Safety Evaluation of Corner Clearance at Signalized Intersections

CHAPTER 5. DATA COLLECTION

The analysis and discussions presented in this study relied on two data sets: one from the State of California and the other from the City of Charlotte, North Carolina. The original plan was to collect data from California with geographical representation from both the northern and southern regions of the State. After the preliminary analysis of California data, the FHWA approved another effort to collect additional data from the City of Charlotte, North Carolina. The data sources for these two study areas differed in many ways and required the research team to develop separate data collection methods for each dataset. The following sections discuss the details of data collection efforts.

CALIFORNIA DATA COLLECTION

The California data for this study came from the following separate sources:

• Prior FHWA study. The research team obtained corner clearance, key geometric features, and operational characteristics from a geographic information system (GIS) database developed under a previous FHWA-funded project entitled Safety Evaluation of Access Management Policies and Techniques.⁽⁹⁾
  
  
• HSIS. The research team obtained intersection, roadway, and 3 years (2009–2011) of traffic and crash data from the HSIS database.
  
  

The current study relied on GIS files compiled under the prior FHWA study to identify candidate intersections for this evaluation. In that study, the researchers collected the original data and developed the GIS files using a combination of tools and techniques, including global positioning system (GPS) location tagging, narrated video logs in the field, and manual measurements in ArcGIS.⁽⁹⁾ The GIS files provided intersection locations, traffic control type (i.e., stop-controlled or signalized), and corner clearance at signalized locations. The HSIS data supplemented the GIS dataset with annual average daily traffic (AADT), reported crashes, number of lanes, lane width, speed limit, and other geometric characteristics. The GIS dataset included California and several other States. The research team initially considered all these candidate States. Ultimately, California was the only dataset collected and used for this study. California HSIS files provided cross-street name for each intersection, a key piece of information to linking GIS and HSIS data.

The research team implemented the following key steps in the data collection effort:

• Step 1—Generate the latitude and longitude of all intersections in GIS using ArcGIS’s Calculate Geometry tool. Export the attribute tables from ArcGIS into text file format, and then import the data into MS Excel and separate the intersections by traffic control type (i.e., stop-controlled or signalized).
  
  
• Step 2—Use Keyhole Markup Language (KML) to convert signalized intersection locations (GPS coordinates) from Step 1 into place markers for Google® Earth™. Import KML files into Google® Earth™.
  
  
• Step 3—Check candidate intersections to determine if they meet the following criteria:
  
  
  • At least 500 ft from another signalized intersection and at least 350 ft from a stop-controlled intersection. This effort used the Ruler tool in Google® Earth™ for distance measurement.
    
    
  • No irregularity in terms of configuration and operation (e.g., no frontage roads, no extreme skew angle) or location (e.g., not at freeway interchange).
    
    
• Step 4—Locate the intersection in the HSIS file, and mark it with the feature identifier (FID) for that same intersection from GIS. The FID is a unique identifier from ArcGIS and shown in the Google® Earth™ KML files. The research team used the cross-street names to relate sites across the two datasets. The street names of the upstream and downstream intersections were available for additional verification. The FID allows data matching from HSIS and GIS. In this step, the analyst also used the Google® Earth™ measurement tool to measure the length of right- and left-turn lanes on the mainline.
  
  

Figure 3 and figure 4 illustrate the process with an example of an intersection on Route 82 in Northern California. In this example, the analyst identified a signalized intersection at Henderson Avenue in Google® Earth™. This intersection is approximately 670 ft from the nearest stop-controlled intersection (Sycamore Terrace); there are no other stop-controlled intersections within 350 ft, and no other signalized intersections within 500 ft of this intersection. It meets the two criteria listed in Step 3 above, and the analyst selected it as a candidate. The cross-street name—Henderson Avenue, as shown in figure 3—was located in the HSIS intersection inventory in figure 4. The nearby intersection, Sycamore Terrace, was used to confirm the location of interest.

©2016 Google®.

Figure 3. Screenshot. Select study location in Google® Earth™ (circle added by research team to indicate intersection of interest).⁽¹⁰⁾

Source: FHWA, data acquired from HSIS.

Figure 4. Screenshot. Locate and verify intersection in HSIS data file.

The research team used milepost, county, and route numbers to identify and link crashes from the HSIS crash data files to each intersection. The team included all crashes that occurred within a 500-ft influence zone from the center of the intersection (i.e., 250 ft upstream and 250 ft downstream). They used the number of vehicles involved and crash severity to develop multiple vehicle and fatal and injury data categories. The research team used accident type (ACCTYPE) and movement preceding accident (MISCACT) to identify crashes for rear-end, sideswipe, right-angle, and turning (left-turn and right-turn) categories.

CHARLOTTE DATA COLLECTION

The data for Charlotte, North Carolina, came from the following two sources:

• HSIS. The research team obtained intersection, traffic, and crash data files from HSIS. The data came in GIS shapefiles that allowed the research team to employ various spatial analysis tools in GIS to process the data. The GIS data also provided intersection location information for data collection from Google® Earth™.
  
  
• Google® Earth™. The research team obtained corner clearance, intersection configuration, number of lanes, driveway density, and the general characteristics of the corridor on which the intersection is located from Google® Earth™ using satellite imagery, Street View™ images, and measurement tools.

These two data sources are further described in the following sections.

