Accelerating declines of North America’s shorebirds signal the need for urgent conservation action

Shorebirds are declining to a greater extent than many other avian taxa around the world. In North America, shorebirds, along with aerial insectivores and grassland birds, have some of the highest proportions of declining species of any group. Here, we apply a new hierarchical Bayesian model to analyze shorebird migration monitoring data from across North America, from 1980 to 2019, and present the most recent available estimates of trends for 28 species. Point estimates for survey-wide trends in abundance were negative for 26 of 28 species (93%). Despite challenges with low precision associated with migration count data, trends for 19 species had 95% credible intervals that were entirely negative. More than half of the species were estimated to have lost >50% of their abundance. Furthermore, estimated rates of decline have accelerated during the last three generations for most species. Point estimates of trend were more negative for 18 species (64%) during the most recent three-generation period in com- parison to the previous three-generation period. Many species now exceed international criteria for threatened species listing. The analytic approach used here allows us to model regional variation in trends, although survey coverage and strength of inference were greatest in the eastern portions of North America (east of 100°W). We found the greatest declines at staging sites along the Atlantic Coast from North Carolina to Nova Scotia, and lesser declines along the Gulf Coast and in the midcontinental United States. The declines in shorebird populations reported here are worrisome and signal the urgent need for conservation action. In addition, it would be beneficial to validate these results through the collection and analysis of com-plementary data, and to initiate demographic studies throughout the annual cycle to determine where and when declines are most likely to originate. This improved information will allow for the development of more targeted efforts to reverse declines through conservation action.


LAY SUMMARY
• Surveys of North American shorebirds during fall migration, carried out largely by volunteers, are used to monitor trends in the abundance of their populations. • Between 1980 and 2019, 26 of the 28 shorebird species analyzed were found to be declining with more than half of the species losing more than half of their abundance. • Declines were greatest along the Atlantic coast from Nova Scotia to North Carolina, and less severe along the Gulf coast and in the Midcontinent. • Declines are worsening in recent years. These large and accelerating declines mean that many species now exceed international criteria for threatened species listing. • Urgent conservation action is needed to slow and eventually reverse declines. Targeted research, and in particular studies of survival throughout the year, could help to pinpoint where shorebirds are most strongly impacted, so that conservation attention can be focused where it is most needed.

INTRODUCTION
Around the globe, approximately one million species are threatened with extinction from a combination of land use changes that degrade habitats, direct exploitation, climaterelated threats, and other factors (IPBES 2019). The conservation of biodiversity has important intrinsic and extrinsic implications (e.g., Cardinale et al. 2012, Sandler 2012, but conserving threatened wildlife species can be costly and difficult (McCarthy et al. 2012). Protecting species before they are threatened with extinction is widely considered to be both more efficient and more effective (Walls 2018). Biodiversity loss is a "wicked problem," with complex and interrelated causes, and solutions requiring a balance of competing interests (Sharman and Mlambo 2012). While the solutions are complex, a necessary first step is an understanding of which taxonomic groups are being lost at the greatest rates and the locations where the losses are occurring. Trends in abundance are among the key parameters being monitored to establish extinction risk and the status of biodiversity (IUCN 2012, WWF 2018, and threshold rates of decline are a primary criterion used to classify species into varying categories of extinction risk in national and international threatened-species evaluation processes, such as the designations of vulnerable, endangered and critically endangered by IUCN (2012).
Birds are a widespread and species-rich component of vertebrate biodiversity. With their comparatively robust monitoring information and their broad public appreciation (Belaire et al. 2015, birds are commonly presented as indicators of biodiversity loss (Rosenberg et al. 2019). Among birds, migratory shorebirds are frequently highlighted as a taxon of conservation concern in global (e.g., Stroud 2003, Butchart et al. 2010, hemispheric (Simmons et al. 2015, NABCI 2016, national (NABCI Canada 2019, Rosenberg et al. 2019) and regional , Warnock et al. 2021 assessments. In the Northern Hemisphere, many species of migratory shorebirds travel long distances from high-latitude breeding grounds to south-temperate or tropical nonbreeding areas. To fuel these long migrations, many species rely on suitable foraging conditions at a restricted network of highly productive staging sites (Myers et al. 1987, Donaldson et al. 2000, Brown et al. 2001). This reliance on a comparatively small area of habitat (Morrison 1983), distributed in space and time across hemispheres, exposes shorebirds to a wide range of threats, from habitat loss or degradation in coastal regions (Pfister et al. 1992, Stroud et al. 2006, to mortality from harvest (Watts et al. 2015, Andres et al. 2022, to disease and pollution (Buehler et al. 2010, McCloskey et al. 2013. This exposure to wide-ranging threats makes shorebird populations susceptible to decline and has led some to consider them sensitive indicators of global environmental change (Piersma and Lindström 2004).
Shorebirds' behavior of congregating at migratory staging sites, however, also presents opportunities for efficient monitoring of their populations. When large numbers of shorebirds aggregate at small numbers of sites in populated coastal regions, it is possible to monitor substantial fractions of the population with relative ease. Because of this, the monitoring of shorebirds at migratory stopover sites has a long history in North America. Coordinated surveys began in the early 1970s in Atlantic Canada, Ontario, and the eastern United States. Initially, these surveys were intended to provide an increased understanding of the distribution and relative abundance, rather than a statistically rigorous estimate of population size or trend (Howe et al. 1989; see also Morrison 2001, Ross et al. 2001. However, they have been used for trend analyses previously, using a variety of analytic methods (Bart et al. 2007, NABCI 2016, 2019. Non-random site selection, limited data on the length of time that individuals remain at each site, and uncertainty about the proportion of each population migrating through different regions all present challenges for estimating population trends from these data. Nevertheless, "migration monitoring" surveys remain the most systematically collected survey data, and indeed the only widespread survey data, for estimating trends for many shorebird species and thus are a critical tool in the status assessment and conservation prioritization for shorebirds. These monitoring data can also point to possible correlates of decline, through the evaluation of patterns in trends across species and regions.
Here, we present updated estimates of changes in population size for migratory shorebirds in North America based on counts at migration stopover sites in the continental USA and southern Canada from 1980 to 2019. In particular, we tested whether previously reported declines in population size have been worsening or ameliorating over time, and consider the recent trends in the context of thresholds for national and international threatened species designations. We develop and present a novel analytic approach that can accommodate the transient nature of the migrants' abundances, the potential spatial autocorrelation among sites, and the varying degrees of survey coverage among regions and time periods. We consider spatial and interspecific patterns in trends and suggest areas for further research into drivers of decline. We also consider the current status of shorebirds with respect to criteria for endangered species listing and discuss the implications for the conservation of North America's shorebirds and other migratory birds.

