Many populations of migratory birds are currently declining. Understanding of space use throughout the entire annual cycle, as well as migratory connectivity (i.e., geographic linkage of individuals and populations across different stages of the annual cycle), can improve our ability to identify factors driving population declines and influencing extinction risk. The main objectives of our study were to i) document the space use and phenology of migration during the non-breeding period and ii) quantify the degree of migratory connectivity across the range of the American Golden-Plover (Pluvialis dominica) breeding across the North American Arctic. American Golden-Plovers breed across their entire breeding range (northern North America) and migrate up to their main wintering site located in South America. We used archival light-level geolocators to track the migration. We quantified migratory connectivity based on the non-breeding range spread of all individuals and the breeding population spread. We used Mantel tests to evaluate whether the relative spatial configuration of the sampled breeding area was preserved on the non-breeding ground. We identified 13 and 7 stopover sites used during the fall (post-breeding, southbound) and spring (pre-breeding, northbound) migrations, respectively, and one main site used during the wintering period. We highlight stopover sites that were previously unknown and show the transatlantic and transpacific routes used by plovers during migration. We found that individuals breeding in proximity tended to be closer to each other during brief and highly limited portions of the non-breeding period. Broadly, individuals from different breeding populations were well mixed during the wintering period and throughout most of the spring and fall migrations. Overall, the migratory connectivity of American Golden-Plovers is relatively low for most of the non-breeding period, suggesting that breeding populations separated by large distances should be similarly affected by disturbances and changes encountered at some migratory stopovers and on the wintering area.

Capturing and marking plovers

Incubating adult American Golden-Plovers were captured at eight study sites distributed across the species’ entire breeding range (Figure 1B). The studied breeding populations were separated by an average distance of 1,864 km, with a maximum separation of approximately 3,800 km (between Nome and Bylot Island; Figure 2). Plover nests were located by searching suitable habitats, and individuals were captured using a 60-cm dome bow net placed over the nest. Birds were fitted with metal and plastic bands, as well as a light-level geolocator (archival data logger) attached to a leg flag, all placed on the tibiotarsus (Figure 1A; Appendix Table S2). Geolocator models deployed included the British Antarctic Survey MK10b (approximately 1.1 g) and the Migrate Technology Ltd. Intigeo W65A9RK (approximately 0.9 g). These devices represented less than 0.8% of the minimum body mass of the plovers (mean mass: 142.4 g; range: 126–162 g; n = 25). Geolocators are archival devices that estimate latitude and longitude by recording ambient light levels (Isakov et al., 2012) and must be retrieved to download data. Geographic variation in the timing of sunrise and sunset is used to estimate the geographic location of individuals (Lisovski et al., 2012). Day and night lengths are approximately equal across latitudes during the spring and autumn equinoxes, which increases uncertainty in latitude estimates during these periods, but not in longitude estimates.

A total of 262 geolocators were deployed from 2009 to 2015, and 45 of them (~17%) were retrieved by recapturing plovers 1 to 4 years after their initial capture (Appendix Table S.1). Distance between nest locations of marked individuals monitored more than one year was 319 m on average (range 54 m to 1,119 m; n = 22). No effects of geolocator on annual survival was detected (Weiser et al., 2016) and the relatively low recovery rate of geolocators likely reflects the difficulty of re-sighting and recapturing marked plovers on their breeding grounds (i.e., the breeding site fidelity has to be high; the individual must initiate breeding and be re-observed in the field; its nest must be located and the individual recaptured prior to nest depredation or hatching). The effort deployed in the field to recapture plovers was also variable between breeding populations. Although 45 geolocators were retrieved (Appendix Table S.2), 9 of them only showed partial migration tracks due to equipment failure, which left 36 loggers with tracks covering most of the non-breeding period. Of these tracks, 33 came from different individuals and were used in the analyses, unless otherwise indicated.

