Outdoor artificial nighttime lighting, from streetlights and other sources, has increasingly been recognized as a significant anthropogenic pressure on the environment (e.g., [1
]). The disruption it causes to natural light regimes has a wide range of impacts, in large part because of the central role that these regimes commonly play in determining the timings of biological activity [4
]. Outdoor artificial nighttime lighting may contribute to a variety of chronic human health conditions ([5
]; but see [8
]), and the loss of views of celestial night skies has been argued to have a significant influence on people’s sense of place [9
], although, in both cases, empirical evidence is limited. The evidence is clearer with regard to other organisms, with impacts of outdoor artificial nighttime lighting on gene expression [10
], budburst and flowering times [11
], hormone cycles [13
], activity patterns [14
], foraging success [20
], growth [22
], mortality risk [24
], trophic interactions [26
], population abundances [28
], migration patterns [30
], community structure [33
] and ecosystem functioning and services [37
Key to determining where the biological impacts of artificial nighttime lighting are likely to be experienced, is an understanding of the spatial occurrence and variation in that lighting, and particularly how this changes with other factors [3
]. To date, from this perspective, this understanding has been rather limited. The most frequently studied correlates of spatial variation in artificial nighttime lighting have been levels of urbanization, human density and economic activity (e.g., [40
]). In part, this focus has been because urban infrastructure is clearly a major contributor to artificial nighttime lighting, and in part because of interest in the potential to use remotely sensed data of such lighting to track or hindcast changes in urbanization and human density, where such information is not directly available.
A few studies, typically at either very large (e.g., global) or quite small scales (e.g., cities, streets or microhabitats), have examined how artificial nighttime lighting changes with ecosystem type, land cover or land use. At a global scale, it has been found to be most prevalent in temperate climate ecosystems, followed by Mediterranean and tropical/subtropical ones [48
], and within urban areas to decline with levels of greenness and to increase with coverage by roads [44
]. In Brazil the most heavily affected vegetation types are terrestrial coastal (restingas and mangroves), semideciduous seasonal forests and mixed ombrophilous forests [51
]. Within individual cities or parts thereof, there have been attempts to determine the different sources of lighting (e.g., [50
]). Research at more intermediate scales has been lacking.
A small number of studies have also examined the levels of artificial nighttime lighting within and around areas protected for biological conservation (global—[52
]). The designation, establishment and maintenance of protected areas is widely considered key to buffering biodiversity from a diverse range of anthropogenic pressures [57
]. However, whilst levels of artificial nighttime lighting tend, overall, to be lower in these areas, many are not escaping such influences [53
], and lighting is particularly increasing in their vicinity, with implications for the lighting of horizons observed from within [55
One can recognize two different forms of artificial light at night: light emissions that come directly from sources, and diffuse skyglow (brightening of the nighttime sky that results from artificial light emissions being scattered by water, dust and gas molecules in the atmosphere). Almost invariably, to date, studies of spatial variation in artificial nighttime lighting have focused on measures of light emissions (but see [59
In this paper, we determine how light emissions and skyglow vary with human density, land use, roads and protected areas for the first time at a national scale, using Britain as a case study. This makes an interesting exemplar, as it is an area in which the environmental impacts of artificial nighttime lighting have attracted a good deal of policy and management attention (e.g., [60
]). We demonstrate that, although artificial light at night increases with human population density, the amount of light per person decreases with increasing urbanization. Overall, the amount of artificial light at night varies significantly across different land uses, and we demonstrate that light pollution is no longer solely an urban issue.
At a spatial resolution of c.500,000 m2
, the variation in both artificial nighttime light emissions and skyglow is marked. Although levels of the two are spatially correlated, inevitably the distribution of light emissions is more localized, whilst some level of skyglow was detected over the entirety of Britain. This latter observation is consistent with recent evidence that skyglow can extend hundreds of kilometers from urban sources (e.g., [87
]). It is also reflected in the longer distances over which a marked spatial autocorrelation was detected in skyglow compared to light emissions.
