[M-IS04] Interdisciplinary studies on pre-earthquake processes

[MIS04-02] Real-Time Earthquake Forecasting in China Using AI

*Yangkang Chen¹、Omar Saad²、Yunfeng Chen³、Alexandros Savvaidis¹、Sergey Fomel¹、Xiuxuan Jiang³、Dino Huang¹、Innocent Oboué³ (1.University of Texas at Austin、2.King Abdullah University of Science and Technology、3.Zhejiang University)

キーワード：Earthquake, Earthquake forecasting, Machine learning, Geoacoustic, Electromagnetic

Earthquake forecasting aims to save human lives and mitigate catastrophic damages by providing early alerts before the occurrence of large destructive earthquakes. It was long considered intractable due to various uncertain factors, including data processing artifacts, unknown physical mechanisms, geological complexity, anthropogenic interventions, etc. With the advent of artificial intelligence (AI) and gigantic datasets from multiple sources, earthquake forecasting has become more hopeful. In this study, we trained an earthquake forecasting model using a classic random forest algorithm and a large-scale dataset from West China, where earthquake activities are prevalent. We obtained encouraging real-time testing results on an independent dataset from the same area. The training data comprises the geo-acoustic (GA) and electromagnetic (EM) data of more than 120 M>3.5 earthquake events recorded by 150 stations from 10/01/2016 to 12/31/2020. Instead of using the continuous waveforms directly, we extract physically meaningful features to lower the freedom and uncertainty involved in the training process. The real-time forecasting performance reaches above 70% accuracy with a distance error close to 200 miles and a magnitude error below 0.5 Ml. This research sheds light on more widely tackling the enigmatic earthquake forecasting problems using AI-assisted data-harnessing technologies.