Redistribution of wintering American Common Eiders ( Somateria mollisima dresseri )

,


INTRODUCTION
Broadscale changes are occurring within marine ecosystems along the Atlantic coastline of North America.The ocean climate of the Northwest Atlantic has seen rapid shifts that include accelerating rises in sea surface temperature (Pershing et al. 2015), increasing frequency of extreme marine heatwaves and intense storms (Scannell et al. 2016, Pershing et al. 2018, Chisholm et al. 2021), and changes in ocean circulation (Gonçalves Neto et al. 2021, Seidov et al. 2021).These hydroclimatic changes, in concert with direct impacts of human activities (e.g., coastal development and resource extraction), are likely to drive a wide-scale redistribution of biodiversity as species track conditions that meet their physiological and ecological requirements (Lenoir et al. 2020).Indeed, changes in abundance and distribution linked to shifts in climate regime have been documented across the food web of the Northwest Atlantic (Pershing et al. 2021), from zooplankton (Morse et al. 2017) and mollusks (Sorte et al. 2016) to fish (Pinsky et al. 2013, Pershing et al. 2015) and whales (Meyer-Gutbrod et al. 2021), with potentially wide-reaching ecological, cultural, and economic consequences.
As higher-trophic level predators, sea ducks (tribe Mergini) play an integral role in coastal marine ecosystems of the Northwest Atlantic, including during the winter months, when they congregate to forage in large flocks.The American Common Eider (Somateria mollisima subspecies dresseri) is an iconic winter sea duck often found closer to shore in shallower water than other congeners (Silverman et al. 2013), from the Gulf of St. Lawrence southward to Long Island, NY, with the greatest numbers in Maine and Massachusetts (Goudie et al. 2020).Common Eiders occupy a high trophic position in the nearshore marine environment, specializing on benthic inter-and sub-tidal prey, and particularly mollusks (Goudie et al. 2020).In addition to the potential stressors of climate regime shifts, dresseri eiders face threats from other anthropogenic sources, such as coastal development, shipping, offshore energy production, aquaculture, and commercial exploitation of preferred prey that may compound, and in turn alter, habitat and prey availability during the non-breeding period, differentially through space and time (Noel et al. 2021).Furthermore, this subspecies is commonly harvested throughout its wintering range for both recreational hunting and for subsistence by Indigenous peoples (Rothe et al. 2015).Based on limited data and solicited expert opinion, localized declines have been indicated in breeding and wintering numbers in Nova Scotia, New Brunswick, and Maine (Bowman et al. 2015, Noel et al. 2021, Giroux et al. 2021, Robertson et al. 2021), where the most dramatic oceanic changes in the Northwest Atlantic are also underway.
Typical for sea ducks in general, Common Eiders have been poorly monitored in both spatial and temporal scope relative to other North American waterfowl (Bowman et al. 2015, Koneff et al. 2017).To date, there has been no range-wide census of eiders across the dresseri wintering range that would allow for comparative estimates of localized changes in abundance over a multi-decadal time series.Key areas within the Canadian portion of the wintering range were thoroughly surveyed by air in 4-6 years over the last two decades, but similar monitoring has not occurred in the U.S. portion of the range.A comprehensive overview of changes in local numbers across the range is necessary to fully understand the geographic breadth of shifts in distribution, to ultimately inform coordination and cooperation on the prioritization of harvest management and habitat conservation efforts among regions.Increasingly, we are recognizing that alternative sources of information should be considered in management to provide insights where baseline data are lacking, including citizen science (Young et al. 2019).Where range-wide targeted monitoring for eiders has not been conducted through the past, general bird census programs like the Christmas Bird Count (CBC) have the potential to capture broad spatial patterns in changes over time, including for waterbirds (e.g., Wilson et al. 2013, Meehan et al. 2021).
