by simulations of the normal behaviour of seismicity, where no distinction between foreshocks, mainshocks and aftershocks can be made. In the former view, foreshocks are passive tracers of a tectonic preparatory process that yields the mainshock (i.e., loading by aseismic slip) while in the latter, a foreshock is any earthquake that triggers a larger one. Although both processes can coexist, earthquake prediction is plausible in the first case while virtually impossible in the second. Here I present a meta-analysis of 37 foreshock studies published between 1982 and 2013 to show that the justification of one hypothesis or the other depends on the selected magnitude interval between minimum foreshock magnitude m(min) and mainshock magnitude M. From this literature survey, anomalous foreshocks are found to emerge when m(min) < M - 3.0. These results suggest that a deviation from the normal behaviour of seismicity may be observed only when microseismicity is considered. These results are to be taken with caution since the 37 studies do not all show the same level of reliability. These observations should nonetheless encourage new research in earthquake predictability with focus on the potential role of microseismicity.

Understanding and forecasting earthquake occurrences is presumably linked to understanding the stress distribution in the Earth's crust. This cannot be measured instrumentally with useful coverage. However, the size distribution of earthquakes, quantified by the Gutenberg‐Richterbvalue, is possibly a proxy to differential stress conditions and could therewith act as a crude stress‐meter wherever seismicity is observed. In this study, we improve the methodology ofbvalue imaging for application to a high‐resolution 3‐D analysis of a complex fault network. In particular, we develop a distance‐dependent sampling algorithm and introduce a linearity measure to restrict our output to those regions where the magnitude distribution strictly follows a power law. We assess the catalog completeness along the fault traces using the Bayesian Magnitude of Completeness method and systematically imagebvalues for 243 major fault segments in California. We identify and reportbvalue structures, revisiting previously published features, e.g., the Parkfield asperity, and documenting additional anomalies, e.g., along the San Andreas and Northridge faults. Combining localbvalues with local earthquake productivity rates, we derive probability maps for the annual potential of one or moreM6 events as indicated by the microseismicity of the last three decades. We present a physical concept of how different stressing conditions along a fault surface may lead tobvalue variation and explain nonlinear frequency‐magnitude distributions. Detailed spatialbvalue information and its physical interpretation can advance our understanding of earthquake occurrence and ideally lead to improved forecasting ability.