Quantifying the attractiveness of broad-spectrum street lights to aerial nocturnal insects.

1. Sodium street lights, dominated by long wavelengths of light, are being replaced by broad- spectrum, white lights globally, in particular light- emitting diodes (LEDs). These white lights typically require less energy to operate and are therefore consid ered “eco- friendly”. However, little attention has been paid to the impacts white lights may have upon local wildlife populations. 2. We compared insect attraction to orange (high- pressure sodium, HPS) and white (metal halide, MH and LED) street lights experimentally using portable street lights and custom- made flight intercept traps. 3. Significantly more (greater than five times as many) insects were attracted to white MH street lights than white (4,250 K) LED and HPS lights. There was no statistical difference in the numbers of insects attracted to LED and HPS lights for most taxa caught. However, rarefaction shows a greater diversity of insects caught at LED than HPS lights. 4. Policy implications . With the current, large- scale conversion to white light- emitting diode (LED) lighting, our results give insight into how changes to street light technology may affect wildlife populations and communities. We recommend avoiding metal halide light installations as they attract many more insects than competing technologies. We high light the need to tailor LED lighting to prevent disturbances across multiple insect taxa.

Following the recent global financial recession, many local authorities have suffered monetary cut-backs and are looking to use their resources more efficiently. Authorities are also required to reduce carbon emissions in line with national and international legislation. Energy-saving, "eco-friendly" street light technologies are being adopted world-wide, with large areas being switched from orange/yellow sodium lights (longwavelength dominated) to white (broad-spectrum) lighting. Switchovers can happen relatively quickly, e.g. in the UK, Cornwall County Council (CCC) planned to replace c. 47,000 sodium street lights with MH in just 3 years as part of their "Invest-to-Save" project (Williams, 2009).
With large-scale installations happening so rapidly, it is important to know how changes in street lighting technology may affect wildlife. Stone, Wakefield, Harris, and Jones (2015), working in conjunction with CCCs "Invest-to-Save" project, found that activity of common pipistrelles, Pipistrellus pipistrellus, increased around MH lights following the switchover from LPS lights. As these bats hunt insects around lights (Rydell, 2006), and many insects are disproportionately attracted to UV light (Barghini & de Medeiros, 2012), this higher bat activity is likely to be a result of higher prey abundance at UV-emitting, broadspectrum MH lights. Relative to sodium lights, white lighting is predicted to: increase the bandwidth of wavelengths to which species are visually sensitive; alter species interactions (Davies, Bennie, Inger, Hempel de Ibarra, & Gaston, 2013); and to "exacerbate ecological impacts" of street lights (Pawson & Bader, 2014).
We compared the relative attractiveness of three common street light technologies (HPS, LED and MH) to volant insects. We tested two hypotheses: 1. MH street lights attract more insects than LED and HPS lighting.
2. Broad-spectrum "white" lights (MH and LED) attract a greater diversity of insects than long-wavelength-dominated (HPS) street lights.

| Study sites
Experiments were carried out at 12 field sites across southern England between 3 July and 10 September 2014. Sites were located an average of 115 km apart (range 1-256 km) to: (1) maximise the diversity of insects caught; (2) reduce any potential impact of the experiment on local insect populations; and (3) generate a clearer picture of how street lights are affecting insect attraction over a wide spatial scale (see Figure S1). Sites consisted of linear woodland edges (n = 10) or hedgerows (n = 2) at least 170 m in length, which adjoined either open meadows or grazed pasture. Woodland edges and hedgerows were selected for the study because they are linear features along which street lighting is often found in suburban areas and along minor roads in semi-rural and rural areas. Each site was sampled for one night and located >100 m from existing artificial lighting to minimise the impact of existing lighting.

