Analyzing car crashes in space and time

Determine where and when driving is most dangerous

A crashed car
A crashed car

Can GIS analysis make our roads safer?

You probably didn't wake up today thinking you would lose a loved one in a car crash. Unfortunately, before this day ends, more than 100 people in the United States will have died, and an additional 12,000 people will have been injured or disabled, as a result of a car crash ( ASIRT, 2021  NSC, 2021 ). 

Dr. Lixin Huang, IT Engineer III, is a GIS analyst for Brevard County, Florida. He knows that Florida's interstates are ranked among the  nation's deadliest  and that the number of traffic accidents in Brevard County is increasing.

Graph showing the number of car crashes 2010 to 2015
Graph showing the number of car crashes 2010 to 2015

Number of car crashes per year in Brevard County, Florida

The costs associated with car crashes are staggering. Beyond the tragic loss of lives, the monetary costs of highway crashes are estimated to be more than  $871 billion each year . The overwhelming majority of these crashes are  preventable .

Lixin hopes that by identifying where and when crashes occur throughout Brevard County, he might be able to help prevent some of them. He begins with a quick exploratory analysis of traffic accident trends around the county. He then focuses on where accidents occur along the road network. Finally, he examines temporal cycles and a 3D visualization of yearly trends. Lixin's analytical workflow is outlined below. A  step by step tutorial  with all the data used for his analysis is also available.

What data is needed?

Lixin obtains crash data from the University of Florida GeoPlan Center. It includes the location, date, and time for every motor vehicle crash in Brevard County between 2010 and 2015. Each crash is shown as an orange point on the map below. Click a point to learn more. See if you can find a crash point where alcohol or distracted driving was a factor.

Brevard County car crashes 2010 to 2015

Notice that it's difficult to discern any kind of pattern from the point locations alone. Lixin decides to restructure the data so he can examine space-time trends.

Traffic at night

Crash trends

Where are car crashes increasing?

Lixin performs a quick exploratory space-time pattern analysis to confirm that the number of traffic accidents is increasing overall, and that the increase is statistically significant.

The number of crashes is different every month, of course. Finding a statistically significant increase in the number of crashes between 2010 and 2015 indicates the increase is not just the result of random fluctuations.

By focusing on different areas around Brevard County, Lixin can interactively explore traffic accident trends and identify broad problem areas.

Space-time car crash trends

The hexagon bins classified as New Hot Spots are locations that had a large number of crashes during the final four months of 2015. Consecutive Hot Spots are locations that consistently had a large number of crashes the last year or two (2014 and 2015). Sporadic Hot Spots are locations that sometimes had a high number of crashes and sometimes didn't. Click the map to see the definition for each type of hot spot trend.

Map legend for select trend categories

In three dimensions, each hexagon becomes a column of stacked bins. Each bin represents a four-month time period, with the most recent time period at the top of the column.

The red bins are statistically significant space-time clusters where large numbers of crashes occurred. The blue bins are statistically significant space-time clusters with very few crashes.

The intensity of the red and blue colors corresponds to clustering statistical significance.

Hot spot analysis map legend

See if you can find new, consecutive, and sporadic hot spot trends by matching what you see in the map to the examples below.

Example images of new, consecutive, and sporadic hot spot columns

Use your mouse to navigate around the 3D map. Tilt the map, for example, by pressing and holding the right mouse button.

Is Lixin done?

There are a couple of important problems with this quick exploratory analysis of traffic accident trends.

  1. The spatial analysis used to assess hot and cold spot areas is based on  Euclidean distance  rather than the actual road network.
  2. The analysis does not consider important temporal cycles such as the workweek rush hour.

Lixin will refine his analysis to address both of these problems.

Car lights reflecting in the wet street

Road network crash hot spots

Where are the most dangerous roads?

Two crashes separated by a river or by a major highway might be close together using a straight line (Euclidean distance), but far away from each other on a road network with few bridges or underpasses. Because hot spot analysis is looking for high crash rates that cluster close together, accurate distance measurements are essential.

