Anticipating Extreme Hurricane Winds in the United States Using Bayesian Hierarchical Models

PI将分析和模拟沿海飓风活动的可用记录（1）提供更准确和相关的估计美国极端飓风的概率在年度到千年的时间尺度，（2）开发计算方法来分析和模拟极端气候事件的变化。建模和分析工具将来自极值理论和贝叶斯推理方法。数据将来自现代记录、整理的历史记录和地质沉积物岩心。沿海飓风是美国严重的社会和经济问题。强风、暴雨和风暴潮造成人员死亡和财产损失。飓风造成的破坏可与地震相媲美。PI将通过将分析纳入贝叶斯框架，推进极值理论在分析气候数据中的应用。这项研究将为飓风风险评估问题提供尖端科学。为实现赠款目标而采取的步骤包括：（1）为19世纪上半叶的飓风登陆事件分配概率，（2）以气候指数值为条件，生成沿海地区年飓风风速过度分布的样本，（3）将贝叶斯预测概率与其他飓风风模型的概率进行比较，（4）检查将空间信息加入风速模式的效用，（5）使用国家大气研究中心/国家环境预测中心再分析数据检查厄尔尼诺-南方涛动和北大西洋涛动沿海飓风联系，（6）开发极端飓风活动地质记录校正软件。总体目标是为气候研究创建一套可访问的分析和建模工具，这些工具利用贝叶斯模拟技术。贝叶斯模拟技术尚未渗透到气候分析和预测界。这项研究的目标之一是扩大这项新技术，使气候界能够使用它。更广泛的影响将包括加强科学，以更好地评估面对气候变化的自然灾害风险。PI将通过在互联网上传播数据、模型和预测来培训研究生并促进公众的气候教育。结果将是相关的私人（再）保险业和政府资助的应急管理。通过推进和促进思维的转变，从使用计算机生成分析解决方案到使用计算机生成解决方案空间的样本（模拟），预计将对气候科学产生重大影响。结果将挑战并最终改善气候和相关科学中的统计教学方式。目标是使贝叶斯模拟成为气候科学分析和预测的常规组成部分。该项目由气候多变性和可预测性方案（CLIVAR）资助。

The PI will analyze and model available records of coastal hurricane activity (1) to provide more accurate and relevant estimates of the probability of extreme hurricane winds in the United States on annual to millennial time scales, and (2) to develop the computational means to analyze and simulate the variability of extreme climate events. The modeling and analysis tools will derive from extreme value theory and methods of Bayesian inference. Data will come from modern records, collated historical accounts, and geological sediment cores. Coastal hurricanes are a serious social and economic concern to the United States. Strong winds, heavy rainfall, and storm surge kill people and destroy property. Destruction from a hurricane rivals that from an earthquake. The PI will advance the use of extreme value theory in analyzing climate data by putting the analysis into a Bayesian framework. The research will provide cutting-edge science to the problem of hurricane risk assessment. Steps to be taken to achieve the grant goals include: (1) Assign probabilities to hurricane landfall events during the first half of the 19th century, (2) Generate samples of annual hurricane wind speed excedence distributions for coastal segments conditioned on values of climate indices, (3) Compare Bayesian predictive probabilities with probabilities from other hurricane wind models, (4) Examine the utility of adding spatial information into the wind speed model, (5) Examine the El-Nino-Southern Oscillation (ENSO) and North Atlantic Oscillation (NAO) coastal hurricane linkages using the National Center for Atmospheric Research (NCAR)/National Centers for Environmental Prediction (NCEP) reanalysis data, and (6) Develop software for calibrating geological records of extreme hurricane activity. The overarching goal is to create an accessible set of analysis and modeling tools for climate research that makes use of Bayesian simulation technology. Bayesian simulation techniques have yet to infiltrate the climate analysis and prediction communities. An objective of the research will be to expand this new technology and make it accessible to the climate community. The broader impacts will include strengthening the science to better assess the risk of a natural disaster in the face of climate change. The PI will train graduate students and promote climate education of the public by disseminating data, models, and forecasts over the Internet. Results will be relevant to the private (re) insurance industry and to government funded emergency management. By advancing and promoting a shift in thinking away from using the computer to generate analytical solutions to using it to generate samples of the solution space (simulations), a large impact on climate science is anticipated. Results will challenge and ultimately improve the way statistics is taught in the climate and related sciences. The goal is to make Bayesian simulation a routine component of analysis and prediction in climate science. This project is supported under the Climate Variability and Predictability Program (CLIVAR).