| Lighting equipment
Three different street light technologies were tested: HPS (50 W SON-T, 4,400 lm; Philips, Amsterdam, The Netherlands); LED (2 × 8 LED Axia module arrays, 3,200 lm, 4,250 K; Urbis Schréder, Basingstoke, UK); and MH (45 W CPO-T, 4,750 lm; Philips). All lights were suitable for installation at a height of 5 m along minor roads or in suburban settings. These lights are deemed to be of similar light output for human needs and therefore are likely to be found on the same types of roads and in similar habitats. It is important to note that these lights vary in intensity as well as their spectral characteristics ( Figure 1). Our goal was to measure relative insect attraction to three commercially available lighting technologies, and so it was not necessary to match lights for absolute intensity. However, lights were housed in matching luminaires (Sapphire 1; Urbis Schréder, Chineham, Basingstoke, Hampshire, UK) to control for any potential differences in insect attraction caused by different housing designs. As many existing street lights are being updated by retrofitting of LED or MH units rather than entire luminaire replacement, this is an accurate reflection of current street lighting practice.
Each light, as well as a fourth "control" (CON) light which remained switched off throughout the study, was top-mounted onto a 5-m high tripod (REF 49-Z; Powerdrive Drum Company Ltd, Leighton Buzzard, Bedfordshire, UK) using a custom-made aluminium adaptor.
This set-up conforms to the mounting specifications of the lights. The four lighting columns were spaced an average of 34 m apart (range 32-35 m; Figure 2), which is representative of the 35 m distance between actual street lights of this type (Fotios et al., 2012). Lights were powered by a portable generator (Eu10i; Honda (UK), Bracknell, Berkshire, UK), positioned a mean of 97 m away (range = 78-100 m) F I G U R E 1 Irradiance and illuminance measures of the three street lights used in this study: high-pressure sodium (HPS), lightemitting diode (LED) and metal halide (MH). Irradiance readings for each street light were taken in a darkened room using a cosine corrector at the end of a 400 μm diameter, ultraviolet-visible, fibre optic cable connected to a spectrometer (USB2000; Ocean Optics, Dunedin, FL, USA) controlled by a PC running SpectraSuite (version 6; Ocean Optics). Each curve represents the average of three scans. Illuminance readings were taken using a digital lux meter ( Illuminance to minimise the risk that noise from the generator would affect animal behaviour (Stone, Jones, & Harris, 2009). An electrical splitter was used to enable all experimental lights to be powered from the same generator. The order in which the four luminaires were arranged was randomised between sites and their relative positions, recorded as either "edge" or "centre" (Figure 2), were included in statistical analyses to account for potential edge effects.
One of the lights (HPS) was fitted with an active photocell (SELC installed halfway between the two central lights.

| Insect trap design
Insects were caught using custom-made flight intercept traps based on the design used by Eisenbeis and Eick (2011).

| Statistical analyses
To test whether MH street lights attract more insects than HPS and LED lights, data were analysed by fitting generalised linear mixed models (GLMMs) with Poisson error structures using the package lme4 (Bates, Maechler, Bolker, & Walker, 2014) in r (version 3.4.0. 2017).
"Light" was included as the sole fixed effect term and "light position" ("edge" or "centre," e.g. Figure 2) nested within "site" was included as a random effect in each GLMM. Response variables included "total number of insects," "Diptera," "Coleoptera," "Lepidoptera," "Erebidae," "Chironomidae," "Noctuidae" and "Psychodidae;" Catches of other F I G U R E 2 A photograph of a typical field site. Main: four street lights are positioned c. 34 m apart along the edge of a hedgerow, approximately uniform in height along its length. The lights were control (CON), light-emitting diode (LED), high-pressure sodium (HPS) and metal halide (MH) and were orientated to illuminate away from the hedge into open grassland. In this example, the positioning of the CON and MH lights were classed as "edge" and the LED and HPS lights as "centre" for the purposes of statistical analyses. Inset: a photograph of one of the custom-made insect flight intercept traps. Traps were designed to channel insects through the lower funnel and into the black collection chamber below. Ethyl acetate in the collection chamber was used to prevent insects escaping [Colour figure can be viewed at wileyonlinelibrary.com]