The maximum measured light emissions were more than 40 times greater than maximum modelled skyglow. To calculate measured light emissions for each pixel, we subtracted the modelled values of skyglow [59
] from the light emissions corrected for albedo. In green spaces located on the outskirts of large urban centers (known in the UK as the green belt), there were high levels of skyglow but few light emissions, which led us to calculate negative values for light emissions (which were subsequently corrected to 0.08 mcd/m2
). This suggests that the values of skyglow modelled by [59
] overestimate the degree of skyglow around urban centers, possibly because a constant value of atmospheric aerosols was used, although these might vary markedly within and beyond urban areas. This overestimation of skyglow is likely to be applicable across the entire skyglow atlas. Median levels of light emissions were greatest in England, with levels in Wales being two-thirds and levels in Scotland a third that of England. Only 7.5% of people in Britain are able to see the Milky Way where they live due to artificial light at night, because levels of both light emissions (8.3% of population) and skyglow (20.5% of population) are below 0.6 mcd/m2
Much of the spatial variation in levels of light emissions and skyglow was accounted for in models by the independent variables analyzed (human density, landcover, National Parks, SSSI and roads). Landcover was the strongest predictor of both dimensions of artificial light at night, followed by human density for light emissions and the road network for skyglow (Table S2
). Across Britain, we found an asymptotic relationship between the levels of light emissions or skyglow and human density (Figure 3
a), with the amount of light emissions per person decreasing with human density. This confirms work by previous studies at the national and regional scale [88
], and is supported here by the negative relationship with the indirect effects in the spatial model, which demonstrated that the more people that live in an area, the less artificial light is produced per person. Indeed, we found that the median light emissions per person were three times greater in rural than urban areas, and skyglow 11 times greater. Overall, we found that the indirect effects were stronger than the direct effects (indicated by a negative total effects), demonstrating that, although there was a positive relationship between human density and artificial light at night, there was a stronger effect of people living in close proximity in urban areas producing less light per person. Thus, resources are used more efficiently than in rural areas (for example, there can be significant variation in the number of people living and working around roads with similar densities of streetlights; [90
]). This can be clearly seen in England, where 84% of people live in urban areas ([91
]). Whereas, for example, in Scotland, which has a more rural population, if modelled in isolation, there were no significant indirect effects between light emissions and human density (Direct 1.5 × 10−3
± 4.8 × 10-5; Indirect −6.1 × 10−4
± 7.1 × 10−4
), while the median levels of both light emissions and skyglow per person were twice those of England or Wales.
As one might expect, the direct effects showed that artificial light at night was least in mountains, heath and bog and greatest in urban areas. The median levels of light emissions were higher than median skyglow for all landcover types, except for grassland and croplands. Central and south England have high degrees of skyglow, while also containing large areas of arable, improved grassland and semi-natural grassland (here termed grassland; Table S1
), habitats characterized by short vegetation structure and large fields while also often being close to urban infrastructure. Indirect and total effects showed that the contribution of neighboring pixels to direct emissions in the focal pixel was least in woodland areas, which is likely due to reduced infrastructure in surrounding woodland shown by reduced direct effects, and because trees physically block the spread of light, thereby diminishing the area it can penetrate into. Indirect effects found that skyglow was least at coastal sites and inland and marine water, and these sites tend to be relatively far from urban areas, and are often surrounded by mountainous topography that is not conducive for the spread of artificial light at night, so these sites have a minimal influence on the focal pixel. The behavior of artificial light at night in different landcover types provides more detailed information on previously reported global trends in different ecoregions [48
There were lower levels of artificial light at night inside of National Parks than outside (shown by negative direct effects). Although the overall levels of light emissions were minimal, we found that there were a number of strong sources of artificial light inside these areas (Figure 3
c). In the UK, National Parks contain villages, towns and industry, although these are subject to planning restrictions, and this result suggests that National Park planning acts to constrain light pollution. Conversely, there was a positive relationship for indirect effects, demonstrating that mean higher levels of light emissions and skyglow in pixels surrounding National Parks are associated with increased skyglow inside of National Parks. This relationship is driven by National Parks in England, which can be surrounded by high levels of urbanization. Indeed, the median skyglow was higher than the median light emissions in five of the ten National Parks in England; in the Peak District National Park, for example, an upland area in the North of England surrounded by major cities, the median skyglow was an order of magnitude greater than median light emissions (see Table S3
). The median skyglow within the National Parks suggests that, on average, it is possible to see the Milky Way in winter (when it is faintest) in all National Parks (skyglow levels are below 0.6 mcd/m2
]), with the best views in the Cairngorms National Park in Scotland (0.01 mcd/m2
), although the Milky Way is likely to be faint in the Peak District (0.44 mcd/m2
) National Park. A cause for concern is that the newly designated South Down National Park Dark Skies Reserve had the third highest median levels of skyglow of the 15 National Parks on mainland Britain. The UK is indicative of an increasing global issue of the pervasiveness of artificial light at night and, in particular, skyglow into key biodiversity areas [92
Direct effects show that both forms of artificial light at night were lower within SSSI than outside. Further, indirect effects for light emissions show that if the mean light emissions are higher in pixels neighboring a SSSI pixel, then there will be lower levels of artificial light in the SSSI pixel, demonstrating that light emissions in and around SSSI tend to be localized. Conversely, if there is less skyglow in pixels neighboring a SSSI pixel, then there will less skyglow inside the SSSI pixel, thereby reflecting the greater dispersion and reduced localization of skyglow compared to direct emissions. A cause for concern is that there still appear to be powerful sources of artificial light at night within or immediately adjacent to (i.e., in the same pixel) some SSSIs (Figure 3
d). These are likely to have significant biological impacts on the ecology of these protected spaces, and are prime locations for targeted management to reduce such impacts.
