Data-Driven Wildland Fire Science with Applications to Fire Management Systems

野火是一个全球性的问题，除了一个大洲外，所有其他大洲都在发生。我提出的研究重点是野地火灾，以支持加拿大野地火灾管理。具体地说，我请求NSERC探索赠款基金来研究加拿大野地火灾制度的关键特征，这些特征可以概括为三个主题。1.在很大程度上(省级或更大规模)改进，精细的空间和时间上明确的野地火灾发生预测；2.模拟野地火灾寿命；3.建立火灾负荷的随机时空模型，即从管理的角度来看，可以在不同的空间尺度上，如在区/部门、区域、省甚至国家层面上查看给定时间点的活跃野地火灾的数量。主题1和主题2提供给主题3，因为它们研究是什么驱动了火灾在时空中的发生和生存，这是理解火灾负荷的关键。主题1和主题2涉及更好地了解是什么驱使整个地区每天发生火灾，以及是什么驱使火灾在扑灭之前持续多长时间(自然或由于灭火努力)。主题1将调查机器学习技术(统计和基于算法)的使用，以加强我们对火灾如何在时空中到达的了解，为适当评估和比较火灾发生的数据驱动模型制定指导方针，重点是在如何使用这种模型为火灾管理行动提供信息的背景下做到这一点，并研究整个加拿大火季的时间是如何变化的。主题2涉及从生存分析到荒地火灾寿命特征的方法的开发和应用，包括火灾寿命的连续组成部分可能如何相关。要调查的问题包括：探测延迟或调度延迟如何影响火灾的未来寿命？是什么驱使灭火事件发生？还有，有没有地方可以修改消防管理工作，以改进关键的绩效衡量标准，如初始攻击努力的成功率。主题3将结合这些其他主题的进展，开发火灾负荷如何随时空变化的模型。这项工作涉及数据科学和分析工具的开发和应用，因为这项研究涉及对大型和复杂的时空数据集的分析。这些数据集将通过融合来自各种来源的数据进行汇编。这项研究将解决加拿大林业局最近出版的《加拿大荒地火灾科学蓝图(2019-2029年)》中确定的差距。与消防管理人员的直接协作将增加对最终用户的知识传授，并产生用于消防管理信息系统和决策支持的实用工具。

Wildland fire is a global problem, occurring globally on all but one continent. My proposed research focuses on wildland fire to support Canadian wildland fire management. Specifically, I am requesting NSERC Discovery Grant funds to study key characteristics of Canadian wildland fire regimes that can be summarized into three themes. 1. Enhancements to large extent (provincial or larger), fine-scale spatially and temporally explicit wildland fire occurrence prediction; 2. Modelling wildland fire lifetimes; 3. Developing a stochastic space-time model for fire load, namely the number of active wildland fires at a given point in time which, from a management perspective, can be viewed on a variety of spatial scales such as at the district/sector, regional, provincial or even national level. Themes 1 and 2 feed into theme 3 as they study what drives the occurrence and survival of fires in space time, which are key to understanding fire load. Themes 1 and 2 involve developing an enhanced understanding of what drives fire occurrence on a daily basis across the landscape and what drives how long a fire survives prior to extinguishment (either naturally or due to fire suppression efforts). Theme 1 will investigate the use of machine learning techniques (both statistical and algorithm-based) to enhance our understanding of how fires arrive in space-time, develop guidelines for the appropriate assessment and comparison of data-driven models for fire occurrence with an emphasis on doing so in the context of how such models are used to inform fire management operations, and study how the timing of the fire season is changing across Canada's landscape. Theme 2 involves the development and application of methods from survival analysis to characteristics of wildland fire lifetimes, including how the sequential components of fire lifetimes may be related. Questions to be investigated include the following: How do detection delays or dispatch delays impact the future lifetime of a fire? What drives extinguishment events? And, are there areas where fire management efforts could be modified to improve key performance measures, such as the success rates of initial attack efforts. Theme 3 will combine advancements from these other themes to develop a model for how fire load varies over space-time. This work involves the development and application of data science and analytics tools since this research involves the analysis of large and complex spatio-temporal data sets. Such data sets will be compiled through the fusion of data from a variety of sources. This research will address gaps as identified in the "Blueprint for Wildland Fire Science in Canada (2019-2029)" recently published by the Canadian Forest Service. Direct collaboration with fire management staff will increase the knowledge transfer to end users and lead to practical tools for fire management information systems and decision support.