Ontogenetic exposure to light influences seabird vulnerability to light pollution

The current increase of pollutants such as artificial light has detrimental consequences for many species (Gaston and de Miguel, 2022). Seabirds, and particularly burrow-nesting seabird species, are among the most endangered avian groups and are highly affected by light pollution (Dias et al., 2019; Rodríguez et al., 2019). Upon their maiden flights to sea, which occur at night, the fledglings of many burrow-nesting seabird species are grounded across coastal urban areas after encountering artificial lights. These are high mortality events (fallouts), which impact over 70 seabird species worldwide, some of which are critically endangered (Rodríguez et al., 2017b; Silva et al., 2020). Despite improvements to urban lighting, which reduce light pollution, the number of grounded birds remains high (more than 4000 Cory's shearwaters were grounded in a single breeding season in the Azores archipelago (Rodríguez et al., 2017b), and despite well-established rescue campaigns, which reduce direct mortality, light pollution plays a role in the decline of seabird populations (Fontaine et al., 2011; Gineste et al., 2016; Raine et al., 2017).

Low levels of moonlight, strong winds or the presence of fog and rain aggravate fallout (Rodríguez et al., 2014; Syposz et al., 2018; Telfer et al., 1987). Fledglings are disproportionately affected by light pollution (68–99% of recorded individuals) while adults are seldom found (reviewed in Rodríguez et al., 2017b). While there has been an increase in the number of studies focusing on fallout events, the biological mechanisms involved remain unclear (reviewed in Brown et al., 2023). For instance, what drives the higher vulnerability observed in fledglings, is there variation in vulnerability at an individual level and how is it shaped, and what is the contribution of light characteristics to the behavioural responses to this pollutant?

Four non-mutually exclusive hypotheses have been proposed to explain fallout events. First, foraging inexperience: fledglings may be innately attracted to the bioluminescence of their prey and may confuse it with artificial light stimuli (Imber, 1975). Second, association of light with food source: chicks may relate the lit entrance of the burrow to the delivery of sustenance (D. Ainley, personal communication, in Rodríguez et al., 2017b). Third, loss of orientation cues: artificial lights may interfere with the orientation mechanisms of birds (Guilford et al., 2018; reviewed in Rodríguez et al., 2017b). Fourth, a more recent hypothesis suggests that a still under-developed and untrained visual system, together with behavioural inexperience at fledging, may drive the observed age bias (Atchoi et al., 2020). Chicks of burrow-nesting species grow with limited exposure to ambient light. These chicks remain in their covered nests for most of the breeding season, only emerging during the last weeks before fledging to train flight motion or stretch, and even then these excursions are conducted at night (Yoda et al., 2017a). While burrows can vary in shape, depth and size, selective pressures (e.g. predation) towards narrow, deep and sinuous burrows (Hervías et al., 2013; Ramos et al., 1997; Rodríguez et al., 2022b) lead to chicks growing in darker environments lacking light stimuli. Such light stimuli are necessary for a proper development of vision, a costly sensory system, and for stimulating visual learning (Mitkus et al., 2018). Delayed development of the visual system has been observed in the burrow-nesting Leach's storm-petrel Hydrobates leucorhous (Mitkus et al., 2018). However, empirical evidence is needed to support the link between vision development and vulnerability to light pollution.

