Hyperabundant herbivores limit habitat availability and influence nest site selection of Arctic- breeding birds

1. Understanding an organism's habitat selection and behavioural flexibility in the face of environmental change can help managers plan for future conservation of that species. Hyperabundant tundra-nesting geese are influencing Arctic environ-ments through their foraging activities. Goose-induced habitat change in Arctic wetlands may influence the availability of habitat for numerous shorebird species that breed sympatrically with geese. 2. We explore whether goose-induced habitat alteration affects shorebird breeding density and nest site selection. Using habitat data collected at sites with High, Moderate and Low goose influence, and samples collected during two periods separated by 11 years, we document the habitat characteristics influenced by geese. We describe the habitat characteristics preferred by shorebirds and relate their availability to goose influence and shorebird density. Finally, we examine whether shorebird nest site selection has changed over time and whether shorebirds select nest sites differently in habitat influenced by geese. 3. We document spatial and temporal changes in sedge meadow habitat and lateral concealment relating to goose influence. The availability of sedge meadow habitat and the degree of lateral concealment declined with increasing goose influence, and also declined at two sites over the 11 years of the study. 4. Densities of both cover- and open-nesting shorebirds were highest where goose


| INTRODUC TI ON
Species vary in their flexibility of habitat selection, from generalists to specialists, with specialists being less flexible and more susceptible to change (Devictor, Julliard, & Jiguet, 2008;Owens & Bennett, 2000). By understanding an organism's habitat selection and flexibility in the face of environmental change, managers can plan for future conservation actions.
Projected climate-related changes such as rising temperatures, longer growing seasons and expanding shrub communities (Post et al., 2009) are expected to profoundly impact Arctic environments, increasing the need to understand habitat selection among northern wildlife. Typically, tundra vegetation changes at a relatively slow rate (Forbes, Ebersole, & Strandberg, 2001;Jorgenson, Hoef, & Jorgenson, 2010); however, in the last 20 years, some North American Arctic and sub-Arctic vegetation communities have experienced significant and rapid change (Abraham, Jefferies, & Rockwell, 2005;Batt, 1997;Kotanen & Jefferies, 1997). In addition to climaterelated changes, increasingly abundant breeding and staging lesser snow (Chen caerulescens caerulescens) and Ross' geese (Chen rossii) have altered wetland habitats through their foraging activities (Abraham et al., 2005;Batt, 1997;Kerbes, Kotanen, & Jefferies, 1990). Intensive grazing of above-ground vegetation, and grubbing (e.g. probing beneath ground to consume the plants' rhizomes) by geese can deplete the energy reserves of forage plants and impede their growth (Batt, 1997;Jefferies, Rockwell, & Abraham, 2004;Srivastava & Jefferies, 1996). In extreme cases, when tundra vegetation is removed by intensive foraging, the result can be an alternate stable habitat state of exposed substrate or moss carpet (Srivastava & Jefferies, 1996).
Understanding how geese influence habitat characteristics, how shorebirds select habitat and the degree to which the two processes interact can allow for better management of these species and their environment. Here, we explore these questions using habitat data collected at three study areas in Arctic North America with High, Moderate and Low goose influence. First, we tested whether geese have landscape-level effects on habitat availability. We predicted that the proportions of moss carpet habitat would be highest and sedge meadow habitat and vegetation concealment lowest in areas with High goose influence. We also tested for habitat change over time, using habitat data collected during two periods separated by 11 years. We predicted that we would find a decline in the proportion of sedge meadow habitat and concealment at the study area with Moderate goose influence but no change at the study area with Low goose influence.
We then used habitat data collected at shorebird Nest, Non-use and Random Sites to identify the habitat characteristics preferred by shorebirds, their availability and the resulting influence on shorebird nest density. Finally, we examined whether shorebird nest site selection has changed over time and whether shorebirds select nest sites differently in habitat influenced by geese. We predicted that the density of shorebirds that select nest sites in vegetation that conceals their nests (i.e. cover-nesters) would be highest in the study area with Low goose influence and that the nests at the study area with Moderate and High goose influence would have lower concealment than at the study area with Low goose influence. By contrast, we expected to see no change in the density of shorebirds that selected exposed nest sites with little vegetation (i.e. open-nesters) and no difference in their habitat characteristics among study areas.

