A global atlas of artificial light at night under the sea

The 21st century will see dramatic increases in coastal urbanization (Neumann et al., 2015) and offshore infrastructure. Lighting from coastal urban centers, oil platforms, and other offshore structures scatters in the atmosphere to form artificial skyglow that increases the extent of light pollution from source to hundreds of kilometers into the surrounding marine habitats. The impacts of this light on marine ecosystems has emerged as a focus for ecological light pollution research (Davies et al., 2014; Ludvigsen et al., 2018; Fobert et al., 2019; Torres et al., 2020).

One of the clearest demonstrations that we have entered another epoch, the urbanocene (West, 2018), is the prevalence of artificial light at night (ALAN) visible from space (Falchi et al., 2016a). Satellite-derived images of the world’s nighttime lights have been successfully coupled with atmospheric light dispersion models to produce new insights into the prevalence and growth of ALAN skyglow on land (Falchi et al., 2016a; Kyba et al., 2017). Much of this light will enter the world’s seas and oceans, which until now has been difficult to quantify. The spectral signature of land-based light sources is detectable on the biodiverse coral reef ecosystems of the Red Sea (Tamir et al., 2017), and radiative transfer modeling of in-water ALAN skyglow transmission suggests that up to three-quarters of the seafloor adjacent to cities can be exposed to artificial light irradiances sufficient to elicit biological responses (Davies et al., 2020). Satellite products of the world’s nighttime lights have proved valuable for quantifying trends in ALAN (Bennie et al., 2014; Kyba et al., 2017), and exposure in terrestrial ecosystems (Mazor et al., 2013; Duffy et al., 2015; McLaren et al., 2018) and protected areas (Gaston et al., 2015). Skyglow-like light levels can affect melatonin in freshwater fish (Kupprat et al., 2020) and more generally impact freshwater ecosystems (Jechow and Hölker, 2019). To date, there is no atlas of the extent of ALAN under the sea, limiting advances in understanding and quantifying the impacts of marine ALAN from the organism to ecosystem level.

We have derived a high-resolution global atlas of ALAN under the sea. The atlas is produced at a spatial scale of approximately 1 km × 1 km by combining the recently available world atlas of artificial night sky brightness (Falchi et al., 2016a, 2016b), satellite-derived in-water spectral optical properties (Lee et al., 2002), sea surface-measured artificial light irradiances (Davies et al., 2020), and hydrological radiative transfer modeling (Mobley, 1995; Figure 1). The atlas is calibrated using derived relationships between artificial sky brightness (Falchi et al., 2016a) and broadband sea surface irradiances (blue, nominally 400–500 nm; green, nominally 500–600 nm; and red, nominally 600–700 nm) collected from waters adjacent to the city of Plymouth, UK (Davies et al., 2020; Table 1). To model attenuation of light in the 3 wavebands, inherent optical properties (IOPs: absorption due to phytoplankton, colored dissolved organic matter [CDOM] and particulates, and backscatter due to particulates), were derived using global monthly climatologies (1998–2017) calculated from the primary IOP data set of the European Space Agency Climate Change Initiative (Sathyendranath et al., 2019a; 2019b). The resulting broadband irradiances were then integrated to produce absolute irradiances of ALAN underwater with depth. This approach allowed insight for the first time into the spatial extent of in-water marine light pollution and its temporal variability on a global scale.

To understand the biological importance of the modeled underwater ALAN field, we calculated the critical depth (Z_c) to which biologically important ALAN penetrates throughout the global ocean estuarine, coastal and nearshore regions, across the area defined by the exclusive economic zone (EEZ) of individual countries. The biological impacts of ALAN are likely to be more severe and widespread in regions experiencing higher values of Z_c across larger spatial scales. Z_c was calculated for each pixel as the maximum depth where underwater irradiance exceeds a predefined threshold for biologically important ALAN (E_lim). We define E_lim using the precautionary principle, following the convention set out by Davies et al. (2020). For integrated absolute irradiance measurements, E_lim is approximated as 0.1 µW m^–2, the minimum irradiance of white light that elicits diel vertical migration in globally widespread adult Calanus copepods (Båtnes et al., 2015).

