Application of machine learning to magnitude estimation in earthquake emergency prediction system

Earthquake early warning (EEW) is an important way for earthquake disaster reduction, and magnitude estimation is an important and difficult part of the entire EEW system. Nowadays, many countries and regions around the world have established their own EEW systems, and two types of magnitude emergency warning methods, characteristic frequency (tau(p) and tau(c) etc. ) and characteristic amplitude (P-d and others) , have been presented. Based on the existing methods and theories, we applied the machine learning algorithm to 55426 records for 843 earthquakes recorded by the KIK and KNET networks in Japan from 2015 to 2017. By using these records as a full data set, a set of machine learning magnitude prediction models have been designed and trained for common magnitude ranges. Compared with the estimated results of the existing methods, the machine learning method may reduce not only the estimated overall error and variance, but also the cross-sectional variance of multiple joint seismic events. The results of this study show that machine learning algorithm has a broad application prospect in earthquake magnitude emergency estimation, and provides a practical basis and research possibilities for end- to-end model of more complex deep learning algorithm framework as well.