Field Survey Methods
The dataset consists of shorebird counts from the International Shorebird Survey (ISS), the Atlantic Canada Shorebird Survey (ACSS), and the Ontario Shorebird Survey (OSS). All have been carried out since the early 1970s, with surveys coordinated by Manomet Inc. in the United States and Environment and Climate Change Canada in Canada (see Howe et al. 1989, Morrison et al. 1994. Surveys are carried out primarily by volunteers, with significant contributions from state and federal protected areas biologists in some regions. Surveyors are asked to visit sites every 10-14 days during the fall migration period between July and November, although not all sites are surveyed with such regularity. Sites are surveyed by one or more observers who count all shorebirds present. Observers are asked to standardize their search effort across visits and for sites with tidal influence, to visit at a tidal stage during which birds are concentrated and easier to count (most often a high and falling tide). Where possible, the boundaries of sites are defined by natural borders to ensure that the area counted in each year is consistent. The boundaries of many sites are described in site catalogs (e.g., Environment Canada 2009) to further ensure this consistency over time.
Survey locations are distributed throughout Atlantic Canada (especially the provinces of Nova Scotia and New Brunswick), the province of Ontario, and throughout the continental United States, with a majority of sites along the Atlantic Coast and far fewer west of the Mississippi River ( Figure 1 and Supplementary Material Figure S1). Although surveys of migrant shorebirds have occurred at stopover sites in Prairie and western Canada and Quebec, the coverage during the period 1980-2019 is insufficient for inclusion in these analyses.
Selection of sites for surveys has traditionally been non-random, often based on convenience of coverage by volunteers, and is likely biased towards sites that are easily accessible and consistently used by shorebirds. Survey coverage and intensity have also varied over time (Supplementary Material Figure S2), and analyses are designed to accommodate these missing data (see below). Several efforts have been undertaken to obtain more systematic or random survey coverage, including a project on the Atlantic coast of the United States in 2011-2014 to ensure improved coverage of sites believed to account for 80% of the bird-use days, and efforts in Ontario and Atlantic Canada beginning in 2013 to increase survey effort at randomly selected locations. These efforts have served to improve coverage in some regions, but non-random site selection remains an important consideration with respect to potential analysis methods to minimize bias.
Since 2008, ISS survey coordinators have requested that data be uploaded to eBird (https://ebird.org/) through a specialized project protocol, available to select as an "observation type" within the eBird data upload process. In Canada, since 2012, survey data are uploaded by surveyors to Nature Counts (https://www.birdscanada.org/ naturecounts/), and, from there, uploaded to eBird. Thus, data are archived and freely available in a standardized format for future analyses.

