Annual Cycle Phenology and Winter Habitat Selection of White-Winged Scoters in Eastern North America

Concern over declining populations of several North American sea duck species has led to research addressing how environmental and anthropogenic factors in various stages of the annual cycle affect survival, habitat use, site fidelity, and migratory strategies. Southern New England provides key wintering habitat for White-winged Scoters (Melanitta fusca). This area has also pioneered the development of offshore wind energy in North America. I deployed implanted satellite transmitters in 52 adult female White-winged Scoters captured during the wintering period in southern New England, and on a molting area in the St. Lawrence River estuary in Quebec between 2015 and 2016. I used winter movement data to determine winter arrival and departure dates, total length of stay, home ranges, and site fidelity for scoters wintering in southern New England. Scoters spent over half of the annual cycle on the wintering grounds and demonstrated a high degree of inter-annual site fidelity to composite core-use areas. Sizes of individual 50% core-use home ranges were variable (x̅ = 868 km; range = 32 to 4,220 km) and individual 95% utilization distributions ranged widely (x̅ = 4,388 km; range = 272 to 18,235 km). More than half of all tagged birds occupied two or more discrete core-use areas that were up to 400 km apart. I combined these home range estimates with biotic and abiotic habitat data to calculate resource selection functions to model predicted relative probability of use for Whitewinged Scoters throughout the southern New England study area. Scoters selected for areas with lower salinity, lower sea surface temperature, higher chlorophyll-a concentrations, and higher hard-bottom substrate probability. Resource selection function models classified 18,649 km (23%) of the study area as high probability of use, which included or immediately bordered ~420 km of proposed Wind Energy Area lease blocks. Important habitats and key environmental characteristics identified by this study should be carefully considered when siting and developing future offshore wind energy areas. Understanding full annual cycle movements of long-distance migrants is essential for delineating populations, assessing connectivity, evaluating crossover effects between life stages, and informing management strategies for vulnerable or declining species. In a complementary second study, I used the same 52 satellitetagged female White-winged Scoters to document annual cycle phenology, delineate migration routes, identify primary areas used during winter, stopover, breeding, and molt, and to assess the strength of migratory connectivity and spatial population structure. Most scoters wintered along the Atlantic coast from Nova Scotia to southern New England, with some on Lake Ontario. Scoters followed four migration routes to breeding areas from Quebec to the Northwest Territories. Principal post-breeding molting areas were in James Bay and the St. Lawrence River estuary. Migration phenology was synchronous regardless of winter or breeding origin. Cluster analyses delineated two primary breeding areas, one molting, and one wintering area. Scoters demonstrated overall weak to moderate connectivity among life stages, with molting to wintering connectivity the strongest. Thus, White-winged Scoters that winter in eastern North America appear to constitute a single continuous population.

Continental Shelf in areas that may provide important staging and wintering habitat for scoters and other species of sea ducks. Concern over the potential impact of offshore wind energy on sea duck populations has led to efforts to develop models to understand their distribution, habitat use and site fidelity. We used satellite telemetry to document winter phenology and site fidelity, as well as fine-scale resource selection and habitat use of 40 White-winged Scoters along the southern New England continental shelf. Scoters spent over half of the annual cycle on the wintering grounds and demonstrated a high degree of inter-annual site fidelity to composite core-use areas. Sizes of individual 50% core-use home ranges were variable (x ̅ = 868 km 2 ; range = 32 to 4,220 km 2 ) and individual 95% utilization distributions ranged widely (x ̅ = 4,388 km 2 ; range = 272 to 18,235 km 2 ). More than half of all tagged birds occupied two or more discrete core-use areas that were up to 400 km apart. Throughout the study area, scoters selected for areas with lower salinity, lower sea surface temperature, higher chlorophyll-a concentrations, and higher hard-bottom substrate probability. Resource selection function models classified 18,649 km 2 (23%) of the study area as high probability of use, which included or immediately bordered ~420 km 2 of proposed WEA lease blocks. Future offshore wind energy developments in the region should avoid key habitats highlighted by this study and carefully consider the

INTRODUCTION
Effective management and conservation of any migratory species relies on a thorough understanding of that species' seasonal distribution and resource use, as well as threats from anthropogenic and other sources. In North America, there is increasing concern over declines in populations of several sea duck species (Sea Duck Joint Venture Management Board 2014, . Reasons for these apparent declines are uncertain, although poor habitat conditions and foraging availability on wintering grounds have been linked to significant mortality events (Camphuysen et al. 2002), reduced annual survival (Petersen and Douglas 2004), and decreased productivity in subsequent breeding seasons (Oosterhuis and van Dijk 2002). Because sea ducks spend much of their annual cycle utilizing habitats in nonbreeding areas where direct anthropogenic threats are often greatest, understanding habitat use dynamics on their wintering grounds is important for conservation planning.
In North America, the first offshore wind energy development (OWED), a 5turbine, 30-megawatt facility off Block Island, Rhode Island, became operational in December 2016. Thus, the potential impact of OWEDs on sea duck populations is a recent conservation concern in the United States, particularly on their wintering grounds because numerous other multi-turbine wind energy leases have been issued for offshore areas in New England and mid-Atlantic states (Manwell et al. 2002, Breton and Moe 2009, Musial and Ram 2010. Potential negative interactions between sea ducks and OWED include collision risk, disturbance, and exclusion from key habitats and prey resources (Fox et al. 2006, Drewitt and Langston 2006, Furness et al. 2013, Dierschke et al. 2016. In Europe where biologists have been investigating the potential impacts of OWEDs on marine birds for over 20 years (Guillemette and Larsen 2002, Desholm and Kalhert 2005, Langston 2013, Bailey et al. 2014, Vallejo et al. 2017, collision risk is likely minimal for sea ducks Kahlert 2005, Bradbury et al. 2014), but avoidance behaviors including displacement from key foraging sites likely have more significant population-level impacts (Hüppop et al. 2006, Furness et al. 2013, Dierschke et al. 2016. Sea ducks are particularly vulnerable because they usually forage in shallow, subtidal areas in substrates that are often favored for OWED (Fox 2003, Kaiser et al. 2006. A review of post-construction studies at 20 offshore wind farms in Europe classified Common Scoters (Melanitta nigra) and Long-tailed Ducks (Clangula hyemalis) as weakly avoiding offshore wind farms (Dierschke et al. 2016). Peterson and Fox (2007) documented short-term displacement of Common Scoters from an OWED in Denmark for three years, though follow-up studies suggest that this displacement may be more long-lasting (Petersen et al. 2014).
This loss of potential foraging habitat, as a result of avoidance and displacement, in areas with large concentrations of wintering sea ducks could have detrimental population-level effects. Habitat conditions and availability during the wintering period may have strong carry-over effects on reproductive success and productivity during the subsequent breeding season (Camphuysen et al. 2002, Oosterhuis and van Dijk 2002, Gurney et al. 2014. Also in Denmark, Common Eiders (Somateria mollissima) avoided flying close to or amongst wind turbines (Larsen and Guillemette 2007), suggesting that habitat use within and around wind farms may be greatly reduced. The cost of avoidance behaviors may be trivial relative to the energetic costs of long-distance migration, but the cumulative impact of avoiding multiple developments along a migration route could be significant (Masden et al. 2009). Thus, identification of important habitats used by sea ducks prior to offshore wind energy development informs the planning process and helps to avoid displacement of sea ducks from preferred habitats.
Satellite telemetry provides an increasingly effective tool for assessing population delineation, movement ecology, and habitat selection of marine birds including sea ducks (Oppel et al. 2008, Berlin et al. 2017  . On their wintering grounds, White-winged Scoters generally feed on benthic mollusks and crustaceans in waters ≤20 m deep (Stott andOlson 1973, Lewis et al. 2007). The continental population of White-winged Scoters has experienced a long-term decline throughout the last half-century , USFWS 2011, with steady rates of decreased annual harvest being recognized on the wintering areas, particularly on the Atlantic Coast (Rothe et al. 2015 Figure 1).

