Aggressive Behavioural Interactions Between Swans (Cygnus Spp.) and Other Waterbirds During Winter: a Webcam‑Based Study Kevin A

Home , Common moorhen, Eurasian coot, Whooper swan

Wood et al. Avian Res (2020) 11:30 https://doi.org/10.1186/s40657-020-00216-7 Avian Research

RESEARCH Open Access Aggressive behavioural interactions between swans (Cygnus spp.) and other waterbirds during winter: a webcam‑based study Kevin A. Wood1, Phoebe Ham2, Jake Scales3, Eleanor Wyeth2 and Paul E. Rose1,2*

Abstract Background: Our understanding of any impacts of swans on other waterbirds (including other swans), and potential efects on waterbird community structure, remain limited by a paucity of fundamental behavioural and ecological data, including which species swans interact aggressively with and how frequently such interactions occur. Methods: Behavioural observations of aggression by swans and other waterbirds in winters 2018/2019 and 2019/2020, were carried out via live-streaming webcams at two wintering sites in the UK. All occurrence sampling was used to identify all aggressive interactions between conspecifc or heterospecifcs individuals, whilst focal observations were used to record the total time spent by swans on aggressive interactions with other swans. Binomial tests were then used to assess whether the proportion of intraspecifc aggressive interactions of each species difered from 0.5 (which would indicate equal numbers of intraspecifc and interspecifc interactions). Zero-infated generalized linear mixed efects models (ZIGLMMs) were used to assess between-individual variation in the total time spent by swans on aggressive interactions with other swans. Results: All three swan species were most frequently aggressive towards, and received most aggression from, their conspecifcs. Our 10-min focal observations showed that Whooper (Cygnus cygnus) and Bewick’s Swans (C. columbianus bewickii) spent 13.8 4.7 s (means 95% CI) and 1.4 0.3 s, respectively, on aggression with other swans. These durations were equivalent± to 2.3% and 0.2%± of the Whooper± and Bewick’s Swan time-activity budgets, respectively. Model selection indicated that the time spent in aggressive interactions with other swans was best-explained by the number of other swans present for Whooper Swans, and an interactive efect of time of day and winter of observation for Bewick’s Swans. However, the relationship between swan numbers and Whooper Swan aggression times was not strong (R2 19.3%). = Conclusions: Whilst swans do exhibit some aggression towards smaller waterbirds, the majority of aggression by swans is directed towards other swans. Aggression focused on conspecifcs likely refects greater overlap in resource use, and hence higher potential for competition, between individuals of the same species. Our study provides an example of how questions relating to avian behaviour can be addressed using methods of remote data collection such as live-streaming webcams.

*Correspondence: [email protected] 2 Centre for Research in Animal Behaviour, Psychology, Washington Singer, University of Exeter, Perry Road, Exeter, Devon EX4 4QG, UK Full list of author information is available at the end of the article

© The Author(s) 2020. This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativeco mmons.org/licen ses/by/4.0/ . The Creative Commons Public Domain Dedication waiver (http://creativeco mmons .org/publi cdoma in/ zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Wood et al. Avian Res (2020) 11:30 Page 2 of 16

Keywords: Aggression, Agonistic behaviour, Bewick’s Swans, Intraspecifc versus interspecifc competition, Remote data collection, Waterfowl, Whooper Swans

Background North America, there have been attempts to eradicate Communities of birds that use aquatic habitats (here- an invasive population of Mute Swans (Cygnus olor) due after ‘waterbirds’) typically contain individuals from to concerns about the efects of swans on native spe- multiple species, many of which overlap in terms of the cies, including other waterbirds. For example, in Mary- resources that they use (Pöysä 1983; Davis et al. 2014). land, USA, a newly established Mute Swan moulting Such overlaps in resource use can result in aggressive fock displaced a mixed breeding colony of Least Tern interactions between individuals within waterbird assem- (Sterna antillarum) and Black Skimmer (Rynchops niger), blages (Wood et al. 2017; Marchowski and Neubauer two species of local conservation concern (Terres and 2019). Aggressive behaviours allow individuals to gain Brinkler 2004). Waterbird managers and conservation- and maintain access to valuable limited resources such ists have also expressed concerns that the invasive Mute as food or preferred breeding or resting locations, and Swans could out-compete the native Trumpeter (Cyg- to deny other individuals access to those resources (King nus buccinator) or Whistling Swans for food and habi- 1973; Amat 1990; Pelligrini 2008). tat (Lumsden 2016). Similarly, conservationists have Among waterbird assemblages, the true swans (Cygnus questioned whether the observed c. 39% decline in win- spp.) are large-bodied, herbivorous waterbirds found on ter Bewick’s Swan (C. columbianus bewickii) numbers all continents except Antarctica (Rees et al. 2019). Previ- in north-west Europe between 1995 and 2010 may be ous authors have highlighted that swans exhibit aggres- at least partially attributable to competition with ris- sion towards both conspecifcs and heterospecifcs ing numbers of Whooper and Mute Swans in the region (Johnsgard 1965). Indeed, there are numerous examples (Rees et al. 2019). Even among conspecifcs, interfer- in the literature of aggression by swans towards other ence competition among foraging swans can reduce food waterbirds (e.g. Johnsgard 1965; Tingay 1974; Ely et al. intake rates (Gyimesi et al. 2010). In the Russian Arctic, 1987; Burgess and Stickney 1994; Gurtovaya 2000). Swans concerns of local hunters regarding the impacts of swans have been observed to threaten other birds with stylised on other waterbirds have been identifed as a motivation displays (Lind 1984), and to attack with their bill, wings for the illegal persecution of Bewick’s Swans (Newth et al. and body (Johnsgard 1965; Burgess and Stickney 1994), in press). Similarly, legal hunting of Whistling Swans which in some instances has resulted in the death of the in parts of the USA has been justifed on the basis that targeted individual (e.g. Stone and Marsters 1970; Dela- competition between the swans and other waterbirds cour 1973). For example, Ely et al. (1987) reported that has been deemed excessive (Sladen 1991). However, our attacks by Whistling Swans (Cygnus columbianus colum- understanding of any impacts of swans on other water- bianus, a subspecies of Tundra Swan) killed two Greater birds (including other swans), and potential efects on White-fronted Goose (Anser albifrons) goslings during waterbird community structure, remain limited by a the breeding season. Similarly, Brazil (1983) observed a paucity of fundamental behavioural and ecological data, Whooper Swan (Cygnus cygnus) attack and kill a juvenile including which species swans interact aggressively with Eurasian Wigeon (Mareca penelope) in Iceland. Despite and how frequently such interactions occur. such reports, recent studies have pointed out that the In this study we used repeated behavioural observa- extent of aggressive behaviours by swans, and the possi- tions at two wintering sites to improve our understand- ble impacts that these may have on other waterbirds, are ing of which species swans interact aggressively with poorly understood (Gayet et al. 2011; Wood et al. 2019a). and how frequently such interactions occur. We used Moreover, two studies of breeding waterbirds found no the behavioural data that we obtained to test three evidence that swans exclude other waterbirds from habi- hypotheses. As Peiman and Robinson (2010) reported tat or reduce breeding densities (Gayet et al. 2011, 2016). that aggression is more likely between individuals with A recent meta-analysis cast further doubt, by show- the greatest overlap in resource use, our first hypoth- ing that swans spent no more time engaged in aggres- esis (hereafter termed the ‘conspecific hypothesis’) sive behavioural interactions than other waterbird taxa was that swans would devote more time to intraspe- (Wood et al. 2017). cific aggression compared with interspecific aggres- Nevertheless, aggression by swans towards other sion. Aggressive behaviours among birds typically waterbirds continues to be relevant for the management become more common as the density of individuals and conservation of waterbirds and their habitats. In within a habitat increases (Metcalfe and Furness 1987; Wood et al. Avian Res (2020) 11:30 Page 3 of 16