The CBC is a land-based winter bird census conducted by volunteers starting at the turn of the 20 th century where observers report the number of birds of each species encountered on a single day within established count circles covering a c. 12-km radius (Dunn et al. 2005).Small parties of observers endeavor to cover all habitats within the circle, including nearshore areas for circles intersecting the coast.In this study, we use coastal CBC circles within the wintering range of dresseri eiders to capture relative changes in local numbers over time and examined the emergent spatial patterns in local trends to assess the extent of regional redistribution.In addition, we use counts from dedicated winter aerial surveys conducted every 3-4 years within the Canadian range in 2003-2019 to evaluate the representativeness of CBC counts for local abundance where surveys overlapped, and to estimate recent trends and contemporary distribution in northern regions where CBC coverage is poor.Overall, we aimed to assess whether declines detected from localized monitoring efforts over shorter time frames are occurring over broader temporal and spatial scales and whether a large-scale redistribution may be occurring across the winter range.

Christmas Bird Counts
Annual count data were obtained for all CBC circles within the dresseri wintering range reporting at least 10 Common Eiders (Somateria mollisima) in total across all years since 1965 (Fig. 1A).The majority of CBC data were derived from the CBC online database (https://netapp.audubon.org/cbcobservation/),with the exception of six circles in New Brunswick.Data from these counts have not been reported to the National Audubon Society in most years but can be extracted from online archives of the "NB Naturalist" (https://www.naturenb.ca/archives-nb-naturalist/).The year of a CBC is defined as the year of December, despite some counts taking place in the first week of January (i.e., the count from 2020 is representative of the winter of 2020/2021).
In years where CBC circles were surveyed but no eider were reported, zeros were assumed.We validated the assumption that eiders would have been counted had they been present by assessing whether counts with assumed zeros reported other coastal or nearshore species; CBC circles with assumed zeros for eider reported on average 13 ± 5 (standard deviation) other species found only in the same marine habitats as eiders, ranging from one to 24 species.Thus, we are confident this is a robust assumption.

Aerial winter surveys in Canada
Wintering Common Eiders within the Canadian range have been aerially surveyed by the Canadian Wildlife Service at varying intervals over the period 2003-2019 (Fig. 1B).The wintering range of the northern Common Eider subspecies (S. m. borealis) overlaps with that of the dresseri subspecies in the Gulf of St. Lawrence and the Newfoundland and Labrador Shelf (Goudie et al. 2020).In particular, about one third of the borealis subspecies winter in the Gulf of St. Lawrence and along the Atlantic coast of Newfoundland, while the rest winter along the western Greenland coast (Mosbech et al. 2006, Gilliland andRobertson 2009).The survey was initiated in 2003 with the objective of surveying all suitable habitats for the borealis subspecies every three years, including the entire Gulf of St Lawrence, the southern coast of Labrador, and large parts of the Atlantic coast of the island of Newfoundland, as well as Saint Pierre and Miquelon, France. In 2006, 2012, 2016, 2018, and 2019, large parts of the coastlines of Nova Scotia and New Brunswick were also surveyed, where most eiders are presumed to be dresseri.
Aerial surveys followed methods initially developed by Bordage et al. (1998).All surveys were conducted from high-wing twin-engine aircrafts (Cessna 337 Skymaster, Partenavia P68 Observer, or Britten-Norman NB-2 Islander) under good visibility conditions, and flown at 165-185 kph at an altitude of c. 300 m above sea level.Surveys were conducted when ice cover was at its maximum, mostly in the last two weeks of February (93% of 37 survey days).The same areas were consistently searched, covering c. 12,000 km of coastline in proximity to potential open water with suitable habitat across surveyed regions.The survey team consisted of an experienced primary observer, a photographer, a recorder/navigator, and a pilot.In the first year of surveys (2003), D. Bordage was primary observer and S. Gilliland was recorder/ navigator, after which S. Gilliland was primary observer for all surveys.The observer made a visual count estimate of the number of white birds and brown birds in each encountered flock using 10x30 image stabilizing binoculars, while the photographer attempted to capture the same flock using a high resolution digital camera with image stabilization whenever possible.The navigator/recorder used a GPS voice recording system to geo-reference the visual estimate and associated image identifiers, while also navigating and spotting for birds.Visual count estimates were generally made using established techniques described in Bowman (2014), and no correction factors were applied (unlike in Bordage et al. 1998), thus we used observer visual count estimates which may result in reduced accuracy of counts at high flock sizes.For analyses and visualization of relative distribution, aerial survey data were limited to areas assumed to comprise at least 10% dresseri (i.e., <90% borealis, surveyed area shown in Fig. 1B encompassing 125 grid cells of 0.25° x 0.25°).

