Foraging habitat and site selection do not affect feeding rates in European shags

Changes in feeding rate can have nutritional effects that impact an animal's energy stores and, ultimately, its fitness (Daunt et al., 2007a; Hassrick et al., 2013; Lescroel et al., 2019). Thus, knowledge of factors affecting feeding rate is essential for understanding demographic consequences of foraging behaviour. Changes in feeding rate can be considered at different temporal and spatial scales. In central-place foragers, individuals make repeated foraging trips out from a fixed point, usually the breeding colony, and on each trip they can potentially use a number of feeding sites and habitats (Monaghan et al., 1994; Boyd, 1996; Sommerfeld et al., 2015). Thus, variations in feeding rate at sites may occur in relation to the habitats used and the order in which feeding sites are visited, and may also influence the decision to change site and/or habitat. At the foraging trip scale, variations in feeding rate may be related to the time spent feeding and travelling, and the number of sites and habitats used.

Foraging theory predicts that an individual is more likely to remain at a feeding site if its feeding rate is high, but move to another site in the same or a different habitat if the feeding rate is low (Stephens and Krebs, 1986). Tests of these predictions have been carried out under experimental conditions (Krebs and McCleery, 1984) and in semi-natural conditions (Werner and Hall, 1988). However, feeding rates at multiple temporal scales have rarely been quantified in free-living animals because of the challenges of estimating the amount of food ingested, the feeding habitats used and foraging activity. Developments of a range of animal-borne data loggers for use on avian and mammalian marine predators have enabled information on habitat utilization and prey capture in free ranging individuals to be quantified using underwater images (Bowen et al., 2002) and accelerometry (Wilson et al., 2007). In avian species that mainly use flapping flight, such as cormorants and auks, new techniques have also facilitated the collection of data on the change in wing stroke frequency before and after foraging, which provide information on changes in body mass (Sato et al., 2008), following the aerodynamic theory that birds adjust stroke frequency proportional to one-half the power of body mass (Rayner, 1987). This technique can therefore be used to quantify food ingestion not only at whole-trip scales, but also at finer temporal resolution when feeding bouts are interspersed by flights.

The European shag (Gulosus aristotelis; hereafter ‘shag’) is a foot-propelled diver that typically feeds benthically on a wide range of fish species (Wanless and Harris, 1997; Howells et al., 2017) and uses flapping flight to travel to and from its feeding areas. The foraging behaviour of birds in the population on the Isle of May off the coast of southeast Scotland has been studied intensively over several decades. During chick rearing, parents typically make several trips per day, each lasting several hours, to feeding areas mainly within ∼10 km of the colony (Wanless and Harris, 1992; Bogdanova et al., 2014). Data collected using VHF telemetry indicate that on a typical trip, a shag makes an outward flight lasting 5–10 min and dives 10–30 times at one or more feeding sites before making a return flight back to the colony lasting 3–14 min (Wanless et al., 1991a, 1998).

The Isle of May is surrounded by a patchwork of sandy and rocky habitats (Wanless et al., 1991a). Previous work has shown that shags may feed entirely in either sandy or rocky habitat during a foraging trip, or use both habitats (Watanuki et al., 2008). Prey taken by this population during the chick-rearing period are predominantly lesser sandeels (Ammodytes marinus) and bottom-living fish, mainly butterfish (Pholis gunnellus), dragonets (Callionymidae) and sculpins (Cottidae) (Wanless et al., 1991b; Howells et al., 2017). Given the marked difference in the habitat preference of these fish species, the assumption is that shags feeding in sandy sites take mainly sandeels, whereas those feeding in rocky areas take butterfish, dragonets and sculpins (Wanless et al., 1991a,b; Watanuki et al., 2008). Therefore, the Isle of May shag population provides an ideal system in which to test predictions from foraging theory and explore relationships between feeding rates, foraging habitats and travel times of a central-place forager.

In this study, we recorded diving behaviour and time spent flying and on land using changes in depth and body acceleration and angle. We used marked differences in dive behaviour to classify dive bouts by feeding habitat (sandy or rocky). We estimated the increase in body mass during individual diving bouts (‘bout food mass’) and over the whole trip (‘trip food mass’) using changes in the wing stroke frequency during the flights before and after each dive bout, and during the outbound and inbound flights at the start and end of each trip. This allowed us to estimate the bout food mass per unit time of the dive bout (‘bout feeding rate’) and the trip food mass per unit time of the trip (‘trip feeding rate’).

