Multispecies sensory networks and social foraging strategies: Implications for population decline in procellariiform seabirds

Significance

Foraging in the open ocean may require species with different sensory adaptations to use each other to find patchily distributed prey. Empirical studies have shown that procellariiform seabirds use different sensory strategies to forage, where some species track prey using olfaction while others rely more on vision. Using agent-based models, we show that social strategies greatly enhance foraging success across sensory strategies. In particular, more visual foragers may benefit from surveying the activity of olfactory foragers. Finally, we show success diminishes as population declines, underscoring the susceptibility of seabird populations to anthropogenic effects. Given the alarming rate at which many seabird species are disappearing, our findings highlight the necessity of better understanding and conserving these social networks.

Abstract

Multispecies sensory networks, where different species prioritize different sensory modalities and then use heterospecific information in a likely noncooperative fashion, may allow animals to improve foraging over large areas for cryptic prey. We test this hypothesis in procellariiform seabirds that forage in mixed flocks, where both prey odors and visual cues provided by other foraging hetero- and conspecifics might improve success rates. Using agent-based models, we explored the impact of social strategies on olfactory foraging for Antarctic krill (Euphausia superba). Our results suggest that social foraging enables species with different sensory adaptations to achieve similar success rates. Additionally, our results indicate that foraging is more successful in mixed-species rather than single-species flocks, where individuals can monitor the activity of other birds that are using different sensory foraging strategies than themselves to find prey. These results suggest that sensory-based foraging networks may be more critical to their survival than previously assumed. Finally, we show that success rates decrease at low population densities. As seabird populations continue to decline, understanding and preserving these social foraging networks may be essential for their conservation and ecological success. Overall, our study provides insights into the critical role of multispecies sensory networks to foraging success, wherein different species have different sensory adaptations for locating prey. While we used empirical anatomical and behavioral data specific to procellariiforms to inform our models, our approach and results may have broader implications for other species as well.

Relative to other habitats, foraging in pelagic (open ocean) environments may present challenges because prey are patchy and ephemeral (1–5). A hypothetical solution is the formation of multispecies aggregations with differing sensory priorities that can cast a multimodal (and likely noncooperative) sensory detection net over a larger region (1, 3, 4, 6, 7). Procellariiform seabirds (or “tubenoses”) are an interesting group in which to examine this, because they are adapted to forage in this environment, perhaps by exploiting social foraging mechanisms across species in ways that are not well understood (4, 5, 8). The procellariiforms are an order of highly pelagic, mostly sub-Antarctic seabird species, well known for their ability to forage over thousands of kilometers of open ocean for ephemeral and patchily distributed prey (reviewed by ref. 9). Many procellariiform species (i.e., shearwaters, prions, storm-petrels) also tend to be social, flocking species (10).

Procellariiforms typically have large olfactory bulbs (11, 12) that support a keen sense of smell. However, results from at-sea and laboratory experiments in the sub-Antarctic suggest that different procellariiform species use different sensory strategies to forage (6, 10, 13–16). Some, typically smaller, more cryptic, burrow-nesting species, track prey using odors such as dimethyl sulfide (DMS), allowing them to locate krill swarms near the ocean’s surface (6, 10, 14, 16–18). Other surface-nesting procellariiform species do not track DMS but respond to other scented compounds released as krill is consumed and to visual cues provided by the accumulation of seabirds (10, 15). These species tend to be larger and less cryptic. Comparative retinal and behavioral investigation of sympatric candidate species have also demonstrated a sixfold increase in visual acuity compared to a DMS-responsive species (10, 15, 19) (SI Appendix). These results, and others, suggest a physiological basis for these different foraging strategies (20), but how these differences influence social foraging success is unknown.