| Study areas and species
We conducted fieldwork from late May to early August, in various years between 2000 and 2017 at three study areas: two within the East Bay Migratory Bird Sanctuary on Southampton Island, Nunavut and one on Coats Island, Nunavut, Canada. Light geese are believed to be the dominant herbivores present in this region. Other herbivores such as Cackling Goose (Branta hutchinsii), caribou and lemming (absent from Coats Island) are present, but in lower numbers relative to light geese. Moreover, the abundance of these other herbivores has not increased enough over time, to explain the reductions in vegetation that we attribute to geese. The principal egg reinvigorate efforts for goose population reduction in order to achieve the population targets articulated by management agencies.

K E Y W O R D S
Arctic, habitat alteration, habitat availability, nest site selection, overabundant, Ross' goose, shorebird, snow goose predators in the region are arctic fox (Vulpes lagopus) followed by Parasitic Jaeger (Stercorarius parasiticus); Glaucous Gull (Larus hyperboreus) are also present, but we have not observed them depredating nests.
The first study area (6 km 2 ) is situated within a Lesser Snow Goose colony (63°57′N, 81°55′W; Figure S1) that has been linked to changes in graminoid-dominated and mesic lowland habitats and their conversion to bare peat and moss (Fontaine & Mallory, 2011;K. Abraham, unpubl. data). The East Bay Mainland (63°59′N, 81°40′W) study area comprises a 12-km² area and is situated approximately 10 km east of the goose colony. At Coats Island (62°51′N, 82°29′W), approximately 135 km away, the 12-km² study area has no permanent goose breeding colony; however, the study area is used for staging and moulting (Kerbes, Meeres, & Alisauskas, 2014, S.A. Flemming pers. obs). Combined, these areas represent a gradient of goose use from High (goose colony) to Moderate (East Bay Mainland), to Low (Coats Island).
We quantified the differences in goose use between our three study areas through a survey of goose faecal pellets. From 2014 to 2017, we counted all goose faecal pellets (old and new) in two 1-m 2 plots within 5 m of the 356 sites selected at random for habitat surveys (see below).

| Study species, nest searching and densities
Across the study areas, eight tundra-nesting shorebird species are To obtain estimates of nesting densities of shorebirds, we intensively searched 300 × 400 m plots throughout the breeding season at each of the study areas (High use area: n = 2 plots, Moderate: n = 3 and Low: n = 3). During intensive nest searches, individual surveyors searched all portions of each plot. Plots were searched during six visits each season and we averaged 4.1 hr per visit; a level of search effort that is considered sufficient to find nearly all nests that persist for multiple visits (Smith, Bart, Lanctot, McCaffery, & Brown, 2009).

| Habitat survey locations and years
We conducted habitat surveys at the study areas over 17 years; however, due to the inter-annual variation in field support and number of nesting shorebirds, our habitat sampling effort varied among years and study areas. From 2015 to 2017, we collected habitat data at locations drawn randomly from within the High, Moderate and Low goose influence areas (Random Sites; see Table 3 for sample sizes). Using these data, we characterized the availability of habitats within each of the study areas to explore how geese influence the habitats important for nesting shorebirds. A subset of 85 of these Random Sites had been surveyed previously, prior to 2005, at the Moderate and Low study areas only (i.e. habitat surveys were not completed at the High study area in these earlier years). We identified temporal changes in habitat that may have resulted from goose foraging by comparing habitat measures between these two sampling periods, and refer to these 85 sites as Temporal Sites when making these comparisons.
To explore spatial and temporal patterns in shorebirds' nest site selection, we completed habitat surveys at all shorebird nest sites found within the Moderate and Low study areas (few nests were found at the High study area) from 2000 to 2004 and from 2014 to 2016 (Nest Sites). In addition to measuring habitat characteristics at Nest Sites, we also conducted surveys at sites within 5 m of the Nest Site to provide a finer scale indication of habitat preference (Jones, 2001), and refer to these as Non-use Sites. In all years, we conducted habitat surveys during the last 2 weeks of July, once breeding efforts were complete.

