A marine predator relies on both social cues and frequently updated memory to search for prey

Abstract

Finding scattered resources is a challenge for marine predators, their search strategies shaping in turn their response to global changes. Three types of search strategies are generally considered: random opportunistic search, reliance on conspicuous social cues or long-term individual fidelity to predictably productive areas. More complex strategies, namely intermediate, composite and/or those varying across time and environmental conditions are likely to prevail but are rarely explored. We investigated memory strategies in Cape gannets, a marine predator heavily relying on social cues, and closely related to the northern gannet which shows long-term individual foraging fidelity. Repeat GPS tracks within seasons revealed that two different individuals leaving the colony more than 3 days apart showed random chances to forage in a similar direction, whereas within-individuals, birds persisted in a same direction for longer timescales, reaching random levels only after 10 days, or 9 consecutive trips. This strategy of transient individual preferences within a breeding season was observed in most individuals every year. These results suggest a consistent and complex search strategy mixing social cues and private information updated every few days. Our simple approach can readily be applied to other central place foragers to better understand the evolutionary ecology of search strategies.

1. Introduction

Finding resources that are scattered beyond the direct perception range is a key constraint in the ecology and evolution of many wide-ranging animals [1,2]. Until they directly perceive resources of interest, animals should follow some rules and/or cognitively process information about where their resources are likely to be found and how they should move to find them [3,4]. These proximal mechanisms of search are at the core of understanding ecological traps that can result from rapid changes in prey resource distribution [5]. They could also determine how likely predators are to switch to conspicuous, often predictable anthropogenic resources causing important conservation issues [6,7]. However, understanding the rules and cognitive strategies used by individuals to search for prey in the wild is scientifically challenging, especially when data about the resources and the behaviour of consumers are often limited [8–10]. As a result, scientists often rely on optimal foraging expectations to assume plausible variations in search strategies, either across species or within populations (i.e. among individuals or through time and changing environmental conditions).

Current optimal behaviour theory predicts that the search strategy used would be primarily determined by prey dynamics and predictability: from opportunistic search for unpredictable prey to long-term memory or even inherited preferences for stable and predictably productive areas [2,11–13]. In between, animals could in theory rely on shorter-term memory (i.e. updated more frequently) [11,14]. Individuals could also rely on social cues and thus take advantage of what other individuals have recently found, especially in very dynamic environments where even short-term memory could rapidly get outdated [15,16]. Relying on social cues to get more recent information than through memory could, however, trade-off with increased costs of competition [15,17]. Considering these costs and benefits, a common simplifying tendency is to assign organisms into one of these strategies: if competition for resources is not a constraining factor, and social cues are available and reliable, then there is limited need to rely on memory (and especially long-term memory). On the other hand, if competition for resources is important, either the environment is predictable and individual memory could help relieve competition [17], or the environment is unpredictable and neither memory nor social information should be used [18]. Unfortunately, data on prey dynamics, especially in the marine environment, are challenging to obtain. This makes it difficult to directly test the accuracy of these predictions on search strategy variations across individuals, species or environments [12]. Alternative expectations could include potential trade-offs with other species-specific traits (e.g. locomotion energetics, sensory capacities), some degree of ‘phylogenetic inertia’ (i.e. strategies not necessarily optimal in the local environment) depending on how long it takes to adapt search strategies to changes in environmental conditions (e.g. interannual variability in the food competition context), or some degree of within- or between-individual variation in search strategies within populations (e.g. associated with traits correlating with competition and dominance effects [19,20], like age, sex or size).

The issue of variations in search strategies is probably best exemplified with wide-ranging marine predators such as seabirds [12]. During the breeding season, individual seabirds make repeated ‘central-place’ trips at sea from their colony searching for food [21]. The long lifespan and slow reproductive rate of seabirds are considered an extreme life-history adaptation to the challenge of finding limited resources at sea [22,23]. Yet the information used by marine predators to find resources beyond their direct perception range has remained elusive for a long time [12]. Progress in bio-logging techniques and advanced oceanographic knowledge have only partially filled the gap [24,25]. We will here consider ‘large-scale’ search strategies (i.e. at the scale of the entire foraging trip, rather than area-restricted-search within a trip [26]). At this large scale, some seabirds are still believed to mainly search opportunistically by minimizing travel costs, rather than using information to anticipate prey locations [12,18]. We know, however, from studies on individual spatial specialization that many seabirds can rely on memory and/or inherited individual preferences to join places and oceanographic features generally associated with prey [27]. In some species, individual specialization can last for years [13,28], while in others, they only last for the duration of a breeding season (e.g. [29]). For most of these species, however, the persistence duration is unknown (see also [30]). We also know from a few detailed studies that some seabirds can use different types of social cues to join distant prey patches [31], and it is suspected that many species could do so [31]. Our understanding of search strategies is, however, still too limited to allow predicting what information different seabirds use when foraging over large scales. Nonetheless, the main recurrent expectation is that tropical oceanic species do not rely much on memory while temperate to polar and/or neritic species do so, potentially over multiple years [12,24,32,33]. Specifically, we empirically know little about variations in the frequency at which different seabirds would update their memory, and it is unclear whether the seabirds that appear to strongly rely on social cues also make use of memory to orient their trips on a large scale [34,35].