The sensory ecology of social foraging in procellariiforms is challenging to study experimentally, yet understanding how these networks operate and the potential impacts of population depletion on foraging success is becoming increasingly important, especially as approximately half the world’s species are declining in population (21, 22). Therefore, we developed an Agent Based Model to explore the role of social foraging strategies in the context of opportunistic olfactory foraging for krill at sea. We examined four foraging strategies, one that only utilizes olfactory-foraging algorithms, and three that additionally use social strategies (Fig. 1). We then use this model to explore how social foraging impacts success rates in four scenarios. First, we explored how varying the range at which birds could visually detect other individuals impacted our results (19). Second, we explored how these results are affected by olfactory ability. Third, we explored how these results change as the population density of the group decreased. Finally, we analyzed how success changes in “mixed species” flocks, where some members of the flock use different sensory-foraging strategies.

Fig. 1.

Graphical representation of foraging strategies. (A) Olfactory search algorithm, where the odor plume is shown in green. (B) Social foraging strategies (for three-bird flocks). On the Left-side of the panels, no birds have detected the odor (crosswind search); on the Right-side of the panels, the top bird has detected the odor (green; upwind cast). Solid black lines show individual headings; black dashed arrows show flock connectivity; solid gray “Group” arrows show the flock’s heading. Flock Foraging: birds search together; however, they do not prioritize movement toward successful birds and thus the flock may fragment if only a small portion detects the odor. Local Enhancement: birds prioritize movement toward successful individuals; however, they search independently, and thus others may be out of range when the plume is detected. Network Foraging: birds flock while searching and prioritize movement toward successful individuals.

Methods

Agent-Based Model.

To explore the impact of social foraging on olfactory-foraging seabirds, we built an Agent Based Model in Netlogo (Fig. 2). This model simulates a scenario in which an ephemeral prey patch (e.g. a small krill swarm) emits a filamentous odor plume that foraging seabirds can detect and track to its source. All individuals in the model (called “birds”) search for the same shared food source, which is nondepleting and provides equal payoff to any bird that reaches it before the simulation ends. The model is informed by flock densities observed during experimental trials (e.g. ref. 14). A full model description, along with the process used to assign values to key model variables, can be found in the SI Appendix, Figs. S1–S4 and Tables S1 and S2.

Fig. 2.

Model details. (A) The model world upon initialization (not to scale). Blue arrows indicate wind direction; the black circle is the prey patch; the odor plume is visualized in green (darker colors = higher concentration); the black square marks a potential bird starting area. (B) Flowchart of the movement algorithm. “Follow” is used only by Local Enhancement or Network Foraging strategies, and “Flock” is used only by Flock Foraging or Network Foraging strategies.

Model Initialization.

The model world (Fig. 2A) is two-dimensional and made up of a turbulent wind field, with wind generally blowing from the top of the map toward the bottom, with some random variation. The prey patch, situated near the top of the world, is 50 m in diameter. At-sea odor slick experiments suggest the plume from a slick of this size can be detected up to 4 km downwind (14). Fifty birds initialize randomly within a bounded starting location with an initial density of 1 bird/m². This starting location is centered 3 km downwind, and 2 km to the left of the food source, to allow flocking behavior to fully form before encountering the odor plume. Within each run, all birds are identical in terms of both their foraging strategy and all other parameters unless otherwise stated.

Odor Plume Model.

Although odor plumes are frequently simplified as Gaussian gradients, this approach can obscure predictions regarding olfactory search (for more detailed discussion, see: refs. 9, 16, 23–26). Turbulent transport is considerably faster than molecular diffusion at these spatial scales, so rather than the smooth gradient that is frequently depicted, the structure of the plume resembles discontinuous filaments that may be transported far from the source depending on the wind. Accordingly, here, we adapt agent-based odor plume models (26–28) to generate a more realistic, filamentous odor plume. Detailed information on this model, including its limitations, can be found in the SI Appendix. In brief, the odor plume is generated via a series of odor puffs which perform a biased random walk downwind over time while diffusing in concentration spatially, thereby generating a virtual approximation of a filamentous odor plume, with high concentration eddies where filaments overlap (Fig. 2A).

Olfactory Search Algorithm.