Factors such as repeated exposure to light during early life or the characteristics of light (e.g. wavelength and intensity) can shape the development of the visual system and visually guided behavioural responses in many taxa (Bateson, 1976; Falcón et al., 2020; Martin, 2017; Nevitt, 2008), thus potentially influencing vulnerability to light pollution. Differences in ontogenetic light exposure can lead to lateralization of vision (in avian species; Rogers, 2002), shifts in photoreceptor sensitivity (in marine crustaceans; Cronin et al., 2001) and changes in eye size (in freshwater crustaceans; Li et al., 2022). Short wavelengths are responsible for the synchronization of circadian clocks and are widely recognized as the most disturbing spectra for the large majority of light-sensitive organisms (Falcón et al., 2020; Longcore et al., 2018). They can disrupt normal flight routes of migratory birds, leading to abnormal concentrations of birds close to the light source (Zhao et al., 2020), or lead to avoidance behaviour in passerines (Goller et al., 2018). In seabirds, shorter wavelengths are preferred by penguins, perhaps because they enhance visual performance during waddling to their nests (Rodríguez et al., 2018); they also repel adult shearwaters from coming to the colony (Syposz et al., 2021a) and lead to higher fallout numbers of shearwater fledglings (Rodríguez et al., 2017a; Salamolard et al., 2007).

In this study, we investigated the drivers of vulnerability to light pollution in burrow-nesting seabirds by conducting a behavioural assay in which Cory's shearwater Calonectris borealis chicks were exposed to contrasting artificial light stimuli across an age gradient. In addition, to investigate how differences in natal environment may influence vulnerability to light pollution at an individual level, we assessed the available ambient light inside each nest. We expected (1) an increase in the magnitude of reactions (here represented by the number, timing and orientation) with chick age, i.e. evidence of a developing visual system (Mitkus et al., 2018), and (2) shorter wavelengths to provoke more responses than longer wavelengths (Longcore et al., 2018; Rodríguez et al., 2018; Syposz et al., 2021a). A functioning visual system needs training to allow the organism to learn how to interpret light stimuli and select an appropriate response (Rabin et al., 1981). By introducing light treatments to groups of chicks at different ages, we expected that (3) chicks exposed more times to our artificial light stimuli would display a greater magnitude of reactions (Bateson, 1976) and, accordingly, (4) chicks from darker nests would react less, as they would have been less exposed to ambient light so that their visual system would have remained less trained (Bateson, 1976).

Cory's shearwater Calonectris borealis (Cory 1881) is a medium-sized procellariiform seabird, which breeds in subtropical north Atlantic archipelagos from February to late October–early November. It is estimated that ∼6% of fledglings in the Azores and ∼14% on the Canary Islands are annually grounded in fallout events by light pollution, prompting rescue campaigns in operation for over three decades (Fontaine et al., 2011; Rodríguez et al., 2017b).

Field experiments were conducted from August to October 2020 at the Cory's shearwater colony of Capelinhos (38°35′N, 28°49′W), situated at the western extremity of Faial Island, Azores, Portugal, and distant from artificial light sources (>4 km from the nearest illuminated port, and >17 km to the city of Horta, the main hub for light pollution on the island). The nests sampled were located on a west-facing slope flanked by the sea. We conducted pre-hatching visits to the study colony every other day from 8 to 25 July 2020 and recorded hatching date for chicks (n=25), allowing age estimation with a maximum of a 2 day error. The first and the last chicks of this study hatched on 15 and 25 July 2020, respectively.

Manipulation and data collection from the individual Cory's shearwaters and the execution of the experiment within the protected area of Capelinhos were conducted under a handling licence from the Azorean government (46/2020/DRA) and an ethical license from University of the Azores (UAC/2020/4182).

The experimental arena (see Supplementary Materials and Methods) was made inside a 60×40×53 cm opaque black polypropylene box with an artificial burrow on one end and the light system on the top (Fig. 1). Black opaque square plastic vases, measuring 24×24×22 cm with a U-shaped 12×10 cm opening on one side, were used to simulate a burrow and its entrance (O'Dwyer et al., 2008). Chick behaviour during both experimental tests and control tests was recorded using video cameras (see Supplementary Materials and Methods, Fig. S1). Burrow-nesting seabirds possess a highly sensitive olfactory system capable of identifying the smell of their own nest and of themselves against that of conspecifics (reviewed in Nevitt, 2008). To reduce possible confounding effects caused by the odours of different individuals remaining inside the arena, we placed a black cardstock at the bottom of the arena and replaced it with a new one for every treatment. The artificial nests were interchanged between each treatment, cleaned with 70% ethanol (Bonadonna et al., 2006) and allowed to dry until ready to be reused (see Supplementary Materials and Methods). The arena was also cleaned with ethanol after each test, i.e. between two different chicks, or after a treatment whenever a chick defecated.