There are different ways social information could be used, at different scales, by seabirds searching for food (recently reviewed in [31]), but formally documenting them is a general challenge [31]. While so-called ‘local enhancement’ (detecting and joining aggregations of feeding predators) is probably widespread, it is thought to mainly act over a few kilometres [36], which is short compared with the entire range of most seabird foraging trips. The mechanisms potentially acting over larger ranges include suspected ‘network foraging’ (cascading reactions to conspecifics changing direction and converging to a distant place [37–40]) and ‘information centre’ mechanisms at the colony (e.g. observing the direction of returning birds and/or the direction of the rafts they form close to the colony where conspecific density is high [9,41]). These larger-scale mechanisms are, however, poorly documented and have been studied mainly in tropical seabirds. One notable exception, however, is the Cape gannet (Morus capensis), a seabird species found in southern Africa and foraging in neritic temperate waters, a type of environment generally considered as relatively structured and predictable [24,33]. Among seabirds, the use of social information is among the most well-documented for this species, with evidence that they heavily rely on local enhancement but also on the direction of returning conspecifics during their departure from the colony and at least in the first few kilometres at sea [36,38,42]. It is also known that Cape gannets can benefit from group rather than solo foraging, to enhance prey capture success [43] or flight energetics [44], at least partially balancing the costs of competition. Under these conditions, it is unclear whether Cape gannets should rely on any form of memory to guide their large-scale search. Individual consistency and memory, however, have never been investigated in this species.

There are many studies of ‘individual foraging site fidelity’ in seabirds and other marine predators, and these are essentially measures of individual specialization in space use (reviewed in [27,32,45]), i.e. a lower variability within than between individuals [46]. For most of these study species, this is the best available evidence to assume the level of reliance on memory and on which time scales. However, individual foraging site fidelity estimates are not strictly equivalent to memory use: consistent differences between individuals could be due to fixed, inherited heterogeneity between individuals rather than memory of past individual experiences, and within-individual consistency could be due to stable external cues rather than memory. Furthermore, individual specialization estimates have some limitations for comparative approaches (including for comparing individual specialization over different time scales) [47,48]. We will thus here adopt a slightly different approach [30,49], independently describing how within-individual consistency on the one hand, and between-individual similarity on the other hand, each change with increasing timescales. Seabirds are colonial foragers so different individuals should be exposed to very similar external cues if they depart ‘simultaneously’ (i.e. not long enough apart for key external cues to change). Two individuals leaving simultaneously will thus be likely to share similar foraging areas, and the time scale at which it remains true will inform on how quickly external (including social) cues change around the colony. Reliance on frequently updated private information (highly suggestive of memory) could then manifest as within-individual spatial consistency decreasing at a slower rate than between-individual similarity (whereas with similar rates of decrease, the use of private vs public information cannot be disentangled).

Here, we investigated search strategies and reliance on memory in Cape gannets, a marine predator renowned for its frequent use of social cues during foraging [36,38,42], but closely related to the northern gannet (Morus bassanus), a species displaying high and long-term individual spatial specialization [13,50]. For this, we used consecutive GPS tracks collected from individuals during a breeding season, over several breeding seasons, but with different individuals being tracked each year. These data were used to investigate (i) within-individual consistency in foraging areas across consecutive trips (within-season), (ii) whether the tendency for within-individual consistency varied across individuals and whether it could be related to intrinsic individual traits such as sex or size (that could be linked, e.g. to individual competitive ability), (iii) for how long on average within-individual consistency persisted within-season (testing for frequency of search information updates), (iv) whether the probability of revisiting a similar area (by the same or different individuals) decreased with time scale, but at a slower rate within- than between-individuals (testing reliance on private information and not just dynamic public cues), and (v) if the within-individual frequency of search information updates varied across years (relating this with measures of average annual foraging effort as an indirect proxy for potential interannual variation in competition for food).

2. Material and methods

(a). Study system

Data were collected on Cape gannets breeding on Bird Island (33°50′ S; 26°17′ E) in Algoa Bay, South Africa (figure 1). This colony comprises approximately 95 000 breeding pairs, and has been relatively stable in numbers since 2001 [51]. Tracking data were collected from adults guarding initially small chicks (December−February), during the period 2010−2021 (years correspond to December). During this stage, adults alternate short trips at sea [52] with the partner staying at the nest for protection. Adults at this colony primarily feed on natural prey such as sardine (Sardinops sagax), anchovy (Engraulis encrasicolus), saury (Scomberesox saurus) and other fish or cephalopods [53]. The data on sex of tracked birds, determined through genetic assessment [54,55], were available up to 2015. All equipped birds were weighed at least once in a season (we used the mean if they were weighed multiple times; so our mass metric reflects both size and body condition) and their culmen length measured (see details in [54]).

