The ecological significance of extremely large flocks of birds

Abstract Population size is generally limited by resource availability during and outside the breeding season. Therefore, maximum size of flocks may provide important information on population regulation and the influence of diet and trophic level on maximal degree of sociality. We hypothesized that (a) flock size should increase with nutrient availability; (b) flock size should decrease with latitude because productivity is higher at lower latitude; (c) aquatic habitats should have larger flocks than terrestrial habitats because the former are less accessible; (d) smaller species should have larger flocks because they require overall less food; (e) human‐impacted species that live in perturbed habitats should have smaller flocks than other species; (f) flock size should decrease with increasing trophic level because there is a reduction in biomass due to conversion at each trophic level; and (g) flocks of species depending on ancestral landscapes should have decreased in size in recent years due to human impact (e.g., land‐use). We obtained 1564 observations of flocks that exceeded 100,000 individuals in order to test the predictions listed above. Most effect sizes were small to medium accounting for 1%–9% of the variance, while large effects accounting for 25% or more were only found for total nitrogen used per km2 and area used for agriculture. Changes in large bird flocks were caused by habitat degradation and persecution, and temporal decline in size of large flocks revealed changes in nutrient use, reductions in nutrient cycling, and changes in flock size linked to trophic level.

regulation and the influence of ecological factors such as nutrient availability, diet, and trophic level for maximum flock size. This raises the question which are the largest bird flocks and colonies ever observed in the world, and which ecological factors are the main determinants? There are numerous studies of the rarest species in the world, but few or none of the most common species. Here, we provide such a study. Seabird colonies, and by inference flocks of birds, are important global drivers of the nitrogen and phosphorus cycles (Mosbech et al., 2018;Otero, Peña-Alberti, Pérez-Alberti, Osorio Ferreira, & Huerta-Diaz, 2018). This role for seabird colonies is due to effects of soil, sediment, and water eutrophication.
We hypothesized that (a) nutrient availability sets a limit to flock size through its effects on primary and secondary productivity (Guignard et al., 2017). We suggested that (b) flock size should decrease with latitude (higher productivity at higher latitudes due to permanent daylight in Polar regions during the breeding season and increased loss of ice cover; Meltofte et al., 2013); (c) aquatic habitats should have larger flocks than terrestrial habitats due to higher productivity. Carbon flows are two to three orders of magnitude lower in terrestrial than in aquatic ecosystems (Gounand, Little, Harvey, & Altermatt, 2018), reducing flock size in terrestrial ecosystems; (d) smaller species should have larger flocks due to smaller food demand per individual; (e) human-impacted species should have smaller flocks than other species due to lower carrying capacity and effects of organic pollutants (Crozier & Gawlik, 2002;Meltofte & Clausen, 2011); (f) flock size should decrease with increasing trophic level due to a reduction in biomass at higher trophic levels (Guignard et al., 2017;Lindeman, 1942); (g) large flocks should decrease in size in recent years due to human impact and exploitation of natural resources (Mokross, Ryder, Côtes, Wolfe, & Stouffer, 2014;Peters, Likare, & Kraemer, 2008); (h) flocks should be larger in more seasonal environments where large flocks can be sustained by seasonal availability of resources (Jakubas, Wojczulanis-Jakubas, Iliszko, Strøm, & Stempniewicz, 2017); and (i) flock size should increase with increasing amounts of fertilizer for species living in farmland (Pretelli, Baladron, & Cardoni, 2018). We collected a total of 1,564 observations of bird flocks or colonies exceeding 100,000 individuals from 154 species in 69 countries in order to test these predictions.

