Breeding Stage and Environmental Factors Affect Prey Selection by Greater Crested Terns Off Southeast Australia

Variation in the diet of marine predators such as seabirds can be used to track environmentally-driven changes in ocean ecosystems. However, studies of predator diet must account for intrinsic inuences on prey selection, such as changing nutritional requirements during breeding. Using digital photography, we investigated how the type and size of prey brought back to the colony by greater crested terns (Thalasseus bergii) changed in relation to breeding stage, and to variation in oceanographic conditions around Montague Island, Australia (36°15’S, 150°13’E). 2469 prey items were identied to species or family level over 35 consecutive days of photo-sampling in 2018. Australian anchovy (Engraulis australis), a surface-schooling clupeid sh, was the most abundant prey returned to the colony during all breeding stages (84.5%). The proportion of anchovy increased from 77.0% when birds were provisioning their adult partners during incubation, to 92.4% when they were provisioning chicks, suggesting selective foraging behaviour on this energy-dense species as a means to facilitate rapid chick growth. Anchovy size was largest during incubation (91.1 ± 14.9 mm), smallest during early chick provisioning (71.8 ± 11.0 mm), and increased slightly during mid-provisioning (79.6 ± 11.9 mm), indicating adaptive prey selection that is matched to the physical requirements of different breeding stages. The proportion of anchovy prey was also inuenced by extrinsic environmental factors, with anchovy becoming more dominant with increasing local sea surface temperatures, up to ~17.5°C. Our ndings highlight the importance of examining both intrinsic and extrinsic determinants of diet composition across breeding stages in seabird populations.