Regional designations
Trends from CBCs and aerial surveys are described or summarized as nine distinct "regions" relative to marine or geopolitical boundaries that best capture the spatial distribution of the available data, while also maintaining relevance to management jurisdictions (shown in Table 1).CBC circles and aerial survey grid cells within the Gulf of St. Lawrence were grouped together into one region due to relatively low CBC sample sizes for any single province (including coastal Quebec, the northern coasts of New Brunswick and Nova Scotia, and the western coast of Newfoundland; Fig. 1).For similar reasons, the region referred to as southeast Newfoundland included circles and cells in the Avalon Area of southeast Newfoundland as well as at St. Pierre and Miquelon, France, but not those included in the Gulf of St. Lawrence region.Similarly, the regions of Nova Scotia and New Brunswick included circles and cells found along the coastlines of those provinces except for the northern portions included in the Gulf of St Lawrence region.All other regions represent single states or multiple states combined.

Trends in two 20-year periods
To model local changes in numbers over time (trend), we employed standard techniques to analyze count data using generalized linear models (GLMs; Zuur et al. 2009) at the circle-level with the software R v.3.6.1 (R Core Team 2020).Many analyses of CBC data evaluate species-specific regional aggregate trends by summing annual counts within years across circles representing a geographic subunit of interest.This necessitates re-expressing counts as means per circle because not all circles are surveyed each year (e.g., Bowman et al. 2015) and becomes problematic for regions with low circle densities.Doing so can result in the loss of transparency in spatial coverage of the underlying data and the dilution of trends from the loss of circle-specific information within regions (particularly when areas of relatively high and low abundance are pooled together; Meehan et al. 2021).In this study, all CBC count data and trends were analyzed and presented at the circle-level, then discussed relative to the emergent regional spatial patterns.
We generated CBC trend models for each of two discrete c. 20year time periods: 1980-1999 and 2000-2020.These periods were chosen to maximize the number of circles (out of n = 165 reporting at least 10 total Common Eider since 1965) that meet filtering criteria while providing two periods of equal length for comparison.For each period, data were limited to only include circles meeting the following criteria: 1) at least one survey reported within four years of the start of the period and within four years of the end of the period (zeros included; n = 83 in 1980-1999, n = 109 in 2000-2020); 2) a minimum count of 25 eider reported at least once (n = 54 in 1980-1999, n = 94 in 2000-2020); 3) no time series gaps greater than five years (n = 53 in 1980-1999, n = 88 in 2000-2020); and 4) a maximum of 50% zeros on reported counts (n = 44 in 1980-1999, n = 76 in 2000-2020).These criteria ensured that our trend models were limited to circles with acceptable temporal coverage and that were likely to provide suitable eider habitat.Sample sizes by region and period are shown in Table 1.We fitted GLMs with negative binomial distributions and year (trend) effects using the "MASS" R package (Venables and Ripley 2002) for counts at each CBC circle in each time period, as this distribution allows a flexible model with a high and flexible variance to mean ratio to address the extra variance typical in wildlife count data, often due to some form of biological aggregation process and natural fluctuations.We applied the same trend modeling approach to annual aerial survey counts at the regional level in the Canadian portion of the range (aerial surveys were conducted in each region in 4-6 winters between 2003 and 2019).In general, waterfowl survey data are often highly rightskewed with a high proportion of zeros and extreme high counts, particularly for sea ducks that form large, highly mobile, foraging flocks (Zipkin et al. 2010(Zipkin et al. , 2014)).Further, winter eider numbers are known to fluctuate widely from year to year due to high variance in productivity, recruitment (Giroux et al. 2021), and shifts in distribution in response to habitat, weather, and prey (Zipkin et al. 2010).