Two light systems were made from red and blue LED strips (12 V, 14.4 W m⁻¹) with a final illuminated area of 13.3×13.3 cm. The spectrum (Fig. S2) and luminous efficacy of the final light stimuli were measured in an integrative sphere (illumia^®pro 500, Labsphere Inc.), and the luminance was measured with a LumiCam 1300 imaging photometer (Instrument Systems GmbH). The measurements allowed calculation of the radiance of the stimuli that was used for further visual modelling and adjustment of the LED lights (the red LEDs were dimmed with a step-down current controller).

After being installed in the arena, the blue LEDs yielded illuminance of 30 lx at the chick's starting position (Fig. 1). This light level corresponds to illuminance at civil twilight (Cronin, 2014) and was considered appropriate for our experiments. Such a light level should not be harmful as it is similar to what the chicks would experience outside the burrow after sunset. The red LEDs were subsequently adjusted (see below). Levels of <0.1 lx were obtained when LEDs were off, i.e. in the control or no-light arena setup. All illuminance values were measured using a Hagner Screen-Master light meter (B. Hagner, Solna, Sweden).

To test the effect of the light spectrum, we aimed to make both LED stimuli as isoluminant as possible for the achromatic visual system, but to have strong contrast for the chromatic visual system of the species tested. As generally accepted, we assumed that achromatic vision is driven by signals from double cones, while single cones mediate chromatic vision in birds (Martin and Osorio, 2008; Olsson et al., 2018).

Photoreceptor spectral sensitivity was modelled in Matlab (R2015b, The MathWorks, Natick, MA, USA) using a visual pigment template (Govardovskii et al., 2000), as described in detail by Lind et al. (2013b). The modelling parameters were based on the visual system of the wedge-tailed shearwater Ardenna pacifica, a closely related species of the Procellariidae family, for which such data exist. The maximum wavelength (λ_max) of photoreceptor pigment absorbances, the cut-off wavelength (λ_cut) and wavelength of half-maximum absorbance (λ_mid) of oil droplet absorbances, and ocular media transmittance (OMT; λ_T0.5 at 335 nm) were taken from Hart (2004). The Weber fraction of the long wavelength-sensitive (LWS) mechanism was set to 0.1 (Olsson et al., 2018). The cone abundance ratio (VS:SWS:MWS:LWS 0.7:0.5:0.7:1) was taken from Hart (2001). The pigment absorption coefficient was set to 0.035 μm⁻¹ (Warrant and Nilsson, 1998) and the cone outer segment length was assumed to be 16 μm; that is, the value assumed for the wedge-tailed shearwater (Hart, 2004) and found in another procellariiforme, the northern fulmar Fulmarus glacialis, by Mitkus et al. (2016).

The final adjustment of the light stimuli yielded a strong chromatic contrast of 60.7 JNDs (just noticeable differences; the discrimination threshold is 1 JND) between them and an achromatic contrast, calculated as Michelson contrast for the double cones, of 4%. As most birds tested to date have very poor achromatic contrast sensitivity (usually birds cannot discriminate contrasts below 8–10%; Lind et al., 2012; Olsson et al., 2018), and contrast sensitivity matches brightness discrimination well (Lind et al., 2013a), we believe that Cory's shearwaters would not be able to distinguish our stimuli based on luminance alone.

The light systems were made to be as similar as possible, to avoid confounding behavioural responses. Very small thermal, acoustic and electromagnetic differences possibly occurred, but we assumed such potential differences to be insignificant.

Chicks (n=25) were randomly and evenly distributed among five groups (hereafter groups 1–5), and repeatedly tested 5 times at incremental ages (hereafter ages 1–5) across the nestling period (Table 1; Table S1). Each group was subjected to either a Control-test or an Experimental-test, except Group1, which was only subjected to Experimental-tests (i.e. from Age1) (Table 1).