We tested five predictions from the optimal foraging hypothesis, with three related to bout food mass or feeding rate and two related to trip food mass or feeding rate: (1) bout durations, bout food masses and bout feeding rates would be greater in sandy compared with rocky habitats as sandeels typically occur at higher densities than other benthic species (Greenstreet et al., 2006; van der Kooij et al., 2008); (2) bout food mass and bout feeding rate would be greater in later dive bouts during a trip as shags would change site if the feeding rate at the initial site was lower than average (Stephens and Krebs, 1986); (3) if shags fed for longer, gained more mass and gained mass at a higher rate, they would make shorter flights or swim to the next feeding site and would not switch habitat; (4) trip food mass would be greater when shags made longer trips and fed for longer, and the trip feeding rate would be greater for parents with larger broods as they have higher energy requirements (Wanless et al., 1993); and (5) trip duration would be longer and trip food mass would be greater following longer periods of nest attendance in order for parents to replenish their energy reserves, as reported in species of Procellariiformes (Congdon et al., 2005).

The study was conducted on the Isle of May National Nature Reserve, south-east Scotland (56°11′N, 02°33′W), in the 2006 breeding season of European shags [Gulosus aristotelis (Linnaeus 1761)]. Twenty-eight known-aged males (sexed on the basis of size and voice) (Snow, 1960), brooding one to three medium-sized chicks were captured between 26 June and 2 July. Our sampling period was 2–4 days for each bird (Table 1) and ranged between 26 June and 5 July. Sampling was restricted to males to reduce variation in key foraging parameters associated with potential sex-specific differences (Daunt et al., 2006; Bogdanova et al., 2014; Lewis et al., 2015; Carravieri et al., 2020). Variation was further reduced as we sampled birds in the age range over which age effects in foraging parameters and breeding success are not apparent (3–17 years old) (Daunt et al., 1999, 2007b). Body mass was taken using a Pesola spring balance (accurate to 5 g), and a depth and acceleration data logger (D2GT, Little Leonardo, Tokyo, Japan; 15 mm in diameter, 53 mm in length, 18 g in mass) was attached to the back feathers with Tesa-tape. We observed behaviour after release and in all cases, birds resumed brooding within 5 mins or adopted the typical behaviour of an off-duty individual if the mate had assumed brooding duties. All individuals were recaptured at the nest 2–4 days later and the loggers retrieved. At recapture, body mass was measured using a Pesola spring balance. Four loggers malfunctioned or exhibited data conversion problems, so our final sample size was 24 males. NatureScot granted permission to work on the island under their former name of Scottish Natural Heritage (Scientific Research Licence 6676; National Nature Reserve Permit MON/RP/69).

Data loggers were set to record surge (tail–head) (Fig. 1A) and heave (dorso-ventral) accelerations at 16 Hz (Fig. 1F) and depth (at 1 m accuracy) at 1 Hz (Fig. 1B). Calibration of acceleration, estimation of logger attachment angle, the filter used to separate stroke-based acceleration from that caused by gravity and the protocol to estimate body angle and heave acceleration are described in Watanuki et al. (2005). To quantify flight duration and wing stroke cycle (1/frequency), we applied continuous wavelet transformation to take into account the non-stationary oscillation of the heave acceleration (Fig. 1C,F,G) using Ethographer (Sakamoto et al., 2009). We used wavelet analyses and calculated the dominant wing stroke cycle for the window of around seven waves (strokes) that corresponded to 1 s.

To minimize the possibility of including high stroke frequency recorded during the short period of take-off, we set the shortest wing stroke cycle as 0.16 s, corresponding to the fastest wing stroke frequency of shags during cruising flight (6.250 Hz; Sato et al., 2008). To minimize the possibility of missing slow-flapping flight but excluding very slow wing stroke during landing, we set the longest wing stroke cycle as 0.20 s, i.e. a little longer than the value (0.19 s) corresponding to the lowest wing stroke frequency during cruising flight (5.219 Hz; Sato et al., 2008). As the difference in the stroke cycles between those calculated using the non-dimensional frequency of Morlet wavelet function (ω0) 10 and those using ω0 30 was small (≤0.5%), we used ω0 10 to optimize the accuracy and resolution of time.

To collect data over a period long enough to characterize the trip and habitat of individual birds (∼3 days), we set the sampling frequency for acceleration as 16 Hz, which was lower than the sampling frequency in the previous study (64 Hz; Sato et al., 2008). To test the reliability of the stroke cycle in our study, we re-sampled the heave acceleration data at 16 Hz in five fly bouts of a shag collected at 64 Hz by Sato et al. (2008) in order to estimate the dominant cycles under two sampling regimes. We found that the difference was small (≤0.3%), which gave us confidence that 16 Hz sampling delivered a reliable stroke cycle.

Shags sometimes made a high-frequency foot stroke (ca. 0.2 s cycles) at the start of a dive, which was similar to the longest cycle of wing stroke during flight. As the duration of this high-frequency foot stroke was short (≤10 s), we defined flight (Fig. 1D) as a continued high-frequency stroke (≤0.2 s cycle, by the definition above) longer than 10 s and when the birds were not diving (Fig. 1B,C). We then calculated the average cycle for each flight.

