Influenza Outbreak Prediction: Applying Data Assimilation Methodology to Make...

描述（由申请人提供）：将开发一种用于生成流行性流感短期（5 天至 3 个月）基于集合的预测的系统。为了对流感感染结果进行熟练的预测，该系统将应用最先进的数据同化技术，类似于数值天气预报中使用的技术，将流感感染的实时估计纳入流感传播动力学的数学模型中。拟议的工作将建立一个便携式、与当地相关的业务疾病预测系统，该系统在定量、统计上严格的方法方面是新颖的。该系统之所以成为可能，是因为最近出现了实时、基于网络的流感感染率估计，并且存在经过观察验证的流感传播动力学模型。该项目的目标是设计一个基于集合的模型/数据同化系统，将这些信息资源汇集在一起​​，以创建对流感结果的熟练的概率预测。预测系统将使用称为集合调整卡尔曼滤波器的同化技术来开发。要回答的问题包括：基于网络的流感估计对于初始化和约束流感数学模型有多大用处？这种模型/数据系统可以在每周和每月的交货时间内提供什么水平的可预测性？流感爆发高峰期人群中流感发生的时间和水平的不确定性界限是什么？可以在季节的早期评估这些指标？这些问题的答案将决定模型/数据同化系统在不同时间尺度上可以提供的可预测性水平。其他学科领域的工作表明，使用集合调整卡尔曼滤波器开发的模型/数据同化系统是可行的，优化模型行为以更好地匹配观测结果，并为量化可预测性提供严格的框架。据推测，对当地流感风险的熟练预测将在一定的准备时间内实现。将明确确定可预测范围，并对当地流感风险进行预测。 公共卫生相关性：仅在美国，流感每年就会夺去 35,000 人的生命，给全球公共卫生带来巨大负担。基于实时数据和环境强迫的本地短期预测将为公共卫生官员提供制定适合当地的及时干预策略的机会、评估正在发展的流行病严重程度的能力以及获得额外医疗和公共卫生资源的时间

DESCRIPTION (provided by applicant): A system for generating short-term (5-day to 3 month) ensemble-based predictions of epidemic influenza will be developed. To make skillful forecasts of influenza infection outcomes, this system will apply state-of the-art data assimilation techniques, similar to those used in numerical weather prediction, to incorporate real-time estimates of influenza infection into mathematical models of influenza transmission dynamics. The proposed work will establish a portable, locally relevant operational disease forecast system that is novel in its quantitative, statistically rigorous approach. This system is possible due to the recent advent of real-time, web-based estimates of influenza infection rates and the existence of observationally validated models of influenza transmission dynamics. The aim of this project is the design of an ensemble-based model/data assimilation system that brings these informational resources together to create skillful, probabilistic forecasts of influenza outcomes. The forecast system will be developed using an assimilation technique called the ensemble adjustment Kalman filter. Questions to be answered include: How useful are web-based influenza estimates for initializing and constraining mathematical models of influenza? What levels of predictability can such model/data systems deliver at weekly and monthly lead-times? What are the uncertainty bounds on the timing and level of influenza in a population at the height of an outbreak, and how early in the season can these metrics be evaluated? Answers to these questions will determine the levels of predictability the model/data assimilation system can deliver at various time scales. Work in other discipline fields has demonstrated that model/data assimilation systems developed using the ensemble adjustment Kalman filter are practicable, optimize model behavior to better match observations, and provide a rigorous framework for quantifying predictability. It is hypothesized that skillful prediction of local influenza risk will be realized over a range of lead times. The limits of predictability will be explicitly detemined, and forecasts of local influenza risk will be made. PUBLIC HEALTH RELEVANCE: Influenza kills an estimated 35,000 people each year in the United States alone and presents an enormous burden on worldwide public health. Local, short-term predictions based on real-time data and environmental forcing would provide public health officials the opportunity to develop locally appropriate, timely intervention strategies, the ability to gauge the severity of the developing epidemic, and the time to garner additional medical and public health resources