基于SKS深度学习识别的河北省上地幔各向异性  

Upper mantle anisotropy in Hebei province based on deep-learning SKS identification

作　　者：吴鹤帅 李泽峰[2,4] 朱俊 WU HeShuai;LI ZeFeng;ZHU Jun(Hebei Hongshan National Observatory on Thick Sediments and Seismic Hazards,Xingtai Hebei 054000,China;School of Earth and Space Science,University of Science and Technology of China,Hefei 230026,China;Xingtai Earthquake Monitoring Center Station,Hebei Earthquake Agency,Xingtai Hebei 054000,China;Mengcheng National Geophysical Observatory,Bozhou Anhui 233500,China)

机构地区：[1]河北红山巨厚沉积与地震灾害国家野外科学观测研究站,河北邢台054000 [2]中国科学技术大学地球和空间科学学院,合肥230026 [3]河北省地震局邢台地震监测中心站,河北邢台054000 [4]蒙城地球物理国家野外科学观测研究站,安徽亳州233500

出　　处：《地球物理学报》2025年第4期1246-1257,共12页Chinese Journal of Geophysics

基　　金：国家重点研发计划(2022YFC3005602);中国地震局监测、预报、科研三结合课题(3JH-202201044)资助。

摘　　要：SKS波分裂是研究上地幔各向异性的重要手段之一,但目前SKS的识别和挑选多依赖人工.为满足日益增长的SKS震相拾取需求,我们以河北红山台(HNS)人工标注的1116条远震波形作为训练数据,构建了卷积神经网络模型,实现了自动化识别SKS震相,且该模型在测试集的准确率达到了88%.利用迁移学习将预训练模型应用到河北昌黎台(CLI),达到了85%的准确率,验证了模型泛化性能.我们将模型应用到河北省71个台站2017年至2021年共计90454个远震事件波形中,通过连接已有的自动化SKS波分裂参数测量算法,实现了从远震事件波形到SKS波分裂参数的全流程化计算,最终获取了河北地区上地幔1173条各向异性参数.结果显示,河北地区上地幔各向异性快波方向基本呈WNW-ESE方向,延迟时间为0.5~1.8 s,主要源于软流圈地幔的流动.这些结果与前人研究一致,证明了深度学习识别SKS震相和全自动流程的可靠性,大大节省了人工标注震相的时间.本研究为未来大规模研究全国和全球上地幔各向异性、实时化获取SKS波分裂参数奠定了基础.SKS splitting is one of the significant methods for studying the anisotropy of the upper mantle.However,the identification and selection of SKS currently rely heavily on manual processes.To meet the growing demand for SKS phase picking,we constructed a convolutional neural network model to automatically identify SKS phases,using 1116 manually labeled teleseismic waveforms from the HNS station in Hebei as training data.The accuracy of this model on the test set reaches 88%.By applying transfer learning,we adapted the pre-trained model to the CLI station in Hebei,resulting an accuracy of 85%,thereby validating the model's generalization capability.We further applied the model to 90454 teleseismic event waveforms of 71 stations in Hebei(2017-2021).By connecting an automated SKS splitting parameter measurement algorithm,we established a comprehensive workflow from teleseismic waveform processing to SKS splitting parameter calculation,and finally obtained 1173 anisotropy parameters in the upper mantle in the Hebei region.The results indicate that the fast-wave direction of upper mantle anisotropy in Hebei region is predominantly WNW-ESE,with delay times 0.5~1.8 s.The anisotropy patterns support that mantle flow in the asthenosphere is the main cause.These results are consistent with previous studies and demonstrate reliability of deep-learning SKS identification and the fully automatic workflow,which significantly reduce the time required for manual phase picking.This study sets a foundation for large-scale research of upper mantle anisotropy on both national and global scales and real-time measurement of SKS splitting parameters.

关 键 词：SKS波分裂 上地幔各向异性 深度学习 迁移学习 

分 类 号：P315[天文地球—地震学]