A common search strategy among species that track odors is depicted in Fig. 1 (9, 29). First, the organism (here, a seabird) moves crosswind to maximize chances of encountering an odor plume. Once a scent is detected, they then fly upwind to stay within the odor filament. Losing the scent triggers a crosswind cast until another filament is encountered. Repeated casts create a zigzag flight path that may eventually localize activity within visual range of the prey (9, 24, 29). In our model, the foraging strategy that utilizes this olfactory search algorithm, with no additional social strategies, is referred to as Olfaction Only.

Social Foraging Strategies.

Social foraging has been studied in various species of visually foraging seabirds including Cape Gannets (1, 5, 30), Northern Gannets (4), and European shags (31). Here, we add the olfactory search algorithm to three common strategies for social foraging in the pelagic environment (Fig. 1): I) Flock Foraging: Movement in a dispersed flock, where individuals within the flock are widely spaced from one another. Here, birds fly in a spatially distributed network, wherein each bird maintains a minimum distance to their nearest neighbor while moving in the same direction as the group. Evidence from radar (2), bird-borne video cameras (1, 30), and field observations (32) has shown that pelagic seabirds forage in dispersed clusters, reacting to one another’s movements. II) Local Enhancement: Movement toward birds that have detected evidence of prey. Here, birds are alerted by the behavior of other birds that have detected evidence of the prey [i.e., they use public information (33)]. In some species, this behavior could take the form of a deliberate signal (for example, producing an audible call), or indirect/unintentional communication (i.e., a change in behavior) (34). For example, a typical scenario often witnessed in pelagic seabirds is a rapid response to circling and feeding individuals (35–37). In our model, local enhancement occurs when one bird witnesses another bird initiating an olfactory search (i.e. movement upwind or zigzags), however, our model is agnostic as to whether this is direct or indirect communication. III) Network Foraging: A combination of Flock Foraging and Local Enhancement. In this strategy, flock foragers prioritize movement toward flockmates that show a behavioral change indicating they have detected an odor. This enables a rapid transfer of information through connected individuals. Evidence from previous modeling studies has suggested that this strategy improves foraging success for birds using visual-based search strategies (3, 4, 38–41).

Movement Algorithm.

The full movement algorithm is graphically displayed in Fig. 2B. Birds follow a flowchart-like series of behavioral decisions depending on which foraging strategy they are using. After each iteration of the movement algorithm, all birds move forward one body-length.

“Crosswind search”: this is the default behavior, run by all birds. Birds set their heading at a 90° angle compared to the direction of the wind at their current location (generally resulting in movement to the right).

“Olfactory search”: all birds run this algorithm. The olfactory search algorithm is graphically displayed in Fig. 1 and is based upon the common search strategy utilized by animals following a turbulent odor plume (9, 42).

“Follow”: this algorithm is only run by birds using the Local Enhancement and Network Foraging strategies. If an observer-bird is not tracking an odor, but another bird within their detection range is, the observer heads toward the odor-tracking bird. If the odor-tracking bird then loses the odor, the observer behaves as if they had detected but lost the odor themselves (i.e., crosswind casts).

“Flock”: this algorithm is only run by birds using the Flock Foraging and Network Foraging strategies. Here, if a bird detects any other nearby birds during their crosswind search, they employ a well-described flocking algorithm (43, 44). Briefly: First, they find all other birds within their detection range. If the nearest one is too close, they turn away from them, so they do not collide. Otherwise, they first turn in the direction of the average of these flockmates’ headings and then turn in the direction that will take them closer to the centroid of the flock.

Species Parameters.

Models were parameterized using empirically measured physical values—such as flight speed and body size—of known procellariiform species (SI Appendix); however, detection range, olfaction, and foraging strategies were treated as tunable traits. This framework was intended to isolate the influence of sensory and behavioral parameters on foraging success while controlling for variation in morphology and flight dynamics. Although no single species encompasses the full range of parameter values explored, all values fall within biologically relevant ranges observed across procellariiforms (SI Appendix). Accordingly, distinct behavioral and sensory combinations here represent interspecific variation, though could be extended to intraspecific applications in other contexts.