Data Preparation and Filtering
We selected all observations made during the fall migration period, which we defined as survey dates between 1 July and 30 November. The migration monitoring surveys began in 1974, but the spatial and seasonal coverages were much more stable after 1980. We therefore limited our analyses to fall migration seasons in all years from 1980 to 2019. Although spring surveys have been carried out at many sites, coverage is far sparser in space and time.
Our primary aim is to describe the long-term trends in abundance of shorebirds. We present results for all species for which we could fit models that reliably converged. To be included in analyses for a given species, a survey site had to have observations of that species in at least 2 years, spanning a minimum time-frame of 10 years between those two observations. In addition, sites were only included if they were contained within a geographic stratum (defined below) that also met criteria for inclusion. Geographic strata were included if the species had been observed at least twice spanning a minimum time-frame of 20 years (half of the 1980-2019 timeseries).

Analytic Methods
Counting migratory animals as they pass through a sampling frame presents a number of analytic challenges, and these challenges are increased when the potential survey sites range in size and density by several orders of magnitude.
Because key coastal staging sites are large and well known, the monitoring programs are likely sampling a relatively large proportion of the available coastal sites and the population of shorebirds using those sites. In comparison, inland sites tend to be much smaller and less well known, and monitoring programs are likely sampling a relatively small proportion of the available sites and populations of shorebirds using them. Because the inland area of the continent is vast, the total number of potentially available inland sites is large. Therefore, despite the small counts of birds observed at the inland sites selected for monitoring, the region may actually hold a large proportion of the total species population (e.g., Skagen et al. 2008). Without further information on movements during migration, we cannot know the most appropriate weighting of sites among the major regions.
We set out to estimate population trends and annual indices of abundance under the assumption that the magnitudes of counts at particular sites were not necessarily representative of the proportion of the continental population migrating through that area. We also allowed long-term population trends to vary among geographic strata (defined below) in a spatially explicit way. We modeled counts at each survey site using a hierarchical Bayesian model that estimated smooth nonlinear patterns of population change with additional random annual fluctuations. The model, implemented in R (R Core Team 2021), is similar to the model for the North American Breeding Bird Survey, described in Smith and Edwards (2021), with two key differences: (1) we derive the survey-wide estimates of population change from the means across all geographic strata (hyperparameters) and not an abundance-weighted sum of the stratum estimates; and (2) we use a spatially explicit intrinsic conditional autoregressive (iCAR) structure to share information on population change among neighboring strata. We describe the key points of the analyses below, with complete details, links to model code, and an assessment of model fit to simulated data provided in the Supplementary Materials.

Geographic stratification.
We grouped survey sites into geographic strata to allow the model to estimate varying population trajectories across the continent while sharing information among sites in relatively close-proximity. Sites were grouped into strata defined by hexagonal grid cells ~300 km in width.
We modeled annual migration counts from each site j, in stratum i, day d, and year t as realizations of an overdispersed, Poisson distribution, with mean λ j,i,d,t . For the year effects, the structure allowed the trajectory of annual indices to depart from the overall smooth or stratum-level smooths in a given year, when such departures were supported by the data. Year effects were not allowed to vary among strata, because there were generally insufficient data to support the extra parameters, in combination with the stratum-level smooths (below), the site-level intercepts, and the relatively unbalanced sampling through time.

Seasonal smooths.
The component of the model that adjusts for variation in mean counts over the course of the fall migration season (ζ(d)) was estimated as a semi-parametric GAM smooth. We used Ω = 10 knots, spread evenly over the 150-day season, to allow the seasonal pattern to be sufficiently flexible to accommodate more than a single peak, if supported by the data. For some species, we estimated two separate seasonal patterns for the northern and southern portions of the survey area, if preliminary investigations of the data suggested that this distinction was warranted (Supplementary Material Figure S3).

Temporal smooths.
The time series components δ i (t) in the model (the smoothed population trajectories through time in each stratum) were estimated with a semi-parametric GAM smooth. We used a comparable approach to the hierarchical GAM used in Smith and Edwards (2021), where stratum-level smooths were estimated as additive combinations of a survey-wide smooth that reflects the species' overall smoothed population trajectory, and stratum-level smooths that reflect local departures from that survey-wide smooth.

Neighborhood relationships.
Although the strata were defined using a regular hexagonal grid, not all grid cells contained survey sites that could contribute to the analysis for any given species. Therefore, we defined spatial neighbors more broadly than just grid cells sharing an edge, to ensure that the adjacency matrix would be fully connected. We defined the spatial neighborhood relationships using a Voronoi tessellation of the centers of the hexagonal grid cells that contributed data to the analysis for a given species (Supplementary Material Figure S4).

