Design Criteria for Adaptive Roadway Lighting

CHAPTER 2. LIGHTING AND CRASH RELATIONSHIP

One of the first tasks of the project was to evaluate the link between lighting and crash rate. As mentioned, previous research has shown roadway lighting will affect crash risk. Specifically, the relationship being investigated is the link between lighting levels and lighting quality and crash rates. Establishing such a relationship would allow for determining the optimal lighting level for roadways under various traffic conditions. It should be noted that the weather conditions were not included in this analysis and provide the basis for future research.

Currently, there are four criteria to consider in the design of roadway lighting: roadway illuminance (both horizontal and vertical), luminance, and uniformity. Horizontal roadway illuminance is the amount of light falling on the roadway surface; vertical illuminance is the amount of light falling on a vertical surface, such as a pedestrian; luminance is the amount of light perceived by the road user; and uniformity is the ratio of illuminance or luminance values, such as maximum to average, average to minimum, or maximum to minimum. The data collection system for this project was able to measure each of these criteria.

HORIZONTAL ILLUMINANCE

The horizontal illuminance was calculated as the average of four illuminance levels measured by detectors at the top of the data collection vehicle. The mean value for a road segment was based on the average of multiple trips for the road segment. The distribution of the mean values is shown in figure 11.

Figure 11 . Distribution of mean horizontal illuminance level.

The mean horizontal illuminance values were grouped into eight categories based on the characteristics of roadway lighting and the distribution of lighting for road segment samples, as shown in table 8. Note that the bins for the values in table 8 are different from those in figure 11 because the histogram focuses on the distribution of the samples.

Table 8 . Horizontal illuminance lighting level.

* 0~0.5: segments with mean horizontal illuminance variables between 0 and 0.5 lux.

Figure 12 shows the relationship between the horizontal illuminance level and weighted mean night-to-day crash rate ratio. (The illuminance values are the average for each bin of data.) The weighing method is discussed in figure 8. This analysis includes all the lighting that was captured by the measurement system. The data below 3 lux can be considered indicative of no roadway lighting. These data can also be considered as maintained lighting levels because the systems were measured as-is with no estimation of the impact of the age of the luminaire system. The primary indication from this figure is that there is a distinct relationship between lighting and the NTDCRR.

The linear model results link the NTDCRR to lighting levels, and the results indicated there is a significant decrease in NTDCRR with the increase of average horizontal lighting levels (model 1: regression coefficient = -0.02, p-value < 0.001). The comparison of NTDCRRs by discretized lighting levels based on model 2 indicated that the NTDCRRs of lighting levels 0 and 1 are significantly higher than other levels. However, there is no statistically significant difference for lighting levels 2 to 7. This is consistent with the observation from figure 12 that an increase in the lighting level from 5 lux to higher levels does not appear to affect the crash rate ratio. There appears to be a further reduction in the crash rate at approximately 16 lux; however, because only 28 mi (45 km) of data are in this category, the result is not statistically significant. These results indicated that although lighting will benefit road safety, increasing the lighting level does not necessarily always lead to a safer road. There is potentially over-lighting under current practice and an opportunity for adaptive lighting design.

Figure 12 . Graph. Relationship between mean horizontal illuminance level and weighted night-to-day crash rate ratio. Best fit line R² = 0.7944.

Note that in all figures that contain error bars, these bars represent standard error, and the curve fits are second order polynomials.

VERTICAL ILLUMINANCE

The vertical illuminance was based on the illuminance detector positioned behind the windshield inside the vehicle. The tint of the windshield can reduce the illuminance value up to 30 percent.⁽¹⁴⁾ To account for making measurements from inside the vehicle, the measured values of vertical illuminance were multiplied by 1.5. As an example, a measured value of 2.5 vertical lux inside the vehicle is corrected to a value of 3.75 lux outside the vehicle. The distribution of vertical illuminance metrics by road segment is shown in figure 13 . According to the relationship between the horizontal and vertical illuminance measures discussed above, the categorization of vertical illuminance is shown in table 9 .