Wood et al. 2015), thus our second hypothesis (here- Data collection after termed the ‘density hypothesis’) was that swans We used two ways of collecting data concurrently from would spend more time engaged in aggressive interac- observations made during periods of one hour in durations when present in higher density flocks. Finally, we tion: one way was used to record aggression between expected that as winter progressed, a dominance hier- any two individuals of any waterbird species to address archy would establish among individual swans within our conspecifc hypothesis, and the second way was used their flock, and so reduce further aggression; indeed, to record swan-specifc aggression to address our den- Scott (1981) reported that the frequency of fights sity and winter decline hypotheses. First we used all between Bewick’s Swans declined progressively over occurrence sampling to record all incidents of aggres- winter months. Therefore, our third hypothesis (here- sion between individuals over the course of the hour- after termed the ‘winter decline hypothesis’) was that long observation period (Altmann 1974), based on an swans would spend more time engaged in aggressive ethogram of the aggressive behaviours observed during interactions in early winter months compared with preliminary observations at both sites in October 2018. later months. Tese preliminary observations indicated a number of aggressive behaviours consistent with previous work (Johnsgard 1965), including strikes made with the bill, Methods wings, or body, as well as chasing and lunging at another Study systems individual. For each aggressive behaviour that was Our study focused on two wetlands used by wintering observed, the species identities of both the aggressor and waterbirds in the UK; the Wildfowl & Wetland Trust its opponent were recorded; all species were identifed on (WWT) Centre reserves at Slimbridge (51°44ʹ29.3ʺ the basis of size, body shape, and plumage characteris- N, 2°24ʹ21.52ʺ W) and Caerlaverock (54°59ʹ2.4ʺ N, tics, with the aid of an online photo-identifcation guide 3°30ʹ0ʺ W), in southwest England and southwest Scot- (RSPB 2018). land, respectively. Both wetland sites feature a mosaic To address our density and winter decline hypotheses of aquatic and terrestrial habitats, most notably small regarding the variations in aggression with swan num- lakes used by waterbirds for feeding and roosting. In bers and between months, we used focal sampling (Alt- recent winters Slimbridge has supported both Bewick’s mann 1974) to quantify the total time that Bewick’s and Swans and Mute Swans, whilst Caerlaverock has sup- Whooper Swans spent engaged in aggressive interactions ported Mute and Whooper Swans (Black and Rees with other swans. During each hour-long observation 1984). Te mean of the peak winter counts of individ- period, an observer selected a swan at random and used a ual swans between 2014/2015 and 2018/2019 was 149 stopwatch to record the duration of each aggressive inter- Bewick’s Swans (range 116–212) and 441 Mute Swans action with another swan in a 10-min observation period; (range 374–500) in the Severn Estuary (which includes hence, six individual swans could be observed during Slimbridge), and 337 Whooper (range 257–487) and each hour-long observation. A focal observation dura- 74 Mute Swans (range 70–83) in the Solway Estuary tion of 10 min was selected in order to make our study (which includes Caerlaverock) (Frost et al. 2020). Swans comparable with earlier time-activity budget of swans at both sites share feeding and roosting habitat with a that used an observation duration of 10 min (e.g. O’Hare range of smaller-bodied waterbird species, including et al. 2007; Tatu et al. 2007; Wood et al. 2019b). Both geese such as Canada Geese (Branta canadensis) and immediately before and after each hour-long observation Greylag Geese (Anser anser), dabbling ducks such as period, the number of swans that could be observed was Northern Mallard (Anas platyrhynchos), Eurasian Teal counted, with the mean average taken as the number of (Anas crecca), and Northern Pintail (Anas acuta), div- swans present during that observation. ing ducks such as Tufted Duck (Aythya fuligula) and All behavioural data were collected remotely via live- Common Pochard (Aythya ferina), Rallidae such as streaming webcams (AXIS Q6035-E PTZ Dome Network Eurasian Coot (Fulica atra) and Common Moorhen Camera), which were fxed in place at both sites. Both (Gallinula chloropus), as well as Common Shelduck webcams faced directly outwards over the study lake (Tadorna tadorna), and gull species (Larus spp.). Te from the shore, and each webcam maintained the same non-migratory Mute Swans are resident at both Slim- zoom so that the feld-of-view was standardised across bridge and Caerlaverock throughout the year, and all observation periods (Additional fle 1: Figure S1). As Bewick’s Swans are typically present at Slimbridge our study was conducted entirely via the webcams, we between November and February, whilst Whooper do not know the precise numbers of birds outside of the Swans use Caerlaverock between October and March cameras feld-of-view, but we suspect it to be low as the (Black and Rees 1984; Rees 2006). cameras covered major proportions of the surface area Wood et al. Avian Res (2020) 11:30 Page 4 of 16

of each lake (Additional fle 1: Figure S1). Webcams have intervals to be estimated for the proportions of intraspe- been shown previously to be useful tools in behavioural cifc and interspecifc interactions (Clopper and Pearson studies, which allow remote collection of data with lim- 1934). In addition, we used two-sample binomial tests for ited disturbance to the focal birds (e.g. Anderson et al. equality of proportions to assess the signifcance of the 2011; Schulwitz et al. 2018). During winter months diferences between species in their proportion of inter- in 2018/2019 (November 2018–February 2019, inclu- specifc aggressive interactions recorded towards other sive) and 2019/2020 (November 2019–December 2019, birds; P values were again adjusted using Holm-Bonfer- inclusive), webcam footage from WWT Slimbridge was roni corrections (Holm 1979). watched for 1 h on an average of 7.5 days per month at To address our density and winter decline hypotheses either 08:30 a.m. to 09:30 a.m. (hereafter “AM”), 11:30 regarding the variations in aggression with swan num- a.m. to 12:30 p.m. (hereafter “MID”) or 14:30 p.m. to bers and between months, we used zero-infated general- 15:30 p.m. (hereafter “PM”); these times were selected to ized linear mixed efects models (ZIGLMMs), using the achieve a balanced study design and to avoid coinciding glmmTMB R package (Brooks et al. 2017). Zero-infated with the periods when food (wheat grains) is provided as models were required because of the relatively high pro- part of a public engagement programme. Winter storms portions of zeros in the data sets (i.e. observations during at the start of January 2020 regrettably damaged the web- which no aggression was recorded). In each model, our cam and prevented further data collection. We aimed response variable was the number of seconds spent in to use the same methodology for WWT Caerlaverock, aggressive interactions by each individual recorded dur- however, the webcam sufered a technical failure after ing our focal observation. As the aforementioned issues data for only November 2018 could be collected. Simi- with the webcam at WWT Caerlaverock prevented data lar technical issues precluded data collection at WWT collection after November 2018, the resulting datasets Caerlaverock during winter 2019/2020. Our sampling for Bewick’s and Whooper Swans were markedly difer- methodology of ten observations per hour, on an aver- ent in terms of the sample size and the temporal repli- age of 7.5 days per month, over 6 months, yielded a total cation. Terefore, we analysed the data for Whooper of 450 observations at Slimbridge; however, swans were and Bewick’s Swans separately. For each species, we ran only present during 282 of these. In contrast, swans were and compared candidate models that comprised all pos- present at Caerlaverock during every observation period. sible combinations of additive and two-way interactions In total, we therefore obtained 282 focal observations between our explanatory variables, as well as the null (of 10 min each) from WWT Slimbridge and 42 focal model. For Whooper Swans, we considered the following observations (of 10 min each) from WWT Caerlaverock. explanatory variables: (i) the time of day of the observa- While we cannot discount the possibility that some indi- tion, a categorical factor with three levels (AM, MID, and viduals were observed on multiple occasions, the afore- PM), and (ii) the mean number of swans present during mentioned large numbers of individual swans present at the observation. For Bewick’s Swans we considered the both sites mean that this is unlikely to have been a major following explanatory variables: (i) the time of day of the issue. observation, a categorical factor with three levels (AM, MID, and PM), (ii) the mean number of swans present Statistical analyses during the observation, (iii) the month of observation, a All statistical analyses were carried out using R ver- categorical factor with four levels (November, December, sion 3.6.3 (R Core Team 2020). To address our conspe- January, and February), and (iv) the winter of observa- cifc hypothesis that intraspecifc interactions would be tion, a categorical factor with two levels (“A” = 2018/2019 observed more frequently than interspecifc interactions, and “B” = 2019/2020). In addition, a categorical variable we used two-tailed binomial tests to assess the statis- unique to each observation block (termed the ‘observa- tical signifcance of the deviation of the proportion of tion identity’), was ftted as a random intercept in each intraspecifc aggressive interactions of each species from model to account for the non-independence of individ- 0.5 (which would indicate equal numbers of intraspecifc ual swans observed within the same hour-long observa- and interspecifc interactions). For each of our focal spe- tion period. Preliminary comparisons of global models cies, separate tests were carried out for aggressive inter- using second-order Akaike’s Information Criteria (AICc; actions (i) targeted at other individuals, and (ii) received Burnham et al. 2011) showed that for the zero-infated from other individuals. Statistically signifcant diferences generalized linear mixed efects models of the Whooper between proportions were attributed where P < 0.05, Swan data, a negative binomial distribution in which the after P values had been adjusted using Holm-Bonferroni variance increased linearly with the mean (Brooks et al. corrections to account for multiple comparisons (Holm 2017) performed better than either a negative binomial 1979). Our binomial tests also allowed 95% confdence distribution in which the variance increased quadratically Wood et al. Avian Res (2020) 11:30 Page 5 of 16