Control tests consisted of three 10 min long bouts without any light inside the arena (control treatments, equivalent to no-light treatment in experimental tests). Experimental tests consisted of three 10 min long bouts (blue light, red light, no-light) assigned in a random order. Before each 10 min bout, the chick was placed within a dark box positioned inside the arena for 5 min to dark adapt, even during the control test treatments. The duration of treatment and dark adaptation bouts was determined based on similar studies, and adapted to fit the experimental time frame while keeping experimental time to a minimum (Brooke, 1998; Mitkus et al., 2018).

Each group was subjected to the experimental tests at progressively older ages, creating a variable based on the number of times chicks were exposed to the light stimuli (hereafter ‘exposure’), thus the total length of exposure to experimental light varied from 100 min for group1 to 20 min for group 5.

For every experimental session, each chick was subjected to a total of 15 min of dark adaptation (in 5 min bouts) and 30 min of treatments (in 10 min bouts), totalling 45 min spent inside the arena per chick per night. We tested every chick at every age group, with either experimental tests or control tests to ensure equal handling time for all individuals. As it would not be possible to test all 25 chicks every night because of the long duration of each session, each set of experiments for a given age group spanned 2–3 nights. On average, chicks spent 80 min outside the nest for each experimental night, from collection to return, leading to a total of 400 min of handling and experimentation per chick, during the entire experimental period.

To minimize exposure to artificial light other than the experimental light – for example, from the researchers' headlamps – chicks were captured at their nests by hand in the dark. Immediately after capture, chicks were placed inside an opaque bird bag, which in turn was placed inside a blackout cardboard box, in which the chick was transported to the arena. When navigating the colony to reach the nest, we avoided using any light, but when needed headlamps were set to red light of the lowest available intensity. Chicks were brought to the arena one at a time.

We obtained three response variables. (1) ‘Reaction’ – a binary variable of reaction (whenever chicks stepped outside the release location, i.e. the middle area of the arena) versus no-reaction (when chicks stayed within the middle area) (Fig. 1, Table 2). We recorded the position of the chick at the end of a treatment and classified the behavioural responses as dark-ward (when chicks moved inside the artificial nest or to the darker area whenever light treatments were on), light-ward (when chicks moved into the brighter area whenever light treatments were on) or no-reaction (Fig. 1). Subsequently, we pooled dark-ward and light-ward behavioural responses, thus creating the binary variable of reaction versus no-reaction. This was done because we wanted to analyse all reactions to treatments, regardless of the type of reaction (light-ward or dark-ward) and regardless of their frequency (the number of light-ward reactions was negligible) (Fig. 2). (2) ‘Timing of reaction’ – the number of minutes a chick took to react (no-reactions were not included), which was extracted from video footage. (3) ‘Orientation’ – the direction a chick was facing at the end of a treatment, in 45 deg angle increments (Fig. 1, Table 2). Orientation was recorded irrespective of whether the chicks had reacted or not, allowing the analysis of all data.

We collected data on four physical attributes of the burrows to generate a nest darkness index, a proxy for the growing chicks' exposure to ambient light during the nestling period. The index was created as an alternative to the more impractical and expensive method of repeated measures inside the nest via a light meter. The four attributes were: burrow depth (the distance from the burrow entrance to the nest chamber), burrow shape (straight or curved), burrow entrance area (Fig. S3) and illumination of the chick while at the nest chamber, regarding direct exposure to daylight (light, half-way and darkness) (see Supplementary Materials and Methods). Each attribute was transformed into a 0–1 scale (Table S2) and summed to create an index ranging from 0 to 4. Lower values indicate light-exposed nests, i.e. with larger entrances and higher exposure of the chick, while higher values indicate dark nests, i.e. with smaller entrances and deeper pathways, where chicks were less likely to be exposed to ambient light.