Figure 13 . Graph. Distribution of mean vertical illuminance level.

Table 9 . Vertical illuminance lighting level.

* 0~0.3: segments with mean vertical illuminance between 0 and 0.3 lux.

Figure 14 shows the vertical illuminance results compared with NTDCRRs. The relationship of these results to the crash rate ratio is similar to that of the horizontal illuminance. This is an expected result; with the typical optics of a luminaire on the roadway, the horizontal and vertical illuminances are intrinsically linked. The results from the regression model indicated that there is a significant decrease with the increase of lighting levels (model 1 coefficient: -0.04; p-value < 0.0001). The comparison using model 2 indicated that the NTDCRRs for lighting levels 0 and 1 are significantly higher than other levels. There is no statistically significant difference among the other levels.

Vertical illuminance, however, has two components that can affect the driver. The first is the highlighting of the vertical surfaces of objects in the roadway, providing contrast and visibility. The second is glare (a bright or disturbing light source that causes discomfort or disables vision), which is caused by light reaching the driver's eye and can limit the driver's visibility. The method used for measuring vertical illuminance more closely represents a measure of glare, however, with the assumption that the vertical illuminance measured from the driver's viewpoint can be related to the visibility provided by the lighting system.

The results of the present investigation show that the minimum required value may be as low as 3 lux (based on a visual inspection of the relationship presented). Previous research has shown that the average vertical illuminance level required for the detection of pedestrians in a midblock crosswalk is 20 lux.⁽¹⁵⁾ This is an interesting comparison in that the lower values of vertical illuminance found in this investigation may indicate that the vertical illuminance level of 3 lux is adequate for drivers to perceive their surroundings, while still not being adequate for the rapid and accurate identification of pedestrians required for the perception of the roadway.

Figure 14 . Graph. Relationship between mean vertical illuminance level and weighted night-to-day crash rate ratio.

VERTICAL-TO-HORIZONTAL ILLUMINANCE RATIO

As mentioned, vertical illuminance has a dual impact on the driver: visibility of objects in the roadway and also as a measure of glare. Drivers perceive glare as the difference between bright light sources in the field of view compared with other sources. One measure of the potential impact of glare on the driver is the vertical-to-horizontal illuminance ratio. The distribution of the mean of the vertical-to-horizontal illuminance ratio is shown in figure 15 . (A few outliers greater than 5 were censored for clarity.) Because there are no historical data to guide the threshold of classification, the classification was based on an approximately equal number of road segments in each class as shown in table 10 .

Figure 15 . Graph. Distribution of vertical-to-horizontal illuminance ratio (censored at 5).

Table 10 . Vertical to horizontal illuminance lighting level.

* 0~0.15: segments with mean of vertical to horizontal level between 0 and 0.15.

Figure 16 shows this relationship between the vertical-to-horizontal illuminance ratio and NTDCRR. As can be seen, there is a general trend of an increasing night-to-day crash rate ratio with an increasing vertical-to-horizontal illuminance ratio. This is expected because the visibility of objects in the roadway can be limited by glare. Because the measured data are limited and do not explore the full potential range of the vertical-to-horizontal ratio, this study cannot fully explore the boundaries of this relationship. However, a maximum ratio of 1.0 vertical-to-horizontal illuminance, and a desired value of less than 0.6 appears to be supported by the data.

The regression analysis indicated that there is no significant association between NTDCRR and vertical-to-horizontal ratio metric (model 1 coefficient: 0.08, p-value = 0.11). No statistically significant differences in NTDCRR were detected based on model 2.

Figure 16 . Graph. Relationship between mean vertical-to-horizontal illuminance ratio level and weighted night-to-day crash rate ratio.

UNIFORMITY

The lighting uniformity measure is the difference in local maximum and local minimum lighting levels caused by the location of a lighting pole or other factors. This is different than the UR typically used in the design of roadway lighting. Because the data were measured continuously, there were points where the values dropped to significantly lower values than may be seen in a calculation environment. (Variations for ideal luminaire performance and pole spacing as well as ambient conditions affect the measurement.) As a result, the UR, as traditionally defined, was not usable because of high variation and potential infinite values. Thus, the uniformity difference was used instead.