with the mean (ΔAICc = 4.31), or a Poisson distribution model with one additional parameter was competitive (ΔAICc = 169.40). For Bewick’s Swans there was little dif- only if the associated AIC c value was lower than the ference between models with either of the two negative more parsimonious model (Arnold 2010). Three fur- binomial distributions (ΔAICc = 0.24), whilst the Poisson ther metrics for each model were used as indicators of distribution did not converge. Terefore, in all subse- the relative strength of support in the data, to facili- quent models we used the negative binomial distribution tate more detailed comparisons among our candidate in which the variance increased linearly with the mean. models: (i) the probability of a model being the best- To ensure that collinearity did not confound our mod- fitting model compared with the best-supported model elling (Dormann et al. 2013), we tested for covariance shown by AICc (Relative Likelihood RL), (ii) the ratio among our explanatory variables. To our knowledge, Var- of ΔAICc values for each model relative to the whole iance Infation Factors (VIFs), which are typically used to set of candidate models (Akaike weight wi), and (iii) identify collinear variables in linear models (Dormann how many more times less likely a model is to be the et al. 2013), are not currently available for ZIGLMMs, best-fitting model compared with the best-supported and so alternative methods were used. One-way Analy- model shown by AICc (Evidence Ratio ER) (Burnham sis of Variance (ANOVA) was used to test for covariance et al. 2011). In addition, to quantify the explanatory between our continuous variable (number of swans pre- power of each model (Mac Nally et al. 2018), the consent) and each of our categorical variables: time of day, ditional and marginal R2 values, which represented month, and winter. Signifcant covariance was inferred the proportion of the between-swan variance in the where statistically signifcant diferences in the mean time spent on aggression that was accounted for both number of swans present per categorical variable level the fixed and random effects combined and the fixed were detected. Values for the number of swans present effects alone, respectively (Nakagawa et al. 2017), were were square-root transformed so that model residuals calculated for each model using the sjstats R package satisfed the assumptions of the ANOVA tests (Zuur et al. (Lüdecke 2020). Finally, Tukey’s post hoc comparisons 2010). Associations between the frequencies of pairs of of the estimated marginal means of variables within categorical variables were tested using χ2 tests where all our best-supported models were carried out using the frequencies were ≥ 5, or Fisher’s exact test where one or emmeans R package (Lenth 2020). more frequencies were < 5 (Crawley 2013). Signifcant covariance between variables was inferred where statistically signifcant associations between the variables were Results found. Using these methods, we found for Whooper Intraspecifc versus interspecifc aggression Swans a signifcant efect of time of day on the num- For 13 out of 14 focal waterbird species, the major- ber of swans present (ANOVA: F2,39 = 47.27, P < 0.001). ity of aggressive behavioural interactions were given by, For Bewick’s Swans we detected covariance between and received from, conspecifcs (Table 1). Overall, we the number of swans present and both the time of day observed aggression by the three swan species towards (ANOVA: F2,279 = 12.24, P < 0.001) and winter (ANOVA: 9 of the 11 smaller waterbird species, whilst in turn F1,280 = 22.81, P < 0.001), as well as between winter and the swans received aggression from 8 of the 11 species month (Fisher’s exact test: P < 0.001); all other tests were (Table 1). non-signifcant (P > 0.05). Consequently, these collinear Te proportion of aggressive interactions directed by variables were not permitted within the same candidate swans towards conspecifcs ranged from 0.589 (95% CI models. Terefore, in total we ran 3 and 11 candidate 0.540–0.637) among Bewick’s Swans to 0.801 (0.733– models for the Whooper Swan and Bewick’s Swan data- 0.858) among Whooper Swans (Table 2; Fig. 1). Similarly, sets, respectively, accounting for all possible non-col- the proportion of aggressive interactions received by linear combinations of additive and two-way interactions swans from their conspecifcs ranged from 0.623 (0.555– between our explanatory variables. 0.687) for Whooper Swans up to 0.912 (0.880–0.938) We compared the relative support of each candi- among Mute Swans (Table 2; Fig. 1). For both the inter- date model using AICc, calculated using the MuMIn actions directed towards, and received from, conspecif- R package (Barton 2019). Typically the model with ics, the proportions of aggressive interactions that were the lowest AICc value is considered to be the best- intraspecifc were signifcantly greater for all three swan supported by the data, but we also considered models species (P ≤ 0.003 in all cases); i.e. intraspecifc interac- to be competitive where AIC c values < 6.0, following tions were more frequent than interspecifc interactions the advice of Richards (2008) for dealing with over- (Table 2; Fig. 1). Among the 11 species of smaller water- dispersion. Furthermore, to avoid selecting models birds, intraspecifc interactions directed towards other with uninformative parameters, we considered that a individuals were signifcantly more frequent in 8 species, Wood et al. Avian Res (2020) 11:30 Page 6 of 16

Table 1 The total numbers of aggressive interactions (n) Table 1 (continued) given and received by each focal waterbird species Focal species Aggression n Aggression from n Focal species Aggression n Aggression from n towards towards Mute Swan 3 Mute Swan 31 Bewick’s Swan Bewick’s Swan 245 Bewick’s Swan 245 Bewick’s Swan 2 Bewick’s Swan 11 Northern Pintail 37 Northern Pintail 11 Whooper Swan 1 Whooper Swan 7 Tufted Duck 32 Tufted Duck 3 Common Pochard 1 Common Pochard 0 Eurasian Coot 30 Eurasian Coot 4 Gull spp. 1 Gull spp. 0 Canada Goose 19 Canada Goose 16 Greylag Goose 0 Greylag Goose 2 Eurasian Teal 17 Eurasian Teal 1 Northern Pintail Northern Pintail 356 Northern Pintail 356 Northern Mallard 11 Northern Mallard 2 Eurasian Teal 38 Eurasian Teal 24 Common Moorhen 9 Common Moor- 1 Northern Mallard 26 Northern Mallard 56 hen Common Moorhen 22 Common Moor- 18 Gull spp. 9 Gull spp. 0 hen Greylag Goose 7 Greylag Goose 7 Bewick’s Swan 11 Bewick’s Swan 37 Mute Swan 0 Mute Swan 23 Eurasian Coot 8 Eurasian Coot 21 Mute Swan Mute Swan 364 Mute Swan 364 Tufted Duck 4 Tufted Duck 10 Whooper Swan 80 Whooper Swan 27 Canada Goose 1 Canada Goose 2 Northern Mallard 31 Northern Mallard 3 Mute Swan 0 Mute Swan 1 Bewick’s Swan 23 Bewick’s Swan 0 Greylag Goose 0 Greylag Goose 2 Canada Goose 19 Canada Goose 5 Eurasian Teal Eurasian Teal 146 Eurasian Teal 146 Northern Pintail 1 Northern Pintail 0 Northern Pintail 24 Northern Pintail 38 Gull spp. 1 Gull spp. 0 Eurasian Coot 10 Eurasian Coot 12 Whooper Swan Whooper Swan 137 Whooper Swan 137 Northern Mallard 4 Northern Mallard 6 Mute Swan 27 Mute Swan 80 Bewick’s Swan 1 Bewick’s Swan 17 Northern Mallard 7 Northern Mallard 1 Canada Goose 0 Canada Goose 1 Canada Goose 0 Canada Goose 2 Tufted Duck Tufted Duck 152 Tufted Duck 152 Canada Goose Canada Goose 190 Canada Goose 190 Northern Mallard 23 Northern Mallard 9 Northern Mallard 28 Northern Mallard 3 Northern Pintail 10 Northern Pintail 4 Bewick’s Swan 16 Bewick’s Swan 19 Bewick’s Swan 3 Bewick’s Swan 32 Mute Swan 5 Mute Swan 19 Eurasian Coot 3 Eurasian Coot 1 Northern Pintail 2 Northern Pintail 1 Common Pochard 1 Common Pochard 3 Whooper Swan 2 Whooper Swan 0 Canada Goose 0 Canada Goose 1 Tufted Duck 1 Tufted Duck 0 Greylag Goose 0 Greylag Goose 1 Eurasian Teal 1 Eurasian Teal 0 Common Pochard Common Pochard 10 Common Pochard 10 Greylag Goose 0 Greylag Goose 3 Tufted Duck 3 Tufted Duck 1 Greylag Goose Greylag Goose 39 Greylag Goose 39 Eurasian Coot 2 Eurasian Coot 0 Bewick’s Swan 7 Bewick’s Swan 7 Northern Mallard 0 Northern Mallard 1 Canada Goose 3 Canada Goose 0 Eurasian Coot Eurasian Coot 61 Eurasian Coot 61 Northern Mallard 2 Northern Mallard 0 Northern Pintail 20 Northern Pintail 8 Northern Pintail 2 Northern Pintail 0 Eurasian Teal 12 Eurasian Teal 10 Tufted Duck 1 Tufted Duck 0 Common Moorhen 4 Common Moor- 3 Common Shelduck Common Shelduck 2 Common Shelduck 2 hen Eurasian Coot 3 Eurasian Coot 0 Bewick’s Swan 3 Bewick’s Swan 30 Northern Mallard Northern Mallard 240 Northern Mallard 240 Northern Mallard 1 Northern Mallard 7 Northern Pintail 56 Northern Pintail 26 Tufted Duck 0 Tufted Duck 3 Tufted Duck 9 Tufted Duck 23 Common Pochard 0 Common Pochard 2 Eurasian Coot 7 Eurasian Coot 1 Common Shelduck 0 Common Shelduck 3 Eurasian Teal 6 Eurasian Teal 4 Common Moorhen Common Moorhen 102 Common Moor- 102 hen Common Moorhen 6 Common Moor- 3 hen Northern Pintail 18 Northern Pintail 22 Canada Goose 4 Canada Goose 28 Northern Mallard 3 Northern Mallard 6 Wood et al. Avian Res (2020) 11:30 Page 7 of 16

