This article delves into a study conducted at two prominent UK wetland centers: the Wildfowl & Wetland Trust (WWT) Centre reserves at Slimbridge in southwest England and Caerlaverock in southwest Scotland. The research aimed to understand aggression patterns among wintering waterbirds, with a particular focus on swans. Both sites provide a diverse array of aquatic and terrestrial habitats, serving as crucial feeding and roosting grounds for various waterbird species, including swans, geese, and numerous types of ducks. Analyzing behavioral data collected through remote observation, the study sought to test hypotheses related to conspecific aggression, the impact of density, and seasonal variations in aggressive interactions.
Study System and Data Collection
The study sites, Slimbridge and Caerlaverock, are characterized by small lakes that attract a significant number of waterbirds during winter. In recent winters, Slimbridge has hosted both Bewick’s Swans and Mute Swans, while Caerlaverock has been a haven for Mute and Whooper Swans. Population estimates from 2014/2015 to 2018/2019 indicated substantial numbers of these species, with mean peak winter counts reaching 149 Bewick’s Swans and 441 Mute Swans at Slimbridge, and 337 Whooper Swans and 74 Mute Swans at Caerlaverock. These swans share their environment with a variety of other waterbirds, including Canada Geese, Greylag Geese, dabbling ducks like Mallards and Teals, diving ducks such as Tufted Ducks and Pochards, Rallidae like Coots and Moorhens, Common Shelducks, and various gull species. Mute Swans are year-round residents at both locations, while Bewick’s Swans typically winter at Slimbridge from November to February, and Whooper Swans are present at Caerlaverock from October to March.
Data collection employed two concurrent methods over one-hour observation periods. The first method, all-occurrence sampling, recorded every instance of aggression between any two waterbird individuals to investigate the “conspecific hypothesis.” An ethogram of aggressive behaviors, developed from preliminary observations, included bill strikes, wing or body lunges, and chasing. The species of both the aggressor and the opponent were meticulously recorded, identified by size, shape, and plumage characteristics, aided by an online photo-identification guide.
The second method, focal sampling, was used to gather data for the “density” and “winter decline hypotheses,” specifically focusing on aggression among swans. During each hour-long observation, a single swan was randomly selected, and the duration of its aggressive interactions with other swans was recorded using a stopwatch over a 10-minute interval. This allowed for the observation of six individual swans per hour. A 10-minute focal observation duration was chosen for comparability with previous studies. The total number of swans present during each observation period was also counted, with the mean taken as the representative number for that observation.
All observations were conducted remotely using live-streaming AXIS Q6035-E PTZ Dome Network Cameras positioned on the shore, facing the study lakes. The cameras maintained a consistent zoom level, standardizing the field-of-view across all observation periods. While the exact number of birds outside the camera’s view is unknown, it is presumed to be low given that the cameras covered significant portions of the lakes. Webcams have been proven effective in behavioral studies, enabling data collection with minimal disturbance to the birds. Data from WWT Slimbridge were collected during winter months of 2018/2019 and 2019/2020, with observations scheduled on average 7.5 days per month at 08:30, 11:30, and 14:30 to ensure a balanced study design and avoid periods when supplementary food was provided. Unfortunately, a winter storm in January 2020 damaged the webcam at Slimbridge, halting further data collection. Similar technical issues affected the webcam at WWT Caerlaverock, limiting data collection to November 2018 only, and precluding any data collection during the 2019/2020 winter. In total, 450 observations were planned, with swans present during 282 at Slimbridge and 42 at Caerlaverock.
Statistical Analyses
Statistical analyses were performed using R version 3.6.3. To test the “conspecific hypothesis,” which posits that intraspecific interactions occur more frequently than interspecific ones, two-tailed binomial tests were employed. These tests assessed the deviation of the proportion of intraspecific aggressive interactions for each species from 0.5. Separate tests were conducted for aggressive interactions directed at and received from other individuals. Statistical significance was determined at P < 0.05, with P-values adjusted using Holm-Bonferroni corrections for multiple comparisons. Confidence intervals for the proportions of intraspecific and interspecific interactions were estimated using the Clopper-Pearson method. Two-sample binomial tests were also used to compare the proportion of interspecific aggressive interactions between species, with P-values adjusted similarly.
For the “density” and “winter decline hypotheses,” zero-inflated generalized linear mixed-effects models (ZIGLMMs) were used with the glmmTMB R package, as the data exhibited a high proportion of zeros (observations with no recorded aggression). The response variable was the duration in seconds spent in aggressive interactions. Due to significant differences in sample size and temporal replication between the Bewick’s and Whooper Swan datasets (resulting from webcam failures at Caerlaverock), these species were analyzed separately. Candidate models were constructed with all possible additive and two-way interactions of explanatory variables, alongside a null model. For Whooper Swans, explanatory variables included time of day and the mean number of swans present. For Bewick’s Swans, variables included time of day, mean number of swans present, month of observation, and winter of observation. An ‘observation identity’ random intercept was included to account for non-independence of swans observed within the same period. Model selection favored a negative binomial distribution with linearly increasing variance.
To address potential collinearity among explanatory variables, ANOVA was used to test for covariance between the continuous variable (number of swans) and categorical variables, with square-root transformation applied to the number of swans for ANOVA assumptions. Associations between categorical variables were tested using chi-squared or Fisher’s exact tests. Collinear variables were excluded from the same models. A total of 3 and 11 candidate models were run for Whooper and Bewick’s Swans, respectively.
Model comparison utilized AICc values, with competitive models defined as those with AICc < 6.0 and satisfying specific criteria regarding additional parameters. Relative Likelihood (RL), Akaike weight (wᵢ), and Evidence Ratio (ER) were calculated to assess model support strength. Conditional and marginal R² values were computed to quantify explanatory power. Tukey’s post hoc comparisons were performed using the emmeans R package for pairwise comparisons of estimated marginal means.