The team developed an algorithm to identify local maximum and local minimum at road segment level, as illustrated in figure 17. The mean lighting level for the road segment was calculated first. The road segment was then divided into shorter subsegments based on whether the lighting level is above or below the mean lighting level. In this setup, a high lighting subsegment (above mean lighting level) is always adjacent to a low lighting subsegment. The difference between the maximum of the high lighting subsegment and the minimum of the adjacent low lighting subsegment is the base for uniformity measure. Furthermore, the local maximum/minimum value is usually surrounded by many non-extreme values. To alleviate potential bias caused by extreme values, the average of the highest five points and lowest five points in each high/low lighting subsegment was used to calculate the uniformity measure. The uniformity for a segment is defined as the mean of all pairs of difference between local maximum and minimum.

Figure 17 . Graph. Uniformity measure. The distance index is a counter used to index the data from the measurement system.

The distribution of the uniformity measure is shown in table 19 (which only shows values up to 20 to avoid outliers). Because there is no information on how the uniformity measure would affect safety, we categorized the uniformity level based on an approximately equal number of segments to ensure a reasonable number of samples in each bin for crash risk evaluation. The categorization of uniformity measure level is shown in table 11 .

Figure 18 . Graph. Distribution of uniformity measure.

Table 11 . Uniformity lighting level.

* 0~0.1: segments with mean uniformity measure between 0 and 0.1.

Figure 19 shows the relationship between the uniformity difference and the night-to-day crash rate ratio. The inference does not show a statistically significant pattern (model 1 coefficient: 0.01, p-value = 0.13). Results from model 2 indicate the NTDCRR for level 0 is significantly higher than the other levels, with no significant difference among levels 1 to 7.

As a result of the nature of the uniformity difference calculation, the higher the uniformity value, the less uniform the lighting on the roadway. These results indicate that as the uniformity difference increases and the roadway becomes less uniform, the crash rate ratio decreases; however, there may be a limit where additional non-uniformity does not assist and may be a detriment. Other investigations have seen this same result: when the lighting on the roadway is not completely uniform, the potential for greater object contrast on the roadway increases and, as a result, so does the potential for object detection and a reduction in crashes.⁽¹⁶⁾

Figure 19 . Graph. Relationship between UR level and weighted night-to-day crash rate ratio.

LUMINANCE

The final metric to consider is luminance. The luminance of the roadway was measured from inside the windshield of the vehicle. The transmissivity of the glass can limit luminance inside the windshield by as much as 30 percent.⁽¹⁴⁾ These data were scaled by this factor to allow for comparison with lighting designs, which are measured external to a vehicle.

The luminance reduction program used allows the area of interest to be selected to allow different objects within the same image to be evaluated. For this analysis, the luminance measures were made for all of the visible roadway area, therefore extending the width of the roadway lane laterally and approximately 50 to 250 ft (15.2 to 72.2 m) in front of the vehicle. As a result, it includes the maximums and the minimums-essentially an average of the roadway space, not just the luminance at 272 ft (83 m) from the vehicle as specified by ANSI/IES RP-8-2005.

The distribution of the luminance is shown in figure 20 , and the categorization based on the principle of equal number of road segments is shown in table 12 .

Figure 20 . Graph. Distribution of luminance measure.

Table 12 . Luminance lighting level.

* 0~0.15: segments with mean of luminance between 0 and 0.15.

The relationship between luminance and night-day crash rate ratio is shown in figure 21 . There is no significant trend in the relationship between NTDCRR and luminance (model 1 coefficient: 0.23, p-value = 0.4). However, comparison with model 2 indicated that the NTDCRR for level 0 is significantly smaller than for other levels.

The relationship of the luminance with the NTDCRR is not what one would expect. The figure shows that an increase in the luminance may increase the crash rate ratio. It is important to note that this luminance measurement includes the vehicle's headlamps, which are not included in a lighting design on a roadway. Because the luminance metric was measured for an area rather than a single point as a design would be, and is likely dominated by the headlamps' performance, it is not clear whether these results can be further evaluated and used as a design recommendation. A significant investigation of the interaction of the roadway luminance with and without headlamps is required to substantiate these values.