| Habitat surveys
During Random, Temporal, Nest and Non-use Site habitat surveys, we measured habitat characteristics by estimating the proportional cover of habitat within a 75 m² circle surrounding the Nest or Random Site and identifying the dominant habitat category (sedge meadow, SM; upland, Upl; scrub willow, SW; exposed substrate, ES; moss carpet, MC; water, Wa; Table S1). Following Smith et al. (2007), we also estimated lateral concealment using 12 cm diameter discs, which were covered in 1 cm grids and placed at the point of interest.
We estimated the proportion (±5%) of the vertically oriented discs obscured when viewed from a distance of 5 m and height of 40 cm (approximating the height of an arctic fox), from each cardinal direction. We also recorded the height (in mm) and type of each ground cover type that was touching the disc at each cardinal direction.

| Goose use and habitat alteration
Using Random Sites data, we tested for differences in goose faecal pellet abundance using an ANOVA with study area (High, Moderate, Low) and dominant habitat type (Upl, SM, MC, SW, ES) as fixed effects. To test for differences in concealment and cover height between study areas, we conducted a MANOVA with study area, dominant habitat type and an interaction between the two as fixed effects. To determine which cover type was driving the observed trends, we conducted ANOVAs on each cover type with study area as the fixed effect and height (mm) as the response variable.
We were interested in determining the effect of goose use on lateral concealment within each habitat type after accounting for between area variations in concealment. To evaluate this, we used a linear mixed effects model with lateral concealment as the response variable, and faecal pellet count, dominant habitat type and an interaction between the two latter variables as predictors. To account for variation between the three study areas, we included study area as a random effect.

| Spatial patterns
To identify the availability of each dominant habitat type, we used Random Sites data and determined and the frequency at which each dominant habitat type occurred within each study area. Proportions of habitats within the 75 m² surveyed at each Random Site were highly correlated, so we used a principal components analysis (PCA) to reduce the number of variables. Principal components analyses revealed three principal components that explained 60.82% of the variation among sampling points ( Table 1). The first principal component (PC1) described a gradient from low sedge meadow and low lateral concealment (positive PC values) to high sedge meadow and high lateral concealment (negative PC values). The second principal component (PC2) explained a gradient of water (negative) to dry upland habitat (positive). PC3 described a gradient of moss carpet (negative) to water (positive). We used a MANOVA to test for differences in the scores derived from the first three principal components with study area (High, Moderate, Low) as a fixed effect.

| Temporal trends
We used Temporal Sites data to identify changes in habitat that occurred between the two study periods (prior to 2005 and revisited 2015-2016). We computed scores for both sets of sites using the PCA described above, and then performed a MANOVA on the scores from the first three principal components. We included study period, study area (Moderate use and Low use only; no habitat surveys were conducted at the High use study area prior to 2005) and an interaction between the two as fixed effects.

| Spatial patterns
Using the Nest Sites data, we classified the preferred nesting habitat of each shorebird species as the dominant habitat type (i.e. greatest proportional cover in the 75 m² surveyed patch) that was used most frequently across all study areas. To identify how shorebirds select nest sites across study areas, we tested for differences in PCA scores computed from Random Sites (responses: PC1, PC2, PC3) between survey scales (Nest, Non-use, Random Sites) of individual shorebirds using MANOVAs.
We were also interested in comparing habitat variables between Nest and Non-use Sites among study areas; however, the number of shorebird nests per species and study area was insufficient for analysis. We therefore lumped species into "cover-nesting" and "open-nesting" categories. Based on our results and previous studies (Cunningham et al., 2016;Smith et al., 2007), we classified Red

