{"title":"基于集合卡尔曼反演的地震移动时间层析成像技术","authors":"Yunduo Li, Yijie Zhang, Xueyu Zhu, Jinghuai Gao","doi":"10.1093/gji/ggae329","DOIUrl":null,"url":null,"abstract":"In this paper, we present a new seismic travel-time tomography approach that combines ensemble Kalman inversion (EKI) with Neural Networks (NNs) to facilitate the inference of complex underground velocity fields. Our methodology tackles the challenges of high-dimensional velocity models through an efficient neural network parameterization, enabling efficient training on coarse grids and accurate output on finer grids. This unique strategy, combined with a reduced-resolution forward solver, significantly enhances computational efficiency. Leveraging the robust capabilities of EKI, our method not only achieves rapid computations but also delivers informative uncertainty quantification for the estimated results. Through extensive numerical experiments, we demonstrate the exceptional accuracy and uncertainty quantification capabilities of our EKI-NNs approach. Even in the face of challenging geological scenarios, our method consistently generates valuable initial models for full wave inversion (FWI).","PeriodicalId":12519,"journal":{"name":"Geophysical Journal International","volume":"34 1","pages":""},"PeriodicalIF":2.8000,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Seismic travel-time tomography based on Ensemble Kalman Inversion\",\"authors\":\"Yunduo Li, Yijie Zhang, Xueyu Zhu, Jinghuai Gao\",\"doi\":\"10.1093/gji/ggae329\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"In this paper, we present a new seismic travel-time tomography approach that combines ensemble Kalman inversion (EKI) with Neural Networks (NNs) to facilitate the inference of complex underground velocity fields. Our methodology tackles the challenges of high-dimensional velocity models through an efficient neural network parameterization, enabling efficient training on coarse grids and accurate output on finer grids. This unique strategy, combined with a reduced-resolution forward solver, significantly enhances computational efficiency. Leveraging the robust capabilities of EKI, our method not only achieves rapid computations but also delivers informative uncertainty quantification for the estimated results. Through extensive numerical experiments, we demonstrate the exceptional accuracy and uncertainty quantification capabilities of our EKI-NNs approach. Even in the face of challenging geological scenarios, our method consistently generates valuable initial models for full wave inversion (FWI).\",\"PeriodicalId\":12519,\"journal\":{\"name\":\"Geophysical Journal International\",\"volume\":\"34 1\",\"pages\":\"\"},\"PeriodicalIF\":2.8000,\"publicationDate\":\"2024-09-11\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Geophysical Journal International\",\"FirstCategoryId\":\"89\",\"ListUrlMain\":\"https://doi.org/10.1093/gji/ggae329\",\"RegionNum\":3,\"RegionCategory\":\"地球科学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"GEOCHEMISTRY & GEOPHYSICS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Geophysical Journal International","FirstCategoryId":"89","ListUrlMain":"https://doi.org/10.1093/gji/ggae329","RegionNum":3,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"GEOCHEMISTRY & GEOPHYSICS","Score":null,"Total":0}
Seismic travel-time tomography based on Ensemble Kalman Inversion
In this paper, we present a new seismic travel-time tomography approach that combines ensemble Kalman inversion (EKI) with Neural Networks (NNs) to facilitate the inference of complex underground velocity fields. Our methodology tackles the challenges of high-dimensional velocity models through an efficient neural network parameterization, enabling efficient training on coarse grids and accurate output on finer grids. This unique strategy, combined with a reduced-resolution forward solver, significantly enhances computational efficiency. Leveraging the robust capabilities of EKI, our method not only achieves rapid computations but also delivers informative uncertainty quantification for the estimated results. Through extensive numerical experiments, we demonstrate the exceptional accuracy and uncertainty quantification capabilities of our EKI-NNs approach. Even in the face of challenging geological scenarios, our method consistently generates valuable initial models for full wave inversion (FWI).
期刊介绍:
Geophysical Journal International publishes top quality research papers, express letters, invited review papers and book reviews on all aspects of theoretical, computational, applied and observational geophysics.