Resilience of road networks to natural hazards: A systematic literature review

• A systematic review on road networks resilience to natural hazards. • Resilience metrics are classified into topological, functional, and hybrid approaches. • Methods include: GIS, network analysis, simulation, machine learning, traffic assignment, digital twin, and uncertainty modeling. • Key gaps involve link-level assessment, dynamic traffic behavior, and multi-hazard interactions. • Integrated hybrid and ML-assisted frameworks are recommended for predictive and real-time resilience evaluation. Road networks are essential to societal functioning yet remain highly vulnerable to natural hazards and cascading disruptions. This study presents a systematic review of road network resilience, synthesizing resilience metrics, assessment methods, and research gaps. Following the PRISMA 2020 guidelines, 109 peer-reviewed studies published between 2010 and 2025 were analyzed from the ScienceDirect, Scopus, and Web of Science databases. The results indicate that resilience assessment is primarily based on topology-based, functional performance-based, and hybrid metrics, with 31% of studies focusing on robustness and 25% emphasizing vulnerability and preparedness, while only 4% adopt fully integrated resilience frameworks. Methodologically, conventional approaches dominate the literature, including network analysis (24%), GIS-based methods (15%), and uncertainty modeling techniques (15%), alongside traffic assignment, traffic simulation, and agent-based modeling. In contrast, emerging approaches such as graph neural networks, deep reinforcement learning, digital twins, and hybrid data-driven frameworks are applied in no more than 14% of the reviewed studies, indicating limited but increasing adoption. Despite methodological progress, persistent gaps remain, including limited link-level analysis, inadequate modeling of spatial and temporal traffic dynamics, weak predictive and real-time capability, and insufficient consideration of multi-hazard scenarios. The study highlights the need for integrated frameworks combining machine learning with analytical and simulation-based methods to enhance dynamic resilience assessment and support proactive decision-making.