Phalarope, Dunlin, Semipalmated Sandpiper and White-rumped
Sandpiper as cover-nesting species because they selected nest sites in vegetation that concealed their nests. We classified American Golden-Plover, Black-bellied Plover, Semipalmated Plover and Ruddy Turnstone as open-nesting species because they selected exposed nest sites with little vegetation. We used a linear mixed effects model with PC1 as the response variable, and study area (Low, Moderate), survey scale (Nest Site, Non-use Site) and an interaction between the two as fixed effects. To account for any species-specific variation in selection, we included species as a random effect.

| Temporal trends
As habitat availability changes, the habitat selected by nesting shorebirds could change or not, depending on shorebirds' ability to mitigate the changes through habitat selection. We tested for temporal change in the nesting habitat selected by shorebirds between periods by contrasting habitat at shorebird nest sites prior to 2005 and 2014-2016. Adequate sample sizes (>20/period) of Nest Sites were available for the four most common species (Semipalmated Sandpiper and Dunlin, at the Low use study area, and White-rumped Sandpiper and Black-bellied Plover at the Moderate area). For each species, we tested for temporal shifts in habitat use between study periods using species-specific MANOVAs, with the PC scores as response variables.

| Shorebird Nest density
We tested for variation in shorebird nest density (nests per ha within the intensively searched plots) among the three study areas using a general linear model with a Poisson distribution. In this model, plot, with an interaction between cover type and study area as fixed effects.
Statistical tests were performed in r version 3.2.4 (R Core Team, 2016). All MANOVAs were significant (Table S2), so we report the results of all post hoc ANOVAs.

| Goose use and habitat alteration
Random Sites data revealed that goose pellet counts differed among study areas and habitat types (  Figure 1c).

| Spatial patterns
The frequency of dominant habitat types within the 75 m² circles surveyed at Random Sites varied among study areas (Table 3).
Upland habitat, exposed substrate, and moss carpet were the dom- Scores suggested that the proportion of sedge meadow and amount of lateral concealment (PC1) were lowest at the High goose area and highest at the Low area (Table 2; Figure 2); however, PC2 and PC3 scores did not differ among study areas.

| Temporal trends
Using the Temporal Sites, and after accounting for habitat differences among study areas, we found that habitat differed between sample periods (Table 2; Figure 3). For both areas, sedge meadow and concealment (PC1) decreased and water (PC2) increased between the two time periods. Unexpectedly however, these differences were not greater at the Moderate use area versus the Low (p > 0.05; Figure 3).

| Spatial and temporal trends
Each shorebird species exhibited clear preference for nesting in one of the six dominant habitat types (Table 3). American Golden-Plover and Black-bellied plover preferred upland, Ruddy Turnstone and Semipalmated Plover preferred exposed substrate, and Red Phalarope, Dunlin, White-rumped Sandpiper and Semipalmated Sandpiper preferred sedge meadow habitat.

| Nest density
Shorebird nest density varied by study area, cover type, and a study area by cover type interaction (

| D ISCUSS I ON
We demonstrate significant impacts of geese on both the abundance and habitat use of nesting shorebirds and add to the growing body of evidence that hyperabundant populations of light geese breeding in North America are significantly altering tundra habitats and impacting sympatric species (Flemming et al., 2016;Sammler et al., 2008;Swift et al., 2017). When preferred habitat is altered or becomes limited, birds may search for suitable habitat elsewhere, resulting in locally depressed nest densities (Martin, Arcese, & Scheerder, 2011).
Cover-nesting shorebird densities were lowest in our study area with the greatest goose influence. Individuals that nested in these goose-affected areas selected nest sites with less sedge meadow and concealment than at the less affected area, possibly suggesting some behavioural flexibility in nest site selection. As concealment of nests can be an important predictor of success for some tundra birds (Bentzen et al., 2017;Walpole et al., 2008), shorebirds nesting in sites with less nest concealment may be more susceptible to predation. These results increase the concern that hyperabundant populations of tundra-nesting geese are negatively affecting productivity of sympatric-nesting shorebirds.

