Black Scoter habitat use along the southeastern coast of the United States

Abstract While the Atlantic Coast of the United States and Canada is a major wintering area for sea ducks, knowledge about their wintering habitat use is relatively limited. Black Scoters have a broad wintering distribution and are the only open water species of sea duck that is abundant along the southeastern coast of the United States. Our study identified variables that affected Black Scoter (Melanitta americana) distribution and abundance in the Atlantic Ocean along the southeastern coast of the United States. We used aerial survey data from 2009 to 2012 provided by the United States Fish and Wildlife Service to identify variables that influenced Black Scoter distribution. We used indicator variable selection to evaluate relationships between Black Scoter habitat use and a variety of broad‐ and fine‐scale oceanographic and weather variables. Average time between waves, ocean floor slope, and the interaction of bathymetry and distance to shore had the strongest association with southeastern Black Scoter distribution.

While most sea duck species winter north of the South Atlantic Bight (Lamb et al., 2020), Black Scoters range from the coast of Canada down to Florida. As their range shifts from northern to southern habitats, the response of Black Scoters to environmental conditions at these southern latitudes differs from their response to conditions at more northern locations (Plumpton et al., 2020).
Additionally, the species exhibits high interannual variation in wintering habitats, making monitoring their wintering populations challenging (Silverman et al., 2013a). A better understanding of the range of conditions that Black Scoters use during the wintering season in the southern portion of their range may provide important insights into management of the species and highlight potential areas of conservation concern, such as development of offshore wind farms.
In an effort to quantify the abundance and wintering distribu- to 2012 along the Atlantic Coast of the United States (Silverman et al., 2013a). These surveys focused primarily on five species of concern due to current population declines, potential harvest implications, or habitat limitations (Sea Duck Joint Venture, 2015 To better understand drivers of wintering distribution and abundance of Black Scoters along the southern Atlantic Coast of the United States, we used aerial survey data and a variety of broad-and fine-scale oceanographic and climatic covariates. Our data incorporate a broad range of latitudes, providing an assessment of Black Scoter distribution in relation to variables in more southernly locations. We built upon Silverman et al. (2013a) to examine both oceanographic and climatic variables at a local and regional scale using all available survey data (including additional intensive surveying at more southern latitudes in 2012). We predicted that some habitats might be similar in more northern areas of the Atlantic (e.g., distance to shore), and unique habitats along the Southern Atlantic Coast would result in habitat differences (e.g., bathymetry and ocean floor slope).

| Habitat variables
We included eight independent predictor variables in the model of Black Scoter distribution, five for modeling effects of oceanographic variables, and 3 for modeling effects of weather variables  (Sun et al., 2020) may result in different habitat use. We did not include covariates related to weather that had previously been shown to not correlate with Black Scoter abundance (e.g., sea surface temperature, Zipkin et al., 2010). All variables were standardized to a mean value of zero and a standard deviation of 1 and collinearity between each pair of variables was less than 0.6. To calculate the value for each environmental covariate of interest, we either used the center of the grid cell (for distance to shore) or calculated the mean of a given variable for each grid cell using the raster package (Hijmans et al., 2017).

| Oceanographic covariates
We obtained bathymetric data from the National Oceanic and Atmospheric Administration's (NOAA) National Geophysical Data, ETOPO1 Global Relief Model (Amante & Eakins, 2009). We calculated ocean floor slope (degrees) by using the bathymetry data (Amante & Eakins, 2009) and calculating the difference of the values between neighboring cells (Table S1). We obtained a shoreline shapefile from NOAA's National Centers for Environmental Information Global Self-consistent, Hierarchical, High-resolution Geography Database (GSSH), version 2.3.6, using the intermediate resolution (i) and the boundary between land and ocean (L1; Wessel & Smith, 1996). We obtained monthly values for NAO from the Climatic Research Unit, University of East Anglia, Norwich, UK (https://cruda ta.uea.ac.uk/ cru/data/nao/).