1 cd/m² = 0.292 ft-lamberts
Figure 21 . Graph. Relationship between mean luminance level and weighted night-to-day crash rate ratio.

IMPACT OF ROADWAY TYPE

To fully explore the requirements for the lighting level on a variety of roadway types, the functional class of the roadway from the roadway data was used to further analyze the horizontal-illuminance-to-crash-rate relationship. A linear regression model was applied to the data to determine the impact of the lighting and the functional class. The results are shown in table 13. Overall, the impact of the increased lighting is significant and follows the trend of reducing crash rates. The functional class does not, however, appear to be statistically significant. Note that the functional class was provided in the roadway design data gathered from the State transportation departments as part of the project.

Table 13 . Linear regression results for functional roadway class and lighting level.

To further investigate, a pairwise comparison of the NTDCRRs by lighting level was performed. For each functional class, the lighting levels were all compared with lighting level 0 or level 1. (The base level was selected to be the one with the highest NTDCRRs.) These results by functional class are shown in table 14 through table 17 along with the corresponding lower confidence limit (LCL) and upper confidence limit (UCL). Here the significant differences can be determined from the p-values (p < 0.05). For the Urban Interstates (table 14), the comparison shows that there is a slight difference from level 1 to 2 but no significant difference beyond level 2. (It is noteworthy that all pairs of levels were compared, but only results with respect to a base level were included in this report.) For the Urban Principal Arterials, the pairwise comparisons show that there seems to be no significant difference between levels 2 and 4. For the Other Principal Arterials, the differences are significant beyond level 5. Finally, for the Minor Arterials, the data are more difficult to analyze because the number of data points was very small, but the difference seems to start at level 6.

Table 14 . Pairwise Comparison of lighting level for Urban Interstates.

Table 15. Pairwise comparison of lighting level for Urban Principal Arterials.

Table 16. Pairwise comparison of lighting level for Urban Other Principal Arterials.

Table 17 . Pairwise comparison of lighting level for Urban Minor Arterials.

Figure 22 shows these results. The dashed horizontal lines in figure 22 highlight the area of similar performance in terms of the lighting system and the relationship to the night-to-day crash rate ratio for each roadway type. Using visual inspection, the results of the pairwise comparison can be verified because this horizontal line can be used to determine the minimum lighting required to optimize safety for each roadway type. Beyond this minimum value, an increase in illuminance does not appear to affect the overall safety of the roadway.

Figure 22 . Graph. Relationship between horizontal illuminance and weighted night-to-day crash rate ratio by roadway functional class.

Table 18 shows the minimum horizontal illuminance requirements for the functional classes of roadway, which is determined using both the pairwise comparisons and the visual inspection analysis method.

Table 18 . Minimum horizontal illuminance limits from the functional class analysis.

Another aspect of figure 22 is the different relationship of the lighting level to the crash rate for the minor arterial. This roadway functional class appears to be more significantly affected by the increase in lighting level than the freeways and the major arterials. The crash rate ratio for the minor arterials drops from 1.5 to 0.5 as the lighting level increases, whereas it only drops to 1.0 for the freeways. This indicates that the minor arterials are safer at night with lighting than during the day. This is likely linked to the lower traffic volume at night and a reduced level of potential conflict with other vehicles at intersections and driveways.

The IES divides roadway lighting recommendations into street and highway, the primary difference being the presence of pedestrians. Applying these classifications to the IES requirements, the minor arterials were classified as streets, and the other functional classifications were classified as highways.¹

Figure 23 shows the relationship of the crash rate ratio with the lighting level for highways and streets. Obviously, the same relationship exists as mentioned above. The comparison of NTDCRR for lighting category stratified by highway and street is shown in table 19.