Table 1 (continued) Swans (mean = 0.411, 95% CI 0.363–0.460) than in Focal species Aggression n Aggression from n Mute Swans, Whooper Swans, Canada Geese, North- towards ern Mallard, Northern Pintail, Eurasian Teal, Tufted Eurasian Coot 3 Eurasian Coot 4 Duck, and Common Moorhen (Table 3). Te diference Bewick’s Swan 1 Bewick’s Swan 9 in the proportions of interspecifc aggression towards Gull spp. Gull spp. 19 Gull spp. 19 other species in Whooper Swans (mean = 0.199, 95% Bewick’s Swan 0 Bewick’s Swan 9 CI 0.142–0.267) and Eurasian Coots (mean = 0.396, Mute Swan 0 Mute Swan 1 95% CI 0.300–0.498) was close to signifcance, with Northern Mallard 0 Northern Mallard 1 an adjusted P value of 0.058. No other diferences between species were found to be statistically signifcant (Table 3). whilst intraspecifc interactions received from other Swan aggression times individuals were signifcantly more frequent in 7 species (Table 2; Fig. 1). Whooper Swans at WWT Caerlaverock spent a mean (± 95% CI) of 13.8 ± 4.7 s engaged in aggressive interac- Species comparisons of interspecifc aggression tions with other swans, based on the sample of 42 focal Te proportion of interspecifc aggressive interactions observations collected during November 2018; this dura- towards other birds was signifcantly greater in Bewick’s tion was equivalent to 2.3 ± 0.8% of their time-activity

Table 2 The intra- and interspecifc aggressive interactions given and received by each focal waterbird species

Species Interaction nIntra nInter nTotal PIntra PIntra 95% CI PInter PInter 95% CI P value

Bewick’s Swan To other 245 171 416 0.589 0.540–0.637 0.411 0.363–0.460 0.003 From other 245 68 313 0.783 0.733–0.827 0.217 0.173–0.267 < 0.001 Mute Swan To other 364 155 519 0.701 0.660–0.740 0.299 0.260–0.340 < 0.001 From other 364 35 399 0.912 0.880–0.938 0.088 0.062–0.120 < 0.001 Whooper Swan To other 137 34 171 0.801 0.733–0.858 0.199 0.142–0.267 < 0.001 From other 137 83 220 0.623 0.555–0.687 0.377 0.313–0.445 0.003 Canada Goose To other 190 55 245 0.776 0.718–0.826 0.224 0.174–0.282 < 0.001 From other 190 45 235 0.809 0.752–0.857 0.191 0.143–0.248 < 0.001 Greylag Goose To other 39 15 54 0.722 0.584–0.835 0.278 0.165–0.416 0.012 From other 39 7 46 0.848 0.711–0.937 0.152 0.063–0.289 < 0.001 Common Shelduck To other 2 3 5 0.400 0.053–0.853 0.600 0.147–0.947 1.000 From other 2 0 2 1.000 0.158–1.000 0.000 0.000–0.842 1.000 Northern Mallard To other 240 96 336 0.714 0.663–0.762 0.286 0.238–0.337 < 0.001 From other 240 136 376 0.638 0.587–0.687 0.362 0.313–0.413 < 0.001 Northern Pintail To other 356 110 466 0.764 0.723–0.802 0.236 0.198–0.277 < 0.001 From other 356 171 527 0.676 0.634–0.715 0.324 0.285–0.366 < 0.001 Eurasian Teal To other 146 39 185 0.789 0.723–0.846 0.211 0.154–0.277 < 0.001 From other 146 74 220 0.664 0.597–0.726 0.336 0.274–0.403 < 0.001 Tufted Duck To other 152 40 192 0.792 0.727–0.847 0.208 0.153–0.273 < 0.001 From other 152 51 203 0.749 0.683–0.807 0.251 0.193–0.317 < 0.001 Common Pochard To other 10 5 15 0.667 0.384–0.882 0.333 0.118–0.616 1.000 From other 10 2 12 0.833 0.516–0.979 0.167 0.021–0.484 0.270 Eurasian Coot To other 61 40 101 0.604 0.502–0.700 0.396 0.300–0.498 0.276 From other 61 66 127 0.480 0.391–0.571 0.520 0.429–0.609 1.000 Common Moorhen To other 102 25 127 0.803 0.723–0.868 0.197 0.132–0.277 < 0.001 From other 102 41 143 0.713 0.632–0.786 0.287 0.214–0.368 < 0.001 Gull spp. To other 19 0 19 1.000 0.824–1.000 0.000 0.000–0.176 < 0.001 From other 19 11 30 0.633 0.439–0.801 0.367 0.199–0.561 1.000

A comparison of the proportions of intraspecifc (PIntra) and interspecifc (PInter) aggressive interactions given and received by each focal waterbird species. All P values associated with our binomial tests were adjusted using Holm-Bonferroni corrections for multiple comparisons. Total numbers of interactions (n) are also indicated Wood et al. Avian Res (2020) 11:30 Page 8 of 16

(See fgure on next page.) Fig. 1 The proportions of intraspecifc and interspecifc aggressive interactions, directed towards or received from, other individuals. Error bars represent the 95% binomial CIs. Common Shelduck and Gull spp. are not shown due to low sample sizes. Statistical signifcance of diferences ± between intraspecifc and interspecifc proportions is shown in Table 2