Where the confidence limits of λ bound 1, the trend direction was considered "stable," or indiscernible from negative or positive growth.The precision of a trend estimate can also provide a measure of survey quality in terms of a survey's ability to detect change (Soykan et al. 2016).We use the one-half width of the 95% confidence confidence interval to calculate whether estimated trends were precise enough to detect a 3%, 2%, and 1% per year change in localized counts of eiders on CBCs.Detailed model outputs are provided in Appendix 1 (CBC 1980(CBC -1999)), Appendix 2 (CBC 2000-2020) and Appendix 3 (aerial survey counts).
CBC count data are also often adjusted for observer effort since the number of volunteers participating in a count varies from year to year and can thus impact the number of birds counted.However, sea ducks are large, easily identified, and concentrate into flocks along coastal areas where they are predictably found by birders, making them more likely than landbirds or cryptic species to be counted in similar numbers regardless of total effort within a count circle (Dunn et al. 2005, Bowman et al. 2015).We confirmed that observer effort did not impact CBC counts of eiders by fitting the same GLMs for the period 2000-2020 with an additional covariate for the number of field observers participating in a given count year; of 76 circles, only two indicated effort as a significant covariate (i.e., estimated effect confidence intervals bounded zero), and the direction of effort effect estimates was inconsistent (53% indicated a positive effect on counts while 47% indicated a negative effect).Further, the inclusion of an effort covariate did not change the estimated trend direction.Thus, we did not adjust for observer effort in our reported trend models.

CBC coverage and representativeness
Sample sizes and geographic coverage of CBC circles used for trend analysis varied between time periods (Table 1, Fig. 2).
Overall, the higher number of circles represented in the more recent period is primarily due to later circle establishment; 47 CBC circles reporting eider since 1965 were not established until after 1999.At the southernmost extent of the range, in New Jersey, Delaware, Maryland and Virginia, no circles had sufficient data to meet the modeling criteria in the period 1980-1999, primarily because there were too many zeros in the time series (i.e., because few or no eiders were observed, not because surveys were not conducted).Similarly, in Rhode Island, Connecticut, and New York, the number of circles with sufficient data increased from four in the earlier period to 12 in the more recent period.CBC coverage in the northern extent of the range in Canada was relatively sparse compared to the U.S. in both periods (Table 1, Fig. 2).
In years when aerial surveys overlapped spatially with CBCs, regression analysis of CBC counts against the sum of all aerial counts within 12 km of a CBC circle center made in the same winter (n = 101 CBC counts in 29 circles) indicated a positive relationship on the log-scale (R 2 = 0.19, slope estimate = 0.32 [0.19, 0.46]); that is, aerial surveys documented more eiders in years when nearby CBC counts suggested more eiders.Further, based on the one-half width of the 95% confidence confidence interval of trend estimates for the period 2000-2020, 80% of estimated trends (61/76) were precise enough to detect a 1% per year change in localized counts, while the remaining 20% were precise enough to detect a 2% per year change.Similarly for the period 1980-1999, 90% of estimate trends (40/44) were precise enough to detect 1% per year change and the remainder 2% per year.
Localized trends in the U.S.
In the southernmost region comprised of New Jersey, Delaware, Maryland, and Virginia, four of six circles with sufficient CBC data to meet criteria for trend modeling over the period 2000-2020 showed rapid positive growth (λ = 1.16 -1.20, Fig. 2, Table 1) with two in New Jersey, one in Maryland, and one in Delaware (Fig. 3).Two additional circles in New Jersey and Delaware showed stable trends.No circles in Virginia met modeling criteria, however, small numbers of eiders have recently been reported on CBC circles in Virginia, as well as on 13 additional CBC circles in this region (Fig. 1).