Modelled direct effects demonstrate that the presence of major roads and motorways are significant contributors to artificial light at night, with motorways emitting two or three times more artificial light at night than major roads. Indirect effects of light emissions show that the higher light emissions in neighboring pixels are associated with higher light emissions in the focal road pixel, demonstrating how, over short distances (30 km neighborhood object in this case), the associated road and urban infrastructure magnify the overall levels of light emissions. Conversely, for skyglow, there was a negative relationship for indirect effects for major roads and motorways, demonstrating that if skyglow is lower in areas surrounding a major road or motorway, then there will be more skyglow associated with the major road or motorway. This is a consequence of major roads and motorways passing through rural areas that have reduced levels of skyglow relative to the surrounding urban areas. Further, to control for spatial autocorrelation, the neighborhood object for skyglow was 130 km, and therefore the mean skyglow of the neighboring pixels will be lowered by the inclusion of larger rural areas, making bright roads brighter. Importantly, artificial lights from vehicles tend to be powerful, usually projected in the horizontal plane and are temporally inconsistent, and therefore are unreliably detected by satellite imagery (see [93
]). The contribution of the road network is thus likely to be greater than detected here.
The VIIRS DNB is the most popular and sensitive remote sensing data available for the study of artificial light at night, however, there are several limitations to its use. The first is that it does not detect light in the blue and violet spectrum, and no public data are currently available that could be used to resolve this issue. Globally, there has been an increasing trend of switching away from traditional sodium lamps to white LEDs, particularly in urban areas, due to their perceived increased energy efficiency [94
]. As a consequence, the actual amount of artificial light at night is likely to be greater than estimated here, and this may reduce the difference between the light emissions produced per person in rural and those produced per person in urban areas. A cause for concern is that there is increasing awareness amongst ecologists of the broader negative environmental impacts of white LEDs compared to sodium lamps [3
]. A second limitation is that VIIRS captures images at c.01:30 local time, a period when many ornamental and commercial lights, along with vehicle traffic, are close to their minimum intensity. Therefore, the use of VIIRS data can be thought to provide a baseline or lower limit of artificial light at night. Third, while there has been much interest in the development of standardized measurement systems from the ground to verify remote sensing nighttime lighting data [95
], the collection of these data in the UK was outside of the scope of this study, and would constitute a major program of work given the size and heterogeneity of the region, and the challenges of addressing the spectral issues highlighted above. Indeed, to date, projects to ground truth remotely sensed artificial lighting data have not resolved this last issue [96
]. In future, it would be interesting to compare how spatial relationships vary through time, but perhaps in a more dynamic region. It is also important to note that cloud cover is known to amplify the intensity of skyglow, as seen from the ground, and to influence its spatial distribution [100
], however, no data are currently available to incorporate cloud cover into measures of skyglow, as calculated by the World Atlas of Artificial Night Sky Brightness. To inform policy and planning, future research needs to focus on determining at the regional level the more detailed spatial distribution of all artificial light at night, including more temporary, mobile forms, such as vehicles, and how these vary both with changing climatic conditions, such as cloud cover and across the course of the nighttime.