Main-text results are presented using the physical parameters of two species: Black-browed albatross (Thalassarche melanophris) and the White-chinned petrel (Procellaria aequinoctialis). These medium-sized species overlap in flight speeds (45), and frequently co-occur in mixed-species foraging flocks (37), but differ in their sensory foraging styles (14), making them well-suited for comparative modeling. To examine the influence of morphology more broadly, additional simulations using species at the upper and lower extremes of size and speed—the Wandering albatross (Diomedea exulans) and Wilson’s storm-petrel (Oceanites oceanicus)—are included in the SI Appendix. The model is also designed to support user-defined parameterization for species-specific applications.

Model Output.

All birds have 1 h to locate the food. If they do so, they are marked as successful. Variables were altered across a set of 500 distinct initial conditions (starting seeds). The number of successful birds is output from each run, and the mean and margin of error is calculated across all 500 seeds for each variable. These success rates are then reported as the percentage of successful birds out of the total group.

Conditions.

First, we explored how success rates changed for birds with varying detection ranges, which is how far away they can detect another bird (20 to 200 body-lengths; SI Appendix). Our model is agnostic to the sensory modality used to accomplish this detection; however, while in other animals this may be accomplished with sound, or via the lateral line, in procellariiforms, it would most likely be accomplished using vision. Flock density scales with detection range, such that birds with larger detection ranges maintain greater distances from their nearest neighbors compared to birds with small detection ranges (SI Appendix). Second, we varied the bird’s olfaction, which is how high the odor concentration had to be for the bird to detect it. For high-olfaction birds, detection occurred wherever local plume concentration exceeded the mean plume concentration, whereas low-olfaction birds required concentrations one SD above the mean. Third, we reran the model with a population size of 20, 10, and 5 birds. For these models, birds have the body size (0.9 m) and flight speed (11.5 m/s) of a Black-browed albatross (Thalassarche melanophris) (45). Finally, we modeled “mixed-species flocks” composed of two forager types: 1) “olfactory foragers”—high olfaction, but low detection ranges (20 body lengths), akin to a DMS tracking species (19). 2) “visual foragers”—no olfactory abilities at all (anosmic) but high detection ranges (200 body lengths) (19). Here, visual foragers had the size and speed of a Black-browed albatross, a species which does not track DMS (14), while the olfactory foragers have the size (0.55 m) and speed (11.5 m/s) of a White-chinned petrel (Procellaria aequinoctialis) (45), a known DMS-tracking species (14). Results for additional species are shown in SI Appendix, Figs. S5–S8.

Results

All three social foraging strategies increased success rates when compared to birds that only used olfaction (Fig. 3A). For both low- and high-olfaction birds, Network Foraging was typically the most successful strategy, followed by Local Enhancement, Flock Foraging, and finally, Olfaction Only. For Olfaction Only, success rates were 10.9 ± 0.9% for low-olfaction birds (mean ± margin of error) and 26.7 ± 1.2% for high-olfaction birds. The highest success rates occurred when using the Network Foraging strategy with a large detection range (95.9 ± 1.3% for low-olfaction birds and 98.4 ± 0.8% for high-olfaction birds).

Fig. 3.

Single species models, for birds with the size (0.9 m) and speed (11.5 m/s) of a Black-browed albatross. Lines represent the mean value calculated over 500 seeds; ribbons indicate 95% CI. Each foraging strategy has a unique color and line-type. The y-axis is the percentage of birds that located the food source. The x-axis shows detection range in units of body-length, corresponding to 18 to 180 m. Leftmost panel(s) shows results for birds with high olfaction, rightmost for low olfaction. (A) Models with groups of 50 birds. (B) Models with groups of 20 (Left subpanels), 10 (center subpanels), and 5 (Right subpanels) birds.

For all social foraging strategies, increasing the detection range improved success rates. Network Foraging and Local Enhancement were the most dependent on detection range, where increasing the detection range tenfold (from 20 to 200 body-lengths) improved success rates by ~69% for low-olfaction birds, and ~60% for high-olfaction birds. Flock Foraging was less dependent on detection range, with success rates only increasing by ~7 to 10% over the same range. The success rates for Network Foraging and Local Enhancement asymptote with detection range, but at different rates depending on olfaction. For example, for Network Foraging, success rates exceeded 95% after detection ranges of ~180 body-lengths for low-olfaction birds, versus ~130 body-lengths for high-olfaction birds.

