Mixing of rescaled data and Bayesian inference for earthquake recurrence times

The limits of a recently proposed universal scaling law for the probability distributions of earthquake recurrence times are explored. The scaling properties allow to improve the statistics of occurrence of large earthquakes over small areas by mixing resealed recurrence times for different areas. In this way, the scaling law still holds for events with M >= 5.5 at scales of about 20 km, and for M >= 7.5 at 600 km. A Bayesian analysis supports the temporal clustering of seismicity against a description based on nearly-periodic events. The results are valid for stationary seismicity as well as for the nonstationary case, illustrated by the seismicity of Southern California after the Landers earthquake.