Only dives deeper than 1 m were analysed because of the accuracy of the depth sensor (±1 m). Image data previously collected by bird-borne camera/depth loggers attached to male Isle of May shags indicated that dives shallower than 5 m were largely associated with washing and/or surface swimming, whereas those taken during deeper dives showed birds actively foraging near the sea floor (Watanuki et al., 2008). However, elsewhere shags will feed in the water column by making short (<27 s) pelagic dives (Grémillet et al., 1998) (equivalent to ∼5 m based on depth–duration relationships from our data) and this behaviour has occasionally been recorded on the Isle of May (Wanless et al., 1998; Carravieri et al., 2020). As such, we classed dives to ≤5 m, which constituted 13.9% of all those recorded (n=10,623), as shallow dives and retained them in the analysis.

Visual inspection of dive profiles showed that shags predominantly made U-shaped dives (Fig. 1B). Thus, different phases of a dive were readily defined from the absolute rate of change in depth (descent, >0.6 m s⁻¹; bottom, <0.3 m s⁻¹; ascent, ≥1 m s⁻¹; Watanuki et al., 2005), enabling descent duration, bottom duration, ascent duration and post-dive surface duration to be estimated using the Macro Program of Igor Pro v4 (WaveMetrics). Dive duration was the sum of descent, bottom and ascent durations. As bottom depth showed very little variation within a dive, the maximum depth and mean depth recorded during the bottom phase were very similar, and maximum depth was therefore defined as dive depth.

Breath-holding divers such as seabirds typically make a series of dives with short surface times, followed by an extended period on the surface or in flight. Conventionally, these series of dives are referred to as ‘dive bouts’ and animals are assumed to feed in a localized site or at a single food patch during each dive bout (Feldkamp et al., 1989; Boyd, 1996). Shag diving behaviour followed this pattern such that dives were grouped into distinct dive bouts during a foraging trip (denoted as ‘dive bout’ in Fig. 2A,B). Dive bouts were determined using bout-ending criteria as the inflection point of the log-survivorship curves, assuming that the surface time is under two random processes (movement between and within feeding sites) (Gentry and Kooyman, 1986). Visual inspection of the log-survivorship curve indicated a change in slope between 200 and 300 s (Fig. S1). Accordingly, we assumed that dives separated by more than 250 s constituted different bouts, a broadly similar value to that found previously for this species by Grémillet et al. (1998). In total, 522 dive bouts were identified. We assumed that dive bouts occurred within a feeding site, as surface times between dives within a dive bout were mostly <1 min (Fig. S1). Birds mainly undertook a flight before commencing a new dive bout (76%; ‘inter-bout flight’), but they sometimes remained on the sea after completing a bout before starting a new dive bout (24%). In the latter cases, we assumed that birds changed feeding sites by active swimming or remained in the same location in order to seek other feeding opportunities. On three occasions, shags switched habitats without undertaking a flight (‘dive bout20sandy’ to ‘dive bout21rocky’ in Fig. 2B). In 9% of dive bouts, birds made one to four flights within a dive bout (‘within-bout flight’, e.g. ‘fly bout30’ during ‘dive bout17rocky’ in Fig. 2A). The duration of each within-bout flight (0.97±0.74 min, mean±s.d., n=57) was shorter than the duration of each inter-bout flight (3.62±3.48 min, n=247) (U-test, P<0.001), so we assumed that these within-bout flights enabled birds to re-locate themselves within a feeding site when they had drifted with the water current. Among 522 dive bouts, 119 dive bouts comprised ≤2 shallow dives (≤5 m, ‘dive bout16.1’, Fig. 2A) and 23 dive bouts comprised ≤2 deep dives (≥5 m, ‘dive bout16unknown’, Fig. 2A). These dive bouts were excluded from further analyses as we considered that they might be primarily concerned with washing, preening or exploratory behaviour.

The bout duration, number of dives, median depth of dives, duration of dives and bout order within the trip were calculated for each dive bout. The duration and wing stroke cycles during flights and switch of habitats (Fig. 2B) between dive bouts were also recorded. Shags often made multiple flights between successive dive bouts (one to six inter-bout flights), between departure from the colony and start of the first dive bout (one to six outbound flights), and between the last dive bout and arrival back at the colony (one to five inbound flights). Duration of inter-bout, outbound and inbound flights was the sum of multiple flights when more than one occurred.