Population trajectories.
The survey-wide population trajectories are the collection of estimated annual indices of abundance (N t ). These values are exponentiated sums of the mean survey-wide temporal smooth (Δ(t)), the species intercept (α), the survey-wide yeareffects (γ t ), the mean of the seasonal effects (ζ = D d ζ (d)/D), </ mathgraphic> and variance components to account for the asymmetries of the log-normal retransformation (σ 2 Accelerating declines in shorebirds require urgent action 5 </mathgraphic> . We also calculated a smoothed population trajectory based on annual indices of abundance without the yeareffects, which includes an additional variance component to account for the variance of the year-effects (σ 2 γ × 0.5). </ mathgraphic> These smoothed population trajectories provide an estimate of the medium and long-term patterns of population change after removing the effects of the random annual fluctuations and form the basis of our trend calculations.
Stratum-level population trajectories (n i,t ) were calculated in the same way except that the time-series smooth component was provided by the stratum-level smooths ((δ i (t)), </mathgraphic> and we accounted for the variance in abundance among sites by generating predictions for the set of sites included in stratum i, (s ∈ S i ) </mathgraphic> then averaging the predicted values across those sites. Similar to the survey-wide estimates, we also calculated smoothed population trajectories, without the annual fluctuations. See Supplementary Material Appendix S1 for additional details.

Trends.
We calculated population trends as geometric mean annual changes (%/year) in the estimated annual indices over specific time intervals, using interval-specific ratios of the smoothed annual indices (similar to the methods used in Fewster et al. 2000, Smith andEdwards 2021).
The interval trend from year t a to year t b for the surveywide population was and the stratum-specific trends were calculated similarly using the values of nSmooth i,t .

</mathgraphic>
These interval trends estimate the average medium-to long-term rates of change in the surveyed portion of the species' population, after adjusting for the random annual fluctuations. This particular estimate of trend is useful because it allows for non-linear patterns of population change, so the estimates of short-term trends can provide additional information to the estimates of long-term trends. In addition, trends are relatively stable between adjacent years, because the smoothed trajectories have already been adjusted for random annual fluctuations.
We report trends over the entire study period, 1980-2019, and also over species-specific periods of three generations. The magnitude of declines observed over a three-generation period (where generation time is defined as the mean age of parents of the current cohort) is an international benchmark for identifying species of conservation concern, used by the International Union for Conservation of Nature (IUCN 2012), the Canadian Committee on the Status of Endangered Wildlife in Canada (COSEWIC 2020a) and other conservation listing organizations. We obtained generation lengths from Bird et al. (2020;Supplementary Material Table S1) and reported trends from the most recent three generations (rounded to the nearest year, to match the trend data) and contrast these with trends from the previous three-generation period. The difference between the recent and the previous three-generation trends was calculated as a derived statistic from the modeled trend estimates for the two different timeperiods (Trend recent -Trend previous ) and the posterior distribution of this difference was used to estimate the probability of a worsening trend (i.e., the proportion of the posterior distri-bution that was negative). The duration of three-generation periods for the species considered here ranged from 10 years for Wilson's Phalarope and Solitary Sandpiper, to 23 years for the Black-bellied Plover and Marbled Godwit, with a mean ± SD of 14.7 ± 4.3 years. For species with three-generation times longer than half of the available time-series (i.e., >20 years), the previous three-generation trend was calculated using the remainder of the time-series (i.e., from 1980 to the beginning of the recent three-generation time period).

Regional patterns in trends.
To explore regional patterns in trends, we used an additional hierarchical model to calculate the mean difference between stratum-level and continental trends in each geographic stratum, across all of the species with trend estimates in that stratum (i.e., grid cells). For each species that occurred in a given stratum, we calculated the difference between the stratumspecific and continental estimates of trend. We used these species and stratum-specific differences in trend to summarize the spatial patterns in trends while controlling for each species' overall trend estimate.
For a given stratum, these species-specific trend differences and the standard deviations of their posterior distributions provided the data for a simple hierarchical model that estimated the mean difference across all species in the stratum, while accounting for the uncertainty of each estimated difference. This post-hoc model was applied independently within each stratum and for each trend estimate (e.g., the mean long-term trend was modeled independently of the mean three-generation trend). We then estimated the probability that this mean trend difference was positive or negative as the proportion of the posterior distribution that was above or below 0. See Supplementary Material Appendix S2 for additional details.

RESULTS
Between 1980 and 2019, there were sufficient data to produce trend estimates for 28 shorebird species. For these 28 species, surveyors observed a total of 69.6 million individuals during 82,843 surveys of 3,657 sites during the fall migration period. Most sites were not surveyed frequently enough to be included, but 668 sites contributed data to trend analyses for at least one of the species included here (Figure 1). Semipalmated Sandpiper was by far the most commonly recorded species, with more than 32 million individuals counted over the course of the study. The next most abundant species, Sanderling and Semipalmated Plover, had ~2 million individuals each, while an additional 5 species had >1 million individuals observed (Table 1).
Survey effort was greatest in the northeastern United States and Atlantic Canada, the regions in which these volunteer survey programs were initiated (Figure 1). Within these two regions, ~200 sites per year were surveyed, which was more than for all other portions of North America combined. Importantly, a majority of the sites with consistent coverage across the duration of the study period are also from these regions.