budget. A comparison of candidate models showed that during observations made during the afternoons in the time spent by individual Whooper Swans in aggres- winter 2019/2020 than during either the afternoons sive interactions with other swans was best explained by of winter 2018/2019 or midday observations in win- the mean number of swans present during the observa- ter 2019/2020 (Table 6; Fig. 3); no other comparisons tion (Table 4). Tis model had the lowest AICc value and were signifcantly diferent. Tis model had the lowest accounted for approximately 51% of the total Akaike AIC c value and accounted for > 66% of the total Akaike weights (Table 4). In this model, the time spent in aggres- weights (Table 4). Overall, the efects of time of day and sive interactions by Whooper Swans increased with the winter together with the random efect accounted for mean number of swans present (Table 5; Fig. 2). Over- 34.3% of the variance in swan aggression times in total, all, the efect of swan numbers together with the ran- with the efect of time of day and winter accounting dom efect accounted for 35.1% of the variance in swan for 23.3% of the variance (Table 4). Crucially, this best- aggression times in total, with the efect of swan numbers supported model had a lower AIC c value than the null accounting for 19.3% of the variance. However, two other model (ΔAICc = 3.24; Table 4). A further three candi- models had associated AICc values within our threshold date models also had ΔAICc values within our thresh- of 6.0 of this best-supported model. Te null model, com- old of 6.0 of the minimum AIC c value, although none of prised of only an intercept and random efect term, had a these performed better than the null model (Table 4). ΔAICc value of 0.45 and accounted for c. 40% of the total Two of these three models comprised the single addi- Akaike weights (Table 4). As the model containing the tive efects contained within our best-supported model, efect of swan numbers performed only marginally better time of day (ΔAICc = 5.94) and winter (ΔAICc = 5.31). than the null model, this suggests that the efect of swan Te third comprised the numbers of swans present dur- numbers was not strong and had limited explanatory ing the observation (ΔAICc = 4.74), but the support power, and hence an efect of swan numbers should be for this model was weak. Te model of the numbers interpreted cautiously. Finally, a model in which aggres- of swans accounted for only c. 6% of the total Akaike sion time varied with the time of day of the observation weights and the evidence ratio value indicated that had an associated ΔAICc value of 3.42; however, the over- it was > 10.68 times less likely to be best-ftting model all support for this model was weak, as it accounted for compared with the lowest AICc model (Table 4). As a only c. 9% of the total Akaike weights and the evidence fxed efect the number of swans accounted for only ratio value indicated that it was > 5.5 times less likely to 0.7% of the variance in the time spent in aggressive be best-ftting model compared with the lowest AIC c interactions (Table 4; Fig. 2), and so overall we con- model (Table 4). Moreover, the model containing time sidered that there was little evidence that the numbers of day performed less well than the null model (Table 4), of swans present had an efect on the time spent on and so overall we considered that there was little evi- aggression by Bewick’s Swans. dence that Whooper Swan aggression time varied among the three times of day (Fig. 3). Discussion Bewick’s Swans at WWT Slimbridge spent a mean In accordance with our conspecifc hypothesis, we found ( 95% CI) of 1.4 0.3 s engaged in aggressive interac- ± ± that aggressive interactions by swans were typically tions with other swans, based on the sample of 282 focal directed towards other swans rather than smaller water- observations collected during winters 2018/2019 and birds. Indeed, across all three swan species intraspe- 2019/2020; this duration was equivalent to 0.2 0.1% ± cifc aggression accounted for between 59 and 80% of of their time-activity budget. Comparison of our candi- all aggressive interactions directed at other individuals. date models revealed that the time spent by individual Previous studies of Mute Swans by Conover and Kania Bewick’s Swans in aggressive interactions with other (1994) and Włodarczyk and Minias (2015) found simi- swans was best explained by an interaction between the larly that 47% and 80%, respectively, of all aggressive time of day and the winter of observation (Tables 4, 5). behaviours were directed towards conspecifcs. Our data Post-hoc testing indicated that swan aggression times showed that all three swan species received the greatest varied between time of day and the winter of obser- proportion of interspecifc aggression from another swan vation such that swans spent less time on aggression Wood et al. Avian Res (2020) 11:30 Page 9 of 16 Wood et al. Avian Res (2020) 11:30 Page 10 of 16 Gull spp. 0.067 0.763 1.000 1.000 1.000 0.328 1.000 1.000 1.000 1.000 1.000 0.150 1.000 – Common Common Moorhen 0.002 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 0.122 – 3.23 Eurasian Eurasian Coot 1.000 1.000 0.058 0.140 1.000 1.000 1.000 0.116 0.106 0.081 1.000 – 10.00 9.58 Common Common Pochard 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 – 0.79 0.03 5.01 Tufted Tufted Duck 0.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 – 0.65 10.82 0.01 3.62 test statistic, while the values in the upper right triangle represent the triangle in the upper right represent while the values statistic, test 2 Eurasian Eurasian Teal 0.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 – < 0.01 0.60 10.31 0.02 3.68 Northern Pintail 0.000 1.000 1.000 1.000 1.000 1.000 1.000 – 0.35 0.45 0.32 10.11 0.66 4.53 Northern Mallard 0.040 1.000 1.000 1.000 1.000 1.000 – 2.27 3.11 3.43 0.01 3.91 3.32 6.06 Common Common Shelduck 1.000 1.000 1.000 1.000 1.000 – 1.08 1.87 2.32 2.39 0.28 0.19 2.58 8.12 Greylag Greylag Goose 1.000 1.000 1.000 1.000 – 0.98 < 0.01 0.26 0.72 0.80 0.01 1.66 1.01 5.05 Canada Canada Goose 0.000 1.000 1.000 – 0.44 2.06 2.45 0.06 0.05 0.08 0.43 9.72 0.23 4.11 Whooper Whooper Swan 0.000 1.000 – 0.26 1.07 2.60 4.04 0.79 0.02 0.01 0.80 11.49 < 0.01 3.35 Mute Swan Mute 0.037 – 5.95 4.23 0.03 0.94 0.11 4.58 4.84 5.30 < 0.01 3.28 4.77 6.58 Bewick’s Bewick’s Swan – 12.36 23.09 23.03 3.02 0.16 12.21 30.21 21.72 22.94 0.11 0.03 18.44 11.20 Statistical comparison of the proportion of aggressive interactions that were interspecifc that were interactions of the proportion comparison of aggressive Statistical Swan Swan Goose Goose Shelduck Mallard Pintail Teal Pochard Coot Moorhen adjusted P values adjusted 3 Table the χ left triangle represent in the lower Values 2 . Table as reported in other species, proportions conducted on interspecifc towards Binomial tests Bewick’s Bewick’s Mute Swan Mute Whooper Canada Canada Greylag Greylag Common Common Northern Northern Eurasian Eurasian Tufted Duck Tufted Common Common Eurasian Eurasian Common Common Gull spp. Wood et al. Avian Res (2020) 11:30 Page 11 of 16

Table 4 Comparison of models of the time spent in aggressive interactions between swans 2 2 Species Model k AICc ΔAICc RL wi ER Rc Rm

Whooper Swans i N (i|D) 2 295.87 0.00 1.00 0.505 1.00 0.351 0.193 + + i (i|D) 1 296.31 0.45 0.80 0.404 1.25 0.215 0.000 + i T (i|D) 4 299.29 3.42 0.18 0.091 5.53 0.142 0.129 + + Bewick’s Swans i T * W (i|D) 12 802.43 0.00 1.00 0.664 1.00 0.343 0.233 + + i (i|D) 1 805.67 3.24 0.20 0.131 5.06 0.226 0.000 + i N (i|D) 2 807.16 4.74 0.09 0.062 10.68 0.236 0.007 + + i W (i|D) 3 807.74 5.31 0.07 0.047 14.26 0.226 0.000 + + i T (i|D) 4 808.37 5.94 0.05 0.034 19.51 0.211 0.017 + + i M (i|D) 5 808.80 6.37 0.04 0.027 24.16 0.224 0.030 + + i T W (i|D) 6 810.44 8.01 0.02 0.012 54.83 0.213 0.019 + + + i N M (i|D) 6 810.74 8.31 0.02 0.010 63.75 0.231 0.030 + + + i T M (i|D) 8 811.00 8.57 0.01 0.009 72.75 0.198 0.070 + + + i N * M (i|D) 10 813.40 10.97 0.00 0.003 241.39 0.195 0.071 + + i T * M (i|D) 20 820.81 18.38 0.00 0.000 9786.90 0.216 0.091 + + A summary of the relative support for each of our candidate models of the time spent by swans in aggressive interactions with other swans. Model parameters: intercept (i), number of swans present (N), month (M), winter (W), time of day (T), and observation identity (D). k refers to the number of fxed efects in the model

Table 5 Efect sizes for our best-supported models of Whooper and Bewick’s Swan time spent on aggression Species Model Parameter Estimate SE Variance SD

Whooper Swans Conditional i 1.68 0.70 – – N 0.04 0.02 – – (i|D) – – 0.17 0.41 Zero-infation i –1.79 0.82 – – Bewick’s Swans Conditional i 1.09 0.19 – –