Capture and Marking
We used floating mist net arrays (Brodeur et al. 2008)    in August 2016 at a prominent molting location in the St.
Lawrence River Estuary, Quebec, Canada (48.69°N, 69.06°W; n = 262). We used 2-4 sets of mist nets (36 m long, 127 mm mesh) in nearshore (<1 km) areas previously identified as consistent feeding locations. We monitored nets with teams of 2-4 biologists in outboard boats from predawn to 3-6 hours after sunrise. Upon removal from the mist nets or gill nets, we determined the age and sex of all captured birds based on plumage characteristics , cloacal examination, and bursal depth . We also weighed each bird with a Pesola spring scale (± 5 g), and uniquely banded each with a United States Geological Survey size 7 aluminum or incoloy butt-end leg band.
We selected 52 White-winged Scoters that were either second-year (SY) or after-second-year (ASY) to receive implanted satellite transmitters (Cape Cod Bay n = 22; Long Island Sound n = 4; Quebec n = 26). Appropriate sample sizes for satellite telemetry studies vary depending on study objectives (Lindberg and Walker 2007).
Hebblewhite and Haydon (2010) suggested that at least 30 individuals were needed to make population-level inferences on resource selection, while Thaxter et al. (2017) reported that area use in seabirds could be reliably characterized by tracking 13-41 individuals for at least 145 days. As this study was part of a larger multi-species project implemented by the Sea Duck Joint Venture to assess population-level linkages between wintering, breeding, and molting areas, we chose to implant only second-year (SY) or after-second-year (ASY) females with satellite transmitters.
Females of many sea duck species including White-winged Scoters exhibit a high degree of natal and breeding philopatry Savard 2015, Mallory 2015), and would thus be more likely to provide consistent breeding location data to achieve those objectives. To increase sample size and improve robustness of resource selection function modeling, these particular analyses also included an additional 16 White-  transmitters were sterilized with ethylene oxide and allowed to de-gas before implanting. All transmitters were implanted by licensed veterinarians with prior sea duck surgery experience using sterile surgical procedures and techniques described by . All birds were administered subcutaneous boluses of lactated Ringer's solution (30 ml/kg). Isoflurane given by mask, followed by intubation was used for the general anesthesia. All birds received a line-block of bupivacaine (2 mg/kg) or bupivacaine and lidocaine (2 mg/kg) at the site of the skin incision. During holding, transport, and recovery, we held birds separately in small pet carriers equipped with padded sides to avoid bill damage and a raised mesh floor above a bed of pine shavings to allow them to remain clean and dry. We allowed birds to recover in their crates for 1-2 hrs after surgery and then released them at or near their original capture location within 11 hrs of initial capture (x ̅ = 7.5, range = 3.0 -11.0). The project and methodology were approved by the University of Rhode Island Animal Care and Use Committee (IACUC #AN1516-002).

Location Data
We used the Argos satellite-based location and collection system (Collecte Localisation Satellites 2017) to receive transmission signals and PTT diagnostic data from all deployed birds. We downloaded and archived transmission data nightly and subsequently filtered data through the Douglas Argos Filter (DAF; ) to remove redundant data and unlikely point locations. Using the DAF, we employed a hybrid filter to retain the single location with the highest accuracy from each duty cycle to reduce redundant daily positional information in our analyses.
Argos processing centers report calculated accuracy estimates for each of the four highest quality location classes (i.e., location classes 3, 2, 1, and 0 had estimated accuracies of <250 m, 250 to <500 m, 500 to <1,500 m, and >1,500 m, respectively; accuracies were not estimated for location classes A, B, or Z (invalid location), however few locations of these classes were used in our analyses ( August 2016 were programmed to begin on the conservative duty cycle and later switch to the intensive cycle after the first winter period. While this was counter to our earlier fall deployments, we projected that the conservative duty cycle would provide an acceptable number of winter locations (~40) sufficient for habitat analysis.
We programmed transmitters deployed in 2010-2012 with a duty cycle of 2 hrs on and 72 hrs off. The shorter on period was thought to increase battery life, although the fewer high-quality locations received from these transmitters resulted in the later use of the 4 hrs on duty cycle. To minimize potential bias in habitat use and movement behavior associated with capture and surgery trauma, we excluded the first 14 days of data collected after release ). For the same reason, we only included birds that transmitted >60 days after release in our analyses.
We used only one year's worth of data for each bird when calculating winter resource selection in order to standardize for mortality and PTT longevity and avoid biasing the analysis towards individuals that have over one year of data. As the potential exists for movement patterns and behavior of birds to be affected by transmitter implantation during the period following capture and deployment (Barron et al. 2010), we preferentially used data for an individual in the second winter they were tracked if such data existed. When calculating movement phenology and inter-annual site fidelity, multiple years of data were used when available. We managed and analyzed all telemetry data, as well as produced all maps, using ArcGIS 10.3.1 (Environmental Systems Research Institute, Inc., Redlands, CA).
We performed all statistical analyses using the statistical software R 3.3.1 (R Core Team 2016).

Winter Phenology and Length of Stay
To determine when White-winged Scoters were present in the study area and potentially vulnerable to proposed and existing OWED, we calculated fall arrival dates, spring departure dates, and overall length of stay following criteria described by . We defined the fall arrival date into the study area as the median date between the last location outside the study area and the first location within the study area during fall migration. Similarly, we calculated the spring departure dates as the median date between the last location within the study area and the first location outside the study area during spring migration. We defined the first winter length of stay as the period between transmitter deployment and the spring departure date. We estimated the length of stay during the second winter period as the difference between the fall arrival date and the spring departure date plus one additional day, to account for the possibility that birds could have been present within the study area on either or both the arrival date and departure date. We report the overall winter length of stay as mean ± SE, whereas we report only the median (range) arrival and departure dates. not typically found on inland freshwater areas during the wintering period, apart from relatively small numbers wintering in the Great Lakes (Prince et al. 1992). We then calculated the total area (km 2 ) of the individual and composite utilization distributions and core-use areas. We reported total area for individual and composite utilization distributions and core-use areas as x ̅ ± SE. We used Wilcoxon rank-sum tests to compare total area of utilization distributions and core-use areas by sex and capture location. For birds with two or more distinct 50% core-use areas, we calculated the Euclidian distance (km) between centroids of each area. For White-winged Scoters that spent two consecutive winters within the study area, we compared total area of utilization distributions and core-use areas between winters using paired t-tests.