| Goose-induced habitat alteration
We documented significant habitat differences across our three study areas and two time periods consistent with an effect of geese. As predicted, the coverage of sedge meadow habitat was highest at the study area with the least influence of light geese, while coverage of moss carpet and exposed substrate was greater at the study area with greater goose influence. The proportional cover of sedge meadow and amount of lateral concealment (driven by changes in graminoid height) was also lowest in our High goose influence area. It is possible that these trends are not gooseinduced and this spatial variation was always present; however, our other findings suggest that this is unlikely. For instance, faecal pellet counts suggest that moss carpet and exposed substrate were the two most used habitats by geese, while sedge meadow was fourth most used (after scrub willow), indicating that these habitats are the most likely to be altered. Sedge meadow areas with more faecal pellets also had less lateral concealment, directly implicating a goose effect.
Similar landscape-level changes induced by grubbing and grazing of graminoids by geese have been reported elsewhere. In the Queen Maud Gulf region, Canada, the proportion of exposed peat increased significantly with proximity to goose colonies (Conkin & Alisauskas, 2017) and the biomass of graminoids was 330 g/m 2 lower within colonies compared to outside (Samelius & Alisauskas, 2009). Over ~23 years, satellite imagery shows that the proportion of exposed substrate present in the Queen Maud Gulf region increased by 410%, which resulted in a ~32% decline in wet sedge meadow, particularly within the goose colony (Conkin & Alisauskas, 2017). In the Hudson Bay lowlands, goose-altered areas have 29% more bare ground, 30% more moss and 32% less sedge cover than intact areas (Sammler et al., 2008), and grazing geese have reduced the proportions of sedge (−37%) and scrub habitat (−61%; Peterson, Rockwell, Witte, & Koons, 2013). At our study areas, intensive use of the goose colony and post-hatching dispersal of family groups to our Moderate use area, along with the preferential use of certain habitats as seen in the faecal pellet counts, can explain the gradient of (presumed) habitat disturbance we see among areas and habitat types.  (Kerbes et al., 2014;J. Leafloor, unpubl. data). Thus, although the population growth in this region is still substantial and ongoing, it is less rapid than the exponential growth that occurred range-wide prior to the 2000s (e.g. Alisauskas et al., 2011).

| Shorebird nest site availability
The densities of "cover-nesting" shorebirds that use vegetation to conceal their nests from predators (Cunningham et al., 2016;Smith et al., 2007) were ~20 and ~15 times greater at the Low goose influence area compared to the High and Moderate areas respectively. Red Phalarope preferred sites that provided significantly higher than average lateral concealment (32% and 59% concealment respectively). We propose that these species are avoiding goose-altered habitat because nest sites with suitable attributes are limited.

Comparable results have been reported in the Hudson Bay Lowlands
where cover-nesting shorebirds and passerines were 3.23 and 2.44 times more abundant in intact versus altered habitat respectively (Sammler et al., 2008).
Similarly, the density of open-nesting species was eight times higher at the Low goose study area compared to the Moderate area, and no open-nesting species were found nesting at the High area.
In the Arctic, "open-nesting" shorebirds select nest sites that provide unimpeded views of predators, relying on distraction displays and crypsis to avoid predation (Smith et al., 2007). Thus, changes in the amount of lateral concealment or sedge meadow prevalence, for example from goose grazing, should not affect nest site choice for these species. American Golden-Plover and Black-bellied Plover primarily nested in upland, a habitat type used infrequently by geese based on faecal pellet counts, while Semipalmated Plover and Ruddy Turnstone preferred sites with exposed substrate and selected nest sites with less sedge meadow and concealment than the mean for Random Sites. Instead of habitat effects, the absence of these open-nesting species within areas heavily used by geese could reflect an increased abundance of predators in these areas. Lamarre, Legagneux, Gauthier, Reed, and Bêty (2017) found that the frequencies of occurrence of American Golden-Plovers were ~40% and ~20% higher far from a goose colony compared to within the centre of the colony during high and low lemming years, respectively, demonstrating a link between occupancy and predation risk, irrespective of habitat.

