Machine Learning for Geophysical Hazard Sequences

许多地球物理灾害(火山喷发、大地震、山体滑坡)在它们之前都有一系列可观察到的事件。专家们经常使用这些前兆进行预测，但许多前兆直到危险事件发生后才被识别出来。随着ML技术的普及和能力的提高，它们越来越多地应用于各种问题。随着人口增长和气候变化，更多的人面临极端地球物理灾害的风险。这给科学家和当局带来了压力，要求他们设计和实施更有效的缓解行动。预测是缓解战略的重要组成部分。现在是研究ML技术可以带来的好处和见解的时候了。学生将通过文献回顾和与监管团队内外经验丰富的科学家的讨论，熟悉地质灾害监测数据、仪器网络和事件序列。他们将习惯使用ML算法，例如：用于识别模式的时间序列分类和神经网络；用于识别最重要数据的降维算法；以及用于评估最佳监测网络的集成方法。学生将用历史数据集训练算法，并将其与专家意见进行比较。算法将成为社区的重要资源。所有输出都将是新数据。该项目的风险很低。GNS Science拥有世界上最先进和最全面的监测网络之一，再加上新西兰丰富的地质灾害事件，提供了丰富的数据。所有数据均可从geonet.org.nz获得。在这一领域已经有了成功使用ML的例子，所有的结果都将是新颖和有趣的，例如，如果项目确认某个已知模式对预测有用，找到一个新模式，或者建议没有有用的模式，这些结果将对该学科和风险缓解有用。该项目包括来自四个不同学校/研究所的监督员。这是该项目的一个优势，因为学生将向一系列专家学习，体验不同的工作和学习环境，以及与更广泛的专业人员进行接触。这名学生将以UEA ENV为基地，在那里他们将成为一个强大的地质灾害小组的一部分。学生将从前沿、多样化和跨学科的地质灾害研究中受益。学生还将参加督导小组中其他机构的研究小组活动，进一步扩大他们的支持网络。在UEA CMP，他们将接受ML技术的设计、实施和使用方面的培训。该项目将包括在新西兰GNS Science的时间，在那里学生将体验在政府机构工作，实时监测地球物理灾害，并向拥有毕生默契的专家学习。主管代表项目各个方面的专业知识，为学生提供探索和专门研究项目中他们最感兴趣的部分的机会。督导团队的多学科性质将为学生提供跨学科边界思考的能力，为他们未来的职业生涯做好准备。

Many geophysical hazards (volcanic eruptions, large earthquakes, landslides) have sequences of observable events that precede them. Experts often use these precursors to make forecasts, but many precursors are not recognised until after the hazardous event. As ML techniques grow in popularity and capability, they are increasingly applied to a diverse range of problems. With population growth and climate change, more people are at risk from extreme geophysical hazards. This puts pressure on scientists and authorities to devise and implement more effective mitigation actions. Forecasting is a major part of mitigation strategies. It is time to investigate the benefits and insights that ML techniques can bring. The student will familiarise themselves with geohazard monitoring data, instrument networks and event sequences through literature review and discussions with experienced scientists within and beyond the supervisory team. They will become accustomed to ML algorithms such as: time-series classification and neural networks to identify patterns; dimensionality reduction algorithms to identify the most important data; and ensemble methods to evaluate the best monitoring networks. The student will train algorithms with historic data sets and compare these to expert opinions. Algorithms will be important resources for the community. All outputs will be new data. Risks for the project are low. GNS Science has one of the most advanced and comprehensive monitoring networks in the world, coupled with the abundance of geohazardous events in NZ, provides an abundance of data. All data are available from geonet.org.nz. There are already examples of successful ML use in this area, and all results will be novel and interesting, e.g. If the project confirms that a certain known pattern is useful for forecasting, finds a new pattern, or suggests that there are no useful patterns, these results will be useful for the discipline, and for risk mitigation. This project includes supervisors from four different schools/institutes. This is a strength of the project as the student will learn from a range of experts, and experience different working and learning environments, as well as making contacts with a wider range of professionals. The student will be based in UEA ENV, where they will be part of a strong geohazards group. The student will benefit from cutting-edge, diverse, and cross-disciplinary geohazards research. The student will also take part in research group activities in the other institutions included in the supervisory team, broadening their support network further. At UEA CMP they will be trained in the design, implementation, and use of ML techniques. The project will include time at GNS Science, NZ, where the student will experience working within a government agency, real-time monitoring of geophysical hazards, and learn from experts who have lifetime's' worth of tacit knowledge.The supervisors represent expertise from all aspects of the project, providing the student with opportunity to explore and specialise in the parts of the project that interest them most. The multidisciplinary nature of the supervisory team will provide the student with the ability to think across discipline boundaries, preparing them for any future career.