Doing tomography differently: building the imaging tools of tomorrow

基于弹性波的直接观测无法获得的成像是一个具有高社会和经济影响的广泛应用的重大问题。在这个项目中，我们的目标是大幅提高地震层析成像的分辨率，以便在两个具有高度社会和经济利益的领域产生增强的精细分辨率图像：具有前所未有的分辨率的区域层析成像和石油工业被动地震成像。我们超越了传统的被动成像方法，如环境噪声层析成像或接收函数偏移，通过对地球浅层或深层进行高频全波形成像，帮助调查大陆造山带的深层根源或大地震的延伸断层源。为了实现这一目标，我们将成像技术扩展到高频，并针对高性能计算中高度非结构化的不规则问题推导出数据驱动的模拟方案和新技术。我们建议的开创性步骤之一是放弃波传播模型中的典型近似，而包括剪切波的全部贡献。在最近的初步但有希望的结果中，我们已经表明，这有时可以导致局部分辨率提高十倍。为此，我们将共同解决两个技术差距：开发改进的混合计算技术，以及使用精心设计和调整的高性能数据驱动计算方法来解决非结构化和/或不平衡问题。它们的结合使用是一个创新的概念，为这两个目标应用的研究开辟了一条新的途径。我们的方法可以被视为对伴随反演技术的补充，并提供了一个重要的优势：通过将反演限制在区域层析成像盒中，我们的工具将不需要领先级别的机器。因此，我们设计了具有质量控制指标的灵活和半自动的反转工作流，以便社区可以在对Tier-2类机器的日常研究中使用它们。为了实现这一目标，数据驱动的高性能计算技术至关重要，特别是对涉及到的海量数据的编排、编组和协调。总而言之，我们在数据分析和高性能计算之间架起了一座桥梁。预期的科学影响很高，因为我们在成像方法领域提出了一种新的范式。它将把这些断层成像问题带到物理分辨率上，这是以前从未达到过的。如果成功，该提议也可能成为一项突破，对工业应用产生中期影响，因为高精度的断层图像对于能源资源的勘探至关重要，例如在石油行业。预期的社会影响也很大，因为我们将帮助发展新的能力，以监测人口稠密地区，评估其安全性，并研究大地震的来源。

Imaging what is inaccessible to direct observation, based on elastic waves, is a major issue with a wide range of applications of high societal and economical impact. In this project we aim at drastically improving the resolution of seismic tomography to produce enhanced finely-resolved images in two domains with high societal and economical interests: Regional tomography at unprecedented resolution, and oil industry passive seismic imaging. We go beyond classical passive imaging approaches such as ambient noise tomography or receiver function migration, by performing high-frequency full waveform imaging of the shallow or deep Earth, to help investigate the deep roots of continental orogens or the extended fault sources of large earthquakes. To achieve this goal, we extend our imaging techniques to high frequencies, and derive data-driven simulation schemes and novel techniques for highly unstructured irregular problems in high-performance computing. One of the ground-breaking steps that we propose is to abandon typical approximations in wave propagation models, and to include the full contribution of shear waves. In recent preliminary but promising results we have shown that this can sometimes lead to a ten-fold resolution increase locally. To do so we will address two technological gaps jointly: developing improved hybrid calculation techniques, and using well designed and tuned high-performance data-driven computing approaches for unstructured and/or imbalanced problems. Their combined use is an innovative concept, which opens a new avenue of research in the two targeted applications. Our approach can be seen as complementary to adjoint inversion techniques, and offers an important advantage: By restricting the inversion to regional tomographic boxes, our tools will not require leadership-class machines. We thus design flexible and semi-automated inversion workflows with quality-control metrics, so that the community can use them in daily research on Tier-2 class machines. To achieve this goal, data-driven high-performance computing techniques are essential, in particular the orchestration, marshalling and coordination of the involved huge amounts of data. In summary, we bridge the gap between data analytics and high-performance computing.The expected scientific impact is high, because we propose a new paradigm in the field of imaging methods. It will bring these tomographic problems to physical resolutions that have never been accessible before. If successful, the proposal can also be a breakthrough with mid-term consequences on industrial applications because highly-accurate tomographic images are of crucial importance for exploration of energy resources e.g. in the oil industry. The expected societal impact is also high because we will help develop new capabilities to monitor populated areas to assess their safety and also study the source of large earthquakes.