T(Mid) –0.29 0.29 – –

T(P.M.) 0.32 0.24 – –

W(2019/2020) 0.27 0.66 – –

T(Mid):W(2019/2020) 0.66 0.82 – –

T(P.M.):W(2019/2020) –2.59 1.03 – – (i|D) – – 0.15 0.38 Zero-infation i 0.33 0.16 – – The mean and SE estimated efect sizes associated with each of the parameters (as defned in Table 4) in our best-supported models of Whooper and Bewick’s Swan time spent on aggression. Additionally, the variance (and SD) associated with the random efect of observation identity is also shown species. For Mute Swans and Whooper Swans, the pro- observed in aggressive interactions with swans, which portion of aggressive interactions directed towards other may have been due to their low relative abundance at birds did not difer from the values observed for smaller the sites. Given that even the low incidences of aggres- waterbirds, although Bewick’s Swans showed higher sion observed in our study could carry the risk of seri- values than 8 of the 13 other species. Taken together, ous injury or death, it may seem counter intuitive that our results were consistent with previous fndings that smaller waterbirds are so often observed to share habitat aggression is more likely between individuals with greater with swans. Terefore, smaller waterbirds must balance overlap in resource use (Peiman and Robinson 2010). potential risks of aggression with the possible benefts of Although the majority of aggressive interactions by sharing habitat or even of associating more closely with swans were directed towards, and received from, conspe- swans. Previous studies have found that some smaller cifcs, we observed some aggression towards 9 of the 11 waterbird species associate with swans for improved smaller waterbird species present at the two sites. Only access to food resources, such as submerged aquatic Common Pochard and Common Shelduck were not plants that swans bring to the surface (e.g. Bailey and Wood et al. Avian Res (2020) 11:30 Page 12 of 16

Fig. 2 The relationships between the number of swans present and Fig. 3 The mean ( 95% CI) duration that swans spent engaged in the time spent in aggressive interactions with other swans. The data ± aggressive interactions with other swans for each a winter month points are indicated by the solid circles, whilst the solid and dashed and b time of day. Data for winters 2018/2019 and 2019/2020 are lines represent the mean and 95% CI relationships estimated by our presented separately best-supported model, where evidence for such an efect was found (where no such efect was detected, no line is presented)

indicated that swan numbers accounted for only 19.3% Batt 1974; Beven 1980; Källander 2005). Foraging swans of the variance in Whooper Swan aggression times. typically bring to the surface more food than they ingest Yet in contrast, Bewick’s Swan aggression showed no (Gillham 1956), which consequently provides foraging efect of swan numbers. Te reason for these divergent opportunities for other species. Gyimesi et al. (2012) fndings may be due to the diferences in the ranges of found that commensal foraging with Bewick’s Swans swan numbers recorded at both sites. Te mean swan doubled the instantaneous food intake rate of Common numbers recorded during observations at WWT Caer- Pochard. In addition to commensal feeding associations, laverock, where large numbers of Whooper and Mute some smaller waterbird species such as Eurasian Coot Swans overwinter, ranged between 9 and 46 individuals, may also feed directly on swan faeces (Vogrin 1997; Shi- whilst the mean recorded swan numbers at WWT Slim- mada 2012). bridge ranged between 1 and 13 individuals; hence, the We found mixed support for our density hypothesis number of potential competitors faced by the Whooper regarding the infuence of swan numbers on the dura- Swans at WWT Caerlaverock was markedly higher tion of aggressive behaviours. As expected, individual than that faced by the Bewick’s Swans at WWT Slim- Whooper Swans spent longer in aggressive interactions bridge (Fig. 2). Aggressive behaviours among birds typi- when more swans were present. However, the relation- cally become more common as the density of individuals ship between swan numbers and Whooper Swan aggres- within a habitat increases (Metcalfe and Furness 1987; sion times was not strong, as the marginal R2 value Wood et al. 2015). It is possible that Bewick’s Swans Wood et al. Avian Res (2020) 11:30 Page 13 of 16

Table 6 Post-hoc contrasts associated with our between the three times of day was observed. For our best-supported model of Bewick’s Swan time spent focal Bewick’s Swans, for which data could be collected on aggression over multiple months in two winters, we found evi- Contrast Estimate SE t ratio P value dence that the time devoted to aggression varied both between times of day and between winters. Te tendency T :W T :W 0.29 0.29 1.02 0.912 (A.M.) (2018/2019) − (Mid) (2018/2019) for Bewick’s Swans to spend less time on aggression T :W T :W – 0.32 0.24 – 1.35 0.756 (A.M.) (2018/2019) − (P.M.) (2018/2019) with other swans during the afternoon observations in T :W T :W – 0.27 0.66 – 0.41 0.999 (A.M.) (2018/2019) − (A.M.) (2019/2020) 2019/2020 could be due to the swans spending less time T :W T :W – 0.63 0.46 – 1.38 0.739 (A.M.) (2018/2019) − (Mid) (2019/2020) foraging, and hence less aggression linked to competition T :W T :W 2.01 0.81 2.49 0.131 (A.M.) (2018/2019) − (P.M.) (2019/2020) for food resources. Previous studies of swan behaviour T :W T :W – 0.61 0.28 – 2.21 0.239 (Mid) (2018/2019) − (P.M.) (2018/2019) have shown that the most intensive foraging periods for T :W T :W – 0.56 0.67 – 0.84 0.961 (Mid) (2018/2019) − (A.M.) (2019/2020) swans typically occur in early morning and before dusk T :W T :W – 0.93 0.48 – 1.93 0.388 (Mid) (2018/2019) − (Mid) (2019/2020) (e.g. Bowler 1996), and thus the afternoon represents a T :W T :W 1.71 0.82 2.10 0.293 (Mid) (2018/2019) − (P.M.) (2019/2020) period of low foraging activity. Future research could test T :W T :W 0.05 0.65 0.08 1.000 (P.M.) (2018/2019) − (A.M.) (2019/2020) for a correlation between foraging activity and aggres- T :W T :W – 0.31 0.45 – 0.70 0.982 (P.M.) (2018/2019) − (Mid) (2019/2020) sive interactions by collecting data on all behaviours as T :W T :W 2.33 0.80 2.91 0.045 (P.M.) (2018/2019) − (P.M.) (2019/2020) part of a time-activity budget. Our repeated 10-min focal T :W T :W – 0.37 0.76 – 0.48 0.997 (A.M.) (2019/2020) − (Mid) (2019/2020) observations showed that Whooper and Bewick’s Swans T :W T :W 2.27 1.01 2.25 0.217 (A.M.) (2019/2020) − (P.M.) (2019/2020) spent means ( 95% CI) of 13.8 4.7 s and 1.4 0.3 s, T :W T :W 2.64 0.90 2.95 0.040 ± ± ± (Mid) (2019/2020) − (P.M.) (2019/2020) respectively, engaged in aggression with other swans. Parameters are as defned in Table 4. Statistically signifcant contrasts are in Tese durations were equivalent to 2.3% and 0.2% of italics the Whooper and Bewick’s Swan time-activity budgets, respectively. In this study we focused our explanatory modelling on the interactions between swans, as for all would show increasing durations of aggression with ris- swan species the interactions between swans were more ing swan numbers if observations could be made over a frequent than interactions between swans and smaller greater range of swan numbers; although with ongoing waterbirds. However, future research could extend our declines in Bewick’s Swan numbers at wintering sites in methodology to examine the time spent by swans on the UK (Beekman et al. 2019), obtaining such data may interactions with all waterbird species, as well as the prove challenging. behavioural context of aggression (for example, whether Further comparisons of our models of the time spent the aggression occurred while both individuals were for- in aggressive interactions with other swans found no aging). Types of food resources represent another poten- support for our winter decline hypothesis, namely that tial area for further investigations. At our sites the swans swans would spend more time engaged in aggressive fed on natural vegetation, supplemented by some wheat interactions in early winter months compared with later grains which were provided as part of a public engage- months. Te more limited data collected for Whooper ment programme (Black and Rees 1984). At other sites Swans did not allow between-month efects to be tested swans are known to feed on food items that are buried for, and so only our Bewick’s Swan data could be used to in aquatic and terrestrial sediment, including pond weed test this hypothesis. Our results contrasted with those tubers (Potamogeton spp.) and root crops such as Sugar of Scott (1981), who found that the frequency of aggres- Beet (Beta vulgaris) (Wood et al. 2019b). Te distribu- sive interactions between Bewick’s Swans at WWT tions of such cryptic food items are more difcult for Slimbridge declined progressively over winter months. the birds to predict, and there may be a higher perceived Te number of Bewick’s Swans overwintering at WWT value of defending proftable feeding patches from com- Slimbridge was lower in our study winters than in the late petitors. Future research could also therefore assess how 1970s and early 1980s (Beekman et al. 2019), and so the the frequency of aggressive behaviours responds to dif- divergent fndings may refect lower levels of competition ferences in the types of food resources that are available. in recent winters. Our study provided an example of how questions Te data that we collected showed that the times spent relating to avian behaviour could be answered using by swans in aggressive interactions with other swans data that were collected remotely via live-streaming showed some variation between time of day and between webcams. Such remote data collection ofers several winters for Bewick’s Swans. Te more limited data col- advantages to researchers, including less disturbance lected for Whooper Swans did not allow between-winter to the focal birds (once the camera has been installed), efects to be tested for, although no consistent variation lower environmental costs (i.e. carbon footprint Wood et al. Avian Res (2020) 11:30 Page 14 of 16