We confirmed the validity of our methodology using data from a field campaign in the Gulf of Aqaba, Red Sea (Tamir et al., 2017). Such data sets are extremely rare due in part to the technical difficulty of measuring multispectral irradiances to the level of µW m^–2 nm^–1 with an associated high signal to noise ratio.

To describe the penetration and subsequent attenuation of ALAN as it propagates through the water column, Beer’s Law was used for a general wavelength λ:

where E is irradiance (W m^–2 nm^–1), K_d the diffuse attenuation coefficient (m^–1), and z is depth (m). Expanding to account for spectral wavebands, λ_i (i = blue, λ_b, nominally 400–500 nm; i = green, λ_g, nominally 500–600 nm; i = red, λ_r, nominally 600–700 nm):

The left side of Equation 2 can be expanded thus:

where the contributions of the surface irradiance in the _b, _g, and _r are expressed as a simple sum:

which results in

Using spectral irradiance fractional (f) contributions to the whole

where, for example,

allows the relative irradiance contributions f_b,g,r to be calculated at the surface.

Broadband (400–700 nm) K_d can be calculated using the following expression at a fixed depth (e.g., for z = 1 m):

Calculating the critical depth is then straightforward, where the irradiance drops below a preset threshold (E_lim):

In order to quantify regionally and globally the amount of ALAN reaching different depths within the water column, 2 different data sets were used: an artificial night sky brightness world atlas (Falchi et al., 2016b) to define the above water light field and a global data set of in-water IOP (Sathyendranath et al., 2019b), the absorption and backscatter components being calculated using the IOP model of Lee et al. (2002). Global monthly mean climatologies (199–2017) were calculated from the primary IOP data set (Sathyendranath et al., 2019b), specifically: the absorption due to phytoplankton at a wavelength of 443 nm, a_ph(443); the absorption due to CDOM and particulates at 443 nm, a_dy(443); and the backscatter due to particulates at 555 nm, b_bp(555).

A sea-surface broadband irradiance data set collected from Plymouth Sound and the Tamar Estuary (Davies et al., 2020) was used to convert the night sky brightness (Falchi et al., 2016b) into spectral waveband irradiances (see Modeling section). For the specification of EEZ, the global data set of the Flanders Marine Institute (2018) was used.

The above water global artificial night sky brightness is given in units of mcd m^–2 and is ostensibly a measure of the broadband signal. As the penetration of light through the water column has a strong spectral dependency, essentially being a function of absorption and scatter of the different optically active components (pure water, chlorophyll [phytoplankton], CDOM, suspended particulates), then the above water light field needs to be split spectrally into different bands. In this article, we use relatively wide spectral bands coincidental with the wavebands on a Skye Instruments multispectral irradiance sensor (blue, 400–500 nm; green, 495–560 nm; red, 620–740 nm) in order to empirically fit the broadband global atlas (Falchi et al., 2016a) values to their spectral signature (Davies et al., 2020). Figure 1 shows a schematic diagram of the model framework, split into 2 stages.

A look-up-table (LUT) of spectral K_d values (blue, 400–500 nm; green, 495–560 nm; red, 620–740 nm) was generated using the HYDROLIGHT in-water radiative transfer model (Mobley, 1995) assuming a totally diffuse sky and a wind speed of 5 ms^–1. A 4 independent component model was used within HYDROLIGHT, these being: (I) pure water, (II) phytoplankton, (III) CDOM, and (IV) particles. For component I, tabulated absorption values (Smith and Baker, 1981) were used together with a pure water scattering phase function (Mobley, 1994). Components II and III are assumed to be nonscattering (i.e., no phase function required) and are described as a continuous function of wavelength. For component II, the normalized absorption coefficient data of Prieur and Sathyendranath (1981) were employed and for III, a simple exponential function was used:

Finally, component IV is assumed to be nonabsorbing and is described by the spectral function for particulate backscatter (Smyth et al., 2006):

together with a particle phase function (Petzold, 1972). The input IOPs to HYDROLIGHT were a_ph(443), a_dy(443), and b_bp(555), based on the global IOP data set wavelengths. Both a_ph(443) and a_dy(443) input values were prescribed at logarithmic intervals between 0.01 and 10.0 m^–1 (n = 28), whereas b_bp(555) ranged between 0.001 and 0.1 m^–1 (n = 19) resulting in 14,896 runs of the HYDROLIGHT model for each spectral band (λ_b, λ_g, λ_r) to create a LUT. The resultant LUT was then used to calculate 12 monthly global climatology maps each of K_d (λ_b, λ_g, λ_r) at a nominal resolution of 4 km on an equal angle projection (8192 × 4096 pixels), using a 3D weighted nearest-neighbor scheme.

The relationships between the global artificial night sky brightness atlas (Falchi et al., 2016b) and sea surface broadband irradiances collected from Plymouth Sound and the Tamar Estuary (Davies et al., 2020) were spectrally resolved using empirical equations of the form:

where m_λ and c_λ are the spectral slope and intercept, respectively (Table 1), and E_F is the night sky brightness (mcd m^–2). The c_λ term also takes into consideration the natural sky brightness as well as compensating for some of the assumptions within the Falchi et al. (2016b) atlas, such as the blue “blindness” of the Day/Night Band of the Visible Infrared Imaging Radiometer Suite (VIIRS) satellite sensor (see section on Limitations of approach) and use of a fixed aerosol optical depth. The surface spectral light field was then propagated through the water column using a broadband formula for K_d (Equation 8), weighting each spectral band by its relative contribution to the surface irradiance (Equations 6–8). The K_d global climatological maps were resampled to match the spatial (1 km) resolution of the global artificial night sky brightness atlas. For each pixel of the regional or global map generated, determining the broadband irradiance was then possible at any given depth, or the depth below which the light levels dropped below a predetermined threshold (E_lim). E_lim was set to 0.1 µW m^–2 based on the light sensitivity of Calanus (Båtnes et al., 2015), a globally widespread key diel migratory species of zooplankton important to fisheries. E_lim was converted from 0.47 × 10^–6 µmol photon m^–2 s^–1 using the approximate conversion factor given by Morel and Smith (1974) for mean frequency averaged lighting conditions (1 W m^–2 ≈ 4.6 µmol photons m^–2 s^–1). The critical depth (Z_c) was calculated using Equation 9.

We used the in-water multispectral irradiance data set generated on a field campaign in the Gulf of Aqaba, northern Red Sea, as reported in Tamir et al. (2017). The multispectral in-water light profiles were measured using a SeaWiFS-compliant, high resolution, profiling reflectance radiometer (PRR-800; Biospherical Instruments Inc., San Diego) with 19 spectral channels in the 300- to 900-nm wavelength range. The spectral irradiances were aggregated into 1-m depth bins, and then the entire spectrum at each depth was interpolated linearly onto a 1-nm wavelength resolution grid, which was then integrated into the broadband ranges for blue (400–500 nm), green (495–560 nm), and red (620–740 nm), for each depth and each station to allow intercomparison to be made with the global model.

Point extractions from the global model were made at the locations of the 19 stations described in Tamir et al. (2017). These were (1) above surface sky brightness (mcd m^–2) from the global atlas of Falchi et al. (2016a), and (2) the in-water broadband spectral K_d (λ_b, λ_g, λ_r) extracted from the global climatological maps. The surface sky brightness was converted to the blue, green, and red wavelengths (µW m^–2) using Equation 12 and coefficients in Table 1, and then propagated through the water column using Equation 1. This approach allowed direct intercomparison with the in situ measured irradiances for each station as a function of depth. In addition, for each station at each 1-m depth interval, the broadband (PAR) downwelling irradiance was calculated from the 3 spectral components. This calculation allowed the value of Z_c to be calculated for each station.