We ran statistical models for each response variable: reaction (0–1), timing of reaction (1–10 min) and orientation (0–180 deg) (Table 2). All three response variables were modelled against four explanatory variables: the factor treatment with four levels (control, blue, red and no-light) and the ordinal variables age, exposure, and their interaction (age×exposure) (Table 3). For reaction, we ran a generalized linear mixed model (GLMM) with a binomial family with logit link, including nest identity as a random factor. For timing of reaction, we ran a negative binomial generalized linear model (GLM). Because there was a smaller sample size (n=76), as many individuals did not react after the 10 min of treatment, we excluded the random factor term (nest identity) to avoid overfitting. For orientation, we fitted a linear mixed model (LMM) including nest identity as a random factor. We tested the same models using data from light treatments only, for comparison (Table S3A). Interactions between age and exposure (age×exposure) were tested but removed from the final models if they were not significant (Table 3; Table S3B). We calculated variance inflation factors (VIFs) and did not identify multicollinearity issues (all VIFs<3.5 in reaction model; VIFs<3 in orientation model; and VIFs<4.3 in timing model).

To remove the confounding effect of exposure when analysing responses against darkness index, we summed the reactions to blue and red light treatments, during the first exposure to experimental light for each chick (i.e. exposure 1). The responses to the no-light treatment were not considered for this analysis. This generated a variable of summed reactions ranging from 0 (no-reaction to any of the light treatments) to 2 (reaction to both light treatments: blue and red). We fitted GLMs (family=Poisson) to explain variation in summed reactions with darkness index, age and their interaction (darkness index×age), as explanatory variables. The best model was selected using the corrected Akaike information criterion (AICc) with the lowest AICc indicating the best model (Table 4).

All statistical analyses and data manipulation were conducted using R version 4.1.2. (http://www.R-project.org/). LMM and GLMM were conducted using R package lme4 (Bates et al., 2015), GLM (family negative binomial) using R package MASS (Venables and Ripley, 2002), and GLM (family poisson) using R core package stats. VIF analysis was conducted using R package car (Weisberg and Fox, 2019), and AIC calculations using R package MuMIn.

Two of the 25 studied chicks were lost. Both were tested at age 1, but not found in their burrows from age 2 onwards; thus, we collected data from 351 out of the planned 375 trials (62.5 h of experimental procedure).

Reactions were very seldom recorded during no-light treatments (3/147 during control tests, 7/68 in no-light treatments during experimental tests). Such instances were recorded for distinct chicks and evenly distributed among the different age and exposure groups.

The chicks predominantly showed dark-ward responses in both light treatments, while the occurrence of light-ward responses was negligible (Fig. 2). The proportion of reactions (dark-ward plus light-ward) versus no-reaction was similar between light treatments (40/68 for blue light and 33/68 for red light) (Fig. 2). The proportion of reactions varied little among ages, but increased with exposure (Fig. 3).

Chicks showed a higher number of reactions and orientated towards the darker area more frequently during light treatments (blue and red) and with increasing exposure (Figs 2 and 3, Table 3). Chicks took increasingly more time to react with increasing exposure (Fig. 4, Table 3). No-light (in experimental tests) and control treatments elicited a low number of reactions and a lower occurrence of dark-ward movements than the blue or red light treatments (Fig. 2; Table S3C). No effect of age was observed for any response variable (Table 3). There was no evidence of a difference between blue and red light treatments, but 95% confidence interval (CI) values support a higher effect of blue treatment in chick reaction, timing of reaction and orientation models (Table 3; Table S3C).

A positive correlation between the number of reactions of an individual chick and the darkness index of its nests was found (Fig. 5, Table 4). Age had no significant effect on this relationship (Table 4).