Site Fidelity
We assessed winter site fidelity of White-winged Scoters between consecutive winter periods by determining the number of second winter (2016-2017) locations within the study area that fell within an individual's first winter (2015-2016) 50% core-use area and 95% utilization distribution as well as those that occurred within the composite 2015-2016 core-use areas and utilization distributions. We measured mean (± SE) geodesic distances between first and second winter core-use areas for each individual.
We also calculated the percentage of second winter points for each individual that occurred within the first winter 50% core-use areas of all other individuals for which we had two winters of data to assess whether birds preferentially occurred within their own core-use area as compared to the core-use areas of other birds in the population.

Resource Selection During Winter
We used the composite 95% utilization distributions and 50% core-use areas to assess White-winged Scoter habitat use and resource selection within the study area. We made no distinction between diurnal and nocturnal locations when calculating individual and composite home ranges, so resource selection estimates were based on a full 24-hr diel period. Following  and , we investigated third-order resource selection (Johnson 1980) by quantifying and comparing habitat covariates within the composite 95% utilization distributions (available) and 50% core-use areas (used; Manly et al. 2002; Sampling Protocol-A).
We used Spatial Analyst in ArcGIS to generate the maximum number of random points within the 95% utilization distribution and 50% core-use area, with a minimum separation distance between points of 500 m to reduce spatial autocorrelation. We did not assess overlap between used and available samples as resource selection functions (RSF) are robust to such contamination (Johnson et al. 2006).
Several studies have shown that distribution patterns of wintering sea ducks are driven in large part by available food resources (Žydelis et al. 2006(Žydelis et al. , Kirk et al. 2008 and ocean bottom geography , Heinänen et al. 2017). We therefore chose a set of eight geophysical and oceanographic habitat variables that we hypothesized could serve as proxies for benthic invertebrate distributions and thus provide significant predictive ability in determining scoter use throughout the study area. To quantify distance from shore, we calculated the Euclidian distance (km) from each resource unit to the nearest segment of the NOAA Medium Resolution Digital Vector Shoreline (1:70,000; NOAA 2017a). We measured water depth (m) and slope (degrees) within each resource unit using the NOAA National Geophysical Data Center Coastal Relief Model (3 arc-second) for the United States (NOAA 2017b). To estimate likely areas of hard bottom occurrence, we used a kernel-based probabilistic model developed by Loring (2012). We obtained sediment grain size data from the Nature Conservancy's Northwest Atlantic Marine Ecoregional Assessment data portal (Greene et al. 2010). These data were interpolated from pointbased sampling and classified based on grain size following the scale developed by Wentworth (1922). To convert this to a continuous dataset, we assigned the median grain size value from each ordinal class to pixels within each category. Following , we used Marine Geospatial Ecology Tools in ArcGIS to create long-term mean raster sets for oceanographic habitat variables including sea surface temperature, sea surface salinity, and chlorophyll-a concentrations.
We obtained smoothed daily sea surface temperature (SST; degrees Celsius) estimates derived from interpolated data from high resolution satellite imagery and floating buoys (Stark et al. 2007). These data are collected at a spatial resolution of Processing System. These data were derived from the Aqua sensor aboard the MODIS satellite system which produces radiometric measurements of chlorophyll fluorescence at a 4 km scale (Mueller et al. 2003). To account for the approximately six years of sampling data included in this study, we calculated six-year mean datasets for each of the oceanographic variables by averaging winter-month (1 October -1 May) raster values. We randomly sampled habitat variables at 25% of resource units from both the 95% utilization distributions and 50% core-use areas to reduce spatial autocorrelation between variables. All habitat data were in raster format and resampled to a standardized 250 m X 250 m cell size (hereafter: resource units) prior to extraction and analysis.
We calculated Pearson product-moment correlations to assess correlations between all possible pairs of habitat covariates and checked for multicollinearity of variables using variance inflation factors (VIF). Within samples throughout the study period, pair-wise correlation among variables did not exceed 0.6 and VIF values were ≤2.0. Therefore, we retained all variables through the modeling step. We used logistic regression to model habitat covariate effects and estimate the parameters for exponential resource selection models (Manly et al. 2002). All environmental variables, including quadratic terms for each to account for possible nonlinear relationships, were included in an initial global model. Non-linear terms for some variables (e.g., water depth, distance to shore) suggested significance in the global model, but after inspection parameter estimates were exceedingly low and not ecologically meaningful, thus only linear terms for each variable were included in further modeling. We performed backwards step-wise model selection, excluding uninformative parameters in order of least significance. We compared each model iteration, as well as an intercept-only and individual-parameter models using Akaike's Information Criterion adjusted for small sample size (AICc). We ranked models using AICc differences (∆AICc) and AICc weights (wi) to estimate the relative likelihood of each candidate model (Burnham and Anderson 2002). Competitive models were considered at ≤2.0 ∆AICc from the best performing model if they contained no uninformative parameters, and we selected the parameter coefficients from the most parsimonious model to calculate the RSF. Model residuals were checked for spatial autocorrelation by using a Moran's I test in the R package spdep (Bivand 2009).
We predicted relative probability of use for 77,390 km 2 of our study area using the RSF derived from our highest ranked logistic regression model. We were unable to predict probability of use for 5,182 km 2 of our study area due to incomplete spatial coverage of the sea surface temperature, salinity, and chlorophyll-a datasets. We calculated the RSF model using equation 5.11 in Manly et al. (2002): where W is relative probability of use, βn are the model coefficients estimated from the logistic regression for each habitat parameter, and xn are the predictor variables.
We used Raster Calculator in ArcGIS to complete the above equation and then reclassified the distribution into 4 quantile bins to characterize relative probability of use from low to high.
We evaluated the predictive ability of the top-ranked RSF using k-fold crossvalidation methods described by Johnson et al. (2006). We used Huberty's (1994) rule of thumb to partition resource units into 3 k-folds with approximately 37% of used resource units being used for model testing against 63% of remaining model training data. We partitioned resource selection functions predicted from the model training data into four quantile bins following Morris et al. (2016), who suggested that RSFs should be validated using the same classification scheme as presented visually. We determined strong predictive ability of the RSF model by a high R 2 value and a nonsignificant χ 2 goodness-of-fit value between observed and expected proportions of use across quantile bins (Johnson et al. 2006). We assumed that areas classified with a high probability of selection in the RSF model were high quality habitat that should warrant conservation from developers when planning and siting future wind energy areas.

