Stochastic Models and Statistical Methodology for Marked Spatio-Temporal Point Processes with Applications to Wildland Fire Management

目标：1)开发影响野地火灾科学和管理的统计工具、方法和模型；2)培训总部总部，强调协作、跨学科、基于团队的数据科学；3)向科学界传播森林统计、环境计量学、野地火灾科学和管理方面的进展；4)通过提供概念框架、软件和决策支持系统工具，将进步转化为火灾管理。 方法：我的研究是由对大型、复杂的标记时空点过程和火灾天气数据集的科学研究推动的。主题包括： 数据可视化：将开发时空点图案及其标记的可视化工具，以探索关键问题。(例如：烧伤如何降低着火风险，以及这种风险如何消散？火的寿命在空间和时间上是如何变化的？点火的主要聚集性是由于热点还是随机聚集性？如何在决策支持工具中最好地可视化基于模型的信息？) 火灾发生：将开发细尺度的火灾发生时空预测模型，用于火灾的存在/不存在、计数和大型逃生火灾。新的设计方案将创建更相关的风险暴露衡量标准，并导致提高统计效率。将创建方法论，以开发大空间范围的模型。在监测气候变化的影响时，探测有效性的变化将被量化，并用于减少混杂影响。 火灾持续时间：将开发单个火灾寿命的模型和大片地形上火灾生存时间的空间模式。这些模型将描述火灾生命周期的关键时期(例如，从发现到报告的时间、最初的攻击逃生时间、扑灭时间)。将汇编一个具有历史火灾时变协变量的大型数据集，并将为火灾管理创建预测未来时空寿命的新模型。 随机模型：将开发时空点过程集群模型来模拟火灾到达。标记将与这些和/或其他点火模型耦合，以模拟火灾状况特征，如燃烧面积或火灾负荷(景观上活跃的火灾数量)。 决策支持系统：上述主题的组成部分将被纳入基于模型的决策支持工具，这将更好地为火灾管理提供信息。 影响：拟议的组成部分将带来重大进展，进一步加深对野地火灾的科学理解。学生将在跨学科、协作的团队环境中接受现代统计方面的培训。这项研究将影响火灾科学和火灾管理、统计学和环境计量学、运筹学，并产生影响保险和自然资源部门的技术。

OBJECTIVES: 1) Develop statistical tools, methodology and models that impact wildland fire science and management; 2) Train HQP in modern statistical methods emphasizing collaborative, interdisciplinary, team-based data science; 3) Disseminate advancements in statistics, environmetrics, wildland fire science and management to the scientific community; 4) Transfer advancements to fire management by providing conceptual frameworks, software and decision support systems tools.  APPROACH: My research is motivated by the scientific study of large, complex marked spatio-temporal point-process and fire-weather data sets. Topics include: Data Visualization: Visualization tools for spatio-temporal point patterns and their marks will be developed to explore key questions. (Eg: How does a burn reduce ignition risk and how does this dissipate? How do fire lifetimes vary spatially and temporally? Is the clustering of ignitions due to hot spots or stochastic clustering? How can one best visualize model-based information in decision support tools?) Fire Occurrence: Fine-scale spatio-temporal fire occurrence prediction models for the presence/absence, for counts, and for large escaped fires will be developed. New design schemes will create more relevant measures of exposure and lead to gains in gain statistical efficiency. Methodology will be created to develop models over large spatial extents. Changes to detection effectiveness will be quantified and used to reduce confounding effects when monitoring impacts of climate change. Fire Duration: Models for individual fire lifetimes and for spatial patterns in fire survival times over large landscapes will be developed. These will characterize the key epochs of a fire's lifetime (eg, time from detection to report, initial attack getaway time, suppression time). A large data set with time-varying covariates for historical fires will be compiled and novel models that forecast future lifetimes over space-time will be created for fire management. Stochastic Models: Spatio-temporal point process cluster models will be developed to model fire arrivals. Marks will be coupled to these and/or the other ignition models to model fire regime characteristics, such as area burned or the fire-load (the number of fires active on the landscape) over space-time.  Decision Support Systems: Components from the above set of topics will be incorporated into model-based decision support tools which will better inform fire management. IMPACTS: The proposed components will lead to significant advances which further the scientific understanding of wildland fires. Students will be trained in modern statistics in an interdisciplinary, collaborative team setting. This research will impact fire science and fire management, statistics and environmetrics, operations research, and produce technology that impacts insurance and natural resources sectors.