This study presents empirical evidence for the contribution of the light available in the rearing environment to seabird visual development. Our results show that Cory's shearwater chicks exposed more often to artificial light during the nestling period display a higher magnitude of behavioural responses to light. Cory's shearwater chicks of all ages sought darkness when exposed to artificial light. These results agree with our initial predictions regarding exposure and light avoidance behaviour but fail to confirm an effect of age on the magnitude of responses. Blue light elicited slightly stronger behavioural responses than red light, although no significant differences were found. Finally, an exploratory analysis revealed a correlation between natal environment conditions and the response of chicks to artificial light, with chicks from darker nests displaying more reactions to the light stimuli.

The rate of reactions observed in our study was 49% when chicks were subjected to red light, and 59% when subjected to blue light. Similar studies obtained different response rates. When presented with a two-choice maze with light stimuli, 90% (n=154) of adult common diving petrels, Pelecanoides urinatrix (Brooke, 1998), and approximately 70–80% (n=401) of loggerhead turtle hatchlings, Caretta caretta (Young et al., 2012), made a choice (sea turtle hatchlings are similarly affected by artificial light during emergence from their buried nests). An inability to exhibit a behavioural response in procellariiform species could derive from an underdevelopment of the visual system as seen in Mitkus et al. (2018), where 55% of younger Leach's storm petrel chicks reacted to a light stimulus, while reaction rates increased to 80% with older chicks.

In our study, we expected younger chicks and chicks less exposed to light to react less. Indeed, the low rate of reactions observed in our study could be partially explained by an underdevelopment of the visual system, as evidenced by the increase in reactions with an increase in exposure to light (Fig. 3B). However, the lack of reactions cannot be explained by the development of the visual system alone, as some individuals never reacted to light, regardless of exposure (Fig. 6), and because contrary to our expectations, older chicks did not react more than younger chicks, i.e. reaction rates or magnitude did not increase with age (Fig. 3A). In our study, chicks took more time to react as they were progressively more exposed to the light stimuli (Table 3). Chicks could take longer to react because (1) their vision has improved so that they have become better able to perceive their surroundings and will look around to assess the information before making a more informed choice (head scanning behaviours were observed during treatments) or (2) they may have become habituated to the light stimuli, identifying them as less dangerous, so taking longer to seek darkness as there is no imminent threat (Bateson and Seaburne-May, 1973). Mitkus et al. (2018) presented evidence of delayed development of the visual system in Leach's storm petrel, a species found in fallout events (Wilhelm et al., 2021). Older Leach's storm petrel chicks displayed more visually guided behaviours than younger individuals. This was not the case in our study, where Cory's shearwater chicks reacted to light at the earliest tested age (2 out of the 5 youngest nestlings reacted when they were 24±2 days old), while some individuals did not react at any tested age (the oldest chick to not react to the experimental stimulus was 81±2 days old). Given the similarity between the two studies and species, the disparity of results suggests that visual development in seabirds might be shaped by a more complex set of factors, such as ontogenetic exposure to light or species-specific traits.

Unexpectedly, no significant differences were found between the number, timing or orientation of responses to blue versus red light stimuli (Fig. 2; Table S3C). However, because both the statistical model estimates for the three response variables (see 95% CI of Table 3) and proportion of reactions (Fig. 2) are compatible with a stronger effect from the blue light treatments, such a difference might be observed using a larger sample size. Indeed, blue light (shorter wavelengths) has been identified as more disruptive to various taxa in general (Longcore et al., 2018) and to seabirds in particular (Rodríguez et al., 2018, 2017a; Syposz et al., 2021a). Spectral sensitivity varies among species, with some wavelengths being perceived more intensely than others. Thus, it is crucial to design experiments which differentiate between intensity and wavelength discrimination for the target species (Young et al., 2012). Our LED light stimuli were prepared based on the visual system parameters of the wedge-tailed shearwater, which has similar ecology, life-history and morphology, and is a close relative of Cory's shearwater. Thus, theoretically, Cory's shearwater chicks should be able to distinguish between the two light treatments based on their spectra, but not on their intensity difference.