To identify the period during which shags were on land, we first excluded periods with dives and then used body angle data to distinguish time on land from time swimming on the surface and in flight. Shags on the sea surface or flying keep the longitudinal axis of the body almost horizontal. Time on land was therefore defined as periods of more than 60 s when the body angle was greater than 45 deg (standing) (Figs 1A,E, 2A,B). When shags land either from the air or from the water and approach or leave the nest, they stand upright and walk (Fig. 1E). However, adults brooding chicks do sometimes have a body angle <45 deg (Fig. 2A,B). Therefore, shags were defined as being ‘on land’ during long dive-bout intervals including standing but without flight. The frequency distribution of durations of on-land time showed three peaks with gaps around 80 and 400 min (Fig. S2). Long periods on land (>1 h) typically included periods during which shags had a body angle <45 deg, consistent with them being at the nest brooding their chicks. Accordingly, periods on land >1 h were assumed to reflect time at the nest, with the start of a trip defined as the bird standing up (body angle >45 deg) and the end of the trip by the bird adopting a body angle of >45 deg. As the birds sometimes spent time on sea rocks when they came back to the island (Wanless et al., 1993), we also used flight after a period of standing at some other location on the island to define the end of a trip (‘fly bout40’, Fig. 2B).

Criteria to identify bottom habitat from diving behaviour and body acceleration were developed using underwater image data from camera/depth loggers deployed on seven male shags over the same period for which the accelerometry data were collected [see Watanuki et al. (2008) for full details on field protocols and data processing]. Using data from 36 dive bouts, we carried out a discriminant analysis of dive bout characteristics to separate habitats into sandy (seabed composed of fine to coarse sand, sometimes with pebbles) and rocky [either bare rock or rock covered with kelp (Laminaria spp.) or soft corals]. Shags used either rocky or sandy habitats and did not switch habitat within a dive bout. Shags foraged mainly at two depths (modes of 24 and 32 m) in sandy habitats, but the depths were more variable in rocky habitats (5–40 m). Furthermore, the proportion of dive bouts with a small coefficient of variation of dive depth (CV<10) was greater in sandy habitats (16/16 bouts) than in rocky habitats (8/20), indicating that shags changed dive depth more in rocky habitats. During the bottom phase, shags kept the angle of the body vertical in sandy habitats but horizontal in rocky ones.

Therefore, to identify the bottom habitat, the dive pattern, mean bottom depth (D), coefficient of variation in the bottom depth (CV), percentage of dives in which the body was horizontal during the bottom phase (pHO), percentage of dives in which the bird's head was angled downwards during the bottom phase (pDO), and trend in depth change (DT) within a bout were calculated. Body angle during the bottom phase was classed as horizontal or vertical based on the proportion of the image occupied by the seabed during the bottom phase of the dive (Watanuki et al., 2008). The trend in depth change was categorized using CV of bottom depth and the regression coefficient (b, P≤0.05) of depth on dive order within a dive bout as decreasing (b≤0, score 2), stable (CV≤10, score 1), increasing (b≥0, score 3) or variable (CV≥10 and b was not significant, score 4). The discriminant function was as follows: score=0.179D+0.733CV+1.159pHO+0.377pDO+0.064DT. Using this discriminant function, the analysis assigned habitat correctly in 33 out of 36 bouts (92%).

For each dive bout for the 24 male shags in the present study, we calculated the discriminant score and assigned these to rocky or sandy habitats. Postures in the bottom phase of each dive were categorized using longitudinal acceleration as ‘vertical’ when the mode of body angle was steeper than −30 deg, or ‘horizontal’ when the mode of body angle was shallower than −30 deg. We chose −30 deg as a threshold because the distribution of the mode of body angle in each dive showed a weak gap around −30 deg (Fig. S3). Using mean bottom depth (D), CV of bottom depth, percentage of the number of dives with horizontal (pHO) or vertical (pDO) posture during the bottom phase and trends in the change of bottom depth during the dive bout (DT), each dive bout was categorized as being in sandy or rocky habitats using this discriminant function. Habitat was identified for 379 of 380 dive bouts with ≥3 deep (≥5 m) dives. In a single dive bout with 12 deep dives, the trend of bottom depth was not determined; hence, habitat was not estimated.

The habitat use of the seven male shags fitted with back-mounted cameras (20 dive bouts over rocky habitat and 16 dive bouts over sandy habitat; Watanuki et al., 2008) was broadly comparable with that of the 24 males in the current study (146 dive bouts over rocky habitat and 102 dive bouts over sandy habitat, see Table 2). The ranges of mean dive depth of dive bouts over rocky (10.9–39.1 m) and sandy habitat (20.6–34.1 m) of the seven males with cameras were comparable with the median depth of dives of our 24 males (24 m over rocky habitat and 26 m over sandy habitat; see Table 2). We were therefore confident that the seven males with cameras could be used as reliable proxies for our study.

We recaptured males to retrieve the loggers when they were at the nest brooding their chicks. We could not, therefore, be confident that they had not already fed the brood, so we could not estimate trip food mass directly from changes in body mass to compare with values estimated from changes in wing stroke cycle.