| Weather covariates
We obtained daily values of wind speed and time between waves for corresponding dates of aerial surveys for all grid cells from NOAA's National Data Buoy Center. We acquired wind speed data from 20 buoys located along the southeastern U.S. coast from the Virginia coast (37°60′ N) to the Florida-Georgia border (30°70′ N). We acquired time between waves data from nine buoys located along the southeastern U.S. coast from the Chesapeake Bay (36°91′ N) to the Georgia border (31°40′ N). For each day of the aerial survey, we averaged the daily wind speed and time between waves and interpolated the average daily wind speed and time between waves across our study area by using inverse distance interpolation over the latitude range of 28° to 39° N and over the longitude range of −82° to −72° W with the gstat package (Pebesma & Graeler, 2017). Then, we averaged survey dates to calculate the average wind speed and average time between waves for each grid cell across the survey period.

| Model fitting
We used Bayesian hierarchical models to evaluate the relationship between abundance of Black Scoters along the southern Atlantic Coast of the United States. We initially compared a zero-inflated negative binomial, a zero-inflated Poisson, and a negative binomial model; however, the negative binomial model was the only model to converge and have adequate model fit based on Bayesian p-values.
Our count data, y i,t , were modeled for each site i in year t with a negative binomial model conditional on site presence where p i,t is specified as p i,t = r r + i,t and r is the overdispersion parameter. Our mean conditional abundance, μ i,t , was then modeled with a log-linear model where β represents effects of covariates.
We used Bayesian p-values based on the Freeman-Tukey statistic to evaluate model fit (Conn et al., 2018). We used indicator variable selection (γ) to select variables for inference (Hooten & Hobbs, 2015). Because our analyses used indicator variable selection for linear, quadratic, and interactive terms, we required the model to incorporate linear terms along with higher order terms (e.g., linear terms for NAO and bathymetry were included in the interaction term for NAO × bathymetry). We incorporated these terms (γ ADJ ) by multiplying all relevant γ values for higher order terms (e.g., value for γ 12 = γ 1 × γ 2 × γ 12 ). Our analyses (Supplement S1) were run in R (R Core Team, 2017) using JAGS (Plummer, 2013) via package runjags (Denwood, 2016). Because we did not have a priori information to support particular effect sizes, we used vague priors specified as σ ~ InvGamma(1,1), and r ~ Unif(0,10).
Our model indicated that Black Scoter abundance was greater in areas with greater time between waves and flatter slopes.
Additionally, the response of Black Scoter to bathymetry was modulated by changes in NAO. In years with lower NAO (cooler, wetter years), abundance was higher in deeper waters, whereas it was higher in shallower waters in years with high NAO (Figure 4).

| D ISCUSS I ON
Few studies have examined sea duck habitat use along the southeastern Atlantic Coast of North America Jodice et al., 2013;Kaplan, 2011 foraging in contrast to the flat topography at lower latitudes that allows scoters to use areas further away from shore, as shown by our negative relationship between slope and relative abundance (Silverman et al., 2013;Zipkin et al., 2010). While the biological mechanism for shifts in Black Scoter use of different ocean depths in response to changes in NAO is unclear, broad-scale changes in environmental conditions driven by the Gulf Stream in response to NAO phase (Sun et al., 2020) may be responsible for shifts in different habitat use by Black Scoters.
In addition to bathymetry, NAO, and slope, time between waves also affected Black Scoter wintering distribution. Greater wind and wave speeds may have negative effects on energetic costs for sea ducks (Žydelis & Richman, 2015), causing them to seek out calmer waters. Additionally, increased time between waves may increase observability of Black Scoters, resulting in greater counts in these conditions.
We did not find strong relationships between Black Scoter habitat use and wind speed, distance to shore, or latitude, though these have been shown to affect Black Scoters in more northern studies using satellite telemetry (Smith et al., 2019). These differences could be due to the nature of the aerial survey and the "snapshot" in time that it captures. Our aerial survey also was unable to account for detection probability which might affect perceived habitat use if the probability of detecting a Black Scoter correlated with our habitat covariates. Additionally, the lack of relationships with these covariates could be due to differences in habitat use at different latitudes (Plumpton et al., 2020).  (Newbold & Eadie, 2004;Rushing et al., 2017).

ACK N OWLED G EM ENTS
Salary for H.M.P. was provided through U.S. Geological Survey and the Department of Forestry and Environmental Conservation at Clemson University. We thank D. Diefenbach, C. Jachowski, P. Jodice, and R. Kaminski for feedback on earlier versions of this manuscript. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

CO N FLI C T O F I NTE R E S T
We declare no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data used for the analysis are uploaded in a Dryad repository (https://doi.org/10.5061/dryad.31zcr jdkj).