In Rhode Island, Connecticut, and New York, the four circles with sufficient CBC data for trend modeling in both periods were in the northern half of this region; these circles shifted from rapid positive growth (λ = 1.25 -1.40) from 1980-1999 to stable or negative trends from 2000-2020 (Fig. 2, Table 1).The eight additional circles with trend estimates only in the period 2000-2020 showed either stable (one circle) or positive trends (seven circles; Fig. 2), six of which were on Long Island, New York (Fig. 3).The raw CBC data from these six circles showed clearly that eiders were rarely, if ever, observed in this area before 2000, and that numbers have grown at average rates of λ = 1.25 since (Appendix 4).
New Hampshire and Massachusetts comprised the only region in which negative trends were detected in the period 1980-1999 (Fig. 2, Table 1).The proportion of circles with positive trends declined from 43 to 27% between the two periods (Table 1), however, those few circles which experienced positive growth between 2000-2020 had large numbers of eiders counted at the start of the period, resulting in substantial gains in the number counted overall for the region.Circles with important growth (Fig. 3) and high counts in recent years (Fig. 1) have been in the Tuckernuck and Nantucket Islands in Nantucket Sound, as well as the eastern end of Cape Cod, Massachusetts, while circles to the west in this region have generally seen losses with negative trends (Fig. 3).
In Maine, trends shifted from stable or positive from 1980-1999 to mostly negative with no positive growth from 2000-2020 (Fig. 2, 3, Table 1).An example is provided from an area along the southern coast of Maine around Portland (from Pemaquid-Damariscotta to Biddeford-Kennebunkport), where five circles showed consistent average declining rates (λ = 0.91 -0.95) since 2000 (Appendix 4).Two of these circles were surveyed consistently

Localized trends in Canada
In New Brunswick in the Bay of Fundy, four circles surveyed consistently in both periods shifted from positive or stable to negative trends (Fig. 2, Table 1).The raw CBC data and predicted relationships captured dramatic declines in the number of birds observed since 2000 (Appendix 4, Fig. 2; average λ = 0.85 -0.93).
Trend analysis of total annual counts from aerial surveys conducted in the same region of New Brunswick in five years during the period 2006-2019 suggested declines, with λ = 0.96 but an upper confidence limit bounding one ([0.91,1.01], Fig. 4).A fifth nearby circle in Maine experienced relatively stable trends over the same time period with overall lower numbers of eiders reported (Appendix 4).
In Nova Scotia, trends were stable across five circles in 1980-1999, but four of 10 circles surveyed in 2000-2020 showed negative trends along the southern and eastern coast while the others were stable (Fig. 2, 3, Table 1).Trend analysis from aerial survey counts conducted along the entire Nova Scotia coastline in four years In southeast Newfoundland, CBC counts in three circles on the Avalon Peninsula and one in St. Pierre and Miquelon indicated trends remained stable in both periods (Fig. 2, 3, Table 1).Total counts from aerial surveys conducted in the same area in six years over the period 2003-2018 also indicated stability (Fig. 4).
In the Gulf of St. Lawrence, CBC circles reporting Common Eider between 1980-1999 (n = 4) or 2000-2020 n = 11) showed trends in both periods as stable or positive (Fig. 2, 3, Table 1).Importantly, the locations of CBC circles in the Gulf of St. Lawrence generally did not thoroughly capture the major wintering areas for dresseri eiders (Fig. 1).Trend analysis of total annual counts from aerial surveys conducted to overlap eider wintering areas in the Gulf of St. Lawrence in six years during the period 2003-2018 indicated an overall increase across the region (λ = 1.1 [1.05, 1.15]; Fig. 4).