associated with sampling) due to not having to under- (Rees et al. 2019). However, our fndings show that take visits to study sites, greater accessibility of research aggressive interactions from Mute Swans accounted for to scientists who cannot physically travel to study sites only 7% of all of the aggressive interactions received, (either due to logistical difculties or disability), and whereas intraspecifc aggression from other Bewick’s the facilitation of citizen science programmes (Eichorst Swans represented 78% of aggression received. Given 2018; Schulwitz et al. 2018). Given these advantages, these fndings, it appears unlikely that aggression from we expect that remote data collection methods will Mute Swans has contributed to the observed decline in become increasingly popular with researchers. How- Bewick’s Swan winter numbers. Similar research is now ever, our study also highlights a key drawback of such needed at other sites used by Bewick’s Swans, includ- remote data collection, namely the reliability of the ing migratory stopover sites and those in the breeding technology involved. Malfunctions and environmental range. damage to the webcams limited the collection of data at both of our study sites, although this did not prevent us Supplementary information from addressing the key questions of our study. Future Supplementary information accompanies this paper at https://doi. org/10.1186/s40657-020-00216-7. studies that aim to use remote data collection methods should consider carefully the reliability of both the Additional fle 1: Figure S1. Screenshots of the live webcam footage, cameras themselves as well as the stable internet con- illustrating the feld-of-view of the webcams at (a) WWT Slimbridge and nections required to stream the camera videos. For (b) WWT Caerlaverock. example, the use of multiple webcams at a single site would provide a bufer against the impacts of the fail- Acknowledgements ure of a single camera. We also believe that waterbirds We are grateful for the contributions of Anna Allis and Esenya Mathews in are useful focal species with which to test remote data observing the swans and helping to collect the fnal dataset. The editor and two anonymous reviewers provided valuable feedback that helped us to collection methods, as the birds typically have relatively improve this manuscript. We thank Dave Paynter, Paul Marshall, John Herivel, large body sizes and use open habitat which provides and Samantha Luxa for assistance with the webcams. researchers with the unobstructed views required to Authors’ contributions make reliable identifcations of species and accurate KAW and PER conceived and designed the study. PH, JS and EW collected the assessments of behaviour (Anderson et al. 2011; Peluso data. KAW carried out the analyses. All authors contributed to the writing of et al. 2013). the manuscript. All authors read and approved the fnal manuscript. Funding This work was supported by the Wildfowl & Wetlands Trust and the University Conclusions of Exeter.

Our study illustrates how detailed behavioural inves- Availability of data and materials tigations can help to improve our understanding of The datasets generated and analysed during the current study are available the prevalence of aggressive interactions within and from the corresponding author upon reasonable request. between species. Spatially- and temporally-replicated Ethics approval and consent to participate data can allow researchers to identify which species are The procedures in this study comply with the current laws of the United King- most commonly involved in aggressive interactions, as dom, where they were performed. This study was carried out with the prior approval of the ethics committee of the College of Life and Environmental well as the frequency and direction of these interac- Sciences of the University of Exeter (eCLESPsy000890v3.3). tions. Our fndings that most aggression was intraspecifc and accounted for a low proportion of the total Consent for publication Not applicable. time-budget, together with the lack of strong density- dependence, suggest an absence of any conservation Competing interests or management issues related to aggression between The authors declare that they have no competing interests. waterbirds at either site. For example, the behavioural Author details data show that Common Pochard, a species listed as 1 Wildfowl & Wetlands Trust, Slimbridge Wetland Centre, Slimbridge, Glouces- 2 Vulnerable that is undergoing declining population tershire GL2 7BT, UK. Centre for Research in Animal Behaviour, Psychology, Washington Singer, University of Exeter, Perry Road, Exeter, Devon EX4 4QG, size (Brides et al. 2017), show relatively few aggres- UK. 3 University Centre Sparsholt College, Sparsholt, Hampshire SO21 2NF, UK. sive interactions with other waterbirds and none with swans. Conservationists have questioned whether the Received: 29 June 2020 Accepted: 2 August 2020 observed c. 39% decline in winter Bewick’s Swan numbers in north-west Europe between 1995 and 2010 may have been at least partially attributable to competition with rising numbers of Whooper and Mute Swans Wood et al. Avian Res (2020) 11:30 Page 15 of 16