Introduction
Upper trophic-level predators such as seabirds can act as sentinels of changes in marine ecosystems, with their demography, behaviour and physiology re ecting variability in oceanographic and ecological conditions (Hazen et al. 2019). For example, temporal changes in the occurrence, abundance and size of prey that seabirds consume can be used as indicators of food availability and ecosystem structure (Suryan et al. 2002;Parsons et al. 2008), and can provide an early warning of anthropogenic impacts on food webs (Green et al. 2015). Monitoring seabird diets can therefore be an important tool to inform conservation biology and ecosystem-based sheries management (e.g. Lorentsen et al. 2018;Thayne et al. 2019;Hazen et al. 2019).
During the breeding season, the foraging range of seabirds is constrained by central-place foraging, which requires individuals to return to the colony between foraging trips to incubate eggs or provision their young (Cairns 1988). This means that birds must forage e ciently on available prey that is within a relatively short distance of the colony. Optimal foraging theory for central-place foraging predicts that single-prey loaders (species including guillemots and terns that carry a single whole prey item per foraging trip back to the colony) should maximise the energy load per unit foraging by providing larger, or energetically richer prey to their chick than they feed themselves (Orians and Pearson 1979). For instance, adult crested terns (Sterna bergii) conduct short foraging trips close to the colony to provision their offspring several times a day with small single prey items (McLeay et al. 2010). This behaviour is assumed to be adjusted to energetic requirements and ingestion capabilities of chicks. Therefore, during provisioning, single-prey loaders should bring back small fatty sh that are energy dense, selecting increasingly large sh as the chick grows (Batchelor and Ross 1984). This contrasts with the foraging strategy of multiple-prey loaders (e.g. auks carry multiple prey items per foraging trip in the bill; Procellariiformes regurgitate stomach oil), which do not have the same constraints on the selection of individual prey items, but should maximise the energy load per foraging trip, especially if they travel large distances from the breeding colony (Orians and Pearson 1979).
While intrinsic constraints imposed by provisioning requirements may in uence prey selection during breeding, the availability of prey also varies with extrinsic environmental conditions. Temporal variation in oceanographic variables can have a strong effect on the foraging success of seabirds, which can impact their body condition, growth and ultimately reproductive performance (Harwood et al. 2015). This is particularly true when birds undertake central place foraging during breeding, as they are restricted to nding prey within a limited distance from the colony and cannot move to more favourable foraging areas. For instance, a long-term study on king penguins suggests that anomalies in sea surface temperature (SST) adversely affected breeding success as birds had to dive deeper and farther from the colony to forage on favoured prey (Bost et al. 2015). Similarly, a shift in the diet of black guillemots (Cepphus grylle) from Arctic cod (Arctogadus glacialis) to demersal sh during a period of decreasing sea ice and increasing SST, led to slower chick growth, low edgling mass and an increase in chick starvation (Divoky et al. 2015). In addition, prey availability may be altered at shorter time scales, such as in response to weather (White 2008). Prey capture under adverse weather conditions can be mediated by poor ight performance, or reductions in prey detectability or catchability (Elliott et al. 2014;Kogure et al. 2016). There is evidence that, for example, high wind speeds in uence foraging success in seabirds, as indicated by a decrease in dominant prey, as well as a reduction in prey energetic value and size (Stienen et al. 2000;Howells et al. 2017). Accordingly, the interplay of both intrinsic (i.e. physical requirements of chicks vs. adults) and extrinsic (environmental variability) processes affecting diet composition should be accounted for when interpreting seabird diet data.
Some seabird species bring whole sh back to the colony in their bill to provision their chick or partner at the nest. In the past, experienced researchers have used binoculars to directly identify prey types, a technique that facilitates large samples being collected over short periods, and allows documentation of seasonal variation in prey choice (Rodway and Montevecchi 1996). While this sampling approach is feasible in low diversity ecosystems where birds bring back only a few types of prey, it is challenging in complex food webs where birds consume a broad suite of prey types or carry multiple prey items in their bill (Barrett et al. 2007). This method is limited by observer bias that can result in the misidenti cation of prey and can undermine taxonomic resolution, especially when there are no independent means by which to verify observations (Larson and Craig 2006). Recent advances in digital photography overcome many traditional observer-based limitations, with the potential to provide a veri able, non-invasive tool to study seabird diets. Photographing birds with prey at the colony allows researchers to identify a large number of prey items with greater taxonomic precision and to estimate the size of prey being selected, while facilitating re-analysis and veri cation of species if necessary (Gaglio et al. 2017).
Greater crested terns (Thalasseus bergii) are small coastal seabirds with an extensive breeding range that encompasses the Australian coastline. Terns are nearshore foragers that mostly acquire food by dipping onto the sea surface or diving up to one metre below the surface (Crawford et al. 2005). During breeding, they are central place foragers with adult pairs taking turns to leave the colony to provision themselves and to bring prey back for their partner or offspring. During incubation, a single-prey item is brought to the partner for display or as a gift in courtship. Assuming that the prey brought back for the partner is equivalent to the prey consumed by birds at sea, this is an indication of the diet composition of the adult population. Once the chick hatches, both parents deliver single-prey items to their offspring. The billloading behaviour of terns makes it possible to quantify trends in prey type and size using a digital camera and a telephoto lens without disturbing the colony. Digital photography has previously been used to describe the diet of T. bergii in South Africa, doubling the known diversity of prey species compared to traditional regurgitate sampling, and providing accurate and precise measurements of prey length (Gaglio et al. 2017).
This study aims to identify intrinsic (i.e. physical requirements of chicks vs. adults) and extrinsic (environmental) determinants of diet composition of greater crested terns at a key southern-eastern Australian breeding site, Montague Island, during the 2018 breeding season. Speci cally, we use digital photography as a non-invasive sampling technique to assess variation in diet composition (prey type and size) between breeding stages (incubation and chick provisioning), and to determine whether short-term variability in environmental covariates in uences diet composition. First, we investigate the trade-off made by single-prey loaders between maximising energy intake per foraging trip, and ensuring that prey can be physically swallowed by the chick (Orians and Pearson 1979). We predict that (1) the proportion of anchovy in the diet will increase during provisioning stages, because this is an abundant, lipid-rich prey that can facilitate the rapid growth of small chicks (Batchelor and Ross 1984). Similarly, we predict that (2) anchovy standard length (SL) will decrease during early provisioning when chicks are small, then increase during mid-provisioning as parents adjust prey size in accord with the ingestion capabilities of growing chicks. Second, we analyse shifts in diet composition of greater crested terns in response to short-term variation in environmental variables. Speci cally, we predict that (3) the proportion of anchovy will decrease under adverse environmental conditions such as strong winds, high SST or high waves, which are factors likely to in uence the availability or catchability of surface-schooling prey.

