Source-linked AI summary
A Multi-Scale Analysis of 27,000 Urban Street Networks: Every US City, Town, Urbanized Area, and Zillow Neighborhood
Geoff Boeing
TL;DR
Street-network research has been constrained by small samples, ad hoc processing decisions, and simplified graph representations. The paper uses OSMnx and OpenStreetMap to analyze 27,009 US networks across three urban scales with reproducible, nonplanar directed-graph methods. It finds that street-network characteristics vary by scale and reflect differences in density, connectedness, resilience, and urban form.
Problem
Street-network studies commonly use small samples, difficult-to-reproduce processing choices, and planar or undirected graph simplifications.
Method
The study uses OSMnx to consistently download, model, correct, analyze, and visualize OpenStreetMap street networks as nonplanar directed graphs across metropolitan, municipal, and neighborhood scales.
Results
Street-network characteristics vary by scale: neighborhood networks have lower median circuity and higher median streets per node than municipal and urbanized-area networks.
Takeaways & Limitations
Multiple scales in concert provide planners and scholars a clearer view of urban form and topological structure than any single definition of city.
Takeaways & Limitations
The study is cross-sectional and emphasizes network structure rather than the temporal complexity of city dynamics and processes.
Abstract
from arXiv · showhide
OpenStreetMap offers a valuable source of worldwide geospatial data useful to urban researchers. This study uses the OSMnx software to automatically download and analyze 27,000 US street networks from OpenStreetMap at metropolitan, municipal, and neighborhood scales - namely, every US city and town, census urbanized area, and Zillow-defined neighborhood. It presents empirical findings on US urban form and street network characteristics, emphasizing measures relevant to graph theory, transportation, urban design, and morphology such as structure, connectedness, density, centrality, and resilience. In the past, street network data acquisition and processing have been challenging and ad hoc. This study illustrates the use of OSMnx and OpenStreetMap to consistently conduct street network analysis with extremely large sample sizes, with clearly defined network definitions and extents for reproducibility, and using nonplanar, directed graphs. These street networks and measures data have been shared in a public repository for other researchers to use.
1. Introduction
Street-network research has faced small samples, difficult reproducibility, and simplifying graph assumptions. This paper addresses these limitations by analyzing 27,000 US networks consistently across metropolitan, municipal, and neighborhood scales.
- Street-network studies often analyze only about 10 to 50 networks, limiting the generalizability and interpretability of findings.
- Ad hoc decisions about spatial extents, topology, nodes, and edges have made street-network analysis difficult to reproduce.
- Prior studies frequently simplify street networks into planar or undirected primal graphs for tractability.
- The paper introduces a methodology for acquiring, modeling, and analyzing large samples of street networks as nonplanar directed graphs.
- It presents descriptive urban-morphology findings for every US city, urbanized area, and Zillow neighborhood, while investigating smaller-sample findings at scale.
2. Methodology
The study uses OSMnx and OpenStreetMap to construct reproducible, spatially embedded street-network graphs at multiple urban scales. It computes metric, topological, connectedness, centrality, and resilience measures across 27,009 networks.
- Street networks are modeled as primal, directed, nonplanar graphs, with intersections as nodes and street segments as edges.
- OSMnx downloads, models, corrects, analyzes, and visualizes OpenStreetMap street networks for places defined by names, addresses, or polygons.
- OpenStreetMap provides collaborative geospatial data through APIs, including detailed street attributes and pathways beyond TIGER/Line’s scope.
- Study boundaries use census urbanized areas, municipal geometries, and Zillow-defined neighborhood geometries at metropolitan, municipal, and neighborhood scales.
- OSMnx buffers each geometry by 0.5 km, downloads nodes and ways, constructs and corrects graphs, truncates them to target polygons, and calculates measures.
- Average segment length proxies block size, while node, intersection, edge, street, and circuity measures characterize network grain and density.
- Average node degree, streets per node, and average node connectivity quantify connectedness and vulnerability to disconnection.
- Betweenness centrality and PageRank assess node importance, while maximum betweenness centrality indicates exposure to shortest-path choke points.
3. Results
Across 497 urbanized areas, 19,655 cities and towns, and 6,857 neighborhoods, US street networks vary substantially in density, connectedness, grain, and resilience. The results reveal regional and neighborhood-scale patterns associated with grid form, development history, terrain, and municipal boundary choices.