Survival and Transmitter Performance
Of three months after deployment while still on the molting grounds, and two transmitters went offline in presumed live birds.
Locational accuracy classes of the best-per-duty-cycle locations used to calculate winter movement phenology and generate winter home ranges ranged from location class (LC) 3 to LC B, with most locations classified as ≥LC 2 (estimated accuracy of ≤500 m; Table 1). Among the randomly-selected locations used to generate composite winter home ranges, 74% of locations were classified as either LC 2 or LC 3.

Winter Phenology
We

Wintering Area Distribution
White-winged Scoters wintering in the study area had 50% core-use areas ranging widely from 32 to 4,220 km 2 (x ̅ = 868 ± 174 km 2 ). Individual 95% utilization distributions ranged from 272 to 18,235 km 2 (x ̅ = 4,387 ± 761 km 2 ). For the 40 Whitewinged Scoters (31 females, 9 males) that spent an entire winter within the study area (including the additional 2010-2012 Quebec-caught birds), the composite core-use area was 2,054 km 2 and the composite utilization distribution was 9,790 km 2 ( Figure  2). Core-use areas were located in Cape Cod Bay, the outer edge of Nantucket Sound between Monomoy Island and Nantucket Island, Buzzards Bay, Long Island Sound and Montauk Point, as well as the Nantucket Shoals south of Nantucket Island. We found no significant difference in individual core-use areas or utilization distribution size based on initial capture location (Wilcoxon rank-sum test, P = 0.6 and P = 0.7, respectively). For White-winged Scoters that spent consecutive winters in the study area, total area of utilization distributions and core use areas decreased by ~30% and ~20%, respectively, though this was not significant (P = 0.2 in both cases). One bird spent the majority of the 2015-2016 winter outside the study area, migrating to Lake Ontario shortly after deployment. Eleven tagged White-winged Scoters spent all or most of the winter outside the study area during the 2016-2017 winter. Alternate wintering areas included Lake Ontario, mid-coast Maine, and coastal Nova Scotia.
During the winter period, individual White-winged Scoters occupied 1-5 distinct 50% core-use areas, with 29 of 40 birds occupying two or more. The mean distance between the multiple core-use areas was 101 km (±16) and ranged from 37 to 404 km. Composite scoter 95% utilization distributions and 50% core-use areas overlapped with or immediately bordered 484 km 2 and 69 km 2 of current wind energy area lease blocks, respectively.

Resource Selection During Winter
Scoter core-use areas within our study area were generally shallower and closer to shore relative to utilization distributions, while bottom slope and sediment grain size were similar throughout (Table 3). The best performing logistic regression model estimating relative probability of use by White-winged Scoters (n = 40; 31 females, 9 males) included four significant parameters (i.e., sea surface temperature, hard bottom probability, sea surface salinity, and chlorophyll-a concentration) and accounted for 49% of Akaike weight (Table 4). The second-ranked model was within 2 ∆AICc but contained an uninformative parameter and thus was not considered competitive. Based on this best model, scoter core-use areas were negatively associated with sea surface temperature and sea surface salinity and positively associated with probability of hard bottom substrate and mean chlorophyll-a concentrations, relative to utilization distributions (Table 5) The top-ranked RSF model was able to predict relative probability of use by White-winged Scoters for 77,390 km 2 of the 82,572 km 2 study area. Throughout the study area, 18,654 km 2 (24.1%) were classified as low probability of use, 19,122 km 2 (24.7%) were classified as medium-low, 20,965 km 2 (27.1%) were classified as medium-high, and 18,649 km 2 (24.1%) were classified as high probability of use ( Figure 3). Approximately 420 km 2 of current wind energy area lease blocks fell within or immediately bordered areas classified as high probability of use.

DISCUSSION
This study is the first to document spatially-explicit resource selection and habitat use of White-winged Scoters wintering on the Atlantic Coast of North America. The resulting estimates of probability of use across the study area provide important insights into specific areas and habitat characteristics that should be considered when planning for and siting offshore wind energy development. Additionally, our study provides important seasonal movement and phenology data on a female-only cohort of White-winged Scoters that can be used to better manage and conserve this species and their habitat during a crucial portion of their annual cycle.

Winter Phenology
The results from this study confirm past survey data , Baldassarre 2014  These instances of long-distance within-winter movements highlight the potential vulnerability of White-winged Scoters to offshore development in the area.
Additionally, a recent study on Black Scoters throughout the migratory and wintering period in southern New England highlighted a tendency to venture outside near-shore core-use areas to locations further offshore, increasing the likelihood of encountering offshore wind energy facilities . While the locations of current wind energy lease blocks in the study area have minimal overlap with scoter core-use areas, the development of offshore structures such as wind turbines could act as an impediment to White-winged Scoters moving between important areas in Cape Cod Bay and Long Island Sound during the winter period.

Site Fidelity
White-winged Scoters in our study were highly philopatric to the broad southern New Knowledge of local prey distributions is one of several advantages that could result from a high rate of site fidelity among sea ducks .
The high rate of population-level site fidelity we observed supports this hypothesis, as many core-use areas we identified were located near high-productivity areas known to be of seasonal importance to sea ducks (i.e., Nantucket Shoals; White et al. 2009. The variability in individual-level site fidelity reported in our study suggests that White-winged Scoters are also able to adjust to changes in local environmental conditions between years to respond to shifting prey distributions and habitat quality.