Study site and species
This study was undertaken on Montague Island (36°15'S, 150°13'E), an 81 ha nature reserve located approximately 9 km off the coast of southeast New South Wales, Australia. Greater crested terns nest in the open on the island at high densities. The tern colony on Montague Island comprises multiple subcolonies; this study was conducted on the largest and most accessible one, consisting of ca. 1000 breeding pairs in 2018 ( Fig. 1a; counted from a wide-angle photograph of the colony taken on 20 October 2018). The study was conducted over 35 consecutive days from 2 October to 5 November during the 2018 greater crested tern breeding season. As chicks hatch synchronously across the colony, the rst chick sighting was selected as the day of hatching and the start of provisioning (here: 23 October 2018study day 22). Breeding stages were classi ed as incubation, early provisioning and mid provisioning, and the study covered the full period from incubation to mid provisioning. During incubation, partners alternatively incubate the nest and bring single-prey items to their mate for courtship and pair-bonding.
Early provisioning comprised the week following hatching during which both parents provision the chick in the nest cup, and mid provisioning the subsequent week when chicks start to move around the nest (Gaglio et al. 2018).

Photographic sampling
Greater crested tern diet was investigated using digital photography to assess differences between adults and chicks in the size and type of prey consumed over the course of the study period. Adult terns carrying prey in their bill were photographed as they returned to the colony (Fig. 1a). Photographs were taken using a Canon (Tokyo, Japan) 50D camera tted with a Canon EF 70-300mm f/4.0-5.6 IS USM tele zoom lens. Camera settings were mostly applied following Gaglio et al (2017), with the following settings: i) shutter speed priority 1/2500-1/4000 s depending on light conditions, ii) automatic ISO, iii) high-speed continuous shooting, iv) autofocus on AI servo (autofocus system for continuous tracking of moving subjects), and v) large Jpeg le format. The tele zoom lens was set to autofocus with the image stabiliser on.
Every day, photos of birds arriving with prey at the colony were taken continuously during three two-hour sessions (morning session from 0700-0900 hr, midday session from 1100-1300 hr and afternoon session from 1500-1700 hr AEST; after daylight savings from 0800-1000 hr, 1200-1400 hr and 1600-1800 hr AEDT). Photos were taken from a different vantage point during each session to account for sun position. Photos were taken as a photo-set consisting of three images, with the sharpest selected for prey identi cation and measurements. The distance between the vantage points and the closest bird was approximately 5 m. The birds did not show any sign of stress or disturbance due to the presence of the photographer.

Identi cation of prey species
All blurred and unclear images (due to e.g. distance, position of prey in the bill, lighting etc.) were excluded from the analyses. From the remaining photographs, prey was identi ed to the highest possible taxonomic resolution, based on morphological characteristics ( Fig. 1b-k). Fish identi cation was undertaken by comparing photographs with published images and sh identi cation guides from the Australian Museum (https://australian.museum/learn/animals/ shes), and subsequently veri ed by multiple species experts from the NSW Department of Primary Industries and the University of New South Wales, Australia.

Estimation of anchovy size
Australian anchovy (E. australis) is a common species in the diet of crested terns in Australia (McLeay et al. 2009a). Therefore, this species was chosen as a standard by which to compare the size of prey brought back to the colony during different breeding stages. However, with prey tending to ex in the bill, direct measurements of standard length (SL; the length from the tip of the snout to the posterior end of the hypural plate) from images may underestimate true sh length (Gaglio et al. 2017). Nevertheless, accurate estimates of SL can be obtained by extrapolating from measurements of individual body parts such as eye diameter, head width and operculum width that are easier to measure from photographs.
Accordingly, for each image, the 'line selection tool' in the image analysis software IMAGEJ (Schneider et al. 2012) was used to estimate eye diameter, head and operculum width of each sh by scaling the pixel length in the image to the mean culmen length of crested terns (61.2 mm; Crawford et al. 2005). Using log linear allometric regressions, three estimates of SL were obtained (see Table S1 in Gaglio et al. 2017 The arithmetic mean of these three measurements was used in further analyses including anchovy SL. An example of the application in IMAGEJ is provided in Fig. 2 and detailed information on the accuracy and precision of the methodology is given in Gaglio et al (2017).
Importantly, Gaglio et al (2017) used the European anchovy (E. encrasicolus) to create regressions for anchovy eye diameter, head and operculum width, and these equations were applied to measurements of Australian anchovy in this study. The morphometric characteristics of both species are very similar, although the European anchovy is slightly larger than the Australian anchovy (average length: 13.5 cm vs. 12 cm; Whitehead et al. 1988). Since this study aims to compare relative sh length within the sample rather than illustrating variability in absolute sh length, regression equations calculated for European anchovy were used with the caveat that these measurements may consistently bias the true length of Australian anchovy photographed in this study.