The wing stroke cycle was calculated for flights longer than 1 min in a previous study of European shags (Sato et al., 2008). Here, we calculated the cycle within a range of 0.16–0.20 s, as in the ‘Fieldwork’ section, for flights longer than 10 s to include as many short inter-bout flights as possible. However, the wing stroke cycle still showed increasing and decreasing trends during the short periods at the start and end, respectively, of some flights (Figs 1C,G, 2). Durations of these take-off and landing periods varied between flights and were difficult to define by acceleration measured at 16 Hz. Then, we calculated the average cycle for each flight and examined the variation of the average cycle across the duration of fly bouts. Variation in the average wing stroke cycle was greater for short flights but there was no trend in variation with flight duration (Fig. 3A). Shorter flights might include proportionally more time for take-off and landing, which may cause potential. However, these should not result in any directional bias in estimates of the duration of flight or in body mass change estimated from wing stroke cycles. To decrease the potential for such errors for inter-bout, outbound and inbound flights during which shags made multiple flights on trips, the average wing stroke cycle weighted by the duration of each flight was used (Fig. 3B). Furthermore, we excluded outlier values (4% of fly bouts as shown in Fig. 3B) where the wing stroke cycle was less or greater than the mean±2 s.d.

Our aim was to estimate the increase in body mass during dive bouts and foraging trips using the estimation of the proportional mass (M₂/M₁) based on the proportional wing stroke frequencies (F₂/F₁). We used the mass at capture for each individual as M₁, i.e. the ‘standard’ mass, and estimated M₂, and, therefore, the increase in body mass during the dive bouts and trips as M₂−M₁, where M₂ equals M₁(F₂/F₁)². Variation in body mass at the start of the dive bout depends on the increase in body mass during the previous dive bout and other factors, and should be small relative to M₁. Similarly, variation in body mass at the start of a trip, which depends on the previous time spent in the colony and other factors, should also be small relative to M₁. We assumed that these variations did not bias the estimate of the increase of the body mass seriously.

To estimate the increase in body mass, assumptions of the constancy of wing stroke frequency with variable wind speed and the constancy of amplitude of wing stroke with variable loads are crucial. A study using GPS tracking data for this population showed that shags flew at a relatively constant air speed (14.7 m s⁻¹ on average), with little change in wing stroke frequency when the wind speeds were between −12 m s⁻¹ (head wind) and +12 m s⁻¹ (tail wind) (Kogure et al., 2016). An experiment in cockatiels (Nyphicus hollandicus) showed that the wing stroke amplitude is not affected by loads (Hambly et al., 2004). Thus, we were confident that this technique was applicable during the relatively calm conditions in the present study.

In great cormorants (Phalacrocorax carbo) that have a wettable plumage (Grémillet et al., 2005), the increase of body mass during diving might not always represent food intake. Using the wing stroke frequency, however, Sato et al. (2008) estimated the body mass change in shags rearing chicks at the Isle of May during each trip and found that the mass change ranged between −30 and 260 g, and the estimated body mass during the final flight back to the colony was comparable with the value directly measured on recapture. These masses were close to the masses of meals for chicks estimated by the water-offloading technique (8−208 g, average 106 g; Wanless et al., 1993). The increases in body mass estimated for 155 sample trips (122±76 g, mean±s.d.) in this study were similar to the previous estimates. Thus, we consider that increases in body mass owing to the plumage becoming wet did not substantially affect our estimation of body mass change.

We estimated the increase in body mass during a dive bout using stroke frequency before and after the dive bout (defined as bout food mass). Similarly, we estimated the increase in body mass during a foraging trip using stroke frequency during the outbound and inbound flights (defined as trip food mass).

Habitats could not be identified for >40% of the sum of dive bout durations in eight trips and these uncategorized trips were excluded from the analyses (Table 1). To examine the effects of habitat and dive bout duration on the bout food mass, 248 dive bouts for which both habitat and body mass change were determined were used. To test the effects of bout duration and body mass change on subsequent movement and habitat switching, dive bouts followed by land bouts were excluded. Hence, sample sizes varied between tests (Table S2). Shags sometimes did not fly to the first dive bout (‘dive bout16unknown’ of Trip14, Fig. 2A) and/or fly after the last dive bout (‘dive bout18rocky’ of Trip14, Fig. 2A; 16% and 10% of trips, respectively). Short periods on land (≤1 h) were observed shortly after leaving the nest, before dive bouts for which habitats were identified (seven trips) and after the last dive bouts for which habitats were identified before returning to the nest (34 trips). Similar behaviour in the Isle of May shag population was described in Wanless et al. (1993) using visual observations and VHF telemetry. Short periods on land were also observed between dive bouts in 29 trips (e.g. between ‘dive bout14rocky’ and ‘dive bout15rocky’ of Trip13, Fig. 2A). We could not exclude the possibility that birds were at the nest during these periods on land between dive bouts. To examine the effects of trip duration, brood size and activity budget on body mass change and feeding rates, 155 trips, excluding the above trips, were used.

