Can We Trust OpenStreetMap to Measure Cycling-Infrastructure Change?

A Reproducible Validation Framework Using Google Street View

Eugeni Vidal-Tortosa, Víctor Gonzàlez-Parra, Oriol Marquet

GEMOTT, Universitat Autònoma de Barcelona
WSTLUR 2026 · Beijing

Project: 101117700 — ATRAPA — ERC-2023-STG

Why does this matter?

INTRODUCTION

  • Cities are rapidly expanding cycling infrastructure

  • We need longitudinal data to understand its impacts

  • OpenStreetMap (OSM) offers global, historical data

  • But can we trust OSM to detect change over time?

Research aim

INTRODUCTION

Quantify the accuracy of OSM for detecting cycling-infrastructure change

Study setting and data

DATA & METHODS

Study setting

  • Barcelona, Spain

  • 2015–2023

Data sources

  • Historical OSM cycling-infrastructure snapshots

  • Census-tract boundaries

  • Historical Google Street View (GSV) imagery

Analytical workflow

DATA & METHODS

  1. Detect OSM cycling-infrastructure change

↓

  1. Stratify census-tracts

↓

  1. Generate network validation points

↓

  1. Inspect and code historical GSV imagery

↓

  1. Calculate validation metrics

1. Detect OSM cycling-infrastructure changes

DATA & METHODS

  • OSM differencing identified:
    • 127 km additions
    • 25 km removals
  • Net network growth:
    • +80%
  • 946 candidate change segments

2. Stratify census-tracts

DATA & METHODS

  • Barcelona stratified by:

    • population density
    • centrality

  • 3 × 3 density–centrality strata

  • 6 census tracts randomly sampled per stratum (54 total)

3. Generate network validation points

DATA & METHODS

  • Validation points sampled within selected census tracts

  • Up to:

    • 2 additions
    • 2 removals
    • 1 non-cycling control

  • Points linked to historical GSV imagery

4. Inspect and code historical GSV imagery

DATA & METHODS

5. Calculate validation metrics

DATA & METHODS



\(\mathrm{Precision} = \frac{TP}{TP + FP}\)


What proportion of detected OSM changes are real?


\(\mathrm{Recall} = \frac{TP}{TP + FN}\)


What proportion of real changes are captured by OSM?

Validation performance

RESULTS

  • High precision and recall for additions

  • Most apparent removals were not confirmed; estimates remain uncertain (small sample)

Limitations

DISCUSSION

  • Historical GSV imagery is only a proxy for ground truth

  • Validation had to be carried out manually

  • Removal estimates remain uncertain

  • Findings may not generalise to other cities

Conclusions

DISCUSSION

Can we trust OSM to measure cycling-infrastructure change?

It depends on the type of change being measured

  • Additions are detected reliably

    • Useful for tracking cycling-network expansion

  • Removal estimates remain uncertain

    • Often reflect mapping edits rather than real infrastructure loss

OSM accuracy should not be assumed everywhere

  • Performance may differ across cities

Main contribution

  • Reproducible framework to measure, evaluate, and calibrate OSM-derived change where suitable validation data are available

Thank you · 谢谢

Questions?

eugeni.vidal@uab.cat

Code and reproducible workflow available upon request