Environmental factors
Key environmental variables that are likely to in uence either the distribution or abundance of prey species in the greater crested terns' foraging range, and/or the terns' ability to catch them were selected for analysis. These included wind speed, sea surface temperature (SST) and wave height. Wind speed (km/h) was recorded by the Australian Bureau of Meteorology weather station located at the lighthouse on Montague Island (www.bom.gov.au). Wind measurements were made using an automatic anemometer with an accuracy of +/-5% of the wind speed for wind speeds greater than or equal to 10 m/s. Wind data were downloaded at a 30 min resolution, each measurement representing the average of the 10 min period prior to the observation time.
Gap-free daily records of satellite-derived SST at a 0.05° x 0.05° spatial resolution were measured by the UK Met O ce's Operational Sea Surface Temperature and Ice Analysis (OSTIA) system (downloaded from marine.copernicus.eu). Based on previous observations of the tern's foraging range from Montague Island (O'Hara 2016), SST values were averaged across an area spanning 35° to 37° north to south, from the coast out to the edge of the continental shelf (i.e. 200 m depth isobath). Because SST was available at a daily temporal resolution, daily values were assigned to multiple photography sessions within a day.
Wave height (m) was recorded by the NSW Department of Planning, Industry and Environment's (DPIE) Batemans Bay offshore Waverider buoy (35°44'25", 150°19'03"; https://mhl.nsw.gov.au/Station-BATBOW; approximately 53 km north of Montague Island). Data were recorded in bursts (lasting approximately 34 minutes long starting on the hour) at 0.5-second intervals. Wave height for each of these periods is de ned as the mean height of the largest 33% of waves.
Previous GPS tracking work on Montague Island showed that tern foraging trips last an average of two hours (O'Hara 2016). Hence any photo taken at the colony during a two-hour sampling period could re ect foraging conditions up to two plus hours earlier than the actual sampling time. Therefore, wind and wave data were assigned to a photo session by taking an average of each variable across a fourhour period starting two hours prior to each session.

Data analysis and statistics
The number of each prey species as a proportion of the total number of prey photographed throughout the study period was calculated to quantify the diet composition of greater crested terns during the 2018 breeding season on Montague Island. Key prey species, i.e. species making up > 1% of the diet, were identi ed to simplify graphical illustrations and statistical analyses. The proportion of each prey type was calculated for each two-hour photo session, and linked to information on breeding stage (incubation, early provisioning, mid provisioning) to compare prey brought back for adults (during incubation), and for chicks (during provisioning stages).
A chi-squared goodness-of-t test was conducted to assess differences in diet composition between incubation and chick rearing. A generalised linear mixed model (GLMM) with beta error structure and logit function was used to investigate the effect of breeding stage on the proportion of anchovy returned to the breeding colony by adult terns. Prior to analysis, the response variable (anchovy proportion) was compressed to avoid absolute values of zero or one by taking y = (y*(n-1) + 0.5)/n, where n is the sample size (Smithson and Verkuilen 2006). The model was tted in R using the package 'glmmTMB' (version 1.0.1; Brooks et al. 2017). Similarly, a linear mixed model (LMM; R package 'lme4'; Bates et al. 2015) was used to investigate the effect of breeding stage on the dependent variable anchovy SL. Each of the models included breeding stage (three levels: incubation, early provisioning, mid provisioning) as xed term and study day (N = 35 days) as crossed random effect to account for the repeated three-session per day design. Finally, a post-hoc test was performed using the package 'emmeans' (version 1.4.6; Lenth 2020) to investigate the effect within each breeding stage category. Results were Tukey-adjusted for multiple testing. For all analyses, model assumptions were checked following Zuur et al (2009).
To visualise differences in prey composition in relation to breeding stage, a non-metric multidimensional scaling plot (NMDS) was generated using the R package 'vegan' (Oksanen 2011). NMDS is an ordination technique to represent the position of data points in multidimensional space using distance measures and a reduced number of dimensions (typically 2). The Bray-Curtis dissimilarity index was used to create a distance matrix that re ects the multidimensional distance between data points with objects grouped closer together being more similar. Ellipses were plotted to depict the centroid and inertia of each breeding stage.
Generalised additive mixed models (GAMMs) using a beta regression and logit function were used to assess the in uence of environmental variables (SST, wave height, wind speed) on the proportion of prey species in the diet. GAMMs, as an extension to GLMMs, allow an estimation of potential non-linear relationships between continuous explanatory variables and the response. The models were tted using the R package 'mgcv' (version 1.8-31;Wood 2006). Smoother terms were applied to explanatory environmental variables and were generated using regression splines. Explanatory variables also included breeding stage as a xed effect and study day as crossed random effect to account for intraclass correlations among day-speci c sampling events. An alpha level of 0.05 was used to determine statistical signi cance. All statistical analyses were carried out in R (v3.5.3; R Development Core Team 2017).