Resource Selection During Winter
Scoter core-use areas within our study area were associated with areas of lower sea presumably foraging on sessile prey, such as blue mussels (Mytilus edulis) frequently abundant in harder substrates (Goudie and Ankney 1986). We found the importance of hard bottom probability in our models surprising, as White-winged Scoters are well documented to prefer prey in predominantly soft-sediment habitats (Stott andOlson 1973, Anderson et al. 2008 Kirk et al. 2008, Loring et al. 2013. Increases in sea surface temperature by only a few degrees, corresponding to a mild vs. cold winter period, were associated with 15-19% body mass loss in blue mussels in the Baltic Sea (Waldeck and Larsson 2013).
Similarly, Lesser et al. (2010) documented blue mussels in the Gulf of Maine exhibiting increased expression of heat shock proteins and antioxidant enzyme activity when exposed to higher seawater temperatures. Such environmental stress has been associated with slower growth and impaired reproductive capacity (Petes et al. 2007).
During the winter period when White-winged Scoters must build energy reserves for migration and breeding, selection for areas of lower sea surface temperature may be indicative of higher quality prey.
Chlorophyll-a, as a proxy for overall levels of primary productivity, and salinity can be important parameters for predicting both seabird and benthic invertebrate distributions (Chester et al. 1983, Balance et al. 1997, Suryan et al. 2012).
We assumed that higher chlorophyll concentrations corresponded to increased primary productivity, and thus higher benthic biomass or food availability to foraging birds (Grebmeier 1993, Phillips et al. 2006. King Eiders in the Bering Sea during the molting period and winter were associated with areas of lower salinity (Phillips et al. 2006) and preference for foraging in low-temperature and high-chlorophyll areas has also been documented in other upper trophic-level sea birds, such as the Cape Gannet (Morus capensis) in the Benguela upwelling zone off South Africa (Grémillet et al. 2008  ). Much of this bias towards areas farther from shore comes from the high abundance of White-winged Scoters, as was documented in our study, which utilize the Nantucket Shoals during winter. This area has been well-documented for its importance to wintering sea ducks (e.g. Sea Duck Joint Venture 2015) and holds high densities of pelagic amphipods and bivalves (White et al. 2009). This area also sits adjacent to a large expanse of wind energy lease blocks that skirt the shoals along their western edge. Future development in this area could pose a high risk of displacement, or act as a barrier to White-winged Scoters moving into or within this important habitat. Core-use areas for Black Scoters in Rhode Island averaged ~15 m in water depth but were much closer to shore (~4 km; . Similarly, Common Eiders in the same study area were found primarily in <20 m of water within 2 km from shore . King Eiders in the Bering Sea during winter were in slightly deeper water (~38 m) but were within 12 km from shore (Phillips et al. 2006). In Europe, Common Scoters primarily forage in waters shallower than 20 m (Fox 2003). Foraging scoters are well documented to tend to congregate in areas with high prey density (Kirk et al. 2008;Loring et al. 2013), which occurs along the southern New England shelf at depths shallower than 26 m (Theroux and Wigley 1998). It is assumed that offshore development within this depth range would have the highest potential for displacement of wintering sea ducks in the study area. While core-use areas were found in areas close to shore, the results of our RSF model likely underestimate or incompletely predict probability of use in nearshore areas due to a lack of spatial coverage of habitat variables used in the model. Finally, we acknowledge that the presence of positive spatial autocorrelation in the model residuals from our top-ranked logistic regression model is a potential limitation of our study that we do not directly address in our analyses. However, we remain confident that our model results are not strongly impacted by these limitations, as the degree of autocorrelation was relatively low and our cross-validation results show very good predictive ability of our top model.

Management and Conservation Implications
In the United States, several sites along the mid-Atlantic Outer Continental Shelf have been proposed for offshore wind energy facilities, and commercial wind energy leases have been issued for offshore areas in Massachusetts, Rhode Island, Delaware, Maryland, and Virginia (Manwell et al. 2002, Breton and Moe 2009, Musial and Ram 2010. Large-scale surveys suggest these areas provide important staging and wintering habitat for several sea duck species , and detailed studies of fine-scale habitat selection have confirmed this importance for multiple species that utilize these offshore waters , Berlin et al. 2017 although some co-authors are associated with state and federal agencies that provided some of the funding and they reviewed the manuscript prior to publication.
Ethics statement: All required permits were secured prior to field work, and the project and methodology were approved by the University of Rhode Island Animal Care and Use Committee (IACUC #AN1516-002).

INTRODUCTION
Monitoring movement patterns of long-distance migratory animals over space and time provides insights into key aspects of their ecology (Trierweiler et al. 2014;Hallworth et al. 2015;Stanley et al. 2015). For example, estimating the inter-annual movements of female waterbirds reveals the extent of immigration and emigration from designated breeding areas (Madsen et al. 2014). Coordinated movements of individuals as they migrate between the same breeding and non-breeding areas suggests strong migratory connectivity (Webster et al. 2002;Moore and Krementz 2017). Quantifying the spatial connectivity of a long-distance migrant throughout their annual cycle can identify key breeding, stopover, molting, and wintering areas (Mehl et al. 2005;Bustnes et al. 2010;Barbaree et al. 2016) as well as delineate population structure. Strong connectivity is often the product of geographically or demographically separate subpopulations (Heath et al. 2006;Fraser et al. 2013). The strength of connectivity (i.e., the extent to which individuals from discrete breeding or non-breeding areas remain in sympatry after migration) can also have critical implications for conservation strategies that consider the full annual cycle of a species.
Understanding migratory connectivity is especially vital for species of conservation concern, as environmental events and stressors during the non-breeding season are well documented to affect population dynamics and productivity during the breeding season (Oosterhuis and van Dijk 2002;Gurney et al. 2014;Sedinger and Alisauskas 2014). Effective conservation and management relies on the identification of distinct population units from which accurate population size estimates and vital rates can be determined (Menu et al. 2002;Swoboda 2007). Similarly, the identification of discrete migratory flyways allows for more effective designation of management regions which may warrant varied harvest regulations (Krapu et al. 2011). A classic example of the importance of population delineation in waterfowl is "migratory" and "resident" populations of Canada Geese (Branta canadensis (L., 1758)) in the Atlantic Flyway of the United States (Heusmann 1999;Sheaffer et al. 2007). Differential survival and harvest pressure led to steep declines in some migratory populations, while sedentary residents exploded to nuisance levels (Heusmann 1999), leading biologists and managers to develop targeted hunting seasons to reduce harvest of the migratory population.
Population structure in waterfowl species is typically female-mediated, as females from most species demonstrate strong natal and breeding site philopatry , whereas males are more likely to disperse depending on their paired status (Anderson et al. 1992). Most studies of waterfowl populations have focused on breeding areas when defining demographic or genetic structure within a population. However,  suggested that other annual cycle stages such as the wintering period are also important to consider. For example, pair formation in many species of waterfowl likely takes place on the wintering grounds (Robertson et al. 1998;Smith et al. 2000), so the proportion of males and females that exhibit site fidelity to certain wintering areas may be more important in determining population structure. Waterfowl are also unique among most migratory birds in that there is a post-breeding flightless remigial molt period in the annual cycle. Some individuals migrate thousands of kilometers to specific coastal or freshwater habitats to undergo a flightless molt for about 34-49 days (Savard et al. 2007;Dickson et al. 2012). These molt migrations add another level of complexity to the standard concept of migratory connectivity that typically only considers breeding and wintering areas as key life stages where individuals from discrete areas or subpopulations may mix or remain isolated.
For several species of North American sea ducks, long-term population declines have increased conservation concern and highlighted the need for focused understanding of annual cycle dynamics Žydelis et al. 2006;De La Cruz et al. 2014;. The causes of these declines remain uncertain in part because sea ducks range across the arctic during summer and inhabit often remote, offshore marine environments during the nonbreeding period. As such, delineating the populations of North American sea ducks is a high priority of the Sea Duck Joint Venture (Sea Duck Joint Venture Management Board 2014), and understanding the migratory connectivity between breeding, molting, and wintering areas for these populations is crucial for species management and conservation (Mallory et al. 2006;Robert et al. 2008;). However, there remains a lack of detailed information about the population structure, migration strategies, and annual cycle movements of priority sea duck species including the White-winged Scoter (Melanitta fusca (L., 1758)).
White-winged Scoters are a long-lived sea duck species that has apparently experienced a long-term decline throughout the last half-century ). White-winged Scoters primarily winter along both the Atlantic and Pacific coasts of North America, with increasing populations overwintering on the Great Lakes, and breed throughout the interior boreal forest from Alaska to central Canada . White-winged Scoters are generally regarded as allopatric eastern and western populations, but studies have suggested some degree of sympatry during the breeding period (Swoboda 2007;Gurney et al. 2014). As in many sea duck species, breeding females show a high degree of natal and breeding area philopatry (Brown and Brown 1981;Traylor et al. 2004