To test effects on the frequency (number of flight or swim/rest events after a dive bout) and the values (durations of dive bout and bout mass, cumulative duration of inter-bout flights, and cumulative duration of outbound and inbound flights), we used non-parametric tests (χ², Wilcoxon's signed rank test, Mann–Whitney U-tests) because of small sample size. For these non-parametric tests, we used SPSS v28 (www.stats-guild.com). For generalized linear mixed models, for which we needed to account for repeated measures in individuals, we used library lme4 in R v3.2.1, (https://www.R-project.org/) and used glmer. We fitted all possible linear mixed models capturing all combinations of explanatory variables, with no interaction terms, and performed model selection based on Akaike's information criterion (AIC) (Burnham and Anderson, 2002) using the library MuMIn. Where there was a single adequate model, it was denoted as the best model and its parameter estimates and significance levels were calculated. When multiple adequate models were apparent (ΔAIC≤2.00), these were treated as equally supported models and parameter estimates and significance levels were given by full averaging. Values are shown as means±s.d. in the text and tables, unless otherwise indicated.

Shags made 146 (59%) dive bouts over rocky habitats and 102 (41%) dive bouts over sandy habitats. Dive bouts over sandy habitat were significantly longer than those over rocky habitat, but median dive depth, wing stroke frequency in the flights before and after bouts, bout food mass and bout feeding rate did not differ significantly between habitats (Table 2). The best model explaining bout food mass included bout duration, bout order and habitat (Table S1A). In this analysis, we did not include the number of dive bouts per trip as an explanatory variable as this was correlated with bout order (r=−0.739). Effects of bout duration and order were significant; the bout food mass was heavier for longer dive bouts (Fig. 4A) and greater in earlier dive bouts (Fig. 4B), but was independent of habitat (Table S1A). Thus, we found no support for prediction 1 (see Introduction) that bout food mass would be greater in sandy habitat nor for prediction 2 that it would be greater in later dive bouts. The latter was because bout food mass was greater in trips with a single dive bout (131.2±8.7 g, n=82) compared with bout food masses in trips with two dive bouts (58.3±12.5 g, n=40), three dive bouts (41.3±26.3 g, n=9) or more than four dive bouts (33.7±27.9 g, n=8). When trips with a single bout were excluded, the effect of bout order on bout food mass was not significant (Table S1C).

Three equally supported models explaining bout feeding rate were the null model and two models including order or habitat, but the effect was not significant (Table S1B). Thus, we found no support for prediction 1 that bout feeding rate would be higher in sandy habitats, nor for prediction 2 that bout feeding rate would be higher in later dive bouts. When trips with a single bout were excluded, the result was qualitatively the same (Table S1D).

Shags were more likely to fly (66% of 218 dive bouts with bout food mass) than swim/rest (34%) between dive bouts (Table S2A). The best model explaining the mode of movement (flight or swim/rest) included the duration of the previous dive bouts, and the effect was significant; birds tended to fly after shorter dive bouts and swim/rest after longer ones (Table S2A). The mean cumulative duration of 143 inter-bout flights was 5.1 min (range 0.5–22.2 min). Cumulative duration of inter-bout flights was not related to the duration, bout food mass or bout feeding rate of the previous dive bout (Table S2B–D). Thus, there was only partial support for prediction 3, with an effect of bout duration on subsequent mode of movement but not on cumulative duration of flight.

Two equally supported models that explained the switch of habitat (switch or no switch) were the null model and the model including duration of the previous dive bouts, but the effect was not significant (Table S2E). The bout food mass and bout feeding rate showed no clear effects on the tendency to switch habitat (Table S2F,G). Thus, there was no support for prediction 3 that dive bout duration, bout mass gain and mass rate of gain affected the likelihood of habitat switching. Birds were more likely to fly if they switched habitat (88%, 21/24 cases) compared with whether they used the same habitat (61%, 39/64 cases) (χ²=5.676, d.f.=1, P<0.05). However, the cumulative duration of inter-bout flights did not differ when shags switched (4.6±2.6 min, n=22) or used the same habitat (3.8±2.7 min, n=38, U-test, P=0.197).

In the 155 trips for which trip food mass was estimated, the mean cumulative duration of outbound flight to the first dive bout was 4.4 min (Table 3). On 80 trips, there was only a single dive bout, and on the remaining 75 trips, birds made two to five dive bouts (1.7 dive bouts on average). In this analysis, two trips that included six and seven dive bouts (Fig. 4B) were excluded as these did not give trip food mass. The mean cumulative duration of the inbound flight from the last dive bout back to the colony was 7.4 min (Table 3). Cumulative duration of inbound flights was slightly longer than that of outbound flights for trips with a single dive bout (0.7±1.6 min difference, n=80, Wilcoxon's signed rank test, P<0.001), and markedly longer for those with two to five dive bouts (5.5±4.5 min difference, n=75, Wilcoxon's signed rank test, P<0.01; U-test, P<0.001) (Bonferroni test, Fig. 5). No difference was found in the cumulative durations of inbound flight among trips with two, three, four and five dive bouts.