Trend Analyses
Trends were estimated for 28 species (Supplementary Material Figure S5), and point estimates for survey-wide trends in abundance from 1980 to 2019 were negative for 26 of 28 species (93%). Estimated total percent change over the 40-year period ranged from a loss of 94% for Red Knot (95% credible interval (CI): -97% to -87%) to a highly imprecise estimated increase of 15% (95% CI: -27% to +95%) for Willet. More than half of the species were estimated to have lost >50% of their abundance. Credible intervals spanned 0 for 9 of 28 species (32%), including both species for which trend estimates were positive ( Figure  2).
The rates of decline in abundance were even more substantial over the most recent three-generation period for a majority of species. Point estimates of trend were more negative for 18 species (64%) during the most recent three-generation period in comparison to the previous three-generation period (Figure 3). Among these species, the evidence of worsening trends was strong (>85% probability, based on the posterior distribution of the difference) for 13 of the 28 species (46%). When converted to total percent change over the duration of the species-specific, three-generation periods, point estimates of decline exceeded a 30% threshold for 19 species (68%) and a 50% threshold for seven species (25%; Figure  4). These decline thresholds, coupled with the prospect of ongoing threats, correspond to key criteria for IUCN listing as "Vulnerable" and "Endangered," and also the Canadian COSEWIC listing criteria of "Threatened" or "Endangered." The declines in the most recent three-generation period are centered primarily in the eastern portion of the study area, from South Carolina on the Atlantic coast of the United States, to Atlantic Canada, and inland to Ontario ( Figure  5A). Lesser declines were observed in the midcontinent of the United States and on the Atlantic and Gulf Coasts of the southern United States. This regional variability is more pronounced than was the case in the previous three-generation period ( Figure 5B). Although survey effort has changed over time, including a reduction of effort at inland and western areas in recent decades, this should not explain the increased certainty of greater declines at eastern sites.

Passage Dates
The timing of migration varied widely among species, and, for some species, between northern and southern portions of the study area. For many species, abundance increased throughout July; beginning in early July for temperate and boreal breeders and in mid-to late July for Arctic breeders. Abundance for many species reached a peak in early August, although some species peaked in abundance much later, as late as late October in the case of Dunlin (which moult at northern breeding or staging grounds). For some species, such as the White-rumped Sandpiper, a second peak a month or more later was observed, presumably reflecting the peak passage for later migrating juveniles. This peak of juvenile migration was not clearly discernible in all species. We identified a later and more protracted period of passage through the southern portion of the study area for 14 of the 28 species (Supplementary Material Figure S3). With one exception, these were all species for which the southern portion of the study area is within the stationary nonbreeding range. The one exception, Hudsonian Godwit, passed through the southern portion of the study area in October, perhaps reflecting a different segment of the population for this species with a disjunct breeding range (Senner 2012).

DISCUSSION
Our results reaffirm the urgent conservation status of shorebirds in North America and provide new insights into the species and regions showing the greatest declines. Importantly, we show that the declines are large and accelerating for a majority of species, signaling the need for immediate attention. Regional variation in migration behavior and trends, with protracted passage periods and appearance of lesser declines in the southern portion of the study area, suggest the possi-bility that some birds are spending the non-breeding period farther north, perhaps as a consequence of climate change. Although these represent the best available estimates of trends for most North American shorebirds, these migration surveys are not without weaknesses. Many of the trend estimates have broad credible intervals, survey coverage is greatest in eastern North America, and several sources of potential bias could influence counts irrespective of population change.
Understanding and addressing these weaknesses should be a priority to increase confidence in the results and target conservation action towards the species and locations where it is most urgently needed.