Capture and marking
We used floating mist net arrays (Brodeur et al. 2008) 1). We used 2-4 sets of mist nets (36 m long, 100 or 127 mm mesh) in nearshore (<1 km) areas previously identified as consistent feeding locations. We monitored nets with teams of 2-4 biologists in outboard boats from predawn to 3-6 hours after sunrise.
We also captured scoters using a submerged gill net, adapted from  in August 2016 at a known molting location in the St. Lawrence River estuary, Quebec, Canada (48.7°N, 69.1°W; Fig. 1). Upon removal from the mist nets or gill nets, we determined the age and sex of all captured scoters based on cloacal protuberance (hatch-year birds), plumage characteristics ) and bursal depth . We also weighed each bird with a Pesola spring scale (± 5 g; Pesola AG, Chaltenbodenstrasse 4A, 8834 Schindellegi, Switzerland), and uniquely banded each with a United States Geological Survey size 7 aluminum or incoloy butt-end leg band.
We selected 52 female scoters that were aged either second-year (SY) or aftersecond-year (ASY) to receive implanted satellite transmitters (Cape Cod Bay n = 22; Long Island Sound n = 4; Quebec n = 26). Females of many sea duck species including White-winged Scoters exhibit a high degree of natal and breeding philopatry  and would thus be more likely to provide consistent breeding location data to achieve our objectives. We chose to implant only adult female scoters with satellite transmitters because survival and likelihood of breeding is higher in adults compared to hatch-year birds (Brown and Brown 1981). USA; n = 21). Veterinarians with prior sea duck surgery experience used sterile surgical procedures and techniques described by Iverson et al. (2006) for all implants. Prior to implantation, the veterinarians wrapped all PTTs in nylon mesh and added a felt cuff at the antenna base to provide additional anchor points to stabilize the PTT within the body cavity and provide additional surface area for adhesion to the body wall (Iverson et al. 2006), and skin (felt cuff) at antenna exit site. After applying these external anchoring materials, we sterilized transmitters with ethylene oxide and allowed them to de-gas before implanting.
During holding, transport, and recovery, we held birds separately in small pet carriers equipped with padded sides to avoid bill damage and a raised mesh floor above a bed of pine shavings to allow them to remain clean and dry. We allowed birds to recover in their crates for 1-2 hrs after surgery and then released scoters at or near their original capture location within 11 hrs of initial capture (x ̅ = 7.5, range = 3.0 -11.0).

The project and methodology were approved by the University of Rhode Island
Animal Care and Use Committee (IACUC #AN1516-002).

Location data
We used the Argos satellite-based location and collection system (Collecte Localisation Satellites 2017) to receive transmission signals and PTT diagnostic data from all deployed scoters. We downloaded and archived transmission data nightly and subsequently filtered data through the Douglas Argos Filter (DAF; ) to remove redundant data and unlikely point locations. Using the DAF, we employed a hybrid filter to retain the single location with the highest accuracy from each duty cycle to reduce redundant daily positional information in our analyses.
Argos processing centers report calculated accuracy estimates for each of the four highest quality location classes (i.e., location classes 3, 2, 1, and 0 had estimated accuracies of <250 m, 250 to <500 m, 500 to <1,500 m, and >1,500 m, respectively).
We did not estimate accuracies for location classes A, B, or Z (invalid location) because these location classes were not used in our analyses and rarely occurred.
Individual location data, internal body temperature, and PTT operational information were transmitted from each unit based on pre-programmed duty cycles.
This project was also a part of a study examining resource selection and winter habitat use in White-winged Scoters (Chapter 1), therefore we programmed PTTs on more We managed and analyzed all telemetry data, as well as produced all maps, using ArcGIS 10.4.1 (Environmental Systems Research Institute, Inc., Redlands, CA). We performed all statistical analyses using the statistical software R v3.3.1 (R Core Team 2016).

Annual cycle phenology and migration strategies
We used the highest quality location collected during each duty cycle to calculate the timing of movements and identify key geographic areas throughout each stage of the annual cycle. To account for location error associated with satellite-derived locations we assigned breeding, wintering, and molting areas to each bird by calculating a centroid from all of that individual's locations that were recorded during each time period. We used temporal life stage criteria adapted from the Sea Duck Joint Venture (2015) to assign locations to each life stage (Table 1). Due to the varying accuracy estimates of each location and the time gaps between consecutive locations during the "breeding" period, we did not attempt to quantify nesting success of birds that migrated to suspected breeding areas. During the wintering period, some birds (n = 2) began the winter in one area before migrating large distances to a new wintering area. In these instances, we classified the individual's wintering area as the area in which it spent the majority of the wintering period. More detailed and robust descriptions of intra-winter movements and home range size are described in Chapter 1. To minimize potential bias in habitat use and movement behavior associated with capture and surgery trauma, we excluded the first 14 days of data collected after release (Esler 2000;). For the same reason, we only included birds that transmitted >60 days after release in our analyses. We summarize movement data collected from 27 October 2015 to 6 December 2017.
We used data collected over a one-year period for each individual to calculate their breeding, wintering, and molting centroids. This approach standardized for mortality and PTT longevity and avoided biasing the analysis towards individuals that had PTTs transmit for longer time periods. As the potential exists for movement patterns and behavior of birds to be affected by transmitter implantation during the period following capture and deployment (Barron et al. 2010;), we preferentially used data for an individual in the second year they were tracked when possible. When calculating movement phenology and inter-annual site fidelity, we used multiple years of data when available.
We calculated the arrival dates to areas within each life stage as the median date between the last location outside that area and the first location within it. Likewise, we calculated departure dates as the median date between the last location within and first location outside of a particular area. We estimated total length of stay within an area during each life stage as the difference between the departure date and the arrival date at each location plus one additional day. We added an additional day to account for biases associated with the length of time transmitters were off during their duty cycles, thus the approach accounted for the possibility that a bird was present in an area on during migration periods. We report the overall length of stay at a location and total migration duration and distance as mean ± SE (range), whereas we report only the median (range) arrival and departure dates. We used one-way ANOVA to test for statistical significance of migration phenology based on wintering location and migration route, as well as to assess differences in migratory duration and distance among different migration strategies. We used Tukey-HSD for multiple comparisons when ANOVA indicated significance. We considered results significant at P < 0.05.
Analyses were replicated using non-parametric Kruskal-Wallis tests and significance was not affected.