Shags used only rocky habitats on 77 trips (50%), only sandy habitats on 61 trips (39%) and both habitats on 17 trips (11%). Trips with multiple dive bouts were longer and were more likely to include both habitats (Table S3A,B), indicating that birds appeared to use rocky and sandy habitats probabilistically.

Four equally supported models explaining the trip food mass included brood size, trip duration, cumulative time of dive bout and cumulative time of inbound flight from the last dive bout (Table S1E). Based on model averaging, effects of brood size and trip duration were significant; trip food mass was greater for larger broods and after longer trips (Fig. 6A,B). The best model explaining trip feeding rate included all factors (Table S1F). However, based on model averaging, effects of brood size were marginally significant (P=0.0426) and those of other factors were not significant (Table S1F). Thus, there was support for prediction 4 in that trip food mass was greater when shags made longer trips and fed for longer, and trip feeding rate was marginally greater for individuals with larger broods.

The duration of previous nest attendance was measured for 143 trips but was not related to the duration of the trip (r²=0.012, n.s.), the bout food mass of the first dive bout (r²=0.007, n.s.) or the trip food mass (r²=0.001, n.s.) of the subsequent trip. Thus, we found no support for prediction 5.

For 125 trips, bout food mass was estimated for all dive bouts. For the 78 trips in which birds made only a single dive bout, trip food mass was the same, by definition, as bout food mass. For the other 47 trips in which birds made two to five dive bouts, the difference between the sum of the bout food mass and the trip food mass was minor (0.9±5.7 g, ranging between −9.2 and 27.4 g).

Although we are unable to discount the possibility that the birds may have been negatively affected by the devices, we consider that the fine-scale measurement of food intake using small, dorsally attached accelerometers causes less impact compared with other techniques, such as temperature recorders in the stomach (Wilson et al., 1995), head-mounted recorders with magnets on the beak (Takahashi et al., 2004) and large dorsally attached video or still-picture camera recorders (Watanuki et al., 2008). Internal tags are likely to cause less impact after deployment (White et al., 2013; Forin-Wiart et al., 2019) but require surgery, so the comparison with our approach is not straightforward. A major advantage of the technique used in this study is that it estimates simultaneous foraging behaviour, thus providing fine-resolution information on foraging success, flight and diving behaviour of free ranging marine predators. Using this technique with European shags, we found that although feeding rates at both the bout and trip scale were highly variable, they were not related to the foraging habitat or the tendency to change feeding site.

Our results confirm that dive bout duration, i.e. time spent feeding at a site, positively affected bout food mass, indicating that shags caught more prey when they dived for longer (Sato et al., 2008). Shags feeding in sandy habitats probe the seabed with their bills to drive out sandeels buried in the sand (Greenstreet et al., 2006; van der Kooij et al., 2008). In contrast, when they feed in rocky habitats, shags swim horizontally over the bottom searching for demersal fish among the rocks (Watanuki et al., 2008). Thus, we expected that bout feeding rate would be greater in sandy habitats. However, our results did not support prediction 1 and bout feeding rate did not differ significantly between habitats. The diet of European shags varies across the breeding range and over time (Cosolo et al., 2011; Hillersøy and Lorentsen, 2012; Howells et al., 2017), suggesting that shags adopt a flexible foraging strategy to exploit various prey types that are available. Although individual birds might differ in their use of rocky and sandy habitats, the sampling period for each bird (2–4 days) was short relative to the chick-rearing period (ca. 55 days; Daunt et al., 2007b), limiting our ability to evaluate the level of individual specialization in this study population. The extent to which the higher energy density of sandeels aged ≥1 year – the main age classes of sandeel that Isle of May shags prey on (4.8–6.5 kJ g⁻¹ wet), compared with demersal fish including flatfish, butterfish, sculpin and blenny (3.3–5.0 kJ g⁻¹ wet) (Harris and Hislop, 1978; Garthe et al., 1996; Anthony and Roby, 1996; Anthony et al., 2000; Takahashi et al., 2001; D.A.D. Grant, unpublished data) – relates to energy-based feeding rate and individual specialization will require further work. It is also possible that shags are more likely to feed with conspecifics when feeding on sandeels in sandy habitats (Watanuki et al., 2008). Thus, intraspecific competition may reduce feeding rates in this habitat.

