Physics of Geochemical Mechanics and Deep Neural Network Modeling with Diffusion Augmentation Applications to Earthquake Prediction

This book provides a new data augmentation method based on the local stochastic distribution patterns in natural time series data of global and regional seismicity rates and their correlated seismicity rates. The augmentation procedure is called the diffusion – denoising augmentation method from the local Gaussian distribution of segmented data of long time series. This method makes it possible to apply the deep machine learning necessary to neural network prediction of rare large earthquakes in the global and regional earth system.

The book presents the physical background of the processes showing the development of characteristic features in the global and regional correlated seismicity dynamics, which are manifested by the successive time series of 1990–2023. Physical processes of the correlated global seismicity change and the earth’s rotation, fluctuation of plate motion, and the earth’s ellipsoid ratio (C20 of satellite gravity change) are proposed in this book. The equivalency between Gaussian seismicity network dynamics and the minimal nonlinear dynamics model of correlated seismicity rates is also provided. In addition, the book contains simulated models of the shear crack jog wave, precipitation of minerals in the jog, and jog accumulation inducing shear crack propagation which leads to earthquakes in the plate boundary rocks under permeable fluid flow.

Introduction.- Physics of Geochemical Mechanics.- Characteristic Microstructures Reated to Multiphase Shear Flow.- Recent Variations of Global and Regional Correlated Seismicity.- Neural Network Modeling of Regression in Nonlinear Dynamics Timeseries.- Augmentation of Timeseries and DNN Modeling of Seismic Activity.- Concluding Remarks.

Mitsuhiro Toriumi is a senior researcher at the Japan Agency for Marine–Earth Science and Technology (JAMSTEC) and an emeritus professor of The University of Tokyo. He is also an advisor of the National Institute of Fusion Science and was the Japan Science and Technology Agency. He was the scientific director of the Institute for Research on Earth Evolution (IFREE) in JAMSTEC from 2011 to 2013. He was a professor of complexity earth science, petrology, and structural geology at The University of Tokyo from 1991 to 2011. His research interests include metamorphic petrology, structural geology, rheology in the earth interior, global seismicity dynamics, and neural network modeling. He has published several books of earth sciences (global seismicity dynamics, data-driven science, neural network modeling of seismicity, and geochemical mechanics) in English.