| Nest site flexibility
To test for behavioural flexibility, we examined whether shorebirds that chose to nest, changed their habitat selection when available habitat became limiting. For the four species for which we could test this, we found that the nest site selection remained constant between 11 years, while the availability of sedge meadow and lateral concealment declined. This suggests that habitat selection is inflexible and changing habitat conditions could lead to fewer individuals finding suitable nesting sites. In the Hudson Bay Lowlands where goose effects have been studied in detail, an 84% decline in shrub habitat was mirrored by an ~80% decline in the density of Savannah Sparrows that prefer to nest in this habitat (Peterson, Rockwell, Witte, & Koons, 2014).
Despite birds' constant habitat preferences, we found differences in nest site habitat across our study areas, suggesting habitat limitation. Cover-nesting species selected nest sites with less sedge meadow and concealment at the Moderate use area compared to the Low, presumably because of a lack of availability of sites with high proportions of sedge meadow habitat and high concealment. Since vegetation height and concealment can be positively correlated with nest survival for Arctic birds (Bentzen et al., 2017;Walpole et al., 2008), cover-nesting shorebird nests in goose-affected areas may suffer higher predation rates. Artificial nest experiments in the Arctic support this relationship between concealment and nest survival; during the early nesting season artificial nests with 0% concealment suffer daily survival rates up to 20% lower than nests with 50% concealment (Bentzen et al., 2017).

| Management and future considerations
Goose-induced habitat alteration appears to be limiting the availability of habitat for shorebirds, resulting in lower nest densities and individuals nesting in non-preferred habitat, potentially leaving them more susceptible to predation. The recent declines in Arctic-breeding shorebird populations (Zöckler, Lanctot, Brown, & Syroechkovskiy, 2013) particularly in the eastern and central Arctic (Andres et al., 2012;Bart, Brown, Harrington, & Morrison, 2007;Smith et al., 2012) underscores the urgency for conservation in these regions. Habitat loss elsewhere in the range and climate change are considered the biggest threats to Arctic shorebirds (Thomas, Lanctot, & Szekely, 2006); however, goose-induced habitat alteration could be a contributing factor that has not yet been assessed adequately to understand its impacts range-wide.
Known lesser snow goose colonies occupy a relatively small proportion of the North American Arctic (Flemming et al., 2016); however, our results indicate that breeding geese can alter habitat well beyond the confines of the colony. Furthermore, nonbreeding light geese may influence habitat anywhere within their range, which covers approximately 26% of the Arctic in North America and includes large fractions of the Arctic's wetlands, where a majority of shorebirds breed (Flemming et al., 2016).
The magnitude of effects that goose-related habitat impacts will have on Arctic-breeding shorebirds depends on the shorebird species' preferred habitat and degree of specialization. Our results suggest that cover-nesting, sedge meadow specialists may be more susceptible to goose-induced alteration, while upland specialists may be less so. Elevated predator numbers in the vicinity of breeding geese may nevertheless impact these upland specialists; a hypothesis that requires further verification. Although a warming Arctic may promote the growth of the graminoid vegetation important for cover-nesting species (Gauthier et al., 2013), goosealtered habitat may be slow or unable to respond (Jefferies, Jano, & Abraham, 2006). Predicting how Arctic habitats change with both climate-and goose-related effects, and how shorebird populations will respond to these changes at broad scales, should be a priority focus for future research. Moreover, management strategies for geese should incorporate the habitat needs of sympatric species and reinvigorate efforts for goose population reduction in order to achieve the population targets articulated by management agencies.