Figure 1.

Map (A) and metrics (B–D) of foraging trips recorded between 2010 and 2021 (December−January each year). In (A), maximum range of each foraging trip is represented with a red dot, the colony is shown as a large yellow dot (see also insert showing the location of study area). Map colours have been chosen to highlight the continental shelf (depth < 200 m) and its slope (depth 200−2000 m); isobaths are shown every 200 m from 0 down to −2000 m (not beyond, for clarity).

(b). GPS deployments

The standard protocol for deployment and retrieval of GPS units on Cape gannets on Bird Island are detailed in [53,54]. In essence, partners leaving their nests to go foraging were captured using a 3 m hooked pole. Birds were then equipped with GPS tags (I-gotU GT-600, Mobile-Action-Technology (39 g); CatLog-S, Catnip Technologies, Hong Kong (26 g); or Axy-trek, TechnoSmart, Italy (30 g)) weighing a maximum of 1.5% body weight and programmed to record a fix ranging from 5 s to 15 min depending on years and devices. Instruments were attached to the central tail feathers using waterproof Tesa tape (Beiersdorf AG, Germany) allowing for easy removal after return to the colony. Handling time did not exceed 10 min.

(c). Individual foraging decisions and similarity between decisions

Foraging trips were defined as recordings of at least five GPS locations covering at least 1 h or more further than 3 km from the colony. For our analyses, we summarized each trip by the direction from colony measured at the point of maximum range, as an index of the decision of a general foraging direction by the bird [6,56,57]. In support of this simplification, a mean of 81.7% of locations recorded during a foraging trip fell within 30° of the direction at maximum range (median 90.4%, first quartile 71.4%, third quartile 96.7%) and a mean of 66.3% of locations within 15° (median 71.4%, first quartile 50.0%, third quartile 86.8%). In other words, most single trips only explored a relatively narrow arc of direction around the colony, compared with the 250° available (figure 1). Similarity between trips could then be assessed by calculating the angle between trip directions at maximum range: birds that reached points diametrically opposed from the colony would show an angle of 180°, whereas two trips reaching distal points in a similar direction from colony would form an angle close to 0°.

For each possible pair of trips in our dataset, we thus determined the angle between their respective directions from colony at maximum range (similarity in direction), the time between their respective departure (delta-t), whether the two trips were by the same individual or by different individuals (within vs between individual, respectively), identity and traits (sex, mean weight, culmen length) of each individual involved, the ‘year’ when the two trips were made the same year, and the number of trips apart if trips were made by the same individual (delta-trip). For instance, there is a ‘delta-trip’ of 2 between the first and the third trip of an individual but also between the second and the fourth trip, the third and the fifth trip, etc. (see table 1 for details on the data structure).

Table 1.

Data structure presenting overall and year-by-year sample sizes of Cape gannet foraging trips at Bird Island (in more recent years, the sex of individuals is unknown), and the structure of repeat data: several foraging trips were often recorded per individual. We consider trips 1 & 2, trips 2 & 3, trips 3 & 4, (etc.) by the same individual as ‘consecutive trips’; whereas trips 1 & 3, 2 & 4 (etc.) by the same individual are trips ‘with ∆trip = 2’, trips 1 & 6, 2 & 7 (etc.) by the same individual are trips ‘with ∆trip = 5’, and so forth.

year	N trips	N indiv	N males	N females	max. range (mean ± s.d.)	trip duration (mean ± s.d.)	N indiv with ≥2 consecutive trips	N pairs of consecutive trips (∆trip = 1)	prop. consecutive trips <30°	N indiv with ≥5 consecutive trips	N pairs of trips t & t+5 (∆trip = 5)	N indiv with≥10 consecut. trips	N pairs of trips t & t+10 (∆trip = 10)	N indiv with ≥15 consecutive trips	N pairs of trips t & t+15 (∆trip = 15)	N indiv with ≥20 consecutive trips	N pairs of trips t & t+20 (∆trip = 20)
2010	26	23	?	?	40.1 ± 22.2	13.3 ± 7.5	2	4	0.0%	0	0	0	0	0	0	0	0
2011	30	29	14	14	75.3 ± 50.4	16.8 ± 11.8	1	1	0.0%	0	0	0	0	0	0	0	0
2012	41	40	18	21	97.1 ± 68.7	22.8 ± 10.9	1	1	0.0%	0	0	0	0	0	0	0	0
2013	30	23	15	8	128.2 ± 86.4	35.2 ± 22.9	10	13	38.5%	0	0	0	0	0	0	0	0
2015	187	18	9	9	104.6 ± 103.3	22.9 ± 19.5	25	183	51.9%	16	108	8	43	5	15	1	2
2016	100	18	? ?	?	147.7 ± 126.9	30.2 ± 27.7	9	95	54.7%	7	63	5	35	2	12	2	2
2017	343	16	?	?	69.1 ± 60.2	15.9 ± 13.4	16	361	44.6%	16	297	15	220	14	145	9	91
2018	55	4	?	?	93.3 ± 99.4	18.4 ± 18.1	4	54	53.7%	4	38	4	18	2	4	0	0
2019	113	10	?	?	61.7 ± 36.1	17.5 ± 9.8	10	111	43.2%	9	74	6	36	3	12	1	5
2020	29	29	?	?	82.2 ± 36.4	20.8 ± 7.2	0	0	NA	0	0	0	0	0	0	0	0
2021	18	18	?	?	141.7 ± 84.9	24.7 ± 10.5	0	0	NA	0	0	0	0	0	0	0	0
All	972	228	56	52	91.0 ± 88.8	20.6 ± 18.1	78	823	47.4%	52	580	38	352	26	188	13	100

