NoVA Vulnerable Road User Crash Atlas (2018-2022)

OverviewDataset Summary

This dataset synthesizes every police-reported pedestrian- and bicyclist-involved crash in Northern Virginia from 2018 through 2022 - 3,179 records from VDOT's statewide crash database - with ACS 2022 walk and bicycle commuter counts at the census tract level. The result is three linked layers: a fully attributed crash point layer (severity, lighting, roadway context, hit-and-run, school zone, DBSCAN cluster label), a 579-tract exposure layer expressing crashes per 1,000 walk/bike commuters, and 107 hotspot cluster footprints ranked by crash, injury, and fatality totals.

A companion temporal animation replays all five years of crashes month by month: open the time animation.

Why This DatasetThe Problem It Solves

VDOT publishes crash records statewide, but no public product isolates vulnerable road user crashes for Northern Virginia, normalizes them by how many people actually walk and bike in each tract, and delineates hotspots with a reproducible clustering method. Raw crash maps always point at dense urban cores because that is where the walkers are. Dividing by exposure reverses the picture: it reveals arterial-heavy suburban tracts where a person on foot faces far higher per-capita risk than in walkable Arlington. That distinction is exactly what Vision Zero programs, the VDOT Pedestrian Safety Action Plan, and county capital budgets need to target countermeasures.

MethodologyHow It Was Built

Crash records were retrieved from VDOT's Virginia Roads open data (BIKE_PED_Crashed_2018_2022 ArcGIS REST service) with a NoVA bounding envelope, then clipped by point-in-polygon join to the exact 9-jurisdiction boundary from TIGER/Line 2023 counties. Points were joined to TIGER 2022 tracts and merged with ACS 2022 table B08301 to attach walk and bicycle commuter counts; rates were suppressed where a tract has fewer than 50 walk/bike commuters. Areas were computed in EPSG:32618.

Hotspots were delineated with DBSCAN on haversine distance (epsilon ~300 m, min_samples = 5), yielding 107 clusters; footprints are convex hulls of member points. Monthly counts were decomposed with STL (period = 12) and months with residuals beyond two standard deviations flagged as anomalies. All layers carry embedded FGDC CSDGM metadata.

Key FindingsWhat the Data Shows

• Arlington's US-50 / Columbia Pike-Rosslyn corridor is the region's dominant hotspot: the largest DBSCAN cluster holds 312 crashes and 24 deaths.
• Fatality burden is disproportionate outside the core: Fairfax County logged 1,150 crashes and 83 deaths; Manassas City's 79 crashes produced 10 deaths - the region's highest deaths-per-crash rate.
• 72% of crashes involved a pedestrian (2,277 of 3,179); 902 involved a bicyclist.
• The pandemic left a measurable signature: STL analysis flags May 2020 (30 crashes vs. a trend of ~49) and June 2021 as significant negative anomalies; June 2019 (76 crashes) is the largest positive anomaly.
• Exposure normalization reverses the map: several suburban tracts with minimal walking populations rank far above dense urban tracts in crashes per 1,000 walk/bike commuters.

Analytical UsesWhat You Can Do With This

• Vision Zero prioritization: a defensible, reproducible shortlist of corridors for crossings, lighting, and speed management.
• Exposure-adjusted equity analysis: find tracts where the few people who walk face the highest risk.
• Safe Routes to School review: school-zone, lighting, and hit-and-run flags support corridor audits.
• Before/after evaluation: the 60-month series and anomaly baseline form the pre-period for countermeasures installed after 2022.

Synthesis OpportunitiesWhat It Combines With

Overlay the hotspot clusters on the Day 31 WMATA walkshed layer to test how much of the crash burden falls inside station walksheds, or join tract exposure rates to the Day 1 IEVI to ask whether high-vulnerability tracts also carry elevated per-walker crash risk. A street-lighting or sidewalk-inventory join (county open data) would let the lighting-condition field drive a systemic darkness-crash screen.

LimitationsKnown Caveats

• Only police-reported crashes are captured; minor bicycle falls and near-misses are absent.
• Locations are geocoded to route/milepost and may be offset tens of meters, occasionally shifting points across tract boundaries.
• ACS commuter counts proxy exposure but exclude recreational, school, and errand walking.
• The source service ends in 2022; post-pandemic recovery years are not included.
• DBSCAN with a 300 m epsilon chains contiguous dense corridors into single large clusters - interpret large clusters as corridors, not intersections.

Data DictionaryFields in the GeoPackages

Jurisdiction BreakdownResults by Jurisdiction

Sources & CitationsAll Data is Openly Licensed

• [1]Virginia Department of Transportation. (2023). BIKE_PED_Crashed_2018_2022 [ArcGIS REST feature service]. Virginia Roads Open Data. link
• [2]U.S. Census Bureau. (2023). ACS 2022 5-Year Estimates, Table B08301: Means of Transportation to Work.
• [3]U.S. Census Bureau. (2023). TIGER/Line Shapefiles: 2022 Census Tracts and 2023 Counties, Virginia.