Statistical prediction of earthquakes

Statistical prediction is an extension of Fourier spectrum analysis of time series. Historical attempts at prediction of earthquakes have centered on the question of periodicity. A survey of work in this field is given, both preceding and following critical investigations by Jeffreys. Present evidence indicates earthquakes to be uncorrelated events; observed deviations from the Poisson distribution, previously attributed to dependence in time, may be an effect of spatial inhomogeneity. It is fallacious to infer properties of the earthquake process on the basis of the shape of the distribution function of number of events in time. Some restrictions are shown to apply to the use of direct extrapolation methods in earthquake time series. Multiple prediction is a possibility; however, the actual measurement of geophysical variables for prediction purposes has progressed very slowly. Direct measurements of stresses and strains at depth are needed; values based on the theoretical release of strains by earthquakes are insufficient. Plotting the square roots of earthquake energies yields no specific strain information of predictive value. The estimation of earthquake risks is different, in principle, from the prediction problem. Mathematical techniques are available for risk estimation. A simplified model of the earthquake process is suggested, which may satisfy elementary requirements for purposes of engineering, city planning, and earthquake insurance.