| Data collection
For practical reasons, we used a lower limit of 100,000 individuals to the definition of a large flock, resulting in the accumulation of a total of 1,564 observations of flocks. Three examples of large flocks are shown in Figure 1. If we had relied on flocks that consisted of 1 million or more individuals, this would have resulted in relatively few observations (in this study 329 flocks), while reliance on 10,000 individuals or more across the world would have resulted in so many observations that it would have been unfeasible to obtain even close to full coverage. This is a study of large flocks and colonies of birds throughout the annual cycle. We used observations of flocks from 1637 to 2019.
We defined flocks as aggregations of individuals that remained for shorter or longer time in a location where they consumed food. Therefore, we included flocks of all species that were competing for limited food. However, we did not include migrants that locally may occur in very large numbers for very short periods of time, but do not exploit local food. The classification of species was made according to the variables listed in the remainder of the Materials and Methods.
Flock size was generally derived from visual estimates, counts of a fraction of a flock or counts from photographs followed by extrapolation. Comparison of counts from airplanes and counts from the ground commonly revealed detection rates exceeding 80% for larger species of waterbirds (Broome, 1985;Laursen, Frikke, & Kahlert, 2008;Savard, 1982). If such estimates were reliable, we should expect that repeatability analyses would be reliable and that multiple estimates for a given species would be consistently more similar than a similar number of estimates based on randomly chosen flocks. Flock sizes can also be estimated using estimators based on  the number of individuals present in the neighborhood of an average individual (Reiczigel, Lang, Rósza, & Tóthmérész, 2008), but we did not adopt this approach because of lack of data.  As variables for the statistical analysis, we have used data on fertilizer use per km 2 and agriculture from EuroStat (2019).
Habitat degradation (defined as 0 for pristine or 1 for degraded by humans) and threat status for different species of birds were obtained from BirdLife Datazone (2019). Threat status ranged from +0 to −4, where 0 is least concern, −1 is near threatened, −2 is vulnerable, −3 is endangered, and −4 is extinct. Thus, more negative values imply higher threat status. Bird population size (defined as the size of the global breeding population) and geographic range (defined as the area of the breeding range in km 2 ) were from Cramp andPerrins (1977-1994). The distinction between aquatic and terrestrial habitats was based on literature information (Cramp & Perrins, 1977-1994del Hoyo et al., 1992del Hoyo et al., -2008. That was also the case for aerially foraging bird species that were scored as aerially foraging (1) or not (0) (Cramp andPerrins 1977-1994;del Hoyo et al., 1992del Hoyo et al., -2008. Trophic levels of all species were categorized as 1-primary productivity, 2-consumers, and 3-hyper-consumers. Finally, body mass was obtained from Dunning (1993), while wing area and aspect ratio were obtained from Vágási et al. (2016) and APM. Summary statistics for the predictor variable and the response variables are reported in Supporting Information Table S1.

| Statistical analyses
We tested for consistency by using repeatability (R) estimates as indicators of bias (Falconer & Mackay, 1996). We used species as a fixed factor and log 10 -transformed flock size as a response variable.
If a given flock size is a property of each species, we should expect repeatability to be statistically significant. We identified large flocks in 21 species in Europe for the same 21 species in North America.
We used a linear regression model to test whether the number of large flocks in Europe was significantly positively related to the number of large flocks for the same species in North America.
We used generalized linear models (GLM) to test for the association between response variables and flock size. Flock size was a normally distributed response variable with an identity link function.
We redid all analyses using GLM with species being a predictor variable, other factors being fixed, and other continuous variables being covariates (SAS, 2012). We tested for over-dispersion of data, but always found adequate support for the models, and there were no cases of statistically significant lack of fit at the level of 0.05.
We used Pearson's product-moment correlation coefficients as estimates of effect size. We calculated effect sizes in terms of Pearson's r. Here, we adopted the guidelines of Cohen (1988)  We estimated effect sizes as Pearson's product-moment correlation coefficients, using the equations in Rosenthal (1991), Cooper and Hedges (1994) and Hedges and Olkin (1985).

