地震定位方法最新进展综述

Review of recent advances in seismic location methods

精确的地震定位对于地球内部结构和地震孕育过程认知、断层精细结构探测、资源能源勘探开发、地震预警研究等众多科学技术问题至关重要.鉴于地震定位方法发展迅速,及时的综述最新进展重要且必要.目前的综述论文主要论述常规地震定位方法的进展,而涉及机器学习的地震定位方法的系统总结却很少.为便于读者了解地震定位方法原理与最新的前沿进展,本文首先介绍了近年来新发展的常规地震定位方法,如震源扫描类方法、双差类定位法、GrowClust等;然后重点介绍了最新的涉及机器学习的地震定位方法,包括完全基于机器学习的地震定位方法和机器学习辅助的地震定位流程.其中,基于机器学习的定位方法按照利用的神经网络的不同进行再次分类,包括了卷积神经网络、图神经网络、循环神经网络.机器学习辅助的定位流程介绍了EasyQuake、QuakeFlow、LOC-FLOW三种较受关注的方法.通过详细阐述LSTM-FCN模型、LOC-FLOW方法流程的实际应用,对比了代表性方法的定位效果.最后,本文对机器学习类的地震定位方法存在的问题和地震定位的发展方向进行分析与展望,指出机器学习模型轻量化是重要研究方向以及多种地震定位方法联合定位是地震定位发展的重要目标.

Precise seismic location is essential for manyscientific and technical problems such as the recognition ofthe Earth's internal structure and seismogenic processes,refined fault structure detection,resource and energyexploration and development,and earthquake earlywarning.Given the rapid development of seismic locationmethods,a timely review of the latest advances isimportant and necessary.The existing review papersmainly cover the progress of conventional seismic locationmethods,but there are few systematic summaries ofseismic location methods involving machine learning.Tofacilitate readers to understand the principles of seismiclocation methods and the latest cutting-edge advances,thispaper first introduces the newly developed seismic locationmethods in recent years,such as the source scanning classmethod,the double difference class location method,andGrowClust,etc.We then focus on the latest seismiclocation methods involving machine learning,including thefully machine learning-based seismic location methods andthe machine learning-assisted seismic location processes.According to the adopted neural networks,includingconvolutional neural networks,graph neural networks,andrecurrent neural networks,the machine learning-basedlocation methods can be further categorized into differentgroups.For machine learning-assisted location process,weintroduce three popular workflows,EasyQuake,QuakeFlow,and LOC-FLOW.By elaborating the practical applicationsof LSTM-FCN model and LOC-FLOW method,the locationresults of representative methods are compared.Finally,this paper analyzes and outlooks the problems andprospects of seismic location methods involving machinelearning,pointing out that the lightweighting of machinelearning models is an important research direction and thejoint of multiple seismic location methods is an importantgoal for the development of seismic location.