Research Article  ·  Applied Network Science  ·  Published 2026

Not All Roads Are Created Equal: Network Science Shows Some Highways Matter More for Brazil's Connectivity

Júlio Taveira  ·  Ronaldo Menezes  ·  Fernando Buarque

Applied Network Science, 11: 25  ·  DOI: 10.1007/s41109-026-00777-4

Road Network 75,000 km of Brazilian federal roads modelled as weighted networks
Geographic Data 5.67M georeferenced positions from DNIT's national road system (2023)
Incident Records 1M+ road accident records spanning 2007–present from Federal Highway Police
Network Diameter 5,838 km longest road path, from Cruzeiro do Sul, AC to Chuí, RS

The Problem

Brazil moves 66% of its freight and 95% of its passengers by road. With approximately 8.5 million square kilometers of territory, its federal highway network serves as the connective tissue of the entire national economy. Yet this infrastructure is poorly understood as a system — planners typically manage individual roads in isolation, without a structural picture of which segments are truly critical to national connectivity.

Road infrastructure growth has slowed to just 2.2% in the last decade, while public investment in federal roads in 2024 covered only 14% of estimated need. With limited resources, the question of where to invest — and what to protect — becomes consequential. Which highways, if disrupted by disaster, floods, or degradation, would most fracture Brazil's connectivity? Traditional engineering approaches cannot answer this at scale.

Previous network science studies of Brazil treated the entire road system as a single, coarse network using cities as nodes. This obscures what happens at the segment level, where the actual physical infrastructure — and its vulnerabilities — reside.

"Road infrastructure growth has slowed, with an increase of only 2.2% in the last decade. Public investment in federal roads amounted to $2.5 billion in 2024, representing only 14% of the estimated need."

The distribution of federal roads is deeply unequal: the South region has a road density of 20.5 km per 1,000 km², while the North — nearly seven times larger in area — has just 2.7 km per 1,000 km². A single model cannot capture both without regional decomposition.

The Approach

The study builds five weighted network variants of Brazil's federal road system at both national and regional levels, then subjects each to structural analysis and simulated disruptions. Data comes from three official Brazilian sources: DNIT (road structure and traffic flow), and the Federal Highway Police (accidents).

Key Findings

The multi-weight, hierarchical analysis revealed structural properties of Brazil's federal road network that coarser models had obscured, including a counterintuitive vulnerability pattern with direct implications for infrastructure investment.

Scale-free nationally, dangerously hub-dependent regionally

National networks are scale-free (power-law exponent γ between 2.17–2.67), indicating moderate hub dominance. The North region is consistently heavy scale-free (γ < 2) across all weight schemes, meaning its connectivity depends critically on a handful of junctions — making it disproportionately vulnerable to targeted disruption.

The longest road path exceeds Brazil's physical width

The distance-weighted network diameter is 5,838 km — from Cruzeiro do Sul, AC to Chuí, RS — 33% longer than the maximum straight-line distance across Brazil (4,394 km). The North's diameter alone (5,112 km) nearly matches the national figure, reflecting the indirect routing forced by rivers, rainforest, and sparse infrastructure.

Roads cluster into isolated corridors

All network variants exhibit modularity above 0.90, indicating that Brazil's federal roads naturally organise into distinct, internally cohesive communities. These road-corridor clusters facilitate regional integration but their weak inter-community connections represent bottlenecks for national mobility — the North has the lowest inter-state connectivity (9%).

Major junctions connect to other major junctions

All networks show unexpectedly high positive assortativity (rw = 0.56–0.82), especially under flow and incident weighting. This contrasts with typical infrastructure networks, which are disassortative. High-degree nodes in Brazil's road system preferentially connect to other high-degree nodes — a hub-rich structure that concentrates traffic and risk simultaneously.

The Northeast is the national linchpin

Removing just 7.75% of the Northeast's road edges (by distance weight) causes the national giant component to collapse by 34%. The region contains a small set of critical long-distance inter-regional connectors; once severed, the network plateaus at 65% of original size regardless of further removals. The Southeast shows a similar sharp transition at 15.1% removal.

Peripheral, not central, nodes create structural fragility

Counterintuitively, targeting high-betweenness nodes is less damaging than random removal. Central nodes in dense urban cores are surrounded by redundant paths; it is the low-betweenness peripheral nodes — the sole connections to remote regions — that are structurally critical. Random disruptions, which hit peripheral and central nodes indiscriminately, fragment the network faster.

Why It Matters

A structural map of Brazil's federal highways, with vulnerability profiles attached, gives planners and policymakers something they have not had before: a ranked list of what must be protected, and what the country can afford to lose.

Infographics

Peripheral nodes create structural fragility
Counterintuitively, removing a high-betweenness central node (left) barely affects connectivity — dense urban cores have many redundant paths. But removing a low-betweenness peripheral bridge node (right) immediately splits the network into two disconnected components.
Central hub removed network stays connected
vs
Peripheral bridge removed network splits
Five network variants — five operational realities
The same 75,000 km of federal roads are encoded five ways: by physical distance (BRN), incident count (BRN-I), vehicle flow (BRN-F), incidents per km (BRN-R), and flow per lane (BRN-T). Each lens reveals different critical corridors — a road that appears minor by distance may be a major bottleneck by flow or a dangerous outlier by incident density.
BRN — Distance edge weight = road length
BRN-F — Vehicle flow edge weight = traffic volume

Paper & Authors

Authors

Júlio Taveira University of Pernambuco, Recife, PE, Brazil
University of the Federal Highway Police, Florianópolis, SC, Brazil
Ronaldo Menezes (corresponding) Department of Computer Science
University of Exeter, UK
Fernando Buarque University of Pernambuco
Recife, PE, Brazil

Cite This Work

@article{taveira2026roads, title = {Not all roads are created equal: network science shows some highways matter more for {Brazil}'s connectivity}, author = {Taveira, J\'{u}lio and Menezes, Ronaldo and Buarque, Fernando}, journal = {Applied Network Science}, volume = {11}, pages = {25}, year = {2026}, doi = {10.1007/s41109-026-00777-4} }

Open access · CC BY 4.0