Other metrics of trip similarity are frequently reported in the literature, such as nearest-neighbour distances (NND) [50] or Earth mover’s distance (EMD) [56]. We would argue that the angle between directions at maximum range provides a simpler and more easily interpretable way of testing whether individuals use memory to decide where to forage. Indeed, both NND and EMD are integrative metrics, which means that they intend to compare trips in their entirety, which is not always an advantage. While low NND or EMD values indicate highly similar trips, high values of these metrics can emerge for multiple reasons (because of different trip directions, of different trip maximum ranges and/or of different trip shapes), but we might not want to interpret the same way, e.g. two trips with different range but same direction versus two trips with same range but different direction. In contrast, interpretability of angles is more straightforward: low angles between directions at maximum range mean that the individual(s) foraged in the same direction, while high values mean that the individual(s) foraged in different directions.

To make our results comparable with other studies, and to investigate the robustness of our results to our methodological choice of focusing on the angle between directions at maximum range, we nevertheless present in electronic supplementary materials results using NND and EMD, but also angles between the respective mean directions of trips rather than their direction at maximum range. We also provide in electronic supplementary materials a more thorough discussion on the advantages and disadvantages of the different metrics.

(d). Statistical comparisons of within-individual consistency and between-individual similarity

Even if we assumed completely random search, we would expect that some pairs of trips have similar directions just by chance. If all directions around the colony were equally likely (i.e. totally random direction choice by individuals), the angle between trips should follow a uniform distribution between 0° and 180°. However, gannets generally do not fly over the mainland and so generally avoid northward directions (figure 1A). Moreover, at the population level, we could rapidly identify a clear and consistent (across years) preference for directions southwest of the colony (figure 1A). Knowing these two constraints, we expected inflated apparent similarity compared with the uniform distribution, not necessarily reflecting use of memory or use of slowly changing external cues.

We thus designed two custom null distributions so that our comparisons with null expectations were more conservative. The first null distribution (accounting for land avoidance) was assessed by simulating the distribution of angles obtained between pairs of directions taken randomly between 0° and 250°. The second null distribution (accounting for consistent preferences for southwest waters) was assessed by measuring the distribution of angles between all possible pairs of trips by different individuals over the study period (referred to as ‘asymptotic’ between-individual similarity in trip directions).

Differences between within-individual consistency levels and between-individual similarity levels, and comparisons with any of the two null distributions were tested with Kolmogorov–Smirnov (K–S) tests (testing overall differences between the full distributions). To test over which time scales ‘simultaneous’ trips were more similar than expected by chance (within or between individuals), we identified the smallest delta-t (and for within-individual, the smallest delta-trip) beyond which K–S-test comparisons with null distributions were no longer significant (time-to-fading). To test that within-individual similarity in direction faded more slowly with delta-t than between-individual similarity in direction, we compared their respective time-to-fading. Hence, these tests are based on differences between the full distributions of angles, but visualization of histograms clearly showed that the main differences between distributions were the level of 0°-skewness.

(e). Statistical tests of variations in within-individual consistency

Variations in the tendency of within-individual consistency in trip direction were investigated using generalized linear (mixed) models to include linear predictors and/or multiple explanatory variables. Individual identity of birds was included as a random factor whenever possible (see below) using package lme4 v. 1.1-25 in R v. 3.6.3. We used a binomial categorization (i.e. angle < 30° or not), because angles could not adequately be modelled with a normal distribution (notably due to the skew towards the 0° boundary) and because differences in angle were less meaningful when far from 0° (i.e. a bird going in a direction 110° or 180° from last trip would equally be interpreted as not returning to the same area). Choosing a threshold lower than 30° would not make sense given the general angular width of entire foraging trips, and choosing a larger threshold would reduce chances to detect subtle changes. In the electronic supplementary materials, we nevertheless show that using a threshold of 60° would not significantly affect our conclusions.

