Colliding cascades model for earthquake prediction

A wide set of premonitory seismicity patterns is reproduced on a numerical model of seismicity, and their performance in the prediction of major model earthquakes is evaluated. Seismicity is generated by the colliding cascades model, recently developed by the authors. The model has a hierarchical structure. It describes the interaction of two cascades: a direct cascade of loading, which is applied to the top (largest) element and transfers down the hierarchy, and an inverse cascade of failures, which goes up the hierarchy, from the smaller to the larger elements. These cascades collide and interact: loading leads to failures, while failures release and redistribute the loading. Three basic types of earthquake precursors are considered: (i) the clustering of earthquakes in space and time, (ii) the intensity of earthquake sequences, and (iii) the correlation distance between earthquakes. Patterns of the first two types are used in intermediate‐term earthquake prediction algorithms. Patterns of the third type are found in the colliding cascades model, although they were hypothesized previously. They have not been validated by observations. For each precursor, we explore what is called an ‘error diagram’ showing the total duration of alarms, the rate of failures to predict, and the rate of false alarms.