Machine Learning for Real Time Volcano-Seismic Monitoring.

用于实时火山地震监测的机器学习。

• 批准号: 2843375
• 金额: --
• 依托单位: Birkbeck, University of London
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2023
• 资助国家: 英国
• 起止时间: 2023 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2843375
• 关键词: Machine; Learning; Real; Time; Volcano

Seismic data analysis is central to most volcano monitoring operations, giving insight into the internal structure and physical processes occurring at depth and enabling the identification of potential eruption precursors which are not visible form the surface. A critical challenge in seismic monitoring is detecting seismic signals (events) from background noise and classifying them based on the physical mechanisms that generate them. At most observatories this classification is still undertaken manually by teams of analysts, making it unfeasible to generate comprehensive volcano seismic catalogues in real-time during periods of unrest when 1000's of seismic events can occur each day. Machine learning (ML) and deep learning (DL) methods have received much attention in recent decades for addressing such 'big data' problems, demonstrating remarkable abilities to rapidly extract patterns and classify data in an automated fashion with high accuracy. However, current applications of ML to address volcano seismic event detection and classification are limited by several factors including their inability to maintain high accuracy over time and generalize between locations. This PhD will investigate novel approaches to develop a more robust generalized ML model for volcano seismic event detection and classification which is grounded in geophysical principles, including the development of a multi-volcano training and benchmarking datasets, and physics informed augmentation and synthetic data production.

地震数据分析是大多数火山监测工作的核心，它使人们能够深入了解火山内部结构和发生在深处的物理过程，并能够确定从地表看不到的潜在喷发前兆。地震监测中的一个关键挑战是从背景噪声中检测地震信号（事件），并根据产生它们的物理机制对其进行分类。在大多数观测站，这种分类仍然是由分析人员手工进行的，因此在每天可能发生1000起地震事件的动荡时期，无法实时生成全面的火山地震目录。近几十年来，机器学习（ML）和深度学习（DL）方法因解决此类“大数据”问题而受到广泛关注，表现出快速提取模式并以高精度自动化方式对数据进行分类的卓越能力。然而，目前ML在火山地震事件检测和分类方面的应用受到几个因素的限制，包括它们无法随着时间的推移保持高精度并在位置之间进行概括。这个博士将研究新的方法来开发一个更强大的广义ML模型，用于火山地震事件检测和分类，该模型基于地球物理原理，包括开发多火山训练和基准数据集，以及物理信息增强和合成数据生产。