Table 2 shows the total area of individual EEZs where Z_c exceeds given depths, ranked first by the surface (<1 m) component and then successively down the water column to greater depths (maximum of 50 m) for the top 20 regions. A comprehensive ranking table (Table S1) shows that around 3.1% of the world’s EEZs, a total area of 1.9 million km², are already impacted by ALAN. In the UK part of the North Sea (also shown in Figure 2), ranked #5, Z_c exceeds 1 m across nearly 65,000 km², and exceeds 40 m across approximately 3,000 km². Greater impacts of ALAN are anticipated at depth in EEZs with clearer waters. To give one example, the largest area where Z_c exceeds 40 m in Table 2 (4,393 km²) can be found in the Malaysian part of the South China Sea (ranked #6).

Table 3 shows the total percentage area of an individual EEZ where Z_c exceeds given depths ranked for the top 20 regions (the global ranking table is given in Table S2). Z_c exceeded 1 m across at least 60% of the surface area of all of the top 20 ranked EEZs. Seven EEZs ranked in the top 20 by percentage area in the Persian Gulf and 2 EEZs in the North Sea. In some of the most impacted regions (Qatari part of the Persian Gulf #4; UAE part of the Persian Gulf #9) Z_c exceeds 10 m across sizable percentages of their area (94% and 79.5% of their surface area, respectively). In these regions, extensive ALAN at the sea surface combines with clear surface waters to produce widespread impact zones at depth.

The month impacted the most by ALAN is shown in Column 3 of Table 2. All regions demonstrate some form of seasonal cycle driven by blooms of phytoplankton (a function of light and nutrient availability, and water column stability) as well as riverine inputs to the marine system through suspended particulates and dissolved organic matter, or physical processes such as winter wind-driven mixing which causes significant resuspension of sediment in shallow coastal waters. From the biological impact perspective, ALAN impacts are thus likely to vary between seasons, and indeed over shorter timescales in tidally dynamic regions.

Figure 2 shows the prevalence and likely impact of ALAN on the marine environment in 2 of the most heavily light-polluted regions on the globe, namely the Persian Gulf and the North Sea. These regions are characterized by extensive offshore development including oil and gas platforms, windfarms, and island development. The urbanized areas around the Persian Gulf fringes, such as Dubai, Abu Dhabi, Doha and Manama, as well as the North Sea fringes, such as Sunderland and Hartlepool in the UK, and the extended Dutch and Belgian coastal conurbations, generally have a smaller impact on the marine ALAN field than remote offshore features of these regions. This difference is due in part to the anisotropic nature of the light pollution at the land–sea interface, that is, a large percentage, depending on the orientation of the coastline, topography, and aspect of artificial light sources, is directed inland. It is also dependent upon the clarity of the water in and around the coastal zone, governed by the IOPs, which is generally more opaque (higher K_d in Equation 9) than further offshore due to freshwater runoff from land, upwelling of nutrients and wave, and tidal mixing of the water column and seabed. Water column opacity increases where sediment loading, CDOM concentrations, and possibly phytoplankton densities are higher.

Figure 3 shows the intercomparison between the global model-derived in-water light profiles and those derived from in situ measurements in the Gulf of Aqaba (Tamir et al., 2017). The purpose of including these data here is to provide quantifiable confidence in our overall methodology. The in situ multispectral measurements (Figure 3b, e, h) have been integrated to produce broadband blue, green, and red irradiances to replicate our modeling (Figure 3a, d, g) approach. The logarithmic root mean square differences between the modeled and in situ data are 0.234, 0.663, and 0.634 in the blue, green, and red parts of the spectrum, respectively (Figure 3c, f, i) and replicate reasonably well the decrease in spectral irradiance with depth: less than 1% of the surface irradiances in the red penetrate below 3–5 m; around 10% of the surface green light arrives at 20 m; and more than 10% of the blue light penetrates depths deeper than 30 m in both the modeled and in situ data. The modeled Z_c (mean ± standard deviation of 24.6 ± 1.7 m, N = 18) is slightly lower than the in situ observations (25.7 ± 1.3 m, N = 18; Station 17 removed as an obvious outlier). Important controls on the vertical light profile are the in-water IOPs: our modeled approach uses climatological monthly values (September, for the data presented) and small changes in the IOPs can have a large impact on the value of Z_c. The annual range of modeled Z_c for the stations presented here is between 21.4 ± 0.3 m (N = 19) in November and 24.6 ± 0.4 m (N = 19) in September; the average for all stations and all months is 23.5 ± 1.1 m (N = 228).