Large and Accelerating Declines
We report declines for nearly all of the species studied; point estimates were positive for only 2 of 28 species, and in both of these cases, the credible intervals were very large. Moreover, the estimated declines were large in most cases, with 16 of the 28 species estimated to have lost 50% or more of their abundance over the 40-year period analyzed. Migration surveys such as these are vulnerable to several important sources of bias, discussed below. However, the ubiquity and magnitude of the declines reported here provide strong evidence for widespread declines among North America's migratory shorebird species.
Alternative sources of monitoring information with which to corroborate these results are few and far between. Previous large-scale analyses reporting broad declines in shorebird abundance, such as in NABCI (2016, 2019), make use of these same ISS data, and agree with the general conclusions reported here. Similarly, site-or species-specific assessments of trends in abundance at migratory stopover sites are often based on subsets of these data (e.g., Harrington et al. 2012). ISS, ACSS, and OSS surveys have also been carried out in spring, and several shorebird species take different migration routes in spring vs. fall ("elliptical migration", e.g., Gratto-Trevor and Dickson 1994). Despite the reduced sample sizes in spring, these spring data offer a largely unexplored opportunity to assess trends at different stages of the annual cycle. However, truly independent data, such as long time-series of observations from the breeding and stationary nonbreeding grounds are rare. For Semipalmated Sandpipers, aerial surveys of nonbreeding birds along the . Annual percent change (means ± 95% credible intervals) in abundance of 28 species of shorebirds for the most recent three-generation period ending in 2019, and the three-generation period prior to that. The rate of decline is accelerating for a majority of species.

P. A. Smith et al.
Accelerating declines in shorebirds require urgent action 9 coast of the Guianas have documented a 79% decrease in abundance between 1982-1986-2011). However, observations on the breeding grounds suggest mixed population trends, with large decadal declines at several sites in the eastern portion of the range, and variable trends including local increases in the western and central portions of the range , Brown et al. 2017. Surveys of nonbreeding Red Knots in southern South America have documented dramatic declines (Morrison et al. 2004, COSEWIC 2020b, of a magnitude consistent with those reported here, but no long-term monitoring data are available for the Arctic breeding grounds. Inuit knowledge suggests large declines in the abundance of shorebirds in the Kivalliq region of Nunavut (Carter et al. 2018), and the number of Ruddy Turnstone nests monitored in a fixed study area within this region (at Qaqsauqtuuq/East Bay, Nunavut, Canada) has dropped from a high of 40 nests in 2002 to 4-11 nests per year in 2015-2019 (P.A.S. personal communication). In general, existing sources of monitoring information for shorebirds from the breeding and nonbreeding grounds are either too short-term or too small-scale to be definitive. However, the reports of declines greatly outnumber the reports of increases, consistent with the results presented here. More worrying still is the observation of accelerating declines. We report that declines are accelerating for 18 of 28 species, with strong evidence of accelerating declines for 13 of these species. The rates of decline in the most recent threegeneration period are rapid, exceeding internationally recognized thresholds for conservation status for 19 of the 28 species analyzed. These threshold rates of population decline were devised by the International Union for the Conservation of Nature (IUCN) to describe the risk of extinction, on the assumption that species showing large declines and ongoing threats are expected to continue their decline to extinction (Mace et al. 2008). These decline-based thresholds for threatened species listings have at times been controversial, for example when abundant and wide-ranging species have been listed based on past declines that are stabilizing (Webb andCarillo 2000, Mace et al. 2008). For the shorebird species discussed here, population sizes still number in the hundreds of thousands or more in most cases (Andres et al. 2012), and the risk of extinction would be eliminated if the declines were halted. However, the accelerating declines observed for a majority of species provide no evidence to suggest that population trends are stabilizing. Early conservation action to reverse declines before species become rare is widely understood to be both more efficient and more effective (e.g., Walls 2018).

