Toward Exascale Global Adjoint Tomography

这项研究涉及开发用于模拟地震波在地球内部传播的开源软件。这种模拟可用于定量的地震危险评估、外展和教育活动的可视化以及地震层析成像的基础。与医学层析成像一样，地震层析成像的目标是在所有尺度上对地球内部进行成像，从勘探地震学中的碳氢化合物气田到全球地震学中的地幔，以帮助我们了解地球的内部运作和演化。作为一个更广泛的影响，这个项目将在普林斯顿大学的近实时地震活动门户网站(GlobalShakeMovie)上近乎实时地在线发布5.5级及以上地震的地震波动画。这项研究的主要目标是通过实现海量数据同化和利用现有最大和最快的计算机，使用完整的地震波形和伴随状态方法进一步发展全球地震层析成像。利用253次地震的先导数据集，建立了基于“伴随层析成像”的第一代全球层析成像模型，建立了迭代全波形反演技术。基于这一经验，将通过吸收宽带全球和区域地震台网很好地记录的数千次地震的全波形数据来构建下一代全球模式。先前NSF的支持证明了区域伴随层析成像对于各向同性非均质性、滞弹性和方位各向异性的可行性。将通过能源部激励计划获得必要的计算硬件，在全球范围内进行类似的反转。这项研究支持所需人力资源的一部分，在这一雄心勃勃的努力中，限速步骤是数据质量控制、I/O和工作流管理。考虑到这些障碍，将解决四项主要任务：1)准备并进一步开发用于亿级模拟的开源频谱元素地震波传播解算器SPECFEM3D和SPECFEM3D_GLOBLE；2)进一步开发用于快速I/O的可适应地震数据格式(ASDF)；3)利用开源结构化查询语言数据库引擎SQLight和现有的工作流管理工具，如Pegasus，进一步开发反演和偏移工具包SeisFlows；4)在S的最短周期内用数千次地震的数据库进行全球伴随地震层析成像。任务4包含了建议的主要目标，但如果不同时处理其他任务，它就不可能完成。

This research involves the development of open-source software for simulations of seismic wave propagation inside the Earth. Such simulations may be used for quantitative seismic hazard assessment, visualizations for outreach and educational activities, as well as a basis for seismic tomography. Like medical tomography, the goal of seismic tomography is to image the Earth's interior on all scales, from hydrocarbon fields in exploration seismology to Earth's mantle in global seismology, to help us understand our planet's inner workings and evolution. As a broader impact this project will publish earthquake wave animations online in near real-time for events with magnitudes of 5.5 and greater at Princeton University's Near Real Time Seismicity Portal (GlobalShakeMovie) .The main goal of this research is to further develop global seismic tomography using complete seismic waveforms and adjoint-state methods by enabling massive data assimilation and harnessing the largest and fastest available computers. Using a pilot dataset of 253 earthquakes, a first-generation global tomographic model based on "adjoint tomography", an iterative full-waveform inversion technique was built. Based on this experience, a next-generation global model will be constructed by assimilating full waveform data from thousands of earthquakes that were well recorded by broadband global and regional seismographic networks. Prior NSF support demonstrated the feasibility of regional adjoint tomography for isotropic heterogeneity, anelasticity, and azimuthal anisotropy. Similar inversions on a global scale will be done through secured access to the necessary computational hardware through the Department of Energy INCITE program. This research supports a portion of the required human resources, where the rate-limiting steps in this ambitious endeavor are data quality control, I/O, and workflow management. Four main tasks will be tackled bearing these obstacles in mind: 1) prepare and further develop the open source spectral-element seismic wave propagation solvers SPECFEM3D and SPECFEM3D_GLOBE for exascale simulations; 2) further develop the Adaptable Seismic Data Format (ASDF) for fast I/O; 3) further develop the inversion & migration toolkit SeisFlows by taking advantage of the open source Structured Query Language database engine SQLight and existing workflow management tools, such as Pegasus; and 4) perform global adjoint seismic tomography with a database of thousands of earthquakes at a shortest period of 9 s. Task 4 encompasses the main goal of the proposal, but it cannot be completed without simultaneously addressing the other tasks.