Population delineation and migratory connectivity
To assess for spatial population structuring on either the breeding, molting, or wintering grounds, we performed cluster analyses on all centroid locations within the breeding, molting, and wintering areas using the OPTICS function in R package dbscan (Hahsler 2016; but see Ankerst et al. 1999). This method uses an ordering algorithm similar to a density-based spatial clustering algorithm (i.e. DBSCAN function) to calculate the number of clusters that best represents the breeding, molting, or wintering area centroids for all individuals combined. The algorithm allowed for the possibility that some centroids would not be assigned to a cluster (Hahsler 2016). The algorithm inputs included an epsilon neighborhood which effectively set a distance threshold for identifying clusters. We determined an appropriate value for the epsilon neighborhood by identifying the "knee" in a plot of calculated k-nearest neighbor distances of our point matrices. We set the minimum number of points allowed for identifying a cluster to 5, as tests with fewer points identified multiple small clusters that we did not consider ecologically meaningful.
We conducted the Mantel test (rM) using the R package ade4 (Dray and Dufour 2007) to quantify migratory connectivity between different life stages. This model did not require an a priori designation of distinct population units and thus served as a null model that only considered distances among individuals during two different life stages. The null hypothesis of random mixing among individuals would thus produce an expected correlation coefficient (rM) of zero (Ambrosini et al. 2009). We constructed distance matrices of centroid locations for the breeding, molting, and wintering periods for birds who provided data between successive life stages (i.e., a bird would not be included if a molting area was known but it did not survive to the subsequent winter). We then computed Mantel test coefficients of connectivity between 1) wintering and breeding, 2) breeding and molting, and 3) molting and wintering periods. We determined statistical significance at P < 0.05 after 9999 random permutations.
To further test whether scoters in the eastern United States behave as multiple distinct sub-units or as a single continuous population, we used linear regression to model the effect of breeding longitude on arrival date to the wintering grounds. We also tested the relationship between spring departure date from the wintering grounds and ultimate breeding longitude, given the hypothesis that birds breeding farthest west from their wintering area would arrive on their wintering area later and depart earlier in the spring than birds that did not migrate as far between breeding and wintering areas.

RESULTS
post-deployment and were not included in any analyses after their first wintering period. We were able to document spring migration routes and breeding areas for 27 individuals. We documented molting areas for 23 individuals and fall migration routes for 17. Five individuals provided data long enough to document breeding locations and migration routes in their second year after deployment.

Annual cycle phenology and migration strategies
We collected movement data of female scoters across a two-year time-period, allowing for the identification of key geographic areas used throughout the annual cycle as well as the phenological patterns underlying each life stage. Annual cycle phenology and longitudinal location data for all birds deployed in this study are presented in Fig. 2. We describe below in more detail the spatial and temporal movements of female scoters within the wintering, breeding, and molting stages of the annual cycle. = 0.57, P = 0.57) of birds among different wintering locations. Total length of stay in the wintering areas was 189 ± 6 (110-225) days. Total spring migration distance was significantly shorter for birds wintering on Lake Ontario than those wintering in southern New England (P = 0.02). However, the low sample size of birds wintering in areas outside of southern New England likely precludes robust analysis.

Spring migration
We were able to determine the spring migration routes of 27 scoters (Fig. 3).
Scoters from all capture locations generally initiated spring migration by either heading northeast along the Canadian Maritime coast (i.e. Nova Scotia and New Brunswick; n = 11) or northwest overland (n = 16). Within the group that undertook the northwest overland route, we identified three distinct spring migration routes to suspected breeding areas including an overland route stopping over at James Bay (n = 5), a direct overland route from the wintering areas to inland breeding locations (n = 8), and an overland route stopping over in the Great Lakes (n = 3). Those that migrated along the coastal route through the Canadian Maritime provinces crossed over the St. Lawrence River estuary before continuing on to eventual breeding areas.
We recorded two individual scoters using different migration routes between years.  Fig. 3). One bird migrated during early June as far west as the southeastern portion of the Yukon Territories, but only remained for ~5 days suggesting this individual did not initiate nesting.

Breeders vs. non-breeders
Eight of the 35 birds (23%) alive during summer did not migrate to the breeding grounds during the first breeding season after deployment. Of these eight birds, only one provided data long enough to determine breeding status in the subsequent year.
This bird migrated to the breeding grounds during their potential second breeding season. We found that non-breeding scoters departed the wintering area an average of six days later than suspected breeders, but this difference was non-significant (F[1,29] = 1.29, P = 0.27) and was likely influenced by a single outlier that did not depart her wintering area until late June. Non-breeding birds migrated directly from the wintering grounds to suspected molting or staging areas (e.g., James Bay, the St. Lawrence River estuary, and mid-coast Maine) until they returned to their wintering areas.

Remigial molt
Molting areas appeared to be directly related to breeding status. Most birds that migrated to suspected breeding areas, and transmitted long enough to record subsequent molting areas, spent the molt period in James Bay (57%; n = 13; Fig. 4).
One bird appeared to molt in Nunavut along the western shore of Hudson Bay, two molted along the southwest shore of Hudson Bay, two molted among the Belcher Islands (56.2°N, 79.4°W) in southeastern Hudson Bay, and three molted in the St.
Lawrence River estuary. One bird that apparently nested near Great Slave Lake, Northwest Territories, appeared to migrate only 50 km west to molt on a small inland pond. Non-breeding females primarily molted in the St. Lawrence River estuary (75%; n = 6), apart from two birds that molted in James Bay and mid-coast Maine, respectively. Two breeding females transmitted long enough to document consecutive molting sites, and both returned to the same location within James Bay in both years.
For birds migrating from suspected breeding areas (n = 21), median arrival date on the molting grounds was 12 August (18 July -14 September; Fig. 2). All birds that molted away from the breeding grounds remained at or near their molting area until fall migration was initiated.

Fall Migration
We documented fall migration routes of 17 scoters, which were less variable than spring migration (Fig. 4)

