Using lamp density as a road lighting metric

Reliable data about road lighting metrics such as illuminance and uniformity usually require manual measurements to be taken with an appropriate light meter, following specified procedures. This can be resource-intensive and time-consuming, and it is therefore difficult to obtain lighting measurements for a large number of locations using this method. An alternative source of available data that may offer an approximation of light metrics is the frequency and distribution of road lighting lamps. Such data is held by Local Authorities in databases that include the spatial coordinates of every lamp within their inventory. To assess whether such data can provide some degree of information about the lighting on a particular street, we carried out a validation exercise. This involved comparing lamp density on a street with an estimate of brightness on that street, obtained from images captured from night-time aerial photography.

In 2009 the Environment Agency carried out a night-time aerial survey of Birmingham, UK, capturing high-resolution images of the city at night from a height of 900 m. Each pixel in these open-source images represents an area of 10 cm² and has an associated intensity value in the red, green and blue channels. Following work by researchers at Birmingham University, we resampled these images from 10 cm² to 1 m² pixel resolution, and converted the colour images to a greyscale image, with each pixel now only providing one intensity value – see Figure 1. This ‘brightness’ pixel value has previously been shown to have a strong association with actual measured illuminance on the ground at the same location (Hale et al, 2013 – 400 measurements provided an R2 value of 0.91).

Using GIS software we drew a 15 m buffer around every road segment within Birmingham and calculated the mean pixel intensity within these buffers – this indicating an average level of brightness on each road. We also plotted the locations of road lighting lamps within Birmingham based on open-source data available from the Birmingham data factory. Using the same 15 m buffer zones, we calculated the lamp density for every road, as the number of lamps per 100 m of road. Visual inspection of the brightness imagery overlaid with lamp positions suggested some roads with a calculated lamp density of zero appeared to actually have some lighting present.

Figure 2 illustrates this. It is evident that the road running from top right to bottom left is lit on the left hand side of the image, but no lamps are shown on this stretch. Google StreetView confirms that lighting is present. This suggests lamp data may be missing for some roads. One reason for this is the lighting may not be maintained by the Local Authority but by another Authority such as Highways England (e.g. on primary trunk roads and motorways), or part of a private estate. An example of this is also shown in Figure 2 – the network of roads in the top left of the image is part of the University campus. These roads appear to be lit from the aerial image, but no lamps within the database we have access to are shown. This is likely because the lighting on the campus is maintained by the University rather than the Local Authority.

As roads with apparently no lighting present (according to the database of Local Authority lighting we are using) may in reality have some form of lighting, such roads were excluded from further analysis as the calculated lamp density of zero may be unreliable. All roads with lamp densities greater than zero were still included. These roads were grouped into 0.5 ‘density bins’ (eg all roads with lamp densities below 0.5 were grouped together, all roads with densities between 0.5 and 0.99 were grouped together, and so on), and the mean pixel brightness calculated for each of these bins. Roads with lamp densities greater than 8 lamps per 100m were excluded due to their small number and potential to skew results (such roads represented only 1% of all roads within Birmingham). Mean pixel values by lamp density bin are plotted in Figure 3. Linear regression suggested a good association between lamp density and pixel brightness (R2 = 0.60, p < .001). For every additional lamp per 100m of road, the mean pixel brightness of the road is predicted to increase by 0.24.

Conclusion

1. There is a strong association between the density of road lighting lamp-posts and road brightness (after dark) as observed using aerial photography.

2. Current road lighting data may contain gaps, and there is a need to collate more comprehensive and detailed data about road lighting.