CON LED HPS MH
taxa were too small for reliable statistical analyses. The goodness-offit of each GLMM was tested using the r package aods3 (Lesnoff & Lancelot, 2013) to ensure data were not overdispersed (residual deviance > degrees of freedom; Crawley, 2008). Each model was then compared to a subsequent model lacking the fixed effect term "light" to examine both the Δdeviance between the corresponding models as well as the difference in Akaike information criterion (AIC) values (Zuur, Ieno, Walker, Saveliev, & Smith, 2009). Pairwise comparisons between the different light types were then conducted using Tukey contrasts via the r package multcomp (Hothorn, Bretz, & Westfall, 2008).
Rarefaction curves were generated to compare insect diversity at all light types. Both sample-based (incidence data) and individualbased (abundance data) rarefaction and extrapolation curves were created using the iNEXT online program (Chao, Ma, & Hsieh, 2016).

| RESULTS
Data from 12 sites were included in the following analyses. The use of the photocell resulted in the lights switching on an average of 15 min after sunset (range = 4-23) and switching off a mean of 22 min before sunrise (range = 9-30), giving an average sampling duration of 498 min (range = 416-626). The spiders (Araneae, n = 6), mites (Acari, n = 119) and springtails (Symphypleona, n = 4) caught in the flight intercept traps were not included in the analyses. Moon illumination ranged from 6% to 100% across all sites (Thorsen, 1995(Thorsen, -2017. At the 10 sites where weather variables were successfully recorded, we recorded an average humidity per night of 92% (range 81%-97%), nightly temperature of 13°C (range 9-17°C) and nightly wind speed of 0.8 km/hr (range 0-2.6 km/hr).  Figure 3. Total insect attraction was significantly affected by light treatment (Table 1). Pairwise comparisons show the MH attracted significantly more insects than all other lights, but there was no significant difference in insect attraction between LED and HPS lights. All lights caught a greater number of total insects than CON (Table 2).  Figure 3).

| Broad-spectrum, "white" lights (LED and MH) attract a greater diversity of insects than long-wavelength-dominated (HPS) street lights
The number of insect orders caught varied with light type; CON = 4, HPS = 8, LED = 8, MH = 10. In total 1,372 insects were identified to family; 10 could not be positively identified beyond order (eight Lepidoptera and two Ephemeroptera) and were therefore omitted from family-level analyses. Abundance data are displayed by family in Table S1.
Sample-based rarefaction curves indicate family diversity was greatest at the MH, followed by LED and then the HPS light. These differences are significant based on 95% confidence intervals ( Figure 4a).
However, individual-based rarefaction curves show a higher species diversity at LED lights compared to HPS and MH lights when rarefied down to the number of individuals observed at the HPS light ( Figure 4b). Extrapolations of the HPS and LED data suggest that the curves for these lights may cross with that of the MH at larger sample sizes, although the 95% confidence intervals for these extrapolated data become very large.

