AUTHOR=Chelidze Tamaz , Melikadze Giorgi , Kiria Tengiz , Jimsheladze Tamar , Kobzev Gennady TITLE=Statistical and Non-linear Dynamics Methods of Earthquake Forecast: Application in the Caucasus JOURNAL=Frontiers in Earth Science VOLUME=8 YEAR=2020 URL=https://www.frontiersin.org/journals/earth-science/articles/10.3389/feart.2020.00194 DOI=10.3389/feart.2020.00194 ISSN=2296-6463 ABSTRACT=

In 20th century, more than 10 strong earthquakes (EQs) of magnitudes 6,7 hit South Caucasus, causing thousands of casualties and gross economic losses. Thus, strong-EQ forecast is an actual problem for the region. In this direction, we developed a physical percolation model of fracture, which considers the final failure of solid as a termination of the prolonged process of destruction: generation and clustering of micro-cracks, till appearance—at some critical concentration—of the infinite cluster, marking the final failure. Percolation provides a model of preparation of an individual strong event (slip or EQ). The natural seismic process contains many such events: the appropriate model is a non-linear stick-slip model, which is a particular case of the general theory of the integrate-and-fire process. Non-linearity of the seismic process is in contradiction with a memoryless Poissonian approach to seismic hazard. The complexity theory offers a chance to improve strong EQs’ forecast using analysis of hidden (non-linear) patterns in seismic time series, such as attractors in the phase space plot. For a regional forecast, we applied the Bayesian approach to assess the conditional probability expected in the next 5 years of strong EQs of magnitudes five and more. Later on, in addition to Bayesian probability assessment, we applied to seismic time series the pattern recognition technique, based on the assessment of the empirical risk function [generalized portrait (GP) method]: nowadays, this approach is known as the support vector machine (SVM) technique. The preliminary analysis shows that application of the GP technique allows predicting retrospectively 80% of M5 events in Caucasus. Besides long- and middle-term forecast studies, intensive work is under way on the short-term (next-day) EQ prediction also. Here, we present the results of multiparametrical (hydrodynamic and magnetic) monitoring carried out on the territory of Georgia. In order to assess the reliability of the precursors, we used the machine learning approach, namely, the algorithm of deep learning ADAM, which optimizes target function by a combination of optimization algorithm designed for neural networks and a method of stochastic gradient descent with momentum. Finally, we used the method of receiver operating characteristics (ROC) to assess the forecast quality of this binary classifier system. We show that the true positive rate statistical measure is preferable for the EQ forecast.