Evidence for the potential impacts of ALAN across marine phylogenies and all levels of biological organization is rapidly emerging. ALAN affects multiple aspects of the life history of marine organisms, including reduced reproductive success (Witherington and Bjorndal, 1991; Fobert et al., 2019), disrupted migration (Ludvigsen et al., 2018; Berge et al., 2020; Torres et al., 2020), altered recruitment (Davies et al., 2015), and shifts in the balance of species interactions (Becker et al., 2013; Underwood et al., 2017; Levy et al., 2020). The extent of ALAN sufficient to elicit such responses in marine ecosystems is not yet known. Our atlas is the first to quantify the global spatial and temporal prevalence of biologically important ALAN in undersea habitats. It has numerous potential applications in the design of ecological experiments, biogeography, and conservation.

The atlas highlights the many areas of the global ocean that are strongly impacted by ALAN from land and offshore infrastructure. ALAN penetrates to significant depths within the water column (>40 m) depending on the clarity of the water, which itself is a function of regional biogeochemistry and season. In the most heavily light-polluted regions (Persian Gulf, Eastern Mediterranean, North Sea), there is likely significant disruption to the natural ecosystems of the region impacting every trophic level from phytoplankton upward. The greatest direct impacts are likely on highly photosensitive species that utilize moonlight to guide migrations and synchronize phenological events (Naylor, 2001; Last et al., 2016; Ugolini et al., 2016; Ludvigsen et al., 2018; Torres et al., 2020), many of which are critical to the wider ecosystem and sustain vital ecosystem services (Moberg and Folke, 1999; Hayes, 2003).

Disruption by ALAN at the geographical scales presented here is likely to have a significant impact on an individual country’s realized economic value of their EEZ. The EEZs are already regions under considerable stress from other factors such as shipping and underwater noise (Simmonds et al., 2014). This work makes a compelling case for undersea ALAN to be included as an additional stressor within this already multistressed environment.

Our atlas combines remote sensing products with radiative transfer modeling to produce first insights into the exposure of undersea habitats to ALAN. The relationship between artificial sky brightness and sea surface irradiance is derived from measurements of artificial skyglow originating from the city of Plymouth, UK, recorded on clear moonless nights when the sun was >18° below the horizon. While the derived relationships are robust (Table 1; Figure S2 in Davies et al., 2020), our modeling approach assumes that the spectral power distribution of artificial skyglow is matched between Plymouth, a predominantly light-emitting diode (LED)-lit city, and the rest of the world. While LED lighting use is growing rapidly—forecast to contribute 97% of the global lighting market by 2025 (Bertoldi, 2018)—the modeled exposure levels may overestimate artificial light irradiances where low pressure sodium (LPS) lighting is still in use, but will remain broadly representative where other modern broad spectrum lights (e.g., high pressure sodium and metal halide) predominate. For an in depth study on spectral differences of artificial light sources and the differences in color between countries, see Levin et al. (2020). The in situ data used here and presented in Tamir et al. (2017) contained a strong peak in the sodium band (589 nm); however, the impact of this band was omitted in the modeling and the in situ numerical integration. We contend that this approximation is acceptable, because the absorption due to pure water (Pope and Fry, 1997) at 589 nm is around 0.135 m^–1, which compares with 0.009 m^–1 (450 nm, mid-blue), 0.04 m^–1 (530 nm, mid-green), and 0.465 m^–1 (680 nm, mid-red). Most of the light in this sodium band will therefore be attenuated in the top few meters of the water column, and consequently contribute insignificantly to the overall broadband calculated Z_c.

