森林火灾对自然生态、社会经济与公共健康产生了深远影响。尤其是在全球气候变暖与城市化进程的共同作用下，各国政府面临的森林火灾防控形势愈发严峻。充分掌握林火燃烧规律与蔓延行为特征，构建大尺度林火动态蔓延预测模型，对于应急管理部门制定科学高效的林火防控决策至关重要。然而，新形势下森林火灾防控的场景日益复杂化、需求逐渐多样化、应用趋向平台化，现有林火模型较难适应上述变化，在蔓延特征方程、火场空间表征模式、动力学推演机制等建模方法上仍亟待深入研究。因此，本文聚焦应急实战中林火动态蔓延建模问题，采取异构元胞自动机、图网络、深度学习、地理信息学等多学科理论方法，构建了场景多元化、表征精细化、计算快速化和服务标准化的大尺度林火蔓延预测模型集，为应急管理部门提供全面且精准的林火态势分析服务。本文的主要内容与创新点总结如下：（1）针对复杂交界域林火蔓延定量化建模问题，提出了森林-城镇交界域火灾蔓延与火场风险建模方法。建立了交界域异质可燃物的互引燃机制，构建了基于异构元胞自动机的林火蔓延预测模型；设计了耦合致灾因子与避险能力的消防员风险指数，揭示了火场空间中灭火风险的时空分布规律；构建了森林火灾案例数据集及模型性能评价体系，为林火模型验证与性能评估提供数据与标准支持。（2）针对林火蔓延动力学推演方法优化问题，构建了基于图网络与深度学习的林火蔓延预测模型。建立了图边异质性检验准则与“栅格-图”映射机制，揭示了不规则图网络空间中林火蔓延规律与特征，形成了基于深度神经网络的精细化林火蔓延推演模型；提出了蔓延时空分布场概念来定义网络框架，建立了“状态-条件”渐进式编码机制，构建了基于卷积神经网络的林火蔓延预测模型，提供了毫秒级计算效率，并保证了模拟准确性。（3）针对应急实战中模型的高效集成应用问题，建立了林火模型标准化管理与集成应用体系。基于面向服务框架的设计原理，构建了灾害管理平台与林火专题系统，定义了多主体的角色功能与交互流程；结合OGC WPS标准与灾害系统理论，提出了灾害模型服务范式与接口规范，提升了林火模型的复用性、兼容性与互操作性；以城市综合场景为例，演示了林火模型在应急管理系统与平台中实战应用过程及效果，为灾害模型管理与多灾种动态分析提供理论与技术支撑。
Forest fires have profound impacts on natural ecosystems, socio-economics and public health, particularly in the face of global warming and urbanization. This presents an increasingly challenging scenario for governments worldwide in forest fire prevention and control. Mastering forest fire behavior and spread characteristics, and constructing a large-scale dynamic spread prediction model, are crucial for emergency management agencies to develop efficient fire reduction strategies. However, the evolving complexities, diverse demands, and platform-driven trends in forest fire prevention and control call for a reevaluation of existing models, prompting a need for in-depth research on modeling methods, including spreading characteristic equations, spatial characterization modes, and kinetic deduction mechanisms. Therefore, this paper concentrates on the dynamic modeling of forest fires in emergency response, employing multidisciplinary theory and approaches such as heterogeneous cellular automata, graph networks, deep learning, and geoinformatics. The outcome is a suite of large-scale forest fire spread prediction models featuring diversified scenarios, refined characterization, rapid computation, and standardized services. These models can provide emergency agencies with comprehensive and accurate forest fire situation analysis services. The main contributions of this paper are summarized as follows.(1) Modeling Interface Areas: To address the quantitative modeling challenge in complex interface areas, we propose a method for modeling fire spread and risk assessment in wildfire-urban interface (WUI) areas. This involves establishing a mutual ignition mechanism for heterogeneous combustibles and constructing a forest fire spread prediction model based on heterogeneous cellular automata. Given the danger of wildfire spread and individual ability of risk avoidance, a firefighter risk index (FRI) is designed to reveal spatiotemporal distribution patterns of fire suppression risk. Additionally, we construct a forest fire case dataset and a performance evaluation system to support standardized model validation.(2) Deep Learning for Dynamics: To optimize the kinetic deduction mechanisms of forest fire spread, graph network and deep learning are introduced to model the spread of forest fire. This involves establishing a graph heterogeneity checking criteria and a “grid-graph” mapping mechanism. A refined forest fire spread model based on deep neural networks (DNNs) is constructed to reveal the fire spreading patterns and characteristics in irregular graph network (IGN) spaces. Furthermore, the concept of spread spatiotemporal distribution field (SSTDF) and the “state-condition” (S-C) progressive coding mechanism are introduced to define the network framework. A convolutional neural network-based (CNN-based) model is constructed to predict the spread of forest fire, ensuring millisecond computational efficiency and competitive simulation accuracy. (3) Standardized Integration: Aiming at the challenge of efficient model integration and application in emergency response, a standardized management system for forest fire models is established. Adhering to the design principle of service-oriented architecture (SOA), the disaster management platform (DMP) and forest fire thematic system are constructed, then the role functions and interaction of “provider-user-platform” is introduced. Leveraging the OGC WPS standard and the disaster system theory, a service paradigm and interface specifications for disaster models are proposed to enhance the reusability, compatibility, and interoperability of forest fire models. Taking the comprehensive urban scenario as a study case, it demonstrates the practical application process and impact of the forest fire model within the emergency management system and platform. This provides essential theoretical and technical support for disaster model management and multi-hazard dynamic analysis.
