Hazard forecasting in real time: from controlled laboratory tests to volcanoes and earthquakes

地震和火山爆发等脆性破坏事件的固有可预测性是重要的，未知的，并且备受争议。我们将建立技术，以确定在受控实验室测试的理想情况下脆性破坏的预测能力，使用NERC已经资助的一系列实验的输出数据，以确定缓慢变形下岩石的流变学。我们将利用信息学的最新发展，在预期模式下，即在故障发生之前，实现可验证的故障预测能力。这很重要，因为后见之明的好处在评估回顾性测试的可预测性方面提供了显著的积极偏差。有了这些经验，我们将把类似的技术应用于自然系统，在更大的空间和时间尺度上量化一个不受控制的、更复杂的系统中可预测性的损失。一个主要的技术目标是开发一个开放获取、自动化、基于网络的平台，用于实时数据整理、分析和信息交换，使相互竞争的物理假设和统计方法能够在一个开放、可测试的环境中以完全前瞻性的模式进行测试和开发，例如可与每日天气预报相媲美。这将需要在用户友好的高性能计算环境中对数据应用最先进的统计方法，包括对模型不确定性及其对预测一致性和质量的影响进行正式量化。为确保所产生的技术切实可行，并在灾害规划和减轻风险方面正式提供使用价值，将与最近的全球地震预报倡议、监测观测站和负责发布地震和火山事件警报的民防机构合作开发这些技术。这些结果将提高我们对实验室和地球上控制物质失效的物理过程的理解，并将为严格的、全概率的火山爆发和地震预测提供一个可持续的、基于经验的工具。

The inherent predictability of brittle failure events such as earthquakes and volcanic eruptions is important, unknown, and much debated. We will establish techniques to determine the forecasting power for brittle failure in the ideal case of controlled laboratory tests, using output data from a series of experiments already funded by NERC to determine the rheology of rocks under slow deformation. We will use recent developments in informatics to enable a capability for verifiably forecasting failure in prospective mode, i.e. before it has occurred. This is important because the benefit of hindsight provides a significant positive bias in evaluating the predictability in retrospective tests. With this experience, we will then apply similar techniques to natural systems to quantify the loss of predictability in an uncontrolled, more complex system at greater spatial and temporal scales. A major technical aim is to develop an open-access, automated, web-based platform for real-time data collation, analysis and information exchange, enabling competing physical hypotheses and statistical methods to be tested and developed in fully prospective mode in an open, testable environment comparable, say, to daily weather forecasts. This will require applying state-of-the art statistical methods to the data in a user-friendly, high-performance computing environment, including formal quantification of model uncertainties and their effect on forecast consistency and quality. To ensure that the resulting techniques are practicable and formally provide value for use in hazard planning and risk mitigation, they will be developed in collaboration with recent global earthquake forecasting initiatives, monitoring observatories and civil defence agencies responsible for issuing alerts on seismic and volcanic events. The results will improve our understanding of the physical processes controlling material failure in the laboratory and in the Earth, and will provide a sustainable, experience-based tool for rigorous and fully-probabilistic forecasting of volcanic eruptions and earthquakes.