Again a pairwise comparison based on model 2 was used to look at the impact of the lighting level for the functional class. For highways, the results were compared with lighting level 3 (6-9 lux). Based on this comparison, significant differences are evident only with level 1 and level 0. This indicates that the minimum required lighting level appears to be in the range of 3-6 lux for highways. Similarly, the street comparison for level 4 shows a significant difference starting at level 5 (12-15 lux), which indicates that the required lighting level is in the range of 12-15 lux. Because the project did not set out to measure the minor arterials, there is not as much data in this category as there is for the highway classification. As a result, the relationship is not as strong as the highway relationship.

Table 19 . Comparison of NTDCRR for lighting category stratified by highway and street.

Figure 23 . Graph. Relationship between mean horizontal illuminance and weighted night-to-day crash rate ratio.

COMPARISON TO IES STANDARDS

As mentioned, the ANSI/IES RP-8-00 standards provide guidelines for selecting roadway lighting levels. Table 20 provides the recommended illuminance levels from the IES. For comparison with the results of the current study, Freeway Class A is comparable to the Urban Freeway functional class, Freeway Class B is comparable to Major Arterial, Expressway is comparable to Arterial, and Major is comparable to Minor Arterial. Figure 24 shows the comparison of these values with those found in this project. The comparison is made for Road Surface Types R2 and R3, and the high and low range is shown for the pedestrian classes.

Table 20 . IES illuminance requirements (ANSI/IES RP-8-00).⁽⁴⁾

Figure 24 . Graph. Comparison of the IES requirements with the results of the crash analysis.

One of the more striking results is that the calculated required lighting level for the urban freeways is significantly less than the IES-recommended level. A level of 4 lux was required, whereas the recommendation is 9 lux. This indicates a possible opportunity to reduce lighting on this class of roadway by more than 50 percent. These results suggest the IES recommended lighting level could be reduced for Urban Interstates, but the existing recommendations are suitable for the other classifications. Additional research is required to fully investigate the requirements for detection of pedestrians.

HOURLY ANALYSIS

The other aspect of the lighting impact that is of interest is the relationship to the time of day. The hypothesis is that as the usage of roadways and streets changes during the night, there is the potential to reduce or adapt lighting to the conditions on the roadway. This might include the adjustment of the lighting to traffic or pedestrian volumes. To more fully investigate this issue, the crash rate by the time of day was examined, along with the hourly traffic volume.

The first step of this analysis was to gather data for the crashes and crash rate for the time of day. Only the crashes that occurred at night were included in this analysis. The nighttime was defined by using the National Weather Service sunrise and sunset times. This analysis was also performed only for the State of Washington, which had the most reliable hourly crash and volume data. Figure 25 shows the total number of crashes during hours of dark for the years 2004 to 2008, separated by hour of the day. As predicted, the total number of crashes diminishes during the nighttime period, with the minimum occurring between 3 and 4 a.m.

Figure 25 . Graph. Total number of crashes for Washington from 2004 to 2008
by time of day.

As the traffic volume diminishes through the night, the number of crashes per vehicle mile varies. Figure 26 shows the average crash rate (number of crashes per million vehicle miles traveled) by the time of day for the same data set. The peak at 2 a.m. is significant and may be caused by several factors external to the lighting condition, such as fatigue and alcohol use.

This relationship of the overall number of crashes and the crash rate is an interesting result because it indicates that minimizing the crash rate for the hours from midnight to 3 a.m. will not have as great an impact on reducing injuries and loss of life as can be achieved by reducing the overall number of crashes in the evening hours.

Figure 26 . Graph. Average crash rate by time of day for Washington from 2004 to 2008.

The relationship between crash rate and time of day indicates that factors other than darkness influence the crash rate. This implies that some level of crashes will occur regardless of the presence of lighting and may be related to fatigue, alcohol, and other factors.

SUMMARY

This analysis shows that there is the potential to reduce lighting on roadways during periods of reduced traffic and potential conflict while maintaining the overall level of roadway safety. The data also show there is a potential to reduce standard lighting levels by as much as 50 percent for the Urban Interstate functional class.

¹ For clarity, this discussion will refer to the IES roadway class as highways so as not to confuse it with the roadway in general.