CAREER: Scalable Computational Seismology for All

计算地震学方法分析测量地球振动的数据。这些方法使科学家能够了解地震危害，测量建筑物下地面的稳定性，监测地下水系统，研究受到威胁的地球系统(如冰川和永冻土)的变化，安全有效地勘探地下自然资源，以及监测民用基础设施的健康状况等应用。地震学最近经历了根本性的转变；新的传感器技术使数据收集变得容易得多，使更大的数据集能够用于更大区域的详细研究，并可在很长一段时间内使用。大多数科学家无法使用这些数据，因为：(1)数据仅在拥有新传感器的小组之间内部共享，(2)公共地震学数据存储设施无法支持如此大的数据量，以及(3)大多数地球科学家没有分析数据的计算资源。由于这三个问题，研究环境不公平，许多数据仍未开发，重要的地球科学发现无法出现。虽然正在努力解决第一个问题，但如果没有解决第二个和第三个问题的重大网络基础设施进展，新获得的数据不太可能得到充分分析。该项目旨在创建开放数据共享的新计算算法、软件和模型，以确保任何地球科学家都能从大规模地震学数据中收集有价值的见解。教育和推广计划将为更多的人创造机会，参与科学和工程应用的数学建模和大规模数据分析。PI项目将发展和加强现有的努力，以支持多样化和包容性的研究和学习环境。她将继续开发一个项目，向女性本科生介绍数学研究，使其成为一门可持续的多教职员工课程，服务于更多来自代表性不足群体的学生。该项目将增加由PI领导的年度数据科学会议的影响。这次会议以女性数据科学家的研究和现代数据科学技术的教程为特色，并将农村校园中的跨学科数据科学社区联系起来。该项目将推导和分析新的地球科学算法，开发社区软件，并探索开放数据分发的模式。该项目的目标是确保任何地震学家都能从现场、计算资源有限的机构和现代高性能计算(HPC)系统的极端规模地震数据中获得宝贵的地球物理见解。大规模地震传感、数学、高通量计算科学和算法设计方面的专业知识是实现这些进步所必需的。该项目提出了一种公共地震学数据存档的新模式，允许存储有损压缩数据和数据产品，从而创造了一种新的能力来存储超高密度和大规模的地震数据，而不会取代现有的高质量地震仪数据系统。为了处理大规模的数据分析，PI之前已经创建了几个可扩展的算法，理论分析表明，可以开发出一套完整的可扩展、可并行化的算法，用于多种类型的被动地震数据处理。许多算法直接对压缩信息进行操作，而不重建原始数据，从而减少了昂贵的数据移动。该项目将开发快速的串行和并行软件算法实现，并研究使用加速器硬件来提高计算效率。对于每个算法，该项目将从理论上推导并通过计算验证控制计算效率、内存占用和地球物理分析特定的端到端精度之间权衡的趋势。这些算法将包含现实的非理想化数据的误差界限，并将被包括在预测软件中，供地球科学家在请求压缩数据或数据产品之前做出知情决定。新方法将通过将它们应用于数十至数百TB规模的尖端被动地震数据来进行测试。这些数据将使城市水文学和岩土工程中的地震学分析成为可能，也将有助于通过改善边界条件和对机理的理解来帮助理解冰川运动以改进气候模型。除了地震地震学、水文学、岩土工程和冰冻圈研究外，开发的方法还可以应用于许多使用稀疏或压缩表示法的高通量计算科学问题(例如，结构健康监测、成像科学、太阳物理学、射电天文学、无线通信、工业设施监测)。为了增加地球科学家对新方法的采用，该项目将开发教程，促进科学界的合作，并组织研究工作坊。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Computational seismology methods analyze data that measure vibrations of the Earth. These methods allow scientists to understand earthquake hazards, to measure stability of the ground underneath structures, to monitor groundwater systems, to study changes in threatened Earth systems such as glaciers and permafrost, to safely and efficiently explore natural resources underground, and to monitor civil infrastructure health, among other applications. Seismology has undergone a radical shift recently; new sensor technologies have made data collection much easier, enabling hundreds to thousands of times larger datasets that can be used for detailed studies of larger regions for long periods of time. Most scientists cannot use these data because: (1) data are only shared internally among groups that have new sensors, (2) public seismology data storage facilities cannot support such large data quantities, and (3) most geoscientists do not have the computational resources to analyze the data. Because of these three issues, there is an inequitable research environment, much data remains unexplored, and important geoscience discoveries cannot occur. While there are ongoing efforts to address the first issue, without major cyberinfrastructure advances addressing the second and third issues newly acquired data is unlikely to be fully analyzed. This project aims to create new computational algorithms, software and models of open data sharing to ensure that any geoscientist can glean valuable insights from large-scale seismology data. The education and outreach program will create opportunities for more people to participate in mathematical modeling and large-scale data analysis for science and engineering applications. The project PI will develop and strengthen existing efforts to support diverse and inclusive research and learning environments. She will continue to develop a program to introduce women undergraduates to mathematics research, growing it to be a sustainable multi-faculty course serving more students from underrepresented groups. The project will increase the impact of the annual data science conference led by the PI. The conference features research by women data scientists and tutorials on modern data science techniques, and connects the interdisciplinary data science community on a rural campus.The project will derive and analyze new geoscience algorithms, develop community software and explore models of open data distribution. The project goal is to ensure that any seismologist can gain valuable geophysical insights from extreme-scale seismic data in the field, at institutions with limited computing resources, and on modern high performance computing (HPC) systems. Expertise in large-scale seismic sensing, mathematics, high-throughput computational science, and algorithm design are necessary to achieve these advances. The project proposes a new model for public seismology data archives that allows for the storage of lossy-compressed data and data products, thus creating a new capacity to host ultra-high-density and large-scale seismic data, without displacing existing systems for high-quality seismometer data. To address large-scale data analysis, the PI has previously created several scalable algorithms, and theoretical analyses suggest that a complete suite of scalable, parallelizable algorithms for multiple types of passive seismic data processing can be developed. Many of the algorithms operate directly on compressed information without reconstructing the original data, which reduces costly data movement. The project will develop fast serial and parallel software algorithm implementations, and investigate the use of accelerator hardware for high computational efficiency. For each algorithm the project will theoretically derive and computationally verify trends governing tradeoffs between computational efficiency, memory footprint, and end-to-end accuracy specific to the geophysical analyses. The algorithms will incorporate error bounds for realistic non-idealized data and will be included in predictive software for geoscientists to make informed decisions prior to requests for compressed data or data products. The new methods will be tested by applying them to cutting-edge passive seismic data at the scale of tens to hundreds of terabytes. The data will enable seismology analyses in for urban hydrology and geotechnical engineering, and also analyses to aid in understanding glacier movements to improve climate models via improved boundary conditions and mechanistic understanding. In addition to earthquake seismology, hydrology, geotechnical engineering, and cryosphere studies, the methods developed can be applied to many high-throughput computational science problems utilizing sparse or compressed representations (e.g., structural health monitoring, imaging science, solar physics, radioastronomy, wireless communications, industrial facility monitoring). To increase adoption of new methods by geoscientists, the project will develop tutorials, promote scientific community collaboration, and organize research workshops.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.