| Reliability of data
The range in the number of individuals per flock was 100,000-20 bil-

| Testing predictions
The 1,564 large flocks were unevenly distributed across the seven continents ( Figure 2; likelihood ratio χ 2 (LR) = 3,590.51, df = 7, Note: Total N/km 2 is the amount of nitrogen per km 2 . LR is the likelihood ratio statistic, N is sample size, and p is probability. Estimates and SE are reported for the two classes of data whenever predictor variables are dichotomous binomial variables. Effect size is Pearson's product-moment correlation coefficient.
significant positive relationship between the size of large flocks and the area used for agriculture (Table 1; Figure 3b). The size of large flocks decreased with increasing habitat degradation (Table 1).
The mean size of large flocks shown in Figure 4a (Table 1).
Bird species with larger flocks had larger population size (Table 1).
F I G U R E 3 (a) log Number of birds per km 2 in relation to log total amount of nitrogen per km 2 . The lines show the predicted linear relationships for different species while dots show individual observations. (b) log Flock size in relation to threat status ranging from 0 which is least concern, −1 is near threatened, −2 is vulnerable, −3 is endangered, and −4 is extinct. We could not develop a statistical model that included all predictor variables included in Table 1 because that would severely reduce the total number of observations. Therefore, we developed a full model that included nine predictor variables covering the same factors as those reported in Table 1 while maximizing the number of significant predictors ( Table 2). Most of these partial effects were similar to those from GLMs with one predictor and one response variable (Tables 1 and 2). Effect sizes were generally small to intermediate explaining 1%-9% of the variance. the first records started, breeding season and trophic level. This first category includes factors such as total nitrogen use per km 2 , area used for agriculture, habitat degradation due to human activity and latitude that are all likely to directly or indirectly predict productivity. Colony size of reproducing animals will be limited because foraging distances will set an upper limit to the number of individuals that can breed in any given location (Ashmole, 1963;Furness & Birkhead, 1984;Jovani et al., 2015). This will mainly be a consequence of intraspecific competition, although effects of interspecific competi-   However, it is possible that other factors than human perturbation may have contributed to this slope because many variables may contribute to temporal trends.

| D ISCUSS I ON
A number of species are treated as pests, and considerable efforts have been made to eradicate these species. They include redbilled quelea (Bruggers & Elliott, 1990) and starling (Feare & Craig, 1999). The red-billed quelea numbered 217 million based only on the flocks included here, while the starling numbered 133 million.
The red-billed quelea is a pest of grain in Africa, while the starling can only be considered a secondary pest due to its consumption of cherries, olives, and other fruit.
Why do some species have large flocks, while others are rare and highly asocial? The evolution of sociality is a significant component of the major evolutionary transitions, and such transitions in sociality can be seen in huge aggregations of individuals in nests of termites, ants, and bees associated with relatedness and evolution of eusociality (Maynard Smith & Szathmáry, 1995). These cases differ from what we have described here because the huge flocks that we report are generally unrelated individuals. Indeed, the frequency of huge aggregations of individuals in eusocial insects is exceedingly rare compared to the more than 1,500 cases of large flocks in birds that sometimes reach as many as the several billion individuals in the now extinct passenger pigeon and the single flock of 61 million bramblings recorded in Switzerland in 1951-1953(Géroudet, 1952.
Here, we have focused on factors that are associated with resource limitation, but also factors that reduce access to resources such as aquatic compared to terrestrial habitats, seasonality that reduces access to resources to a few months of peak food availability during the main growing season, habitat degradation and aerial foraging that are linked to superabundant aggregated food.
It is obvious that this study suffers from a skewed distribution This study has a number of perspectives. While it was entirely descriptive, we emphasize that several of the ideas presented here may be tested experimentally. For example, a reduction in fertilizer use estimated as the total amount of nitrogen released per km 2 should result in a reduction in the number of large flocks.
In addition, changes in threat status as affected by protection should result in an increase in flock size. Finally, change in fertilizer use should differentially affect the number of large flocks at the lowest trophic levels.
In conclusion, we have shown that the size of large flocks of birds can be considered to represent ecological indicators of a number of different environmental conditions including the level of total nitrogen fertilizers, threat status of bird species, and several others.

ACK N OWLED G M ENTS
We thank the numerous organizations, the staff of these organizations, and the many amateurs who contributed data to this study.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
APM conceived and designed the experiments. APM and KL performed the study. APM analyzed the data. APM and KL wrote and edited the manuscript.

E TH I C A L A PPROVA L
No approval was required for performing these entire observational studies.

DATA ACCE SS I B I LIT Y
The data are available at Dryad (https ://doi.org/10.5061/dryad. q0p0t3n).