- Metropolitan-scale results: 497 urbanized areas span 26 to 8,937 km2, producing substantial variance in density and count-based measures.The data set covers urbanized areas from Delano, California, to New York–Newark.
- Metropolitan-scale results: Average circuity is 7.4% above straight-line distance, and circuity has a clearer negative relationship with four-way intersections than intersection density does.The relationship between intersection density and four-way intersections is weak (r2 = 0.17), while circuity’s relationship has r2 = 0.43.
- Metropolitan-scale results: The typical urbanized area has 26 intersections per km2, with values ranging from 12.5/km2 in Gainesville to 49.4/km2 in New Orleans.The highest intersection densities concentrate west of the Mississippi, while the lowest form a belt from Louisiana through New England.
- Metropolitan-scale results: The typical urbanized area averages 2.8 streets per node, including 18% four-way intersections, 59% three-way intersections, and 21% dead-ends.San Angelo averages 3.2 streets per node, while Lexington Park averages 2.2 outside Puerto Rico.
- Municipal-scale results: Street-segment lengths generally follow lognormal-style right-skewed distributions rather than power laws, except in consistently sized orthogonal grids.Great Plains states particularly feature orthogonal, reasonably dense networks associated with flat terrain and rapid nineteenth-century platting.
- Municipal-scale results: Great Plains states have the highest state-level medians for mean streets per node, indicating especially gridlike municipal networks.Nebraska, Kansas, South Dakota, Montana, North Dakota, Oklahoma, and Iowa rank highest on this measure.
- Municipal-scale results: Municipal density comparisons are constrained by boundary variation: Alaska averages 1.3 intersections/km2 because some municipal extents reach thousands of square kilometers beyond built-up areas.Rhode Island averages 56 intersections/km2, illustrating how municipal boundaries affect aggregation.
- Neighborhood-scale results: Neighborhood networks show greater variance, including intersection densities of zero where Zillow-defined boundaries contain no intersections.Smaller network sizes and neighborhood boundary definitions produce more heterogeneous measures than metropolitan and municipal scales.
4. Discussion
US street-network characteristics vary systematically with neighborhood, municipal, and urbanized-area scale, reflecting both boundary choices and urban form histories. Across these scales, the analysis links network structure to planning legacies while identifying limits of cross-sectional, US-focused evidence.
- Urban-form patterns: Orthogonal downtown grids show high density, connectedness, and order but low resilience when one-way streets create chokepoints.The reported resilience measures increase when one-way streets are switched to bidirectional streets.
- Scale matters: Neighborhoods have lower median circuity, higher median streets per node, and about 83% higher median intersection density than broader scales.These differences reflect Zillow neighborhoods’ concentration in older, denser core cities, whereas urbanized areas include sprawling suburbs.
- Scale matters: Multiple scales jointly capture planning authority, metropolitan built form, and local lived experience more clearly than any single scale.Municipal boundaries capture legal and planning jurisdictions, urbanized areas aggregate wider systems, and neighborhoods focus on local environments.
- Cross-scale structure: Total street length and node count have a strong linear relationship invariant across scales, while street-segment lengths are typically right-skewed.Exceptions include highly uniform orthogonal grids in Portland downtown and some Great Plains towns.
- Historical legacies: Planning eras, design paradigms, regulations, and instruments remain visible in contemporary US street networks.Great Plains networks show low circuity, many streets per node, short segments, and high intersection density, consistent with development followed by limited growth.
- Limitations and future work: The study’s cross-sectional US findings do not necessarily generalize worldwide, and temporal analyses remain a future research need.The authors also propose dimensionality reduction and broader spatial-geometric analyses for subsequent work.
5. Conclusion
The paper combines large-scale empirical analysis of US street networks with OSMnx methods for consistent, reusable urban network research.
- The study analyzes every US city and town, urbanized area, and Zillow neighborhood using metric and topological measures of density, connectedness, and resilience.
- OSMnx demonstrates methods for acquiring and analyzing street-network data consistently and at scale for urban planners and scholars.
- Datasets for 497 urbanized areas, 19,655 cities and towns, and 6,857 neighborhoods, with attribute data and morphological measures, are available in a public repository.
Notes
The paper provides online access to both the OSMnx toolkit and the analyzed street-network data repository.
- OSMnx is freely available online as a street-network research toolkit.
- The analyzed street-network data are available through an online public repository.