The statistical design of assessments of impacts from explosive volcanic eruptions

火山爆发产生了深远的影响，特别是火山灰在大气中的扩散及其在地面上的沉积，这可能对农业、基础设施和人类健康产生重大影响。火山灰在大气中的扩散取决于喷发的强度，并由高度可变和难以预测的风场驱动。由于这种喷发事件的罕见，以及缺乏对这种喷发期间足够高的空间和时间分辨率的火山灰浓度的直接观测，对大气中火山灰扩散的预测进一步复杂化。本项目将侧重于火山灰扩散业务评估的统计设计，包括：1.概率火山灰危险评估的统计一致性设计。目前进行概率火山灰危险评估的做法是使用由时间或计算限制设定大小的模拟集合，其结果是各个影响阈值的方差在评估内和评估之间有所不同。这项研究的这一部分将使用基本的统计学方法来标准化概率火山灰危险评估的设计，并探索(1)通过将天气模式作为集合的强迫数据进行分层来减少方差，以及(2)对更大(因此更罕见)的喷发采用适当的统计方法；2.用贝叶斯统计方法来表征估计爆炸性喷发强度和集合预报设计中的不确定性。目前在对火山喷发的业务反应和喷发预报设计方面的做法是使用对喷发羽流高度的观测，通过半经验关系估计质量喷发率。研究的这一部分将使用应用贝叶斯回归方法来表征高度-喷发率半经验关系中的不确定性，从而在总体预报和业务火山灰扩散模型的设计中。

Explosive volcanic eruptions generate far-reaching impacts, in particular the dispersion of volcanic ash in the atmosphere and its deposition on the ground which can have significant effects on agriculture, infrastructure, and human health. The dispersion of ash in the atmosphere is dependent on the magnitude of the eruption, and is driven by wind fields which are highly variable and difficult to forecast. Prediction of atmospheric ash dispersion is further complicated by the rarity of such eruption events, as well as the lack of direct observations of ash concentration at sufficiently high spatial and temporal resolution during such eruptions. This project will focus on the statistical design of operational assessments of volcanic ash dispersion, including:1. Statistically-coherent design of probabilistic volcanic ash hazard assessments. Current practice in producing a probabilistic ash hazard assessment uses an ensemble of simulations with sizes set by time or computing constraints, with the result that the variance of individual impact thresholds vary within and between assessments. This component of the research will use fundamental statistical approaches to standardise the design of probabilistic ash hazard assessments, and explore (i) variance reduction through stratification via the use of weather patterns as forcing data for ensembles, and (ii) appropriate statistical methods for larger (and consequently rarer) eruptions;2. Bayesian statistical methods to characterise uncertainty in estimation of the intensity of explosive eruptions and ensemble forecast design. Current practice in operational response to volcanic eruptions and in eruption forecast design uses observations of the eruption plume height to estimate the mass eruption rate through a semi-empirical relationship. This component of the research will use applied Bayesian regression methods to characterise the uncertainty in the height-eruption rate semi-empirical relationship, and thus in the design of ensemble forecasts and operational ash dispersion modelling.