Lixin aggregates all of the crash and fatality data between 2010 and 2015 onto Brevard County roads so that individual segments of the road network get a count representing the number of crashes and the number of fatalities that have occurred there. For each count, he computes the per mile, per year rate. Next, he connects all of the road segment crash and fatality rates using restrictions imposed by the actual road network. When he runs hot spot analysis, he can now see and compare locations on the road network where high crash rates and high fatality rates cluster spatially.

The red sections of the road network are locations with statistically significant clustering of high rates. The top map below shows hot spots for all car crashes. The bottom map shows hot spots for fatal car crashes.

Road hot spots for all crashes and for fatal crashes

Compare hot spots for all crashes (top) to hot spots for fatal crashes (bottom)

These maps provide specific target locations where traffic safety can, and should, be evaluated. They indicate where remediation measures may help prevent future crashes and save lives.

Abstraction of traffic among NYC high rise buildings

Cyclical patterns

When are the most dangerous times to drive?

The number of car crashes increases with the number of drivers on the road. Lixin decides to look for cyclical patterns in the crash data. He creates a graph showing the number of crashes by day of the week and by hour of the day. Several peaks emerge, but the strongest is associated with the workweek between 3:00 p.m. and 5:00 p.m. (between hours 15 and 17).

Line graph showing days and times of peak car crashes

The number of car crashes is highest during the workweek between 3:00 p.m. and 5:00 p.m.

Where do workweek 3:00 p.m. to 5:00 p.m. hot spots occur?

Lixin wonders whether the locations of car crashes associated with the afternoon workweek are the same as those on other days and at other times. He compares a map of the crash hot spots for all crashes (left below) to a map of the crash hot spots for crashes occurring between 3:00 p.m. and 5:00 p.m. Monday through Friday (right below). There are some similarities and some differences.

Side by side maps comparing all car crash hot spots to peak day/time car crash hot spots

Compare hot spots for all crashes to hot spots for weekday 3 p.m. to 5 p.m. crashes.

Lixin notices, for example, that US Route 1 just south of Florida State Road 404 (Pineda Causeway) is not a hot spot area for crashes overall; it is, however, a statistically significant hot spot location on weekdays between 3 p.m. and 5 p.m. He examines the traffic accidents in this area and learns that distracted driving was a factor in a number of the crashes. Increased ticketing for cell phone use while driving may help reduce accidents.

What are the trends for particular peak crash days and times?

Next, Lixin examines weekday 3 p.m. to 5 p.m. crash trends in space and time using a 3D visualization. By stacking road segment crash hot spots for each year, he can identify locations that are persistent problem areas during the workweek afternoon commute.

Workweek 3 p.m. to 5 p.m. crash rate trends

The bottom layer of red ribbons reflects crash hot spots for 2010. The top layer of ribbons reflects crash hot spots for 2015. Lighter red ribbons are still statistically significant (road segments where high crash rates cluster), but the clustering is less intense than the darkest red hot spot ribbons.

Use the mouse to navigate around the map and explore other high crash areas.

Geometric shapes

Workflow summary

What has Lixin accomplished?

Lixin's workflow has answered the following questions.

  • Which intersections and roadways in Brevard County have the highest crash rates?
  • When and where do most crashes occur?
  • How does the spatial pattern of fatalities differ from the spatial pattern of car crashes overall?
  • How does the spatial pattern of crash rates occurring during the workweek afternoon commute differ from the overall pattern of crash rates?
  • Over time, which intersections or roadways are persistent problem areas for car crashes?

The same workflow may be extended to answer additional questions.

  • Where are the hot spot areas for crashes involving elderly drivers, teenage drivers, or crashes in which alcohol is a factor?
  • When and where do accidents involving elderly drivers, teenage drivers, or alcohol cluster spatially and temporally?

By understanding where and when collisions occur throughout the county, Lixin can make more informed recommendations for policies and other measures that can help reduce car crashes and save lives in the future.

Number of car crashes per year in Brevard County, Florida

Compare hot spots for all crashes (top) to hot spots for fatal crashes (bottom)

The number of car crashes is highest during the workweek between 3:00 p.m. and 5:00 p.m.

Compare hot spots for all crashes to hot spots for weekday 3 p.m. to 5 p.m. crashes.