DISCUSSION
We present the first quantitative evaluation of spatial and temporal changes in abundance across the winter range of American Common (dresseri) Eiders, thus providing empirical insight into the geographic extent of suspected shifts in winter distribution.The regional patterns in localized trends we observed by combined data from CBC and aerial surveys collected over the past 40 years were consistent with those suggested by expert opinion (Noel et al. 2021).Wintering populations appeared to have remained stable or increasing in the northern portions of the winter range in southeast Newfoundland and the Gulf of St. Lawrence; shifted from stable before 2000 to widespread declines since 2000 through the mid-latitudes of the winter range in southern Nova Scotia, the Bay of Fundy, and south along the coast of Maine; turned from increasing or stable before 2000 to mostly declining or stable since 2000 around the core of the wintering range in Massachusetts; and have continued to be either stable or increasing at the southern terminus of the winter range from New York south to Virginia.Importantly, our analyses provide spatially explicit trend estimates which confirm the relatively recent occurrence of a large-scale redistribution away from the center of the winter range based on localized population changes at a relatively fine spatial scale.Our confirmation of suspected spatial patterns in trends will play an important part in informing coordination and cooperation on the prioritization of harvest management and habitat conservation efforts among state, provincial, and federal regulators for this species.
It has been suggested that recent declining numbers of eiders in the mid-latitudes of the winter distribution may reflect a redistribution whereby eiders are becoming displaced to more southern wintering areas and may also be "short-stopping" in more northern wintering areas (Noel et al. 2021, Robertson et al. 2021).A similar study using CBC data to assess trends in western grebe (Aechmophorus occidentalis) numbers concluded that declines in the Salish Sea were explained at least partly by a southward non-breeding range shift of nearly 900 km during the previous three decades (Wilson et al. 2013).More recently, Meehan et al. (2021) used CBC data to examine state-and province-level trends relative to changes in winter temperatures for 16 waterfowl species within the Atlantic and Mississippi flyways, finding strong evidence of climate-influenced winter range shifts with the most positive trends in colder regions.For dresseri eiders, increasing numbers wintering in the Gulf of St.
Lawrence and steady numbers in southeast Newfoundland support the short-stopping hypothesis.Regardless of whether short-stopping is occurring, it appears that eiders have begun to abandon the Gulf of Maine and surrounding ecosystems as wintering areas while also mostly declining or remaining stable farther south in Massachusetts, and remaining relatively steady or increasing farther north (Fig. 3).Local trends in the far south of the winter range have been positive (Fig. 3) but abundance remains relatively low (Fig. 1).Overall, it seems possible that increasing numbers of dresseri wintering in the far northern extent of the range combined with large numbers congregating in key Most dresseri eiders continue to congregate for the winter period in the core of the wintering range centerd along the coast of the Cape Cod area and Nantucket Sound (Bowman et al. 2015), where aggregations of hundreds of thousands have been periodically seen from land since the 1950s (Veit and Peterson 1993).Most eiders recently tracked from breeding colonies in Nova Scotia, Maine, and Massachusetts migrated to a core winter area that stretched from Boston, Massachusetts east through the Cape Cod and Nantucket Sound region, and southwest to coastal Rhode Island (Mallory et al. 2020).In the opposite direction, wintering eiders equipped with tracking devices in Massachusetts and Rhode Island returned to breeding areas in Maine, Nova Scotia, and the St. Lawrence Estuary (Beuth et al. 2017).Thus, there is high migratory connectivity across breeding colonies of dresseri eiders within this primary wintering region.Only three circles in Massachusetts showed definitively positive growth rates, the Cape Cod and Tuckernuck Islands CBC circles, and one circle close to the New Hampshire border (the nearby circle in coastal New Hampshire also showed positive growth).The remaining circles in Massachusetts, as well as Rhode Island, showed stability or negative growth.The overall spatial patterns in abundance and trends revealed by CBCs support the notion that Cape Cod and Nantucket Sound are of increasing importance to wintering eiders, since large numbers continue to congregate here, while declines perpetuate to the north along the Atlantic U.S. coast and Maritime Canada (Fig. 1, 2, 3).
