Design Criteria for Adaptive Roadway Lighting

CHAPTER 1. DATA COLLECTION APPROACH

The development of design criteria requires establishing a link between the lighting level and roadway safety. This relationship will establish the possibility of defining the design criteria as well as determining the important parameters that may affect the lighting design level. To establish these criteria, a significant amount of data was obtained from participating States and through field measurements. A significant amount of data was also gathered from the Highway Safety Information System (HSIS), which contains linked crash and roadway data for State-maintained roadways in participating States.

STATE DATA

Data collected from the States included crash data, lighting design, traffic data, and roadway data. Each of these data categories was loaded into a geographic information system (GIS) data system to geographically relate all the data measures. Because of inconsistencies in data format and variables from multiple data sources, separate data structures were maintained for each data type and each State. A geocoded system relates and compares the collected data.

Based on geographic coverage as well as availability of crash and lighting pole location data, seven States were selected for in situ data collection. Participating States in the study were California, Delaware, Minnesota, North Carolina, Vermont, Virginia, and Washington.

The sources for each State are shown in table 4.

Table 4 . Primary State-level data sources.

Caltrans = California Department of Transportation
DelDOT = Delaware Department of Transportation
MnDOT = Minnesota Department of Transportation
NCDOT = North Carolina Department of Transportation
VDOT = Virginia Department of Transportation
VTrans = Vermont Agency of Transportation
VTTI = Virginia Tech Transportation Institute
WSDOT = Washington State Department of Transportation

LIGHTING PERFORMANCE DATA

The lighting performance data were collected in the field and convey detailed information on the performance of the lighting systems at the time of collection. This study represents the first national-level effort to collect such data, which are a major contribution of this research.

Lighting performance was measured in situ using a mobile data collection system developed by the research team. The Roadway Lighting Mobile Measurement System (RLMMS), developed by the Virginia Tech Transportation Institute (VTTI), has the ability to collect data regarding horizontal illuminance, roadway luminance, glare from oncoming traffic and other external light sources, and Global Positioning System (GPS) position. The system also has input buttons to flag special features in the data stream.

The RLMMS contains four waterproof illuminance detectors. Each detector is positioned on the end (facing upward) of each of four arms of an apparatus (also known as the "Spider"), which is mounted on the roof of a vehicle. Positioned in the center of the four arms is a GPS receiver. The GPS's Universal Serial Bus cable and the detector leads are routed into the vehicle via a rear window data collection box. A fifth illuminance detector is mounted on the windshield, facing forward, to measure vertical illuminance on the windshield. This value is a surrogate for oncoming glare.

The illuminance measurements are coupled with luminance and color cameras that take snapshots of the roadway. The cameras are connected to their own stand-alone computer for housing the images, which can also be viewed on a single laptop screen along with the other data being collected in the vehicle by using a Controller Area Network reader. Validation of the cameras included rigorous calibration and tuning. Further details of the analyses of the camera's output are also detailed in Meyer et al.⁽¹²⁾ The system layout is shown in figure 3. The GPS unit and the illuminance meter network are placed on the roof of the vehicle on an aluminum framework, which is attached using suction cups. The color camera, the luminance camera, and the vertical illuminance meter are attached on the inside of the windshield with suction cups.⁽¹³⁾ The data from the RLMMS have been found to be both repeatable and reproducible through standard repeatability and reproducibility techniques.⁽¹²⁾

Figure 3 . Diagram. Roadway Lighting Mobile Monitoring System (RLMMS).⁽¹⁾

Members of the project team drove on the routes selected and collected lighting data at a frequency of 20 Hertz (Hz) and at a video capture rate of 3.75 frames per second (fps). The data collection vehicle adhered to the posted speed limit and followed the direction of roadway signage. Because of the significant number of miles over which the lighting performance data were collected, multiples files were generated.

The acquired data were integrated with ARC-GIS^™ software with the capability of converting the roadway lighting data gathered and forming a visual database.

As shown in Figure 4, the measured lighting level is captured along the roadways by the four detectors (left, front, right, and rear). Because the angle and length of the line of sight between each detector and each luminaire are different, there are differences in the detector readings. As a result, detectors were processed separately in the data reduction. Note the data have been smoothed using a moving window algorithm to mitigate the fluctuation in lighting caused by locations of lighting poles.

1 mi = 1.6 km
Figure 4 . Graph. Field-collected lighting data (1-mi segment).

Another issue in field data collection is that the location of the lighting pole varies from segment to segment, for example on the left or right side. To minimize the resulting discrepancy in lighting measurement and to increase the accuracy of the measurements, data were collected in the outermost and innermost lanes for roadways consisting of at least two lanes in each direction. The only time these lanes were not used was for construction or cases in which emergency vehicles were parked on the shoulder.

DATA ANALYSIS

The data analysis focused on the relationship between lighting level and crash risk during nighttime. Specific emphasis was on whether roadway and traffic characteristics affect this relationship. The analysis was conducted using the in situ lighting data collection and historical data, including traffic and crash information.

The nighttime crash risk can be measured by several metrics. Simple crash rate for a road segment, as measured by number of crashes per million vehicle mi traveled, is shown in the equation in figure 5.

Figure 5 . Equation. Crash rate for States with traffic volume information.