Regional Variability in Trends
The declines we report here are widespread but are most pronounced in the eastern United States and parts of Atlantic Canada. This pattern is particularly evident in the most recent three-generation period, where much of the Atlantic coast of the United States north of the state of North Carolina, eastern Ontario, Canada, and parts of New Brunswick and Nova Scotia, Canada, showed trends that were at least 2.5% per year more negative than the mean trends. This pronounced regional variability could arise from at least two sources: (1) Figure 4. Total percent change in abundance over the most recent three-generation period ending in 2019 (±95 credible intervals) for 28 species of shorebirds surveyed during migration through North America in fall. Thresholds of decline of 30% and 50%, used in international conservation prioritization processes, are shown for reference. redistribution in response to regional sources of disturbance, threat, or environmental change; and (2) different trends among distinct segments of the population passing through these regions.
Redistribution of birds among sites is known to occur as a consequence of disturbance. For example, in Atlantic Canada between 1974 and 2017, migrant shorebirds increasingly aggregated at the largest and safest stopover locations, as pressure from introduced Peregrine Falcons (Falco peregrinus) increased (Hope et al. 2020). Changes in food resources can also affect distribution among sites. For example, Red Knots in Delaware Bay are well known to alter their distribution within and among seasons in response to the local abundance of the eggs of horseshoe crabs (Limulus polyphemus), a preferred food (Karpanty et al. 2006). This among and within site redistribution could lead to potential bias in our trend estimates (see below) but would not explain the regional-scale variation in trends that we observed.
Evidence for redistribution of birds at regional or flyway scales is far less common. Afro-Eurasian Ruffs (Philomachus pugnax) have shifted their global range eastwards in response to reduced habitat quality at staging sites in the western por-tion of their migratory range (Rakhimberdiev et al. 2011, Verkuil et al. 2012. The Richard's Pipit (Anthus richardi), a migratory songbird breeding in Siberian grasslands and traveling to Southeast Asia, has recently adopted a new migration route and is now regularly sighted in western Europe, perhaps as a consequence of large-scale changes in climate (Dufour et al. 2021). Similarly, Rufous Hummingbirds (Selasphorus rufus), which breed in the western United States and Canada and migrate to Mexico, were observed only as migrants in western North America until the 1970s, after which time they became increasingly common as wintering birds along the coast of the Gulf of Mexico and the Atlantic coast of the southern United States (Hill et al. 1998). While similar large-scale shifts in migration routes are possible for the shorebirds studied here, there is little evidence to support this possibility. And regardless, because most regions are declining (albeit to varying extents), the potential for redistribution does not alter the interpretation of population-level declines.
Perhaps more likely is that the regional patterns in declines arise from differing population trends in the subpopulations, or segments of the populations, using these areas. Previous studies have shown larger declines in more easterly breeding Figure 5. Regional variability in trends across species, expressed as the estimated mean of the species-level differences in stratum-level trend, relative to the species' continental estimate (expressed in % difference per year, where +2.5% can mean 2.5% less negative). The size of the icons reflects the posterior probability that the mean deviation across species is positive (cool colors) or negative (warm colors). Results are presented for the most recent three-generation period (A) and the previous three-generation period (B).
populations of Semipalmated Sandpipers . The breeding range of Semipalmated Sandpipers in the Eastern Canadian Arctic overlaps to a large extent with hyperabundant populations of Snow Geese (Chen caerulescens). The habitat damage caused by overabundant geese could be limiting the sandpipers' survival or reproductive output from this portion of the range (Flemming et al. 2016). Several other threats are also concentrated in eastern or coastal regions. For example, shorebird species are exposed to potentially unsustainable rates of harvest in the Caribbean and northeastern coast of South America (Watts et al. 2015, Watts andTurrin 2016). Declines in the Lesser Yellowlegs (Tringa flavipes), for example, have been suggested to be associated with the exposure to this harvest in the Atlantic Flyway (McDuffie et al. 2021). Beaches and other coastal areas in the Atlantic Flyway are also subject to significant human disturbance (Pfister et al. 1992), habitat modification such as shoreline hardening (Prosser et al. 2018), overfishing with direct or indirect effects on resources for foraging shorebirds (Niles et al. 2009), and other threats. Loss and degradation of key habitats at migration staging areas in the Yellow Sea have been shown to underlie the large declines in the abundance of shorebirds in the East-Asian Australasian Flyway (Studds et al. 2017), and similar mechanisms could be operating to create disproportionate risks for shorebirds using the Atlantic flyway. However, other flyways are not without threats, such as the ongoing loss of wetland and native grassland habitats in the midcontinent (Environment Canada 2013), and a comparison of relative threats or mortality among flyways has not been attempted.
Declines were less pronounced in the most southern portions of our study area, in the Southern United States and Gulf Coast. We also identified a later and more protracted period of passage through this portion of the study area for 14 of the 28 species. With one exception, these were all species for which the southern portion of the study area is within the stationary nonbreeding range. The protracted passage and lesser declines in the south could suggest birds lingering or spending the boreal winter farther north as the winter climate warms (La Sorte and Thompson 2007). This range shift, with more birds wintering farther north and therefore remaining within our study area, could give the appearance of a lesser decline in abundance for these species, irrespective of the overall rate of population decline.

