UC Santa Barbara

Fire is not an enemy to be eradicated, but a natural process with which human societies must coexist. For millennia, fire has sustained ecosystems, regenerated landscapes, and supported human civilization through warmth, agriculture, and industry. Yet, when uncontrolled, fire transforms from a regenerative force into a destructive hazard—threatening homes, infrastructure, and lives. As human settlements expand into fire-prone environments and climate extremes intensify, the challenge of uncontrolled fire resilience has become one of the defining spatial problems of our time. Fire resilience is a multidimensional challenge shaped by the interaction between environmental, infrastructural, and human systems. Among uncontrolled fires, vehicle fires, wildfires, and structural fires together constitute the majority of fire-related damages to life, property, and the environment (Hall, 2023). Yet, they differ fundamentally in spatial dynamics: structural and vehicle fires arise within built environments requiring micro-scale access and preparedness, whereas wildfires unfold across landscapes demanding meso- to macro-scale response and regional planning. Despite this distinction, both domains share a common imperative, anticipating and mitigating fire risk through spatially explicit, data-informed decision-making.This dissertation advances an integrated geographic framework for enhancing fire resilience through advanced decision-making and risk management, uniting two complementary research directions. Chapters 2 and 3 address structural fire resilience through spatial optimization methods that operationalize the principles of access and coverage in built environments. A novel extended convexpath algorithm is developed to compute unobstructed Euclidean access paths between hydrants and buildings, overcoming the limitations of network-based proximity measures. This framework quantifies local vulnerability and code compliance at the structure, street, and neighborhood scales. Building on this foundation, the Generalized Euclidean Shortest Path (GESP) and its multipart extension (MGESP) advance classical accessibility theory to continuous geometries, enabling analytical evaluation of complex urban forms. Together, these models establish a prescriptive foundation for optimizing spatial infrastructure under hazard risk.Chapters 4 and 5 shift focus to landscape fire resilience. Chapter 4 develops a multimodal spatio-temporal GeoAI model integrating Sentinel-1 SAR, Sentinel-2 multispectral imagery, GRIDMET climate data, and static geographic metadata to generate highresolution (10 m) burned-area probability surfaces through ConvLSTM temporal reasoning and FiLM-based feature fusion. Chapter 5 discusses how to operationalize these predictions within a bi-objective fire response prepositioning model that jointly optimizes visibility over high-risk landscapes and accessibility to potential ignition points. The model embeds viewshed-based coverage within a Weber-type access formulation, capturing the fundamental trade-off between seeing a fire early and reaching it quickly. By integrating dynamic GeoAI risk surfaces into a continuous-space, dual-objective optimization problem, the framework enables proactive, transparent, and regionally scalable decision-making associated with prepositioning fire personnel under extreme conditions. Collectively, this work conceptualizes risk management of uncontrolled fires as spatial optimization problems. By linking optimization models with big data driven solution approaches and AI-driven prediction systems, this dissertation offers a unified framework for adaptive, interpretable, and actionable fire resilience management. The approach strengthens both scientific understanding and practical capacity for managing fire risk in complex human–environment systems.