For Network Foraging and Local Enhancement, success rates decreased as the population size decreased, especially for birds with a large detection range (Fig. 3B). For birds with high olfaction and large detection ranges, reducing the population by an order of magnitude decreased success rates by up to ~34% for Network Foraging and ~39% for Local Enhancement, but only ~4% for Flock Foraging. For birds with low-olfaction and large detection ranges, the same population reduction led to ~48 and ~47% decreases for Network Foraging and Local Enhancement, respectively, with negligible impact (~0%) on Flock Foraging.

Finally, a mixed species version of the model was also run (Fig. 4). Here, a portion of the group (out of 50 birds total) were “olfactory foragers,” while the other portion was “visual foragers.” Our results showed that when utilizing social foraging strategies, over 94% of the group could be visual foragers and the entire group could still achieve, and often surpass, the success rates of a group where all members were olfactory foragers. The maximum success rate for flocks using Olfaction Only (~26%) occurred when all birds were olfactory foragers and decreased linearly as the number of visual foragers increased. In comparison, the success rate for birds using Flock Foraging was ~30% when all birds were olfactory foragers, stayed relatively flat until ~30% of the group were visual foragers, and then dropped to be equivalent to the corresponding Olfaction Only success rate. For Network Foraging and Local Enhancement, the success rate when all birds were olfactory foragers was ~36 and ~34%, respectively. For both strategies, the success rate then increased with the number of visual foragers, peaking at ~65% for Network Foraging and ~52% for Local Enhancement when around ~60% of the group were visual foragers. For both strategies, success rates then dropped rapidly but were still higher than the maximum success rate for Olfaction Only up until ~94% of the group were visual foragers.

Fig. 4.

Mixed Species models. Lines represent the mean value calculated over 500 seeds; ribbons indicate 95% CI. Each foraging strategy has a unique color and line-type. The y-axis is the percentage of birds that located the food source. The x-axis shows the number of birds in each “species” (out of 50). The top value indicates olfactory foragers, which have the size (0.5 m) and speed (11.5 m/s) of a White-chinned petrel, High Olfaction, and a 20 body-length/11 m detection range. The bottom value indicates visual foragers, which have the size (0.9 m) and speed (11.5 m/s) of a Black-browed Albatross, No Olfaction (anosmic), and a 200 body-length/180 m detection range.

Discussion

Procellariiforms are known for their highly developed sense of smell; however, our results suggest that social strategies may be just as critical for improving olfactory foraging success in some species. Our results show that while high-olfaction flocks initially outperformed low-olfaction flocks, social strategies substantially improved foraging success in both groups. Specifically, Network Foraging increased success rates ~3.5-fold for high-olfaction birds, and ~8.7-fold for low-olfaction birds. These results also imply that conditions or adaptations that improve the ability to spot flockmates (such as flight height, bird contrast, visual capabilities, or light levels) should increase social foraging success. For example, our results show that increasing detection range when using social strategies allowed low-olfaction birds to perform as well as high-olfaction birds, without requiring additional investments into their olfactory capabilities. While exact numerical outcomes in the model varied depending on the initial parameter settings, the overall patterns and relative trends remained consistent across the range of tested values (SI Appendix, Figs. S5–S8).

Our model results show that social foraging strategies can compensate for different sensory-foraging strategies, allowing birds with different sensory-adaptations to achieve similar success rates. Specifically, success rates in our model asymptote more quickly with increasing detection range for high-olfaction birds than low-olfaction birds. This means that high-olfaction birds with lower detection ranges (and therefore denser flocks) can perform as well as low-olfaction birds with higher detection ranges (and more dispersed flocks). These results accurately predict observed patterns in foraging behavior in procellariiforms. For example, many DMS-tracking species such as prions (Pachyptila sp.) and blue petrels (Halobaena caerulea) often form dense flocks, while species that do not track DMS, such as albatross, are thought to rely more on visual cues and often forage in more dispersed networks, responding rapidly to the behavior of other birds (6, 8, 39). It has been proposed that these different foraging strategies (DMS-tracking versus more-visual foraging) reflect a trade-off in the development of the olfactory and visual system linked to rearing environment (10, 20) (SI Appendix). Our results add to this by suggesting that social foraging allows birds with diverging sensory adaptations to be competitive, even in a low-contrast visual seascape.