The current terrestrial world atlas of Falchi et al. (2016a) uses the Day Night Band of the VIIRS satellite sensor which is “blind” in the blue part of the spectrum and cannot spectrally distinguish between sodium-based lights and the longer wavelength component of the LED emission spectrum. The outputs of our methodology could therefore be improved by new satellite missions that could spectrally quantify the ALAN signature of individual countries and urban areas.

The base data set of the world atlas of artificial sky brightness (Falchi et al., 2016a) is limited to clear sky conditions, hence so are the values of underwater irradiance we present here. Cloud cover and cloud base height can amplify light pollution (Davies et al., 2020) by an order of magnitude in luminance (Kyba et al., 2011). The reported exposure irradiances and values of Z_c will therefore be higher when clouds are present. However, this effect may be offset by the clouds, particularly when there is a low altitude cloud base, rapidly attenuating the upward propagation of the surface light field and limiting the far field effects of ALAN. The near-field effects of individual street lights, although considerable on scales <10 m, rapidly decrease with distance and, when interacting with the aquatic environment, much of the light is reflected at the surface (Jechow and Hölker, 2019); the far-field skyglow term will tend to dominate the in-water impacts at distances >100 m as point sources appear closer to the horizon.

The coverage from ocean color satellites used to generate the in-water IOP fields does not include high latitudes (>50°) during the winter months (boreal, December–February; austral, June–August). This limitation is of most relevance in the northern hemisphere, as there are few population centers poleward of 50°S.

The 0.1 µW m^–2 value for E_lim was selected based on the importance to fisheries of globally pervasive diel migratory species such as Calanus. However, other values of E_lim could have been chosen and alternative values are listed extensively elsewhere (Davies and Smyth, 2018; Tidau et al., 2021). Figure 4 of Davies and Smyth (2018) plots the light sensitivity of several different marine invertebrates as a function of light penetration depth. When comparing the sensitivity to types of light source (artificial skyglow and point source) and in-water optical properties, Calanus is shown to be particularly sensitive, which in turn relates to deeper values of Z_c. As not all sensitivity thresholds can be accommodated in any one system, even if available, we adopted a precautionary approach in using the most sensitive species. The value of E_lim also depends upon assuming the mean frequency of averaged lighting conditions which changes depending upon the light source itself (e.g., daylight or artificial light). McCree (1981) showed that assuming this mean frequency gives a variation of around ±8.5%.

By using monthly climatologies of IOPs to modulate the ALAN field through the water column, our atlas adds novel insight into temporal variability and hence seasonal impacts of ALAN. This higher temporal resolution in our approach is limited by a lack of similar temporal resolution in the above surface ALAN. For long-term studies (multiannual to decadal) as urbanization accelerates and the spectral signatures inevitably shift, changes to the above surface ALAN are likely to become increasingly important. Again, improvements in tools to quantify above sea surface ALAN levels will improve our understanding of ALAN impacts underwater.

The ocean color sensors used to derive IOPs as part of the Climate Change Initiative data set (Sathyendranath et al., 2019a) measure top-of-the-atmosphere radiances. These measurements are then used to derive water-leaving radiances, L_w(λ), which involves use of an atmospheric correction procedure (Steinmetz et al., 2011): typically the atmospheric signal contributes 80% or more to the top of the atmosphere radiances (Müller et al., 2015). L_w(λ) in turn is controlled by the in-water IOPs of absorption, a(λ), and back-scattering, b_b(λ). The Lee et al. (2002) IOP model used to invert L_w(λ) into a_ph(443), a_dy(443), and b_bp(555) employed a quasi-analytical algorithm which sought to limit the amount of empirical relationships contained within its framework. This approach minimizes any likely errors caused by region-specific data sets used to derive IOP relationships and also makes the algorithm more generally applicable to coastal (Case 2) as well as open-ocean (Case 1) waters (Morel and Prieur, 1977). We therefore sought to minimize any limitations in the IOP data set by using products that utilize algorithms suited to atmospheric correction (Steinmetz et al., 2011) and IOP retrieval (Lee et al., 2002) in the regions close to shore.

