Geophysical Research Letters, a world-renowned geoscience journal, has recently published a paper by a researcher in Hefei, capital of East China’s Anhui province, which is believed to be conducive to early earthquake warnings. In the paper, Li Zefeng, a researcher at the University of Science and Technology of China, explores the characteristics of source time functions of more than 3,000 5.5- and- above- magnitude earthquakes around the world using machine learning method. The paper presents a panorama of similarities and diversities of the rupture process during earthquakes to deepen the understanding of modes of energy liberation and contribute to early earthquake warnings. Major earthquakes have caused nearly one million casualties and innumerable economic losses in the world over the past two decades. The rupture process vary from earthquake to earthquake, and studying their similarities and differences helps understand the physical process of earthquakes and predict the magnitude early. However, previous researches either superimposed the average rupture process of multiple earthquakes, which failed to measure to what extent global earthquakes differ, or were based on statistics of some rupture features, instead of a systematic comparison of the whole process. Li carried out the research using Variational Autoencoder, a kind of generative deep learning model capable of unsupervised learning. It is part of a serial research called AI for Science, featuring the application of artificial intelligent to science.
|