Due to the heterogeneity of our dataset, we had to design different models depending on the availability or compatibility of explanatory variables. Models 1–2A–D (but not model 3) included as a covariate delta-t (the absolute time lag between the departure date of the two trips considered). Model 1 (GLM) was used to investigate individual trait influence (sex, culmen length as an index of body size and residuals of weight regressed against culmen length as a proxy for variations in body mass after accounting for variations in body size) on the probability that two consecutive trips by the same individual (delta-trip = 1) shared a similar direction. It used a restricted sample size because individual traits (notably sex) were not available for all individuals. Model 1 could not include a random factor because individual ID was too redundant with explanatory variables. For simplicity, we did not include a year effect (see models 2A–D), but including year in any form within model 1 did not change conclusions (not reported). Finally, using weight instead of residuals of weight regressed against culmen length also did not change conclusions (not reported). Models 2A–D (GLMM) used the widest possible dataset to explore different hypotheses all related to the effects of year on the probability that two consecutive trips by the same individual (delta-trip = 1) share a similar direction: model 2A used year as a linear predictor (testing a linear chronological trend across years); models 2B and 2C replaced year by the mean maximum range of trips that year or by the mean duration of trips that year, respectively (to test the effects of these proxies of interannual variations in foraging effort in the population); model 2D used year as a categorical variable to test all possible pairwise comparisons between years (using post hoc Tukey tests corrected for multiple testing). Last, model 3 (GLMM) was used to assess how within-individual consistency levels decreased across multiple consecutive trips (delta-trip from 1 to 10) and to compare decreasing slopes between years (interaction factor between delta-trip and year as a categorical variable). Again, pairwise comparisons of slopes were tested with post hoc Tukey tests correcting for multiple testing.

3. Results

Between 2010 and 2021 (covering all breeding seasons except for 2014), a total of 972 complete trips at-sea were recorded (figure 1A) from 228 different individuals (the structure of the repeat data is provided in more detail in table 1). Most trips (93.0 %) lasted less than 48 h (mean ± s.d. 20.2 h ± 17.4 h, range 1−145.6 h, median 18.1 h; figure 1B) and most (91.3%) remained within 200 km of the colony (maximum range from colony: mean ± s.d.: 88.7 km ± 84.5 km, range: 7.5−541.8 km; median 63.9 km; figure 1C). Overall, birds explored a 250°-arc of foraging directions around the colony, mostly away from the mainland (thus excluding northern directions where the coastline is about 7 km from colony), with a clear preference for waters to the southwest of the colony over the continental shelf (figure 1D).

(a). Comparison of within-individual consistency with between-individual similarity

Multiple trips per individual (from 2 to 38, mean ± s.d.: 11.6 ± 9.2) were recorded during 9 of the 11 breeding seasons (excluding 2020−2021 and 2021−2022), for a total of 78 different individuals (table 1). On average, 33.7 h ± 20.2 h separated two consecutive departures to sea by the same individual (i.e. including first trip duration and time at nest before the second trip began).

Individuals tended to return in a consistent direction over consecutive trips: the angle between trips was <30° in 47.4% of consecutive foraging trips by the same individual (figure 2A; random expectations if all directions away from the mainland coast were equally favoured: 22.2%, K–S test: D = 0.266, p < 0.001). The fact that the population showed a consistent preference for waters southwest of the island over the study period (figure 1D) was not sufficient to explain within-individual persistence in foraging direction across consecutive trips (expected proportion of pairs of trips <30° under this hypothesis: 34.1%; figure 2B; K–S test comparing distribution of angles between consecutive trips by the same individual with asymptotic distribution of angles between trips of different individuals regardless of time elapsed between trips: D = 0.138, p < 0.001; figure 2A–B).

Figure 2.

Individual specialization strategies for the choice of trip direction (measured at max. range from colony). (A) Within-individual consistency across two consecutive trips.(B) Between-individual similarity between trips.(C,D) Rate of decay across several consecutive trips for (C) only within-individual consistency or (D) with time elapsed between trip departures (within-individual consistency and between-individual similarity). In (A,B), the number of observations (frequencies) was converted into density to standardize the y-scale despite different sample sizes. In (C,D), blue (and black) arrows indicate ‘time-to-fading’, when within-individual similarity (and between-individual similarity, respectively) stops being significantly different from asymptotic between-individual similarity (K–S tests), vertical lines indicate 95% CI of the estimated proportions. ‘Asymptotic between-individual similarity’ refers to the proportion of trips with angle <30° in (B), i.e. regardless of time lag between trips by different individuals. In all panels, the red lines indicate null expectations if all directions away from the mainland were equally favoured. CI, confidence interval.