Results from our mixed species model show that social foraging strategies can lead to higher success rates in flocks composed of a mixture of visual and olfactory foraging specialists, compared to single-species flocks. Additionally, we show that high success rates can still be reached in groups where most individuals are visual foragers provided they can see a few olfactory foragers. Extending our results to the real world, Black-browed albatross (Thalassarche melanophris) and White-chinned petrels (Procellaria aequinoctialis) are common to mixed-species feeding aggregations (14, 15, 37, 46). Black-browed albatross are assumed to find prey by monitoring other species (32), but the extent to which their success depends on sensory-based networks is rarely explored (6). Results from our model applied here suggest that their foraging success may be improved by hunting in a network that includes monitoring the activity of olfactory-tracking krill specialists, including vulnerable species, such as White-chinned petrels (17, 18, 21). These results imply that maintaining a variety of sensory-specialists within these foraging networks may be more critical for the survival of the entire foraging community than has been previously assumed. Conservation efforts often focus on these larger, charismatic species, but our results indicate that olfactory-tracking specialists, including smaller, more cryptic, burrow-nesting species, may play a significant role within mixed-species groups. Although burrow nesters can be challenging to monitor due to their inaccessibility, our findings highlight the need for increased efforts to protect them.

The diminishing success rates in our models for social foraging strategies at lower population sizes underscores the potential vulnerability of social foraging networks as many procellariiform species face population decline (21, 47). Our model suggests that this vulnerability may be most pronounced for species with relatively lower olfactory tracking abilities, where there may be the greatest reliance on social strategies. While all strategies showed decreased success rates with lower population sizes, Flock Foraging proved to be the most resilient strategy, with success rates decreasing the least at reduced flock sizes. This resilience is critical in the context of the alarming decline in procellariiform species. Previous models have suggested that for animals relying on collective/social behaviors, populations could be subject to an Allee effect, wherein a precipitous decline can occur after the population drops below a certain level (39, 47–50). Our findings suggest that conservation efforts should consider the preservation of social networks and population densities to maintain the effectiveness of these adaptive foraging strategies.

From a broader perspective, our study offers insights into how information transfer within a group impacts sensory-based searches. For example, we found that Flock Foraging still generally outperformed Olfaction Only, even though agents had no access to public information. In other words, by merely staying close to flockmates, regardless of those flockmate’s performance, success rates improved. Even though this strategy did not perform as well as Local Enhancement or Network Foraging (both of which rely on birds responding to specific behaviors), it may be advantageous in real-world scenarios not addressed here. For example, consider a situation where animals must signal deliberately to use Local Enhancement (i.e., howling or barking in canine packs). In this scenario, Flock Foraging carries the additional advantage that it removes the energetic cost of signaling and reduces the risk of a signal alerting the prey, predators, or competitors (7). Results from the mixed-species model also imply that Flock Foraging may be less susceptible to exploitation, wherein members of the group respond to signalers but do not signal themselves, or even search for food (7). While our model did not specify whether birds were deliberately signaling, the results suggest that such signaling could increase both group and individual success rates (SI Appendix, Fig. S9). However, in situations where there is high competition at the food sources, this could result in a tradeoff between increasing group success rates versus increasing competition.

Overall, our study provides insights into the interplay between social and olfactory foraging strategies in Procellariiform seabirds. It underscores the adaptive potential of social foraging strategies and highlights the critical role of population density in maintaining these networks. As seabird populations continue to decline, understanding and preserving these social foraging networks may be essential for their conservation and ecological success.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Code, Model, Data Generated from Model data have been deposited in Dryad (https://doi.org/10.5061/dryad.kd51c5bkh) (51).