It has been proposed that light stimuli could disrupt the magnetoreception systems and lead to the disorientation observed during fallout events (Brown et al., 2023; Guilford et al., 2018); however, evidence for magnetoreception function in Procellariiformes, e.g. how it may be expressed in chicks or its spatial and intensity thresholds, remains uncertain and largely untested. Disruption of the magnetoreceptors of Manx shearwater (Puffinus puffinus) fledglings had no effect on the direction of flight routes (Syposz et al., 2021b), but evidence for compass orientation was observed during maiden flights of streaked shearwater (Calonectris leucomelas) (Yoda et al., 2017b). Further research is necessary to disentangle the mechanisms regulating visual and magnetoreception systems regarding behavioural responses to light in seabirds.

We consistently observed darkness-seeking behaviour, as chicks moved and faced away from the light, in agreement with similar studies (Mitkus et al., 2018) (Fig. 2, Table 3). Such behaviour is expected in growing chicks and is consistent with the life-history of Cory's shearwaters, as darker environments should be associated with safety and shelter provided by the burrows, especially for chicks, which are unable to fly to escape predators. However, petrel fledglings seem to be attracted towards lit environments after fledging, as evidenced by fallout events (Imber, 1975; Reed et al., 1985; Rodríguez et al., 2022a, 2015). Such a reversal in responses to light could be an indication of a behavioural switch related to the developmental stage. This has been observed in the cavity-nesting common starling, Sturnus vulgaris, which shows negative phototaxis as young chicks and positive phototaxis as chicks approach fledging (Minot, 1988). In our study, we did not observe evidence for such a switch as chicks got older, but our experimental design may not have enabled us to detect a sudden or temporary switch close to fledging as the oldest chicks tested were still 2 weeks before fledging date. A behavioural switch also contrasts with the avoidance behaviour observed for adult shearwaters, which avoid the area whenever it is illuminated (Syposz et al., 2021a). A more likely explanation could be that behaviour towards light is activity dependent. Chicks, fledglings and adults look for darkness while on the ground. This is often observed during Cory's shearwater fallout rescue campaigns, when grounded fledglings are found under cars or other dark small spaces (Rodríguez et al., 2017b). In contrast, fledglings flying towards the sea display tortuous paths and appear trapped by lit urban areas, apparently being attracted to artificial lights (Rodríguez et al., 2022a).

We predicted that repetitive exposure to light during chick growth, either exceptionally via our experimental setup (intentional exposure to artificial light) or cyclically at their nest (differential exposure to natural light depending on the shape or depth of their burrows), should stimulate the development and training of the visual system, and thus influence behavioural responses (Bateson, 1976; Gunnarsson et al., 2008). Our results show that even a short exposure to artificial light stimuli throughout the nestling period (<100 min) can increase the magnitude of reactions to light (Fig. 3). Similarly, other studies have shown that even a brief exposure to continuous light alters visually led behaviours in dark-reared poultry chickens (Bateson and Seaburne-May, 1973; Cherfas, 1977), which was attributed to the non-specific activation of visual pathways (Bateson, 1976). Our results also show that individuals exposed to experimental light fewer times displayed a very low number of reactions, even in older chicks (Fig. 6). Under natural conditions, chicks of burrow-nesting seabirds grow in dark environments where they might lack the necessary light conditions to train their vision. Thus, their visual systems remain underdeveloped at fledging, at least to some degree, which could be a main driver of their disproportionate vulnerability to light pollution when compared with adults (Atchoi et al., 2020). Yet, despite the predominance of dark environments during growth, the chicks of burrow-nesting seabirds should still face differential exposure to ambient light as a result of the variability of burrow structure, orientation or location. This differential exposure could then lead to individual differences in visual system development and to subsequent differences in vulnerability to light pollution, depending on factors such as burrow type and light pollution incidence at their colonies.