Processing Geolocator Data

Plovers that were recaptured had their geolocator leg flag removed, and light data were downloaded using the Communicate program in BASTrack for BAS geolocators (Fox, 2010) or Intiproc for Intigeo geolocators (Fox, 2018).  Data were processed through GeoLight 2.0 (Lisovski et al., 2015) to convert light-level data to location estimates. Following Finch et al. (2015), we used a light threshold of 3 for BAS devices and of 2 for Intigeos, as light data output is different between models of geolocators. We converted light data into two daily locations (morning and evening values in latitude and longitude) for each bird from geolocator deployment to recovery, except when devices malfunctioned during the non-breeding period. For Arctic-breeding birds exposed to full daylight when at high latitude, the dataset was restricted to dates when nights were detected. To reduce light noise during nighttime, the lightFilter function in GeoLight 2.0 was used (Lisovski et al., 2015). Most loggers were calibrated using the rooftop method (n=27, 64%) to provide a start angle for the Hill-Eckstrom calibration (mean=‑6.63° range [‑7.33, ‑2.73], n=27). Residency periods were identified with the ChangeLight function in GeoLight 2.0 (quantile= 0.9) with a specified minimum residency length of 2 days. For movement periods or if no optimal sun elevation angle could be obtained by either rooftop or Hill-Ekstrom calibration, we used the angle provided by the rooftop calibration, and if unavailable, the civil twilight (i.e., 6°, Lisovski et al., 2015, n = 9). We filtered the estimated locations obtained with a loess filter (k=2) to remove outliers.

Following Hobson & Kardynal (2015), the migration track of each plover was smoothed with a state-space Kalman filter and the most probable path was obtained with kftrack (Sibert and Nielsen, 2002) in R. Kalman filtering provides the most probable track from location data and reduces observer bias when dealing with raw location estimates obtained with geolocators (Gow, 2016; Hobson and Kardynal, 2015). As the estimated flight speeds of American Golden-Plovers vary widely (see Johnson et al., 2020), we used a relatively high flight speed estimate (104.6 km/h, Johnson and Morton, 1976), corresponding to a maximum of 2,510 km per day, to set the diffusion component of the model. Kftrack uses an asymmetric error structure peaking on the winter side of the equinoxes, which is typical for geolocator data. Location precision was calculated from calibration data and yielded a mean error of 163 ± 75 km (n = 20), comparable to prior studies (Lisovski et al., 2012 and references therein).

As outlined by Knight et al. (2018), the use of geolocators for determining how migratory birds are spatially connected between breeding and non-breeding periods has some important limitations. The uncertainty in location estimates obtained using geolocators can be up to 300 kilometers, particularly around the equinox. Hence, this method is not adequate for detecting small-scale spatial segregation between individuals from different breeding populations. Moreover, our conclusions apply only to individuals who successfully returned to their breeding site and initiated nesting in more than one year. We cannot exclude that those not recaptured on their breeding site may have used different migratory strategies.

Non-breeding site use

We combined geolocator data obtained from all individuals to characterize non-breeding site use. We first defined the location of ‘non-breeding sites’ by inspecting each track and by identifying clusters of locations where movements had lost directionality and became erratic for at least 5 consecutive days. Hence, the minimum stopover duration at a given non-breeding site could be 5 days. The contour of a cluster associated with a given individual was first defined using a minimum convex polygon (MCP 95%) (Ghetta et al., 2022). Clusters obtained for different individuals were merged into a single cluster when the contour lines overlapped with the contour of an adjacent cluster. All locations assigned to the same cluster were subsequently used to define the boundary of each non-breeding site using Kernel density estimates (75%; (R library ks - function kde, Duong, 2018). As American Golden-Plovers are terrestrial birds, non-breeding sites were clipped to remove any areas over oceans. The southernmost site used the longest by an individual was designated as its wintering site, while stopovers were designated according to the direction of the migratory path (fall stopovers for Southward movement; spring stopovers for Northward movement). Departure date was the date of the first location out of a given site when birds initiated unidirectional movement away from the site, and arrival date was the date of the first location within the site when movement was reduced, lost directionality, and became erratic.

Quantitative measure of migratory connectivity

We investigated the temporal change in migratory connectivity during the non-breeding period using the Mantel test (Ambrosini et al., 2009; Cohen et al., 2018; Knight et al., 2021; Vickers et al., 2021). We characterised the migratory connectivity relative to the breeding locations of individuals by estimating the correlation between two distance matrices (Goslee, 2007). The Mantel correlation coefficient (r_M) can range from −1 to 1, with 0 indicating random mixing of individuals when comparing their breeding spatial distribution with their spatial distribution during the non-breeding period, 1 indicating that individuals retain their relative spatial positions across seasons (Ambrosini et al. 2009). A positive r_M value does not inherently imply that individuals from a breeding population are in close spatial proximity to one another during the non-breeding period. Rather, it indicates that the relative spatial configuration of the sampled breeding area is maintained at non-breeding sites.