Individuals persisted in a similar foraging direction for multiple and up to nine consecutive trips (angle distribution summarized by proportion <30° in figure 2C): it is only after delta-trip values above 8 that the distribution of angles between trips of the same individual was no longer significantly different (i.e. more 0°-skewed) from the asymptotic distribution of angles between trips from different individuals (for 1 to 8 trips apart, K–S tests: D > 0.102, p < 0.05, n > 434 pairs of trips by the same individual, from >42 different individuals, see also table 1 for sample sizes). In terms of time scale, on average (mean ± s.d.), it took 10.5 ± 2.6 days for an individual to depart on eight consecutive trips. Consistent with this, for delta-t between trips up to 10 days apart, the distribution of angles between trips from the same individual was significantly different from the asymptotic distribution of angles between trips from different individuals (for delta-t of 0−11 days: K–S tests: D > 0.073, n > 297 pairs of trips by the same individual, p < 0.02; figure 2D). As a comparison, similarity in trip direction from different individuals decreased much more rapidly with delta-t between them (figure 2D), reaching asymptote levels after ca 3 days (for delta-t of 0−3 days, K–S tests: D > 0.04, p < 0.02, n > 2985 pairs of trips from different individuals).

(b). Variations in within-individual consistency strategies

We had records of 10 or more consecutive foraging trips for 40 different individuals (see table 1). Among these 40 individuals, the proportion of consecutive trips (delta-trip = 1) sharing a similar direction (<30°) ranged from 15.4% to 91.7% (mean 46.6%, median 45.6%, quartiles 32.8–52.4%, with no sign of bimodality), thus spreading widely around the population average. Nonetheless, the majority of individuals showed consistency levels above the asymptotic between-individual similarity in direction (34.1%), suggesting widespread use of this individual consistency strategy in the population. Among these 40 individuals, when considering pairs of trips with delta-trip = 5 (e.g. trip 1 with trip 6, etc.), the proportion of these pairs of trips with angle <30° ranged from 0.0% to 87.5% (median 41.0%, mean 43.3%, quartiles 33.3–55.2%), suggesting most individuals persisted in a similar direction more than expected by chance for at least five consecutive trips.

We then explored factors potentially explaining why some individuals tended to return more often in a consistent direction across consecutive trips (delta-trip = 1) than others. When considering only individuals for which the information was available (table 1), sex, a mass-independent proxy for body size (culmen length) or a proxy for body condition (residuals of mass regressed against culmen length) did not explain variations in the probability to return in the same direction over two consecutive trips (delta-trip = 1; binomial GLM model 1: n = 198 pairs of trips from 37 indiv.; sex: z = 0.668, p = 0.50; culmen: z = 1.066, p = 0.29; residuals of mass regressed against culmen length: z = 0.377, p = 0.71). In addition, the time lag between two consecutive trips did not affect the probability to return in a similar direction (z = −1.052, p = 0.29).

Considering all individuals with repeat trips (table 1), we did not detect a linear chronological trend over the different years in the probability of consecutive trips to share a similar direction (delta-trip = 1, binomial GLMM model 2A, z = 0.464, p = 0.64). The annual mean maximum range or annual mean trip duration, as proxies of annual foraging effort, had no linear effects on the probability of consecutive trips to share a similar direction (delta-trip = 1; binomial GLMM model 2B on max range: z = 1.671, p = 0.10; binomial GLMM model 2C on trip duration: z = 1.357, p = 0.18), despite substantial variability of maximum range and trip duration between years (figure 3A,B). In fact, considering all pairwise comparisons between years, we did not detect significant differences between any two years in the probability of consecutive trips to share a similar direction (delta-trip = 1; contrast method on binomial GLMM model 2D: z < 0.2461, p > 0.85 for all pairs of years). In all these models, we controlled for the time lag elapsed between consecutive trips, which never had a significant effect (|z| < 1.353, p > 0.18) on the probability of consecutive trips to share a similar direction. For years where sample size permitted (2015–2019; table 1), slopes measuring the decay of the probability of trips from the same individual to share a similar direction in function of the delta-trip (delta-trip from 1 to 10) were not statistically different between years (GLMM model 3, n = 6872 pairs from 64 individuals and 5 different years, z < 2.194, p > 0.18 for all pairwise comparisons between years; figure 3C).

Figure 3.

Observed differences across years in (A) maximum range, (B) trip duration and (C) within-individual consistency in direction across consecutive trips by the same individual for different years (shades of grey, slopes non-significantly different, see main text). In all panels, different grey shades are used for different years with a large sample of repeat-trips per individuals; in (A,B) other years are in white. In (C), average within-individual consistency levels across all years (blue line, with 95% CI shown by vertical lines), asymptotic between-individual similarity levels (black dotted line, see definition in figure 2) and random expectations accounting for land avoidance (red line) are also indicated.