References Gyimesi A, Stillman RA, Nolet BA. Cryptic interference competition in swans Altmann J. Observational study of behavior: sampling methods. Behaviour. foraging on cryptic prey. Anim Behav. 2010;80:791–7. 1974;49:227–66. Gyimesi A, van Lith B, Nolet BA. Commensal foraging with Bewick’s Swans Amat JA. Food usurpation by waterfowl and waders. Wildfowl. 1990;41:107–16. Cygnus bewickii doubles instantaneous intake rate of Common Pochard Anderson MJ, Urbine JL, Wilson C, Callabro L. Employment of web-based Aythya ferina. Ardea. 2012;100:55–62. images and a live web cam in the examination of lateral neck-resting Holm S. A simple sequentially rejective multiple test procedure. Scand J Stat. preferences in the American famingo (Phoenicopterus ruber). J Caribb 1979;6:65–70. Ornithol. 2011;24:41–7. Johnsgard PA. Handbook of waterfowl behavior. New York: Cornell University Arnold TW. Uninformative parameters and model selection using Akaike’s Press; 1965. Information Criterion. J Wildl Manage. 2010;74:1175–8. Källander H. Commensal association of waterfowl with feeding swans. Water- Bailey RO, Batt BDJ. Hierarchy of waterfowl feeding with Whistling Swans. Auk. birds. 2005;28:326–31. 1974;91:488–93. King JA. The ecology of aggressive behavior. Annu Rev Ecol Syst. Barton K. MuMIn: Multi-Model Inference. Version 1.43.15. 2019. https://CRAN.R- 1973;4:117–38. project.org/package MuMIn. Lenth R. Emmeans: Estimated Marginal Means, aka Least-Squares Means. Beekman J, Kofjberg K, Wahl= J, Kowallik C, Hall C, Devos K, et al. Long-term Version 1.4.5. 2020. https://CRAN.R-project.org/package emmeans. population trends and shifts in distribution for Bewick’s Swans Cygnus Lind H. The rotation display of the Mute Swan Cygnus olor: synchronised= columbianus bewickii wintering in northwest Europe. Wildfowl. 2019; neighbour responses as instrument in the territorial defence strategy. Special Issue 5:73–102. Ornis Scand. 1984;15:98–104. Beven G. Coot feeding on weed disturbed by Mute Swans. Brit Birds. Lüdecke D. Sjstats: Statistical Functions for Regression Models. Version 1980;73:219–20. 0.17.9. 2020. https://CRAN.R-project.org/package sjstats. Black JM, Rees EC. The structure and behaviour of the Whooper Swan popula- Lumsden HG. Trumpeter Swans and Mute Swans compete= for space in tion wintering at Caerlaverock, Dumfries and Galloway, Scotland: an Ontario. Ontario Birds. 2016;34:14–23. introductory study. Wildfowl. 1984;35:21–36. Mac Nally R, Duncan RP, Thomson JR, Yen JD. Model selection using Bowler JM. Feeding strategies of Bewick’s Swans (Cygnus columbianus bewickii) information criteria, but is the “best” model any good? J Appl Ecol. in winter. PhD Thesis. Bristol: University of Bristol; 1996. 2018;55:1441–4. Brazil MA. A case of unusual aggression by a Whooper Swan. Tori. 1983;32:155. Marchowski D, Neubauer G. Kleptoparasitic strategies of Mallards towards Brides K, Wood KA, Hearn RD, Fijen TPM. Changes in the sex ratio of the conspecifcs and Eurasian Coots. Ardea. 2019;107:110–4. Common Pochard Aythya ferina in Europe and North Africa. Wildfowl. Metcalfe NB, Furness RW. Aggression in shorebirds in relation to fock den- 2017;67:100–12. sity and composition. Ibis. 1987;129:553–63. Brooks ME, Kristensen K, van Benthem KJ, Magnusson A, Berg CW, Nielsen A, Nakagawa S, Johnson PC, Schielzeth H. The coefcient of determination R2 et al. glmmTMB balances speed and fexibility among packages for zero- and intra-class correlation coefcient from generalized linear mixed- infated generalized linear mixed modeling. R J. 2017;9:378–400. efects models revisited and expanded. J R Soc Int. 2017;14:20170213. Burgess RM, Stickney AA. Interspecifc aggression by Tundra Swans Newth JL, McDonald RA, Wood KA, Rees EC, Semenov I, Chistyakov A, et al. towards Snow Geese on the Sagavanirktok River Delta, Alaska. Auk. Predicting intention to hunt protected wildlife: a case study of Bewick’s 1994;111:204–7. swans in the European Russian Arctic. Oryx. (in press). Burnham KP, Anderson DR, Huyvaert KP. AIC model selection and multimodel O’Hare MT, Stillman RA, McDonnell JO, Wood LR. Efects of mute swan graz- inference in behavioral ecology: some background, observations, and ing on a keystone macrophyte. Freshwater Biol. 2007;52:2463–75. comparisons. Behav Ecol Sociobiol. 2011;65:23–35. Peiman K, Robinson B. Ecology and evolution of resource-related hetero- Clopper CJ, Pearson ES. The use of confdence or fducial limits illustrated in specifc aggression. Quart Rev Biol. 2010;85:133–58. the case of the binomial. Biometrika. 1934;26:404–13. Pelligrini AD. The roles of aggressive and afliative behaviors in resource Conover MR, Kania GS. Impact of interspecifc aggression and herbivory by control: a behavioral ecological perspective. Dev Rev. 2008;28:461–87. mute swans on native waterfowl and aquatic vegetation in New England. Peluso AI, Royer EA, Wall MJ, Anderson MJ. The relationship between envi- Auk. 1994;111:744–8. ronmental factors and famingo aggression examined with internet Crawley MJ. The R Book. 2nd ed Chichester: Wiley; 2013. resources. Avian Biol Res. 2013;6:215–20. Davis JB, Guillemain M, Kaminski RM, Arzel C, Eadie JM, Rees EC. Habitat and Pöysä H. Resource utilization pattern and guild structure in a waterfowl resource use by waterfowl in the northern hemisphere in autumn and community. Oikos. 1983;40:295–307. winter. Wildfowl. 2014;4:17–69. R Core Team. R: A language and environment for statistical computing. Delacour J. Waterfowl in large mixed collections. Int Zoo Yearb. 1973;13:15–9. [3.6.3]. Vienna, Austria: R Foundation for Statistical Computing; 2020. Dormann CF, Elith J, Bacher S, Buchmann C, Carl G, Carré G, et al. Collinearity: a Rees EC. Bewick’s Swan. London: T & AD Poyser; 2006. review of methods to deal with it and a simulation study evaluating their Rees EC, Cao L, Clausen P, Coleman JT, Cornely J, Einarsson O, et al. Conser- performance. Ecography. 2013;36:27–46. vation status of the world’s swan populations, Cygnus sp. and Cosc- Eichorst B. Internet webcams provide opportunities for college student oroba sp.: a review of current trends and gaps in knowledge. Wildfowl. research on animal behavior and ecology: an example with birds. Am Biol 2019; Special Issue 5:35–72. Teacher. 2018;80:680–5. Richards SA. Dealing with overdispersed count data in applied ecology. J Ely CR, Budeau DA, Swain UG. Aggressive encounters between Tundra Appl Ecol. 2008;45:218–27. Swans and Greater White-fronted Geese during brood rearing. Condor. RSPB. Birds A- Z: Bird Guides. Royal Society for the Protection of Birds. 1987;89:420–2. https://www.rspb.org.uk/birds-and-wildlife/wildlife-guides/bird-a-z/. Frost TM, Calbrade NA, Birtles GA, Mellan HJ, Hall C, Robinson AE, et al. Water- Accessed October 2018. birds in the UK 2018/2019: The Wetland Bird Survey. Thetford: BTO/RSPB/ Schulwitz SE, Spurling DP, Davis TS, McClure CJW. Webcams as an untapped JNCC; 2020. opportunity to conduct citizen science: siz years of the American Kes- Gayet G, Guillemain M, Mesléard F, Fritz H, Vaux V, Broyer J. Are Mute Swans trel Partnership’s KestrelCam. Glob Ecol Conserv. 2018;15:e00434. (Cygnus olor) really limiting fshpond use by waterbirds in the Dombes, Scott DK. Social behaviour of wintering Cygnus columbianus bewickii. In: Eastern France? J Ornithol. 2011;152:45–53. Matthews GVT, Smart M, editors. Proceedings of the Second Inter- Gayet G, Calenge C, Broyer J, Mesléard F, Vaux V, Fritz H, et al. Analysis of spatial national IWRB Swan Symposium, Sapporo, Japan, 1980. Slimbridge: point pattern shows no desertion of breeding Mute Swan areas by the International Waterfowl Research Bureau; 1981. p. 211–25. other waterbirds within fshpond. Acta Ornithol. 2016;5:151–63. Shimada T. Ducks foraging on swan faeces. Wildfowl. 2012;62:224–7. Gillham ME. Feeding habits and seasonal movements of mute swans on two Sladen WJL. Swans should not be hunted. In: Sears J, Bacon PJ, editors. south Devon estuaries. Bird Study. 1956;3:205–12. Proceedings of the Third International IWRB Swan Symposium, Oxford, Gurtovaya EN. Aggressive interactions between Bewick’s Swans and other 1989. Slimbridge: International Waterfowl Research Bureau; 1991. p. Anseriformes in the breeding period. Casarca. 2000;6:167–76. 368–75. Wood et al. Avian Res (2020) 11:30 Page 16 of 16

Stone WB, Marsters AD. Aggression among captive Mute Swans. New York Wood KA, Ponting J, D’Costa N, Newth JL, Rose PE, Glazov P, et al. Understand- Fish Game J. 1970;17:51–3. ing intrinsic and extrinsic drivers of aggressive behaviour in waterbird Tatu KS, Anderson JT, Hindman LJ, Seidel G. Diurnal foraging activities of assemblages: a meta-analysis. Anim Behav. 2017;126:209–16. mute swans in Chesapeake Bay, Maryland. Waterbirds. 2007;30:121–9. Wood KA, Cao L, Clausen P, Ely CR, Luigujõe L, Rees EC, et al. Current trends Therres GD, Brinkler DF. Mute Swan interactions with other birds in Chesa- and future directions in swan research: insights from the 6th International peake Bay. In: Perry MC, editor. Mute Swans and Their Chesapeake Bay Swan Symposium. Wildfowl. 2019a; Special Issue 5:1–34. Habitats: Proceedings of a Symposium. Virginia: US Geological Survey; Wood KA, Hilton GM, Newth JL, Rees EC. Seasonal variation in energy gain 2004. p. 43–6. explains patterns of resource use by avian herbivores in an agri- Tingay A. Aggression in the Black Swan. Emu. 1974;74:35–8. cultural landscape: insights from a mechanistic model. Ecol Model. Vogrin M. A Coot Fulica atra eating waterfowl droppings. Butll GCA. 2019b;409:108762. 1997;14:63–4. Zuur AF, Ieno EN, Elphick CS. A protocol for data exploration to avoid common Włodarczyk R, Minias P. Division of parental duties confrms a need for bi- statistical problems. Methods Ecol Evol. 2010;1:3–14. parental care in a precocial bird, the mute swan Cygnus olor. Anim Biol. 2015;65:163–76. Wood KA, Stillman RA, Goss-Custard JD. The efect of kleptoparasite and host numbers on the risk of food-stealing in an avian assemblage. J Avian Biol. 2015;46:589–96.

Ready to submit your research ? Choose BMC and benefit from:

• fast, convenient online submission • thorough peer review by experienced researchers in your ﬁeld • rapid publication on acceptance • support for research data, including large and complex data types • gold Open Access which fosters wider collaboration and increased citations • maximum visibility for your research: over 100M website views per year

At BMC, research is always in progress.

Learn more biomedcentral.com/submissions