Using machine learning method to detect slow slip events in ocean bottom pressure data

该项目旨在提高对俯冲带地震和海啸的认识。在海底可以测量到的特殊构造信号可能代表了俯冲带构造应力的释放。如果是这样，海底压力传感器的测量结果可以用来估计地震和海啸的风险。然而，来自海洋过程的噪声使得准确检测该信号变得困难。该项目将利用计算机技术——机器学习的最新进展，开发一种更好的这种信号探测器。该项目将支持STEM领域的早期职业科学家和弱势群体（拉丁裔和女性）。它还将资助一名研究生和几名本科生。该项目将在研究生、本科、高中和初中阶段开发机器学习的教学模块。本项目将在项目完成后立即在公共领域发布代码并在社区内共享教学模块。浅层慢滑事件为俯冲带浅层应变释放提供了一种机制，对海啸灾害评价具有重要意义。对于大多数俯冲带，海沟远离海岸，不清楚是否存在浅层慢滑事件。即使在检测到这些事件的地方，持续时间和震级等关键数量也没有得到很好的限制。因此，浅层俯冲带的锁定状态和浅层慢滑事件的发生机制尚不清楚。为了回答这些问题，该项目将利用机器学习的最新进展和海底压力数据集的积累来提高我们探测俯冲带浅层慢滑事件的能力。对新西兰海底压力数据的初步分析表明，机器学习可以成功识别已知的慢滑事件，并进一步降低海底压力数据中的海洋噪声。利用来自几个俯冲带的可用数据，该项目将进一步改进机器学习检测器，以估计浅层慢滑事件的持续时间、幅度和时间。该项目还将开发一种改进的方法，通过使用机器学习来捕获海洋中可测量量的复杂关系，减少海底压力数据中的海洋噪声。总的来说，该项目将为测量浅层慢滑事件和评估浅层俯冲带的锁定状态提供更好的工具。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This project seeks to improve the understanding of earthquakes and tsunamis in subduction zones. Special tectonic signals that can be measured at the seafloor may represent the release of tectonic stress in subduction zones. If so, measurements from pressure sensors on the seafloor could be used to estimate earthquake and tsunami risks. However, noise from ocean processes makes it difficult to detect this signal accurately. This project will take advantage of recent advances in a computational technique, machine learning, to develop a better detector of this signal.. This project will support early career scientists and people from underrepresented groups (Latino and Female) in STEM fields. It will also support a graduate student and several undergraduates. This project will develop teaching modules of machine learning at the graduate, undergraduate, high school, and middle school levels. This project will publish code in the public domain and share the teaching modules within the community immediately after the project finishes. Shallow slow slip events provide a mechanism for strain release at the shallow part of subduction zones, which is important for tsunami hazard assessment. For most subduction zones, the trench is far from the coast and it is unclear whether shallow slow slip events exist. Even in places where these events were detected, key quantities such as the duration and magnitude were not well constrained. As a result, the locking state of shallow subduction zones and the mechanism of shallow slow slip events is still unclear. To answer these questions, this project will take advantage of recent advancement in machine learning and the accumulation of seafloor pressure datasets to improve our ability to detect shallow slow slip events in subduction zones. Preliminary analyses of seafloor pressure data from New Zealand have demonstrated that machine learning can successfully identify known slow slip events and further reduce ocean noise in seafloor pressure data. Using available data from several subduction zones, this project will further improve the machine-learning detector to estimate the duration, amplitude, and timing of shallow slow slip events. This project will also develop an improved way to reduce ocean noise in seafloor pressure data by using machine learning to capture the complex relationship of measurable quantities in the ocean. Collectively, this project will provide better tools to measure shallow slow slip events and assess the locking state of shallow subduction zones.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.