Diet composition
Over the 35 consecutive days of photo sampling, 3280 photo-sets were taken, yielding images of 2469 prey items identi able to species or family level. 1265 prey items were identi ed during incubation and 1204 items during provisioning (early provisioning: n = 482, mid provisioning: n = 722 prey items). Single prey items brought back to the colony were almost exclusively sh, with only one cephalopod species identi ed during the 2018 breeding season (Table 1).
Seven main prey types were identi ed of which Australian anchovy (84.5%) was the dominant species followed by barracouta (Thyrsites atun; 4.9%), trevally spp. (Centrolophidae spp.; 3.5%), goat sh spp. (Mullidae spp.; 1.9%), bluebottle sh (Nomeus gronovii; 1.4%), southern calamari squid (Sepioteuthis australis; 1.1%) and eastern sea gar sh (Hyporhamphus australis; 1.1%; Table 1). Remaining prey items making up a total abundance of < 1% over the breeding season were summarised as 'other sh' to simplify graphical illustration. Table 1 Numbers and proportions of prey species photographed in the bills of greater crested terns returning to the colony on Montague Island in 2018 according to the breeding stage of the terns. A total of 2469 prey items were identi able to species or family level of which 1265 prey were identi ed during incubation (adult diet) and 1204 during provisioning stages (chick diet). Anchovy was the most common prey during all breeding stages but the diet was more diverse during incubation compared to early provisioning (Fig. 3-4). During incubation, in addition to anchovy, other prey such as barracouta, trevallies and southern calamari squid were frequently brought back to the colony but their abundance decreased once chicks hatched. In contrast, species such as bluebottle sh and goat sh occurred more frequently in the diet of crested terns provisioning chicks, compared to incubation (Table 1).
There was a signi cant difference in the proportion of anchovy between incubation (adult diet) and chick provisioning (Chi-square test, X = 13.72, P < 0.01), with terns showing a lower preference for anchovy when bringing prey back to the adult partner for courtship or display. Speci cally, the mean proportion of anchovy in the diet increased signi cantly from 77.0% to 92.9 % once chicks hatched (early provisioning).
During mid provisioning the proportion of anchovy was 91.9%. This was signi cantly higher than during incubation, but not signi cantly different from early provisioning (Fig. 5a). Statistical results for differences in anchovy proportions between breeding stages are summarised in Table 2.