Sources of Potential Bias
These trend estimates are the best available evidence and suggest large and widespread declines. However, there are important sources of potential bias that should not be ignored, including changing lengths of stay at stopover sites and redistribution of birds from surveyed to non-surveyed areas. Changes in the length of time that individuals remain at a stopover site directly influence the counts at that site, irrespective of actual trends in population abundance. Counts of Western Sandpipers at a study site in British Columbia declined by 18% per year between 1992 and 2001. However, this dramatic apparent decline could be explained by the observed four-fold decline in the length of stay at this site in response to increased disturbance and predation risk from an increasing population of Peregrine Falcons (Falco peregrinus; Ydenberg et al. 2004). Peregrine Falcons have also increased in abundance in coastal regions of eastern North America, and are altering the behavior of migrant shorebirds there. At coastal sites in Virginia, for example, Red Knots avoid using stopover sites in the vicinity of active Peregrine Falcon nesting platforms (Watts and Truitt 2009).
Reduced lengths of stay in response to the increasing abundance of Peregrine Falcons could create the false appearance of population declines. However, the small amount of evidence available at present does not support this hypothesis. Radio-telemetry studies in the Bay of Fundy showed that stopover duration of Semipalmated Sandpipers has not decreased, but rather increased from ~2 weeks in the 1980s to 3 weeks in the 2010s, during which time passage population estimates have declined by 50% (Neima 2017). Additional work is needed to estimate current lengths of stay at key stopover sites, for example using available MOTUS telemetry ) data or other tracking technologies, and to compare these values to the fragmentary historical data.
Redistribution of birds is another important source of potential bias, but as for length of the stay, current information does not support that this bias is leading to a false impression of population declines. A net shift of birds away from surveyed sites to non-surveyed sites over time would lead to reduced survey counts and appear as a population decline. This pattern could arise if, for example, habitat change or disturbance caused birds to shift away from large, traditionally important sites in favor of smaller, more dispersed sites that are less likely to be monitored (Bart et al. 2007). While this bias is plausible, there is no evidence to suggest that it is occurring. For northeastern portions of the study area, birds in fact appear to be increasingly aggregated at larger sites (Hope et al. 2020). Conversely, birds passing through interior areas such as the prairie potholes have always exhibited variable site use, in response to the availability of wetlands Knopf 1994, Skagen et al. 2008), and there is no indication that this variability has changed over time.
Survey coverage in the midcontinent region, and especially in regions farther west, is sparser in space and time than in more eastern portions of the study area. While this could add uncertainty to the overall trend estimates, the hierarchical nature of the model safeguards against bias or undue influence of these more sparsely sampled regions. Nevertheless, improved coverage of these migration surveys in the west would serve to provide a more complete picture of rangewide population status.
These and other sources of potential bias should be considered carefully. However, with large declines observed in the counts of 26 of the 28 species studied, the weight of evidence unequivocally suggests widespread population declines.

The Need for Urgent Action
We suggest that our results indicate a clear need for urgent conservation action. The worsening rates of decline for most species show that our existing conservation efforts are insufficient. The success of the American Oystercatcher Recovery Initiative in halting the decline of this species, and then achieving a 23% increase in the population along the eastern coast of the United States between 2009 and 2019 (Macdonald 2021), shows that focused conservation action based on a clear understanding of factors limiting population growth can successfully restore shorebird populations.
Many of the potential conservation actions needed by North America's shorebirds will require additional investment. Flyway-scale conservation plans have been developed to guide these investments (e.g., AFSI 2015, Senner et al. 2016), but none of these plans has yet succeeded in garnering the financial support necessary for the proposed actions. Taking action earlier helps to avoid the much larger costs associated with attempting to recover a species after it becomes severely depleted (e.g., Martin et al. 2012, Walls 2018.
The species considered here are on a trajectory towards endangerment, with 19 of the 28 species considered showing declines that exceed At Risk listing thresholds for IUCN and COSEWIC. Formal evaluations should be carried out by the endangered species programs in Canada, the United States, and other countries within these species' ranges, to increase the available funding and legal mechanisms for conservation. However, conservation action in the interim could lead to more effective and efficient outcomes. Targeted research is also critically important to improving confidence in monitoring results, including better estimates of stopover durations and their potential contribution to observed counts, and demographic research to pinpoint locations and timing of population bottlenecks. This latter aspect, in particular, demographic studies throughout the whole annual cycle, could help to focus conservation action in space and time and make the challenge of conserving these globe-spanning species more tractable.

Supplementary material
Supplementary material is available at Ornithological Applications online.

Acknowledgements
First and foremost, we thank the many volunteers who contributed their time, in some cases over a period of decades, to collect the survey data analyzed here. Support and coordination of the surveys were provided by Environment and Climate Change Canada and Manomet Inc., with additional significant contributions from the United States Fish and Wildlife Service. We thank the ISS Coordinators Lisa Schibley, Arne Lesterhuis, and Juliana Almeida for their leadership in the collection and management of the data used in this manuscript. Brandon Edwards assisted with aspects of model development.

Funding statement
Funding partners include the National Fish and Wildlife Foundation, United States Fish and Wildlife Service, BAND Foundation, Bobolink Foundation, R. Howard Dobbs, Jr. Foundation, Knobloch Family Foundation, Nuttall Ornithological Club, Environment and Climate Change Canada, and Wader Quest. PAS is supported through a grant from Natural Sciences and Engineering Research Council of Canada (NSERC).

Author contributions
P.A.S., B.H., R.I.G.M., S.B., and A.C.S. conceived the idea, design, experiment (supervised research, formulated question or hypothesis). C.F., J.P., and B.W. collected the data and con-ducted the research. All authors wrote the paper. A.C.S. and P.A.S. analyzed the data.