Resilience of Urban Transport Network-of-Networks under Intense Flood Hazards Exacerbated by Targeted Attacks

1. Introduction

• Flooding and opportunistic cyber-attacks threaten interdependent transport networks, where disruption in one system can spread into severe loss of lifeline functions.The paper describes growing precipitation and flood risks alongside increasingly connected and autonomous transport systems.Source: §1 · pp. 1–2
• Resilience analysis of multimodal urban transport networks remains limited, especially for real spatial networks and compound hazards that combine natural events with targeted attacks.The paper frames its hypotheses around topology-related vulnerability and disproportionate failures under combined hazards and attacks.Source: §1 · pp. 2–3
• Network-science research has examined robustness, attacker strategies, and recovery, while prior transportation studies have focused mainly on single networks and disruptions.The cited work also includes theoretical frameworks for interdependent networks and cascading failures.Source: §1 · pp. 2–3
• The study presents a quantitative framework for resilience, robustness, and recovery in spatial urban transport network-of-networks, demonstrated on the London Rail Network.It tests random, targeted, natural-hazard, and compound failures.Source: §1 · p. 3

2. Results

• The two-layer robustness equation represents changing functionality as nodes fail, with adjacency, failure, and interdependency terms updated to capture cascading failure.For compound threats, a subsequent targeted failure is applied after initial failures, using the updated network state.Source: §2 · p. 4

2.1 Network Vulnerability

• The LRN is least robust under random failure, and under targeted failure is less robust than Random networks but more robust than Scale-Free networks.Its sparse, spatially constrained topology and narrow degree distribution are associated with vulnerability, while identical dependent nodes boost robustness.Source: §2.1 · p. 7, Figure 2 · p. 8
• Increasing dependency in coupled networks makes failure faster; above a threshold, failure becomes complete and the transition changes from second- to first-order.Below a certain dependency ratio, the final giant connected component does not reach zero.Source: §2.1 · p. 7, Figure 2 · p. 8

2.2.1 Identifying Flooded Nodes

• Flood damage is concentrated early: approximately 80% occurs within the first third of node failures, consistent with flooded nodes having higher average degree than the network overall.The analysis identifies 65 stations in the 100–1000-year flood-risk zone.Source: Figure 1 · p. 5, §2.2.1 · p. 9

2.2.2 Compound Disruptions – Flood and Targeted Attacks

• 5% targeted node removal after flooding can cause network-wide failure comparable to purely targeted failure, when total failed nodes are held equal.The analysis models degree-targeted removal after a proportion of flooding has occurred.Source: §2.2.2 · p. 10

2.2.3 Recovery Strategies

• Degree- and betweenness-based recovery is fastest in the LRN, while eigenvector-based recovery is not much better than random recovery.Restoring high-betweenness nodes re-enables more shortest paths; centrality measures remain preliminary guidance given practical cost and resource constraints.Source: §2.2.3 · p. 10, Figure 3 · p. 11

2.3 Realistic Scenario – Flooding of the Underground Network

• When flooding is restricted to the Underground, indirect failures can be disproportionate because a node’s criticality differs across dependent networks; a 5% targeted attack amplified damage by over 300% in the Underground and could completely fail the DLR.This scenario models flood propagation through inter-network dependencies.Source: §2.3 · p. 12, Figure 4 · p. 12, Figure 5 · p. 13
• The claim that compound damage exceeds the sum of separate hazards needs further examination, and broader generalization requires more transport systems and lifeline sectors.The paper identifies further study as necessary for the superadditivity hypothesis.Source: §2.3 · p. 13

Discussion

• Spatially embedded urban transport networks have topological vulnerabilities that differ from non-spatial counterparts, so non-spatial theoretical frameworks may not directly translate to them.The authors call for theory, empirical work, methods, and tools tailored to spatially embedded infrastructure.Source: Discussion · p. 14
• The authors argue that efficiency-focused design and legacy station placement near water bodies rendered the London Rail Network highly vulnerable to flood-induced failure.They recommend embedding resilience considerations in new and existing urban lifeline infrastructure.Source: Discussion · p. 14
• Flooded low-degree connector stations can trigger breakdown of an interdependent network into subcomponents, contrary to a focus only on conventionally high-degree nodes.The authors suggest modifying topology, increasing modularity, and redistributing critical dependencies as avenues for resilience.Source: Discussion · pp. 14–15
• A limited targeted attack can cause disproportionate damage when network capacity has already been weakened, but compound-attack analysis remains an early exploration.The authors call for deeper study from network-science and engineering perspectives.Source: Discussion · p. 15
• The study’s conclusions need broader testing across urban systems and lifeline sectors, and future work should incorporate system dynamics and explore alternatives to giant-component size.The authors also call for analysis of additional failure schemes and contingencies.Source: Discussion · p. 15

Dataset Availability

• The London metro dataset is attributed to De Domenico et al., the flood map to the UK Environment Agency, and Python code is available on the lead author’s GitHub account.The code is provided to facilitate reproduction of results.Source: Dataset Availability · p. 16

• 24 stations are common to the Underground and Overground, 5 to the Underground and DLR, and 1 to the Overground and DLR.Source: Common Stations · p. 16

• Flood failure removes stations in river-proximity categories, with nearer-to-Thames nodes failing first and random order within categories across 20 independent runs.Source: Failure Scenarios · p. 16
• Random failure removes nodes randomly from the overall network.Source: Failure Scenarios · p. 16
• Random-local failure samples 100 clusters of 65 geographically closest stations and removes nodes randomly within each cluster across 20 independent runs.Source: Failure Scenarios · p. 16
• Targeted failure removes nodes in decreasing order of degree or betweenness centrality.Source: Failure Scenarios · p. 16
• Compound failure introduces a degree-based attack after some flood failure while keeping total failed nodes equal to the number flooded.The targeted attack removes n% of nodes based on degree centrality.Source: Failure Scenarios · p. 17
• The cascading-failure algorithm alternates removal of failed and non-GCC nodes with removal of dependent nodes across network layers, returning the GCC sizes.Source: Failure Scenarios · p. 17
• The compound-failure algorithm applies flood and targeted removals through cascading failure when enabled; otherwise, it removes flooded nodes directly.Source: Failure Scenarios · p. 18

• Node centrality measures quantify differences in importance expected among nodes in heterogeneous complex networks.Source: Centrality Measures · p. 18
• Degree centrality is the number of nodes linked to a particular node.Source: Centrality Measures · p. 18
• Betweenness centrality counts shortest paths between node pairs that pass through a given node.The passage defines σDE as the total shortest paths from s to t and σDE(v) as those passing through v.Source: Centrality Measures · p. 18
• Eigenvector centrality treats a node as important when connected to other important nodes and defines its score through the adjacency matrix eigenvector.The matrix relation is λx = Ax; x is the left-hand eigenvector associated with eigenvalue λ.Source: Centrality Measures · pp. 18–19

• The cascade schematic depicts direct failure, dependent-node removals across layers, and feedback removals, with green nodes marking functional nodes in the giant connected component.Source: Figure S6 · p. 23
• The London Rail Network degree distribution is plotted against equivalent Erdős–Rényi and scale-free distributions, with the scale-free plot curtailed at degree 14.Source: Figure S7 · p. 24
• The flooded-network degree distribution is compared with the total network; higher average degree in the flooded area indicates more critical nodes there.Source: Figure S8 · p. 24