Anchovy size
Based on measurements from 423 suitable images, the standard length (SL) of anchovies brought back to the colony ranged from 19.7 mm to 143.4 mm. Anchovies were largest during incubation when birds were bringing sh back for their partner (mean ± SD = 91.1 ± 14.9 mm, n = 193). There was a signi cant decrease in anchovy SL to a mean size of 71.8 ± 11.0 mm, n = 84 during early provisioning (Fig. 5b), suggesting a preference for smaller sized sh when provisioning small chicks. Anchovy SL increased slightly to 79.6 ± 11.9 mm, n = 146 during mid provisioning which was signi cantly smaller than incubation, but not signi cantly different from the early provisioning stage (Table 3). Table 3 Effect of breeding stage on anchovy standard length (mm) estimated from photo-samples (n = 423) of greater crested terns in 2018. Mean (SD) and Tukey test adjusted p-values from comparisons across three breeding stages, incubation (n = 193), early provisioning (n = 84), and mid provisioning (n = 146) are presented. Mid prov. 79.6 (11.9) -6.17 6.15 -1.00 --

Diet composition in relation to environmental factors
The GAMM revealed that the proportion of anchovy in the diet of greater crested terns increased with warmer SSTs, with the rate of increase slowing slightly at SSTs above 17.5°C (Fig. 6a). None of the environmental variables wind speed or wave height had a signi cant in uence on the proportion of anchovy returned to the breeding colony (Table 4). Table 4 Effect of environmental variables (SST, wave height, wind speed) on the proportion of anchovy in the diet of greater crested terns using generalised additive mixed models (GAMMs). GAMM results include breeding stage as xed effect. Shown are the estimated degrees of freedom (edf), chi square and pvalues for each environmental parameter, as well as the percent deviance explained as measure of model performance.

Discussion
Variation in the diet of marine predators can give important insights into variability in marine ecosystems (Hazen et al. 2019). However, when making inferences from predator diets, it is important to disentangle intrinsic drivers such as provisioning requirements on prey selection from extrinsic drivers such as environmental in uences on prey availability. This study quanti ed the type and size of prey exploited by a marine predator, the greater crested tern, during the 2018 breeding season on Montague Island, Australia using photography of prey items brought back to the colony. In order to understand the interplay of intrinsic and extrinsic processes determining diet composition of this colonial seabird, prey selection relating to constraints imposed by breeding stage were investigated alongside the effects of daily variability in environmental conditions.

Prey composition and size
Consistent with ndings on diet composition of crested terns from diverse geographic locations and oceanographic systems, small surface-schooling clupeids, in this case anchovies, were the most abundant prey returned to the colony during the breeding season (Walter et al. 1987;Chiardia et al. 2002;McLeay et al. 2009a;Gaglio et al. 2018). Despite the dominance of anchovy in the diet during this study, greater crested terns still varied the type of prey caught at different stages of the breeding season. This exibility and the differences in diet composition between adults and chicks are likely to re ect the need for terns to select appropriately sized prey to feed chicks during provisioning, as well as variation in the availability of different prey types within the terns' foraging range at distinct times during the study period.
Barracouta -the second most abundant prey in the diet in our study -is also an important prey species of crested terns in South Australia (McLeay et al 2009a). Barracouta form large schools that feed on krill, squids and small shes such as anchovy (Bray and Schultz 2020). Barracouta in southern NSW spawn in late winter (July-August) which may explain the occurrence of smaller (juvenile) barracouta in the diet of adult crested terns at the onset of the spring breeding season, but they may be too large for small chicks to swallow later in the season. Some demersal sh species, like bluebottle sh, are pelagic as juveniles, occurring near the sea surface where they are captured by crested terns at this life stage (Bray 2020). The southern calamari squid, in contrast, is a demersal species found in shallow inshore waters from the surface to 10 m depth (Norman 2000), making them available to seabirds foraging near the shore.
The proportion and size of anchovy returned to the breeding colony varied between incubation and chick provisioning, with chicks consuming a signi cantly higher proportion of anchovy than adults during early and mid provisioning. The high nutritional quality of clupeids like anchovy (Batchelor andRoss 1984, Pichegru et al. 2007) has been suggested to explain ontogenetic differences in diet composition between adults and chicks in greater crested terns in South Australia and the Benguela system in southwest Africa (McLeay et al. 2009b;Gaglio et al. 2018). Similar trophic segregation between adult and chick diets is found in other seabird species including great skuas (Stercorarius skua), sooty terns (Onychoprion fuscatus) and cape petrels (Daption capense) (Cherel et al. 2008;Fijn et al. 2012;Votier et al. 2003). We assume therefore that crested terns in this study selectively foraged higher trophic level (high quality) prey to provide their offspring in order to facilitate chick growth, which was different from the prey selected for an adult partner during incubation.
The differences in prey selection in this study highlight the ability of seabirds to make exible foraging decisions based on prey quality, in order to maximise the growth and survival of their offspring. Previous work has shown that diet composition had a signi cant effect on chick growth and survival in kittiwakes with higher proportions of lipid-rich sandeel positively affecting daily growth rates in chicks (Christensen-Dalsgaard et al. 2018). Furthermore, an experimental study showed that feeding nestlings a lipid-poor diet impaired chick mass and cognitive abilities which likely accounted for an increase in mortality and low recruitment (Kitaysky et al. 2006). Similarly, the "junk-food hypothesis" has been proposed as the cause of breeding failure in a colony of common guillemots (Uria aalge) in the North Sea, following a switch in chick diets from high quality sandeel to sprat which is of signi cantly lower energy density (Wanless et al. 2005). In order to determine whether anchovy served as high-quality food facilitating offspring growth and survival in terns on Montague Island, future research should investigate chick body condition in relation to the proportion of anchovy returned to the colony each day. Although poorly studied and applied, methods of quantifying body condition non-invasively have been tested in free-living birds, for instance using thermal cameras to remotely link body temperature to physiological state (Jerem et al. 2018).
Anchovies were smallest during chick rearing. Due to their relatively small gullet, hatchlings are limited in the size of prey they can swallow and this means parents need to select prey according to the chick's capability. As chicks grow, parents may need to adjust to the increasing energetic requirements. As predicted by central-place foraging theory for single-prey loaders, adults may either increase the delivery rate or provide larger prey to growing chicks. The observed increase in anchovy size with chick age supports previous ndings that there is size-selective predation by adult terns through the breeding season in response to changing energetic demands of nestlings (Hulsman et al. 1989;Shealer 1998;McLeay et al. 2009a;Gaglio et al. 2018). However, while these ndings suggest active prey selection by adult terns in response to breeding stage, the in uence of spatio-temporal availability of prey on diet composition should not be discounted.