Deficiencies in long-term, baseline data for waterfowl can be addressed in part by taking advantage of citizen science programs like the CBC, which certainly have the potential to capture broad spatial patterns in trends (e.g., this study, Wilson et al. 2013, Meehan et al. 2021).Given the CBC is a general bird census program not designed to monitor sea ducks, the use of these data requires transparency regarding underlying biases or shortcomings.Spatial coverage of surveys is dictated by CBC circle locations, which are not random and do not evenly sample the range of our target species (Dunn et al. 2005).Thus, we elected to evaluate trends at the circle-level and present individual λ values for each circle with estimates of uncertainty, to maximize transparency in coverage and confidence (Gutowsky et al. 2022), rather than pooling data across circles into larger spatial units to generate aggregate regional trend estimates.While the latter approach has advantages (e.g., maximize the sharing of information across space and time to fill data gaps, incorporate multiple correlates to explore drivers, generate predictive models or simulations, Ethier et al. 2020;Meehan et al. 2021), our relatively simple approach focused only on extracting trends from negative binomial models of counts allows the reader to fully appreciate the state of the underlying data and the structure of the fitted models.
Effective waterfowl management can benefit greatly from quality estimates of abundance and rates of change over time, with measures of uncertainty, as well as an understanding of relative distribution and subsequent differential pressures faced across the annual cycle and geographic range.For dresseri eiders, we lack accurate abundance estimates at most regional scales and at the continental scale, but we have strong evidence that wintering numbers have experienced a widespread redistribution across a large portion of the range, and that this spatial shift has likely occurred within the past two decades.Waterfowl managers must now decide whether the empirical evidence of dresseri declines and winter distribution shifts provided here is sufficient to warrant and inform specific conservation, harvest, or habitat management actions or whether more information regarding the complex drivers of these changes are required for management decisions.
Eiders represent another position in the trophic food web of the Northwest Atlantic underdoing recent changes in abundance and distribution (Pinsky et al. 2013, Pershing et al. 2015, 2021, Morse et al. 2017, Meyer-Gutbrod et al. 2021), likely at least in part linked to rapid hydroclimatic changes (Pershing et al. 2015, 2018, Scannell et al. 2016p. 20, Chisholm et al. 2021, Gonçalves Neto et al. 2021, Seidov et al. 2021).Winter distributions of dresseri may be shifting both south and north, as eiders abandon traditional overwintering areas in the mid-latitudes of the range and shift to congregate in higher densities in select areas of southern New England while also becoming increasingly abundant in the northern extent of their Canadian range.One driver of this redistribution may be loss of eiders' preferred blue mussel (Mytilus edulis) prey in the Gulf of Maine, which have exhibited significant declines of more than 60% since the 1970s (Sorte et al. 2016), in part attributable to much greater mussel harvest by humans but also some of the fastest recent rates of warming in the global ocean (0.23 °C per yr during 2004-2015; Pershing et al. 2015).Eiders are also a significant source of topdown pressure on the benthic intertidal ecosystem, where they can severely reduce local mussel abundance and impede recovery following disturbance (Hamilton 2000).Unfortunately, monitoring of nearshore benthic habitats across the dresseri winter range is severely lacking, limiting our ability to disentangle the causes of shifting distributions.Overall, our work further emphasizes the need for international cooperation in monitoring and managing eiders and the marine ecosystems they rely on, and for developing a comprehensive management plan to identify, monitor, and address the stressors contributing to now welldocumented localized recent declines of this culturally and economically important species.

Fig. 1 .
Fig. 1.Contemporary distribution of Common Eider (Somateria mollisima) within the wintering range of the American Common Eider (S. m. dresseri) subspecies.(A) From all Christmas Bird Count (CBC) circles reporting at least 10 total Common Eider since 1965, maximum reported count during the period 2016-2020 (n = 127 circles reporting at least one annual count between 2016-2020, circles counted but zero eider reported shown as open circles).Acronyms reflect provinces and states (NL -Newfoundland, QC -Quebec, NS -Nova Scotia, NB -New Brunswick, ME -Maine, NH -New Hampshire, MA -Massachusetts, RI -Rhode Island, CT -Connecticut, NY -New York, NJ -New Jersey, DE -Delaware, MD -Maryland, VA -Virginia) with the exception of SPM (St.Pierre and Miquelon, France).(B) Counts from aerial surveys within the Canadian portion of the range conducted in the Gulf of St. Lawrence (GSL) and southeast Newfoundland (SE NL, eastwards from St. Pierre and Miquelon (SPM), France) in 2018 and from Nova Scotia (NS) and New Brunswick (NB) in 2019 shown summed on a 0.25° grid with the same legend as (A).