4. Discussion

Our study presents clear evidence that memory influences the general foraging direction selected around the colony, in a marine predator known to heavily rely on social cues for this decision. Our approach also shows that apparent individual preferences for a foraging direction are frequently updated within a breeding season, every few days. Most gannets tended to return in a consistent direction over two consecutive trips, with no detected differences in consistency strategy among individuals of different sex or size. Finally, despite some interannual variations in apparent foraging effort, the strategy of individually returning in a consistent direction for several consecutive trips, but eventually updating the preference within the season, was consistently observed across years, with no detected differences neither among years nor as a function of our annual proxies of population-level foraging effort. Altogether, these results suggest a widespread and consistent reliance on frequently updated memory within this population.

Although the theory in behavioural ecology emphasizes the trade-off between memory and social cues [15,34], our results indicate that categorizing search strategies as primarily based on social cues or on memory may be a misleading simplification. Previous studies quantified the use of social cues in this population: all sampled individuals reacted at least once to a form of social cue during a few hours of bird-borne camera video-recordings, and on average did so five times per hour at sea [36]. Furthermore, these recordings showed that two-thirds of individuals flew to their first foraging patch as part of a cohesive flying social group rather than alone [38], with benefits for flight and prey captures [43,44]: their use of social cues appeared widespread and important throughout the entire foraging trip [36]. Reliance on these social cues and/or on other non-social but ‘real-time’ external cues could explain our observation of rapidly fading between-individual similarity in foraging direction (i.e. on average after 3 days, these cues would change which direction is indicated). These ‘real-time’ cues are then insufficient to explain that within-individual consistency persisted for up to 10 days. The most parsimonious explanation (see also [58]) is that individuals use a form of memory to influence their foraging direction, in addition to the social cues used throughout the foraging trip. This joint use of several sources of information was postulated by some theoretical simulation models [59] and seemed necessary to explain that neighbouring colonies segregate in their foraging grounds [59]. Our study thus provides strong empirical support for this mechanism of joint information use when searching for prey. More work is, however, needed to elucidate how exactly these sources of information are combined.

There are many possible hypotheses regarding the nature of information foraging individuals may be memorizing, and the duration and use of this knowledge [60], let alone how they could combine it with social cues. Foragers can alternate various behaviours throughout their foraging trips (e.g. travel and search, feeding, resting, etc.) [61,62], and their search can be a multi-scale process [63,64], where individuals first scan vast areas by rapid directed movement, then switch to area-restricted search (ARS) in areas where prey are more likely to be found (reviewed across various animals in [26]). In that context, individuals might memorize distant and exact locations where they previously fed and/or performed ARS, or places identified as potentially promising during their previous travel phases [26]. In addition, individuals could preferentially return to the last, the best or the closest of these past locations, and might also remember (or not) the route to join these places, to name only a few simple possibilities. Alternatively, predators may use their memory to identify a general, presumably profitable direction at departure rather than an exact distant location, for instance, based on a weighted average of different memories across several previous foraging trips, and then would opportunistically search for prey and/or for cues encountered on the way [13,60]. Unfortunately, the number of different theoretical hypotheses on memory means they cannot be distinguished by only using large-scale metrics of similarity between entire trips, whether angular (as in our study) or integrative (as for NND [50] or EMD [56]). Here, the exact estimate of the frequency at which memory was updated varied among the explored similarity metrics (electronic supplementary material). Most metrics confirmed that within-individual similarity decreased with time scale (within season), and more slowly than between-individual similarity, but either not at the same rate or not reaching a clear asymptotic phase. Yet the integrative nature of non-angular metrics (EMD or NND) made it impossible to tell which aspects of the foraging trips (i.e. range, shape and/or any combination of these with direction) are updated on a different frequency. More complex and detailed analyses would be necessary to start testing some of the hypotheses suggested above (e.g. [6]). We can still conclude from our results that whatever it was that individuals memorized, and whatever the exact sites they visited during a foraging trip, their use of memory influenced the general direction that individuals selected around the colony, and this direction preference was updated on average at most every 10 days.

It is generally accepted that search strategies should be adapted to the dynamics of the environment and of the prey distribution [11–13]. It is unclear, however, how and how fast search strategies could change to adapt to environmental changes. Our study provides several new elements on this question. First, a consistency strategy was observed in most individuals, and with all due caution linked to our annual sample sizes, persistent individual preferences lasting for several consecutive trips were observed every year. It is not sure if sex or size of individuals can be reliable indicators of their competitive ability within feeding aggregations, but either way, these factors did not noticeably affect individual consistency strategies. Similarly, our observations of interannual variations in mean and variance of foraging trip duration and range indirectly suggested that resources close to the colony were more quickly depleted in some years than others, forcing some individuals to forage further [65,66]. Yet these interannual variations in foraging conditions did not seem associated with clear and major variations in memory strategies. All this points to a rather widespread and consistent strategy within the population.