Environmental factors
Environmental characteristics, such as SST, drive the distribution and abundance of schooling sh, and thus the availability of prey to upper trophic levels in the ocean (Bertrand et al. 2008). In this study, SST was positively correlated with the proportion of anchovy returned to the breeding colony. In an earlier study at the same site, Carroll et al (2016) concluded that reduced prey capture of forage sh by little penguins (Eudyptula minor) associated with very low SSTs might re ect a period early in the season when water temperature had not been warm enough to facilitate phytoplankton growth, resulting in a lower local abundance of planktivorous sh. Likewise, SST > 21°C impaired prey capture success at the end of the penguin's breeding season, related to a stronger in uence of the warm East Australian Current at this time (Phillips et al. 2020). Montague Island penguins have been shown to match the distribution of schooling forage sh like anchovy around the island when foraging, indicating that their prey capture success re ects local changes in prey availability that might be important for other sympatric predators such as greater crested terns (Carroll et al. 2017). Range-restricted species, such as breeding seabirds, may thus be particularly sensitive to the local availability of key prey species at crucial times in the breeding cycle (Crawford et al. 2006).
Breeding phenology in birds is likely timed to coincide with periods of peak resource availability. The prevalence of anchovy in the diet increased with SST in this study, and SST was seasonally elevated during chick provisioning stages (see online resources Fig. 1). It is possible that the timing of breeding is linked to elevated levels of production, indicated by greater proportions of energy-rich prey (e.g. anchovy) in the diet during chick provisioning. In support of the latter, anchovy abundance and availability to Peruvian seabirds increased during the period of chick provisioning with a peak around the time of edging, suggesting breeding timing to be adjusted to optimal environmental conditions facilitating chick or edgling survival (Passuni et al. 2016). Importantly, the extent to which a dietary shift in this study can truly be linked to prey availability and its in uence by environmental covariates such as SST would require additional research on estimates of sh stock composition and abundance around Montague Island by, for example, local shery assessments or acoustic surveys (e.g. Green et al. 2015;Carroll et al. 2017;Thayne et al. 2019).