Fig. 2 .
Fig. 2. Trends of Common Eider (Somateria mollisima) on Christmas Bird Count (CBC) circles within the wintering range of the American Common Eider (S. m. dresseri) subspecies for two 20-year time periods: (A) 1980-1999 (n = 44 circles) and (B) 2000-2020 (n = 76 circles).Trends are shown as mean rates of population change per year (λ), derived from negative binomial models of counts by year fitted for each circle with sufficient data in each period.Circles are ordered by latitude, and those represented in both periods are at the same y-axis position.Estimates of λ are shown with unique color-shape combinations relative to marine or geopolitical boundaries separating key portions of the wintering population and are presented with 95% confidence intervals.Region acronyms reflect provinces and states (NS -Nova Scotia, NB -New Brunswick, ME -Maine, NH -New Hampshire, MA -Massachusetts, RI -Rhode Island, CT -Connecticut, NY -New York, NJ -New Jersey, DE -Delaware, MD -Maryland, VA -Virginia) with the exception of GSL (Gulf of St. Lawrence) and SE NL (Southeast Newfoundland, eastward and inclusive of St. Pierre and Miquelon, France).

Fig. 3 .
Fig. 3. Spatial distribution of localized trends of Common Eider (Somateria mollisima) on Christmas Bird Count (CBC) circles within the wintering range of the American Common Eider (S. m. dresseri) subspecies for the periods 2000-2020 (n = 76 circles).Trend estimates are reported as λ from negative binomial models of counts by year fitted for each circle with sufficient data.Color of the trend estimate indicates direction of change and size indicates magnitude of change.Trend estimates with confidence intervals bounding one are identified by an "X," indicating relative stability.Acronyms reflect provinces and states (NL -Newfoundland, QC -Quebec, NS -Nova Scotia, NB -New Brunswick, ME -Maine, NH -New Hampshire, MA -Massachusetts, RI -Rhode Island, CT -Connecticut, NY -New York, NJ -New Jersey, DE -Delaware, MD -Maryland, VA -Virginia) with the exception of SPM (St.Pierre and Miquelon, France).

Fig. 4 .
Fig. 4. Trends of Common Eider (Somateria mollisima) counted on aerial surveys flown within the Canadian wintering range of the American Common Eider (S. m. dresseri) subspecies between 2003-2019.Trends are shown as mean rates of population change per year (λ) with 95% confidence interval (lower confidence limit, upper confidence limit), derived from negative binomial models of total counts (not adjusted for subspecies ratios) by year fitted for each region: (A) Gulf of St. Lawrence (GSL), (B) southeast Newfoundland (SE NL, eastwards from St. Pierre and Miquelon (SPM), France), (C) Nova Scotia (NS) and (D) New Brunswick (NB; Fig. 1B).Raw counts are shown as open circles and the predicted negative binomial relationships with 95% confidence interval are shown by the solid and dashed lines.

Table 1 .
Number of surveyed Christmas Bird Count (CBC) circles with positive, stable, and negative trends of Common Eider (Somateria mollisima) within the wintering range of the American Common Eider (S. m. dresseri) subspecies for two 20-year time periods: 1980-1999 and 2000-2020.Trends were derived as λ estimates generated by negative binomial models of counts by year fitted for each circle with sufficient data in each period.Stable trend estimates are those with 95% confidence intervals which bound one.Trends are shown summarized relative to regions which represent marine or geopolitical boundaries separating key regions of the wintering population.Region acronyms reflect provinces and states (NS -Nova Scotia, NB -New Brunswick, ME -Maine, NH -New Hampshire, MA -Massachusetts, RI -Rhode Island, CT -Connecticut, NY -New York, NJ -New Jersey, DE -Delaware, MD -Maryland, VA -Virginia) with the exception of GSL (Gulf of St. Lawrence) and SE NL (Southeast Newfoundland, eastward and inclusive of St. Pierre and Miquelon, France).