| DISCUSSION
The MH street lights attracted significantly more insects than LED and HPS street lights, as predicted in our first hypothesis. Analysis of incidence data, which accounts for sample (site) heterogeneity, indicates that broad-spectrum, white lights attract a greater diversity of insects than long-wavelength-dominated (HPS) street lights (hypothesis 2).
However, when controlling for the number of individuals caught at each light type, the rarefied family diversity is significantly higher at the LED than at the HPS and MH lights. Regardless of rarefaction method, the LED attracted a greater number of families than the HPS light (Figure 4a,b). However, as rarefaction curves for all lights have not reached a clear asymptote these results should be treated with caution. Insect diversity estimates for each light may alter relative to one another with greater sample size, as suggested by extrapolation of our data.
Light intensity was not equal for all street lights, as we compared lighting technologies based on their real-life application for human needs. However, these differences in intensity are unlikely to have influenced insect attraction as much as spectral differences (Longcore et al., 2015). Despite both emitting white light, the MH caught approximately five times as many insects as the LED light. This may be, in part, explained by the presence of UV light in the MH spectrum and its absence in the HPS and LED spectra (Figure 1). Many insects find  (Tovée, 1995). However, if UV, or wavelengths of light adjacent to UV, disrupt natural behaviour by enticing insects towards artificial lighting, then survival and reproduction may be negatively affected (Frank, 2006).
Even reductions in flight-to-light behaviour of urban moths, predicted to increase survival and reproduction, may reduce moth mobility, with subsequent negative connotations for foraging, colonisation and pollination (Altermatt & Ebert, 2016). Disruption of ecosystem services, such as pollination by Lepidoptera (see Macgregor, Pocock, Fox, & Evans, 2015), is likely to induce trophic cascades and may have implications for human food security.
All lights were mounted at the same height within the same luminaire model, yet there were still differences in the spatial distribution of the emitted light. Qualitatively, the HPS and MH (both gas discharge) lights had similar, diffuse, light distributions, whereas the light from the LED (solid state lighting) appeared to focus light in two distinct planes (Figure 2). Given that flying insects can approach the vicinity of a street light from any direction, the downward-focused LEDs should, on average, be less visible than those which shine light both downwards and at higher elevations. Consequently, the difference in insect attraction between the LED and MH lights may be, in part, due to a difference in the spatial distribution of emitted light. Full shielding of lights has been recommended to limit the impact of light pollution on the environment (Falchi, Cinzano, Elvidge, Keith, & Haim, 2011), but we suggest that further quantitative study investigating how light distribution impacts taxa would have useful policy implications. Degen et al. (2016) have estimated an attraction radius of 23 m for moths at HPS lights. This suggests that for our light separation distance T A B L E 2 Results of multiple comparison tests using Tukey corrections applied to GLMMs for insect catches (12 sites). In all models, "light position" nested within "site" was included as a random effect term and "light" as the only fixed effect term. Lights were control (CON), high-pressure sodium (HPS), light-emitting diode (LED) and metal halide (MH) T A B L E 2 (Continued) of 34 m (typical for UK street lights), there will be overlap of attraction radii for moths. We chose to compare lights together as ratios of moths caught at spectrally different lights have been found to be consistent regardless of whether the lights were presented "alone" or in "competition" with one another (Somers-Yeates, Hodgson, McGregor, Spalding, & ffrench-Constant, 2013). Testing in "competition" allowed us to control for environmental variables, but it should be noted that, generally, different types of street lights are not mixed.
Excluding Coleoptera, there was no statistically significant difference in the number of insects caught at HPS and LED street lights.
Experiments conducted in New Zealand (Pawson & Bader, 2014) found that LEDs attracted 48% more insects than HPS street lights, whereas a study in Germany (Eisenbeis & Eick, 2011) found that LEDs attract significantly fewer insects than HPS. LEDs can vary considerably, with high "correlated colour temperature" (CCT) rated LEDs emitting relatively more blue light than low CCT LEDs. The LED light we used was rated as "neutral white" at 4,250 K, slightly "cooler/bluer" than the 4,000 K LEDs used by Pawson and Bader (2014). Differences in CCTs are unlikely to be the main cause of disparity here as abundances of insects caught at LEDs varying in CCT did not differ from one another statistically (Pawson & Bader, 2014;Wakefield, Broyles, Stone, Jones, & Harris, 2016); although see Longcore et al. (2015) concerning LED spectral-tuning. Therefore, we predict that differences in our results are more likely to be the result of variation in habitats surveyed and the associated variation in insect assemblages that were sampled, as well as the aforementioned differences in light distribution between the lights tested.