Second, we can compare our results with search strategies reported in other species. The very closely related northern gannet is ecologically and morphologically very similar to the Cape gannet [67]. Previous studies in northern gannets found long-term individual consistency across seasons, with little evidence for any fading within a season [13,50], but provided less detailed information about social foraging in this species [68]. Other temperate or polar neritic marine predators were shown to display individual consistency across years (e.g. albatrosses [28]; pinnipeds [69]), while some marine predators in oligotrophic waters (Mediterranean shearwaters or tropical seabirds) showed very little individual consistency even across two consecutive trips [12,70]. Our present results, obtained in a temperate neritic forager, might not seem to conform to this general pattern, with frequent memory updates apparently being commonplace (see also a recent study in North Atlantic [30]). However, the metrics to evaluate the similarity between trips varied among studies, and many previous studies were based on NND measures that present several limitations for comparative approaches (electronic supplementary material). It is possible that northern gannets update their preferences within a season if we consider their general foraging direction [49, p. 99] and/or that Cape gannets would also display signs of individual consistency across seasons (this could not be tested here). A subantarctic neritic albatross actually showed both long-term, interannual individual spatial consistency [28] and short-term memory, from one foraging trip to the next, when related to interactions with vessels [6]. The present study further highlights how little we seem to know about variations in search strategies, and at which time scales different components might be memorized. Our study also illustrates that given the amount of tracking data collected in colonial foragers nowadays, quick progress on this comparative question could be achieved following very simple methods, even without direct data on resources.

More generally, our study contributes to the now widespread studies on individual specialization or animal personality [46,71]. These studies usually report repeatability indices to assess the ratio of within-to-among individual variance in phenotype. It is well established that individual specialization or animal personality tends to be lower when measured over longer time scales [71,72]. Yet the proximal mechanisms, the theoretical consequences and the comparative biology of this temporal scale pattern in comparison remain poorly understood. First, while ultimate ecological and evolutionary explanations for these patterns are often considered (e.g. [11,73,74]), proximate mechanistic questions (e.g. the links between cognition, memory and personality/individual specialization) are rarely discussed explicitly [75]. Yet the proximate mechanisms might not conform to optimality predictions, for instance, they might not evolve at the pace that the environment is changing at [5,19], so the step of empirically describing these mechanisms should not be neglected. Second, recent studies have highlighted how reliance on repeatability indices has limited comparative approaches in personality and individual specialization research [47,48]. Our separation of how within-individual and between-individual variation components each changed with time scale might inspire similar approaches in the broader individual specialization literature. Third, theory in animal personality and individual specialization have addressed why individual heterogeneities should matter in ecology and evolution [76], but these theoretical developments mostly considered stable individual heterogeneities (i.e. fixed individual ‘preferences’). There is no clear theoretical work synthetizing why and how an understanding of the time scale at which preferences change could matter. Models in movement ecology start to provide some ideas of the complex ecoevolutionary feedback loops arising from the time scales of learning and spatial memory [11,77], but whether and how they can inspire transfer to the broader individual specialization research remains unclear. Overall, we thus hope our study will encourage renewed interest in merging memory research with individual repeatability research.

To conclude, it might be no surprise that seabirds can rely on complex search strategies mixing different cues, potentially memorizing different components over different time scales. Even animals with much smaller nervous systems can cognitively process information in sophisticated ways to join distant food resources (e.g. butterflies and bees [78,79]). Some scientists had previously argued that it was not the brain but the unpredictability of the marine environment that might limit complex cognitive search strategies in marine predators, a theoretical view that might be appealing for many other apparently ‘featureless’ environments (see discussions in [12,80]). We hope our study helps stimulate more scientific debate and lead to improved methods to test these thoughts on how organisms search for food, to better document the complexity of search mechanisms and their variations. Biologging now offers powerful ways to more directly document and compare these cognitive processes empirically, even without data on resources.

Ethics

Research was undertaken under animal ethics approval by the Research Ethics Committee at the Nelson Mandela University (A10-SCI-ZOO-008 and A18-SCI-ZOO 009).

Data accessibility

Data and codes to replicate this analysis have been deposited in Dryad [81].

Supplementary material is available online [82].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors’ contributions

J.C.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, visualization, writing—original draft, writing—review and editing; A.T.: data curation, investigation, resources, validation, writing—review and editing; A.-S.B.-L.: conceptualization, formal analysis, methodology, software, validation, writing—review and editing; Y.T.: data curation, funding acquisition, investigation, writing—review and editing; T.C.-K.: data curation, investigation, writing—review and editing; D.K.: data curation, investigation, writing—review and editing; P.P.: conceptualization, funding acquisition, methodology, project administration, resources, supervision, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

Field work was funded by the South African National Research Foundation (NRF) through the DST/NRF Centre of Excellence at the FitzPatrick Institute for African Ornithology with a contribution from the French National Research Institute for Sustainable Development in 2010. JC was supported by a postdoctoral fellowship granted by Nelson Mandela University, and by a CPJ grant by the French Ministère de la Recherche et de l’Enseignement Supérieur (ANR-22-CPJ1-0017-01).