During the non-breeding period, daily distance matrices (measuring distances between all individuals on a given day) were generated, and a new r_M value was calculated for each date, alongside a confidence interval that was computed using a bootstrap approach (Goslee, 2007). The r_M thus represents the degree of spatial organisation of individuals at a specific date relative to their breeding distribution. As geolocators can provide up to two locations per day, we used the centroid of those locations to calculate the great circle distance between two individuals positioned on a given date.

Statistical significance of the Mantel correlation coefficient was determined by random permutations. We randomly permutated the position of individuals at the breeding grounds 9999 times; for each permutation, a distance matrix was calculated, and its correlation coefficient with the actual distance matrix of individuals at a given date of the non-breeding period was calculated. The significance of the observed Mantel coefficient was assessed by comparing its rank among the coefficients generated through the randomization procedure (Ambrosini et al. 2009). When significant connectivity was detected (r_M > 0), we investigated the process that generated it by assessing the number of potential clusters in the case of migratory structuring (using the ‘pamk’ function in the R package fpc (Henning, 2025); see Ambrosini et al. 2009 and Ramos et al 2015). The number of clusters was identified as the number that maximized the overall average silhouette width (oasw), a measure of the goodness of fit of the overall classification of points in a given number of clusters (Rousseeuw, 1987).

To quantify the strength of migratory connectivity and to facilitate the interpretation of temporal variation in r_M, we extracted, for each distance matrix calculated for different dates, the average distance observed between all individuals. This provided a proxy of the non-breeding range spread of individuals at a given date (Finch et al., 2017) and an indication of the variation through time in the scale of the spatial structure. (Cresswell and Patchett, 2024). We also calculated for each date the mean distance among individuals originating from the same breeding area to examine the temporal variation in breeding population spread across the non-breeding period (Finch et al., 2017). For this purpose, we used breeding areas represented by >2 individuals. The Ikpikpuk River and Utqiaġvik study sites were grouped into a single Alaskan breeding population as they are separated by only 111 km. Finally, we measured the distances between locations obtained on the same date but during two different years for the same individuals to assess individual consistency in space use and timing (Bauer et al., 2015).

Individual locations could not be estimated every day during the non-breeding period because of extended daylight periods at high latitudes or imprecise estimations of latitude around the equinox. Hence, we could not compare the distance among individuals on those dates. We first ran the analyses using all individuals (n = 33), which included birds tracked during most of the non-breeding period but located less regularly (total of 115 days, spread from October 7 to May 3; Nome: n=2, Utqiaġvik: n=1, Ikpikpuk River: n=1, Caw Ridge: n=1, Churchill: n=4, Coats Island: n=1, Igloolik: n=6, Bylot Island: n=17). We also repeated our analyses (variation of r_M and of mean distance between individuals through time) using a subset of 20 individuals that were regularly located at the same dates over most of the non-breeding period (total of 149 days, spread from August 27 to May 14). All studied breeding locations were represented (Nome: n=1, Utqiaġvik: n=1, Ikpikpuk River: n=1, Caw Ridge: n=1, Churchill: n=3, Coats Island: n=1, Igloolik: n=3, Bylot Island: n=9). Results were similar in these two analyses (see Results), and thus the description in the main text is based on those obtained with the sample of 33 individuals. Because geolocators were deployed over different years, the 33 individuals were not tracked over the same annual cycle. Therefore, we also ran our analyses separately for each annual cycle and compared the results with those obtained using pooled years. All distances provided in the results are great-circle distances, and all analyses were performed using R version 3.5.3 (R Core Team, 2019).

Ethics statement

All methods, including capture, banding, and tagging, were approved by the Committees for Animal Care relevant to a particular study site.  Also, federal (Parks Canada, Environment and Climate Change Canada, US Fish and Wildlife Service, US and Canadian Bird Banding Offices) and state and provincial permits were obtained for attaching tags and recapturing birds to remove tags.