Population delineation and migratory connectivity
We identified two disjoint clusters of eight and ten breeding centroids, respectively, with nine breeding centroids not assigned to any cluster. The two identified breeding clusters were located southwest of Hudson Bay and immediately surrounding Great Slave Lake in the Northwest Territories of Canada (Fig. 5).
Analysis on molting area centroids revealed one single cluster encompassing all of James Bay. Five additional molting areas inland within the breeding grounds, as well as Hudson Bay and the St. Lawrence River estuary were unclassified. Cluster analysis on wintering areas identified a single cluster encompassing all locations within southern New England. Additional wintering areas in the Great Lakes (n = 2), Long Island Sound (n = 4) and Canadian Maritimes (n = 3) were not assigned to any cluster ( Fig. 5).
We found weak, non-significant migratory connectivity between wintering and breeding areas (rM = 0.13, P = 0.15) among the 27 females where both locations were known within the same year. Connectivity between breeding and molting areas exhibited a weak, though slightly more positive correlation, although this relationship was not statistically significant (n = 21, rM = 0.24, P = 0.08). Connectivity between molting areas and wintering areas exhibited the most positive correlation among life stages, exhibiting moderate but non-significant connectivity (n = 20, rM = 0.46, P = 0.07).
We found no relationship between either breeding longitude and wintering arrival  Petersen (2009) attributed to differences in migration distance and length of stay at stopover locations.
Breeding areas identified in this study represent much of the known breeding range for White-winged Scoters in eastern North America Lepage et al., unpublished data), with probable breeding birds ranging as far west as Great Slave Lake in the Northwest Territories. Two scoters were found as far west as Great Bear Lake in the Northwest Territories and the southeastern portion of the Yukon Territories, although they did not remain long enough to be classified as nesting in this region. Four scoters migrated to suspected breeding areas in northern Quebec, much farther east than the known breeding range for the species. Lepage et al.
(unpublished data) also documented two White-winged Scoters breeding in Quebec further west in the coastal lowlands of northeastern James Bay which support the largest known breeding concentration of White-winged Scoters in Quebec (Benoit et al. 1994(Benoit et al. , 1996. Eastern scoters breeding in the Northwest Territories and as far east as northern Saskatchewan likely overlap with breeding scoters from Pacific and Alaskan wintering areas. For example, White-winged Scoters breeding at Redberry Lake in northern Saskatchewan represent wintering populations from both the Atlantic (25%) and Pacific (75%) coasts based on stable isotope analysis (Swoboda 2007). Despite this overlap, satellite telemetry studies have yet to document scoters wintering on opposite coasts in different years. This pattern of east-west segregation is likely the result of historic population isolation during the last glaciation (Talbot et al. 2015) and is likely maintained due to pair formation during the non-breeding period followed by We identified two principal molting areas for eastern White-winged Scoters. Most birds that migrated to suspected breeding areas apparently molted within James Bay or Hudson Bay based on the timing and length of their stopovers. All scoters that molted away from the breeding area migrated to molting areas in a seasonally-appropriate direction (i.e. along fall migration routes) rather than undergoing a true molt migration in a different direction from the expected fall migration route as has been observed in some species of ducks and geese (Yarris et al. 1994). Timing of arrival on molting grounds typically varies by age, sex, and reproductive status (Savard and Petersen 2015), with males, sub-adults, and non-breeding females undertaking remigial molt before breeding females (Petersen 1980(Petersen , 1981Savard et al. 2007;Dickson et al. 2012 their molting areas were approximately three weeks later than arrival dates of males reported in that study. We did not investigate differences in migration chronology between non-breeding and breeding females in our study, as non-breeding scoters migrated directly to their eventual molting areas after the wintering period, thus we were unable to determine the approximate date that molt was initiated. Molting areas for many sea ducks often also serve as fall staging locations (Petersen et al. 2006;Savard et al. 2011;Savard and Petersen 2015). Scoters in our study remained at or near their molting areas throughout the fall until migrating relatively quickly to their wintering area. Similar to most scoters in our study, Harlequin Ducks (Histrionicus (L. 1758)) in eastern North America migrated directly to their wintering areas without utilizing stopover locations in between (Robert et al. 2008). In contrast, Common Eider have a protracted fall migration that can last several weeks and include several stopover locations along the route (Savard et al. 2011).
King Eiders in Alaska typically take 3-105 days during fall migration to reach wintering areas, with 60% of birds taking longer than three weeks to complete migration after utilizing several stopovers for up to six weeks (Oppel et al. 2008). The phenology of fall migration for female scoters originating from different molting areas showed little to no variability, suggesting that annual harvest along migration does not disproportionately target any segment of the population.

Population delineation and migratory connectivity
Population delineation and migratory connectivity for species of waterfowl have usually relied on band recovery data (Madsen et al. 2014;Guillemain et al. 2017), genetic markers (Fleskes et al. 2010;Liu et al. 2012), stable isotopes (Swoboda 2007;Caizergues et al. 2016), or some combination thereof (Pearce et al. 2008(Pearce et al. , 2014. Assessing stable isotopes or genetic markers, such as nuclear or mitochondrial DNA, can reliably identify overlap in population units and estimate gene flow between discrete breeding locations (Mehl et al. 2005;Sonsthagen et al. 2009). However, information from tracking individuals provides insights into whether such population delineation has resulted in coordinated movements across the annual cycle and thus strong connectivity (Webster et al. 2002;Moore and Krementz 2017), and can identify key breeding, stopover, molting, and wintering areas used (Mehl et al. 2005;Bustnes et al. 2010; Barbaree et al. 2016).
We identified two primary breeding regions for White-winged Scoters in the Northwest Territories and the lowlands southwest of Hudson Bay. These two areas corresponded with areas of high scoter density identified by the Waterfowl Breeding Population and Habitat Survey conducted by the U.S. Fish and Wildlife Service , though this survey does not distinguish between the three scoter species. As population structure in sea ducks is heavily female-mediated due to strong natal and breeding area philopatry , one would expect any spatial population structure to be evident within the breeding areas, though pair formation during the non-breeding season is common (Robertson et al. 1998;Smith et al. 2000) and should ensure genetic mixing as males disperse based on their paired status (Anderson et al. 1992). We estimated weak connectivity between breeding locations and other life stages, and thus little evidence of population delineation among eastern White-winged Scoters, although we recognize that these calculations are based on tracking relatively few females (n = 27) captured at wintering and molting areas. Scoters captured at the molting site in the St. Lawrence River estuary subsequently occurred across the same east-west extent as those captured in southern New England. However, some scoters captured in the St.
Lawrence migrated to breeding areas in areas of northern Quebec that the birds captured in southern New England did not. Our conclusions regarding population delineation and migratory connectivity were supported when only scoters captured in Quebec were included in the analyses, which provides some validation that birds captured on the St. Lawrence molting grounds provide an adequate representation of the eastern population of White-winged Scoters. Future capture efforts of eastern White-winged Scoters should consider this area, as winter conditions and seasonallyvariable scoter distributions make capture efforts during the winter more unpredictable.
Studies of migratory connectivity typically describe movements of individuals between breeding and non-breeding areas (Webster et al. 2002). However, studies of connectivity in waterfowl species must also consider the post-breeding flightless remigial molt period as an additional critical life stage where population structure and mixing may differ from either the breeding or wintering periods. In our study, the strength of migratory connectivity was dependent on which life stages were being compared. Though all were non-significant, connectivity was weakest between winter and breeding, and strongest between molting and wintering sites. This highlights the importance of accounting for the entire annual cycle when assessing migratory connectivity and population delineation in waterfowl.
This study has important implications for conservation and management of eastern White-winged Scoters and provides new insights into their life history. We identified probable breeding locations in Nunavut, northern Ontario, and Quebec that fall outside of published breeding range maps and could warrant further refinement of species range maps and expansion of breeding survey areas. Additionally, this study documented the importance of James Bay and the St. Lawrence River estuary as prominent molting and staging areas for this population, corroborating findings also reported by Lepage et al. (unpublished data). As in many other bird species, these staging and molting areas often act as geographic bottlenecks where large numbers of birds congregate for extended periods of time and thus present unique implications for conservation and management (Leu and Thompson 2002;Lok et al. 2011;Fox et al. 2014; Barbaree et al. 2016). Conservation efforts should consider the value these molting areas provide to White-winged Scoters.