In contrast to prediction 2 that bout food mass and bout feeding rate would be greater in later dive bouts during a trip, we found that in trips during which shags made more than one feeding bout, there was no support for a progressive increase in either bout food mass or bout feeding rate (Fig. 4B). We did not have independent fine-scale information on prey abundance in the study area, but the data from our shags suggest that abundance might be temporally and/or spatially variable. Elsewhere cormorant species have been shown to deplete fish around the breeding colony (Birt et al., 1987). In our study the inbound flight was significantly longer than the outbound flight on trips with multiple dive bouts. This pattern indicates that shags moved progressively further away from the colony on successive dive bouts, as would be predicted by the Storer–Ashmole ‘halo’ hypothesis (Storer, 1952; Ashmole, 1963). Alternatively, individuals might remember sites with high density and/or abundance of prey near the colony and visit these first. In accordance with this, birds fed more during dive bouts in trips with a single dive bout than those with multiple dive bouts. Little penguins (Eudyptula minor) made trips repeatedly to the same sites when feeding success was high (Carroll et al., 2018). However, we did not have location data to allow us to test this possibility. The pattern we observed might also be explained by shags visiting sites probabilistically and spending more time at the first feeding site in order to fulfil their own energy requirements, before starting to catch food to bring back to the brood. However, the lack of relationship between the duration of the previous period in the colony and the bout food mass of the first dive bout of the subsequent trip (prediction 5) does not support this explanation.

Prediction 3 was that if shags fed for longer, gained more mass and gained mass at a higher rate, they would make shorter flights or swim to the next feeding site, and would not switch habitat. We found partial support for this prediction in that shags tended to fly rather than swim or rest after shorter dive bouts, indicating that they moved further, but flight duration after shorter dive bouts did not differ significantly compared with longer bouts. Bout feeding rate also did not affect the flight time to the next dive bout. Although shags tended to fly when they switched habitat, they did not show a propensity to switch habitat after shorter dive bouts when they gained less mass. Bout feeding rate also did not affect the switch in habitat. Thus, our findings indicated that shags did not actively move further and/or switch habitats after they fed at sites of lower quality where density and/or abundance of prey was lower. This provides further support for shags visiting sites probabilistically. Other factors potentially include unpredictability of prey at a fine temporal scale. Shags do not have discrete feeding territories; instead, they share their foraging environment with conspecifics and other avian, mammalian and fish predators. Particularly in sandy habitats, shags feed communally (Watanuki et al., 2008), and conspecific interference competition and/or facilitation may affect prey availability. Thus, it could be that optimal foraging rules break down under these conditions or, at least, are much weaker. However, collecting data on predator interactions under field conditions to test this assertion would be very challenging.

In accordance with prediction 4, we found a positive effect of brood size and trip duration on trip food mass (Fig. 6). Wanless et al. (1993) found that the mass of stomach contents at the end of a trip was positively correlated with the total mass of chicks of each brood. These results indicate that adults adjust trip food mass according to the food requirements of their chicks. However, the effect of brood size on trip feeding rate was marginal (Table S1F). This indicates that individuals with larger broods increase the food brought back to the colony mainly by increasing the trip duration rather than increasing feeding rate.

The duration of the previous nest attendance did not relate to the subsequent trip duration and trip food mass, so prediction 5 was not supported. A previous study on the Isle of May noted that shags often spent time on the sea rocks at the end of a trip and speculated that they were digesting prey for their own requirements before returning to the nest to feed the brood (Wanless et al., 1993). Interestingly, we found that in trips with multiple dive bouts, the sum of bout food masses was almost identical to the trip food mass. This indicates that chick-rearing shags do not digest food during inter-dive bout time during trips. Rather, they retain all the food in the stomach until they return to the colony and then either allocate it for their own energy requirements or use it to provision the brood. In pygoscelid penguins, parents possibly feed and digest food for themselves in the early part of trips and store food in their stomach for chicks thereafter by regulating pH in the stomach (Peters, 1997). However, foraging trips of penguins are substantially longer than those of shags (∼24 h compared with ∼2 h), which may enable shags to retain all the food caught during a trip without the need for this regulatory process. Species of Procellariiformes that typically feed much further away from the colony than European shags regulate the allocation of food for chicks and for themselves by alternating long and short trips (Weimerskirch et al., 1994), wherein long trips are used to feed and digest food for replenishing their energy reserves following long nest attendance (Congdon et al., 2005). Our results therefore indicate that inshore feeding and offshore feeding species may adopt different energy allocation strategies.

In conclusion, male European shags on the Isle of May did not follow classical optimal foraging rules and specifically did not regulate foraging according to the feeding rate in the previous feeding site. Thus, it appears that they feed probabilistically in habitats of varying capture rate affected by prey density and conspecific competition or facilitation. This strategy, wherein average rates of feeding were similar in the two main habitats used, enabled shags to largely achieve their primary objective to sustain energy requirements for themselves and their brood.

We are grateful to Mike Harris for his advice at all stages of the project and two anonymous reviewers for their invaluable comments. The National Institute of Polar Research provided data loggers. We thank NatureScot for access to the Isle of May National Nature Reserve.