In theory, the amelioration of ALAN impacts upon marine, and indeed terrestrial, ecosystems could be achieved instantly: Simply turning off the lights at night would have an immediate effect. Such immediacy contrasts starkly with the multifaceted challenges posed by anthropogenic emissions of greenhouse gases into the atmosphere, which typically have decadal to centennial residence times and are intimately linked to global economic growth and prosperity. However, achieving a direct response is highly unlikely and runs counter to the observed trend toward increased use of artificial light in urban areas due to various other societal concerns. A staged approach to limiting ALAN is therefore more realistic and pragmatic, particularly as many regions of the world have only recently adopted LED technology to replace older LPS lighting. LED lighting peaks more strongly in the blue end of the spectrum, compared with the orange–red of LPS, and penetrates deeper into the water column by virtue of the optical properties of pure water (Smith and Baker, 1981; Pope and Fry, 1997). This deeper penetration is offset somewhat in more productive coastal and estuarine waters where contributions to absorption by phytoplankton and CDOM (Prieur and Sathyendranath, 1981) become significant in the blue. However, even in these more optically complex inshore waters, blue and green light originating from ALAN penetrates much further into the water column than red (Davies et al., 2020). Therefore, rather than the Spartan approach to coastal urban lighting of turning everything off at night, likely solutions could employ some form of filtering or spectral tuning of the LED lighting (i.e., low Kelvin) in order to offset some of the worst impacts.

We have calculated the critical depth (Z_c) to which biologically detectable light can penetrate the water column for the global coastal ocean, as represented by exclusive economic zones. We have quantified the regions of the global coastal ocean which are already heavily impacted by ALAN, namely areas characterized by widespread offshore developments as well as large coastal urban areas. The global transition toward white LED lighting will likely exacerbate in-water light pollution and its impact across species, populations, and marine ecosystems. This atlas on global marine light pollution can provide a powerful tool to guide research and conservation to mitigate the ecological impacts of ALAN.

Global fields of Z_c in netCDF format can be accessed via https://doi.pangaea.de/10.1594/PANGAEA.929749. Code used to create the global fields can be accessed via https://github.com/timjsmyth/ALAN.

The supplemental files for this article can be found as follows:

Figures S1–S11. Docx

All Global EEZ ALAN impact data set. Tables S1 and S2. xlsx

The authors would like to thank Fabio Falchi for making the entire floating-point data set from the 2016 World Atlas of Artificial Night Sky Brightness available to us upon request.

UK Natural Environment Research Council grant NE/S003568/1 (TS, TD, DM, AW, ST). Israel Ministry of Science, Technology, and Space, Office of Energy and Water, grant number 3-11110 (RT, DI, ZD).

The authors have declared that no competing interests exist.

Contributed to conception and design: TS, TD, AW.

Contributed to acquisition of data: TS, TD, DM, RT, ZD, DI.

Contributed to analysis and interpretation of data: TS, TD, AW.

Drafted and/or revised the article: TS, AW, TD, ST, DM, RT, ZD, DI.

Approved the submitted version for publication: TS, AW, TD, ST, DM, RT, ZD, DI.

How to cite this article: Smyth, TJ, Wright, AE, McKee, D, Tidau, S, Tamir, R, Dubinsky, Z, Iluz, D, Davies, TW. 2021. A global atlas of artificial light at night under the sea. Elementa: Science of the Anthropocene 9(1). DOI: https://doi.org/10.1525/elementa.2021.00049

Domain Editor-in-Chief: Jody W. Deming, University of Washington, Seattle, WA, USA

Associate Editor: Kevin R. Arrigo, Department of Earth System Science, Stanford University, Stanford, CA, USA

Knowledge Domain: Ocean Science