INTERSECTION, TRAFFIC, AND CRASH DATA

The GIS shapefiles were a part of a raw dataset processed from HSIS. The roadway shapefiles included all roadway segments in Charlotte, North Carolina. Key attributes of each segment included AADT and number of lanes. Intersection shapefiles have information on location (GPS coordinates) and traffic control types (e.g., signalized and stop-controlled). Crash data shapefiles had location information (GPS coordinates) and key crash characteristics to identify and separate crashes by crash type and severity. The research team imported these data files into ArcGIS as separate layers and used spatial and analytical tools to perform the following tasks:

• Determine intersection type. The research team used the type of traffic control in the attribute table of the intersection data layer to separate all signalized intersections. These candidate study locations went through a second round of screening, removing candidate intersections within 500 ft of another signalized intersection or within 350 ft of another stop-controlled intersection. The research team extracted identification number, location information (GPS coordinates), and intersection description (names of intersecting routes) for the final list of candidate intersections for supplemental data collection using Google® Earth™ (discussed in the next section).
  
  
• Determine number of legs, number of lanes, and AADT for each approach. The research team overlaid intersection and roadway layers, and used spatial analysis tools in ArcGIS to create a 10-ft buffer around each intersection, represented by the center of the intersection. The number of roadway segments within each 10-ft buffer represented the number of legs. In this process, the research team determined the number of lanes by approach and the maximum, minimum, and average AADT values associated with the roadway segments. The AADT values included 3 years of data (2009–2011). The research team used the AADT and number of lanes for classifying the mainline and cross street (i.e., the approach with more lanes and larger AADT was designated as the mainline).
  
  
• Identify and count crashes for each intersection. The research team used spatial and analytical tools in ArcGIS to count and assign crashes to each intersection. Specifically, they used a 250-ft buffer around each intersection and tallied the crashes within the 50-ft buffer assigned to each intersection. The Charlotte Department of Transportation recommended a 150-ft radius for assigning intersection crashes; however, for consistency with the California dataset, the research team decided to use a 250-ft radius from the center of each intersection. In addition to the total crash count, the research team used key crash characteristics, including severity, crash type, and light condition to identify fatal and injury, rear-end, right-angle, sideswipe, and nighttime crashes. The dataset included 3 years of crash data (2009–2011).

Figure 5 shows a screen capture of ArcGIS, illustrating these tasks. The lines represent roadways, and each circle represents the 250-ft radius from the center of an intersection. Each dot represents a crash. If a dot falls within a circle, that crash is counted and assigned to the intersection. It is also worth noting that crashes are assigned to intersections based solely on location (within 250 ft from the center of intersection).

Source: FHWA, data acquired from HSIS.

Figure 5. Screenshot. Example of Charlotte data layers in ArcGIS.

CORNER CLEARANCE, INTERSECTION, AND CORRIDOR CHARACTERISTICS

The research team used KML to create place markers in Google® Earth™ for all candidate study intersections exported from ArcGIS, as described in the previous section. Intersection location information (GPS coordinates) was used to place a marker at the center of each intersection. Intersection identification numbers and descriptions were coded to attach to each marker for easy identification and verification of the location. After creating and importing the KML file into Google® Earth™, the research team manually collected and confirmed the following data elements:

• Corner clearance. The research team used the measurement tool in Google® Earth™ to measure the distance from the corner to the nearest driveway.
  
  
• Number of driveways. The research team counted the number of driveways on both sides of the road and the length of the segment in which these driveways were located. The count and measurement extended two to three traffic signals upstream and downstream from the signalized intersection of interest. Number of driveways and distance were used to calculate the driveway density.
  
  
• Median type. The research team visually determined the type of median in the vicinity of the intersection.
  
  
• Presence and lengths of turning lanes. The research team collected both the presence and lengths of exclusive left- and right-turn lanes.
  
  
• Type of land use. The research team used Google® Street View™ to visually determine the land use type (i.e., residential, commercial, or mixed-use) in the vicinity of the intersection.

In this process, the research team also verified number of legs, number of lanes, and the designation of the mainline and cross streets collected using the GIS tools described in the previous section. In some instances, the research team identified discrepancies between GIS data and Google® Earth™ related to intersection configuration and number of lanes. For discrepancies, data from Google® Earth™ were used.

Figure 6 shows the use of the measurement tool for collecting corner clearance from Google® Earth™. At this location, there are no driveways or access points within 250 ft of the signalized intersection along the mainline. As such, this site was a candidate reference site.

©2016 Google®.

Figure 6. Screenshot. Measuring corner clearance in Google® Earth™.⁽¹¹⁾

DATA SUMMARY

The research team collected and aggregated 3 years of data for the analysis. Table 3 presents the summary of the final dataset with 275 signalized intersections included in the analysis. The final dataset accounts for the dataset corrections discussed in chapter 6 and propensity score matching. Indicator variables are either 0 or 1, indicating the absence or presence of the characteristic, respectively. The mean value of an indicator variable represents the proportion of sites for which the indicator is 1. For example, the indicator for 50 mph or higher posted speed on the mainline in table 3 has a mean value of 0.44. This implies that 44 percent of locations have a posted speed of 50 mph or higher (indicator value = 1) and 56 percent of locations have a posted speed of less than 50 mph (indicator value = 0). It is worth noting that there are overlaps between turning crashes and other crash types (e.g., a rear-end crash can be related to a turning maneuver, so it was also coded as a turning crash).

Table 3. Data summary for signalized intersections and corner clearance.