Scale Invariance of Earthquake Parameters and a Possible Algorithm for Their Prediction

Abstract

An algorithm is described, proposed by the author for predicting the coordinates of sources, energy classes, and time of realization of expected strong earthquakes within given energy and spatiotemporal limits. It is based on the self-similarity of a seismic process in a wide energy range and includes the formation of a sampling of relatively strong earthquakes from a seismic catalog for a monitored area. In earthquake preparation zones, the nature of the epicentral distribution of weak earthquakes of representative classes that occurred during the last tenth of the seismic cycle is determined: feature vectors. They are reduced to a scale-invariant form and serve as “samples” for comparison with feature vectors of predicted (virtual) earthquakes determined from the catalog. If sufficient information is available, the parameters of the tensors of the average focal mechanisms for each preparation zone are added to the feature vectors, taking into account their weight coefficients. The prediction is assumed to be done by the least squares method (LSM), based on the criterion of best fit of all parameters for virtual earthquakes and the samples. The algorithm envisages testing by retroprediction and the creation of a computer program with machine learning for its implementation. During testing, the expected errors in the estimates of the predicted parameters are determined.