Contrary to our expectations, however, we found that chicks from darker burrows displayed more reactions when exposed to the artificial light stimuli for the first time (Fig. 5). Such a contradiction could be explained by a stronger initial reaction towards a disturbing stimulus by chicks less habituated to the presence of a lit environment (chicks from darker nests), as observed in domestic chickens (Bateson and Seaburne-May, 1973). It also provides evidence of the ability of Cory's shearwaters to detect and react to light stimuli even at a young age, when their visual systems are less developed or trained. This is further supported by the fact that the eyes of Cory's shearwater chicks become open as early as 2 days post-hatching (E.A., observation), and chicks in our study did react to light from 24 days old onwards (Fig. 6).

Ontogenetic exposure to artificial light in songbird nestlings can lead to changes in breeding phenology (purple martin, Progne subis; Assadi and Fraser, 2021) and physiology (great tit, Parus major; Raap et al., 2016). Burrow-nesting seabirds will most likely grow in increasingly light-polluted environments, due to the spread of artificial light across both natural and urban areas (Falchi et al., 2016; Garrett et al., 2019). How this change will affect individuals and populations remains to be determined. For instance, artificial light might act both as a pollutant and as its own mitigation factor. Light pollution at the colonies may stimulate the development of vision in chicks from the most exposed nests, rendering them less vulnerable to light pollution at fledging. However, it may also postpone fledging dates (purple martin; Assadi and Fraser, 2021) or reduce chick feeding frequency (Scopoli's shearwater, Calonectris diomedea; Cianchetti-Benedetti et al., 2018) with potentially adverse effects on the fitness of the chicks. In colonies close to artificially lit urban areas, the vulnerability of chicks from less exposed nests, which are less habituated to light (Senzaki et al., 2020) and with less developed visual systems at fledging, could be exacerbated as a result of a high contrast between environments, from a dark burrow or dark-sky colony to a highly lit environment immediately after fledging in coastal urban areas. Indeed, within the same colony, seabird chicks from nests in areas facing an urban environment are at greater risk from light pollution than chicks from nests situated in areas that face away from artificial light (Wilhelm et al., 2021). A similar effect can be observed in fledglings from colonies with minimal exposure to light pollution, but whose paths to the ocean cross over lit urban areas (Rodríguez et al., 2015).

Our study did not account for other factors that regulate behaviour, such as parental care (which determines fitness) or individual characteristics (e.g. body condition or personality). Further research with longer study periods, larger sample size and more variables is needed to better characterize individual vulnerability to light pollution: how exposure to light within the natal environment affects the development of the visual system, how light pollution affects parental care and consequently how fitness and individual characteristics affect visual behaviour.

Understanding the effects of artificial light at night (ALAN) on biodiversity is critical given the recent global transition of public lighting to white LEDs (Kyba et al., 2017), which brought with it a large increase in the radiance of the visible spectrum (de Miguel et al., 2021), as well as an increase of light pollution in natural areas (Garrett et al., 2019). Our results present evidence that repeated exposure to light influences the development of the visual system in seabird nestlings. This demonstrates that light pollution not only affects seabirds throughout urban areas during the fledging period but also potentially interferes with the development and fitness of individuals at their colonies and across the breeding season. These results highlight the importance of integrating a broader temporal and spatial scale when designing management protocols for urban lighting, expanding mitigative actions beyond the limits of fallout events, and taking into account the potential array of impacts that altered nightscapes at breeding colonies may have on seabird populations.

The Institute of Marine Sciences - OKEANOS provided logistical (office, experts, network and transportation) support to the study. We thank Martyna Syposz for improvements to the design of the experimental assay, Observatório do Mar dos Açores for providing logistical support, and field assistant Hannah Gretham for contributing to data collection. Yasmina Rodríguez provided helpful constructive comments on earlier drafts of the manuscript. We are grateful to Faial Natural Island Park and its rangers (in particular Valter Medeiros and Dejalme Vargas) for logistical support. We thank Lorenzo Kunze for providing expert technical assistance. Finally we are grateful to Olle Lind and Peter Olsson for advice on visual modelling.