The nighttime crash risk can be affected by many factors other than lighting, for example geometric design features and traffic control. The daytime crash rate at the same road segment was used as a control for these potential confounding factors. For each road segment, the night-to-day crash rate ratio (NTDCRR) was calculated as shown in the equation in figure 6. The NTDCRR reflects the relative magnitude of nighttime crash risk compared with daytime crash risk. Because daytime and nighttime crashes shared the same road design features and traffic control features, the rate ratio directly reflects the factors that only differ by day and night, with visibility level being the primary one. Therefore, the NTDCRR indicates the impact of lighting and light levels while controlling for the impact of the roadway design, traffic concerns, and other factors associated with road segment characteristics. As such, the NTDCRR was considered the primary metric to evaluate the effects of roadway lighting.

Figure 6 . Equation. Night-to-day crash rate ratio for States with day/night traffic volume information.

The calculation of night-to-day crash rate ratios requires night and day traffic volumes. In this study, we used continuous hourly traffic count stations in close proximity to the study road segments and average daily traffic (ADT) on the study road segments to estimate hourly traffic. The National Oceanic and Atmospheric Administration (NOAA) sunrise and sunset times were used to classify the natural lighting into day, night, and twilight conditions. Twilight was defined as 30 minutes before sunrise and 30 minutes after sunset. The nighttime traffic volume was estimated based on the time of nighttime and hourly traffic volume.

The data availability varies by State. As listed in table 5, not all States provided hourly traffic count station data and thus no night and day traffic volume information. For States without day/night traffic volume information (i.e., Delaware, North Carolina, and California), the night-to-day crash ratio as shown in the equation in figure 7 can be used.

Figure 7 . Equation. Night-to-day crash ratio for States without day/night traffic volume information.

The night-to-day crash ratio does not factor in the differences in day and night traffic volumes. To make the comparison among road segments meaningful, the underlying assumption is that night-to-day traffic volume ratio should be similar across road segments. Therefore, the results in the following analysis were based on the four States with night/day traffic volume (i.e., Washington, Minnesota, Virginia, and Vermont).

Table 5 . Day and night traffic volume indication per State.

The analysis is based on road segments. All major data sources, including crash, traffic volume, road characteristics, and lighting data, were integrated based on spatial relationship, the processing of which was only feasible in a GIS environment. In a GIS base map, roads are represented by a series of arcs, which is the minimum spatial unit with homogeneous characteristics. As a consequence, the segmentation of the road was determined by the minimum arc unit on GIS base maps. The length of the road segment is typically short to ensure relatively homogeneous characteristics for the segment. The summary statistics of the road segments by State are shown in table 6.

Table 6 . Road segment summary statistics.

1 mi = 1.6 km

For a specific lighting level, there are multiple segments. Longer segments should have more exposure to crashes and more stable estimation. A weighted rate ratio was used to estimate the average NTDCRR for a given lighting level, as shown in the equation in figure 8.

Figure 8 . Equation. Weighted crash rate ratio.

Statistical Models

The crash rate ratio is a random variable and has its range from 0 to positive infinity. Through log transformation, it can be modeled as a normal distribution as shown in figure 9.

Figure 9 . Equation. Normal distribution equation for the crash rate ratio.

Here Y_i is the night-to-day crash rate ratio for road segment i ; λ_i is the expected log crash rate ratio for segment i. The length of segment serves as a weighting factor, such that longer segments would have a larger impact on inference results. The λ_is are linked to lighting levels as shown in figure 10.

Figure 10 . Equation. Regression link for lighting.

The lighting_i is the measured lighting level for segment i, and the β values are regression results. There are a couple of ways to define the lighting_i variable: model 1, the average lighting measure for segment i (continuous), or model 2, the discretized average lighting measure. Model 1 allows the evaluation of the overall trend of NTDCRRs with increased lighting levels. Model 2 allows the comparison of NTDCRRs among different discretized lighting levels.

Lighting Metrics

The comprehensive lighting data collected in this study allow multiple characteristics of the lighting to be evaluated. In this study, five metrics were used: 1) horizontal illuminance, 2) vertical illuminance, 3) vertical-to-horizontal illuminance ratio, 4) lighting uniformity measure, and 5) luminance. Detailed definitions for each metric are provided in later sections.

The original lighting data were collected based on time (20 Hz), with the spacing of data points dependent on the vehicle's speed. For example, there could be hundreds of data points corresponding to one location when the vehicle was stopped at an intersection. In contrast, when the vehicle was traveling at 65 mi/h (105 km/h), two adjacent data points could be more than 3.3 ft (1 m) apart. Thus, a reasonable approach is to base the analysis on distance instead of data points.

An algorithm was developed to calculate the distance between two data points based on the projected coordinates. The data were then grouped based on spatial distance. The segments were divided into 13.1-ft (4-m)-long sections. The median of all data points within a 13.1-ft (4-m) section was used to represent the lighting level in the section. This approach effectively addressed the uneven spacing caused by vehicle speed and provided a more accurate lighting level measurement.

The mean lighting metrics for the horizontal illuminance, vertical illuminance, vertical-to-horizontal illuminance ratio, and lighting uniformity measure were calculated as the average metric values from the distance-sampled lighting points. Because there were at least two trips for each road segment, the average of these trips was used to represent the mean lighting level in the segment.

Crash Data

Crash data were aggregated into nighttime, daytime, and twilight periods, according to the location, date, crash time, and sunrise/sunset time calculated from the NOAA chart. Only daytime crashes and nighttime crashes were used in the analysis. The summary of crashes by State is shown in table 7.

Table 7 . Summary of crash data by State.