Analysis of sample-based (incidence) rarefaction curves show significantly more families were caught at the MH relative to the LED light, which itself caught significantly more families than the HPS (comparisons made at n = 12 sites, Figure 4a). As insect catches varied considerably between lights, and a larger sample from an assemblage is statistically more likely to contain more families, we also generated individual-based rarefaction curves. These interpolate (and extrapolate) results while standardising catches for abundance.
These individual-based curves differ from the sample-based results in all but one relationship-a significantly higher diversity of insect families at LED relative to HPS lights (comparisons made at n = 136 individuals in Figure 4b). As these curves, especially those for the HPS and the LED, had not reached a clear asymptote, this indicates that many other families are likely to be sampled at all lights with greater sampling effort.
As a single entity, broad-spectrum white lighting (i.e. LED and MH) did not attract a greater number of insect families relative to longwavelength-dominated HPS lighting. We highlight that differences may occur at a finer taxonomic resolution which could have implications for conservation work often carried out at the species level.
Typically, insect vision is di-or trichromatic, with peak sensitivities at shorter wavelengths including UV (Land & Nilsson, 2012). Sodium lights predominantly emit longer wavelengths (Figure 1) but do still emit light throughout the rest of the visible spectrum and attract more insects than monochromatic long-wavelength lighting, e.g. LPS (Rydell, 1992).
The MH light caught approximately twice as many families of Coleoptera and Lepidoptera as the LED light, despite both appearing white to the human eye. Similar observations have been made in other studies (Nabli, Bailey, & Necibi, 1999;Somers-Yeates et al., 2013). Differences in visual capabilities between insect species and the complexity of ecological networks make it difficult to predict exactly how changes in lighting spectra will affect insect populations. Effects other than phototaxis, such as disrupting sex pheromone production (Van Geffen et al., 2015), diapause inhibition and sex-specific life-history changes (Van Geffen, van Grunsven, van Ruijven, Berendse, & Veenendaal, 2014), are also dependent on spectral compositions, being most affected by white and green light. Therefore, the use of white lighting may affect a wider range of insects, and other wildlife, via trophic cascades.
Contrary to Coleoptera and Lepidoptera, Diptera were most diverse around the LED light (see Table S1). Flies can be attracted to short wavelengths as well as green and red light (Green, 1985). Differences in the range of visual spectra between insect orders may be the cause of these differing trends. This suggests that different insect orders will be affected to differing extents by future street light installations/ conversions. Certain LEDs, particularly those with greater short wavelength emissions, may well exacerbate the ecological consequences of artificial lighting at night (Gaston, Davies, Bennie, & Hopkins, 2012;, although effects on moth populations have yet to be observed (Spoelstra et al., 2015). Advances in LED efficacy and decreased product costs are likely to result in illumination of Family diversity previously unlit areas world-wide. The implications of this for dipteran vectors of disease are discussed by Wakefield et al. (2016).
Our finding that MH street lights attract significantly more flying insects than sodium lights compliments research investigating bat activity around street lights . It is likely that the higher bat activity recorded by Stone et al. (2015) around MH relative to LPS lights was in response to higher densities of prey around the former.
As well as attracting insects and creating local abundances of prey for predators such as bats (Rydell, 2006), white street lights can also interfere with the predator avoidance behaviour of a number of moths, reducing their ability to avoid hunting bats (Svensson & Rydell, 1998;. The underlying causes of insect attraction to light remain unclear, but spectral changes to street lights will have significant impacts on various taxa, altering species distributions, wildlife communities and predator-prey interactions.
White lighting is not always as "eco-friendly" as advertised and thus energy credentials should not be the sole focus for defining how "eco-friendly" a product is. Greater numbers of insects were attracted to MH street lights and a greater diversity of insects were attracted to white LEDs compared with long-wavelength-dominated HPS lights.
Placing these results alongside the existing literature we conclude that whole-scale conversion to broad-spectrum, white, street lights is likely to have negative effects on wildlife. Highly focused/shielded LEDs designed to filter out short wavelengths of light may attract relatively fewer insects and warrant further investigation.