Data collection
Study species
Piping plovers are small-bodied (43–63 g) migratory shorebirds. On the Atlantic coast of North America, where the species is federally listed as Threatened, their breeding range extends from the Canadian Maritimes to North Carolina, and their wintering range occurs from the mid-Atlantic United States to the Gulf of Mexico and the Caribbean [10]. In coastal areas, piping plovers typically nest on open sandy beaches and forage on a variety of invertebrates along the shoreline and intertidal zone [10].
Common Terns are globally-distributed, mid-sized (110–145 g) seabirds that breed throughout the Northern Hemisphere and winter along tropical and subtropical coasts [11]. Along the northern Atlantic coast of North America, Common Terns nest colonially on islands and barrier beaches from Newfoundland and Labrador, Canada to South Carolina, USA [11]. During their northerly migration, common terns have been observed flying over open ocean more than 50 km off the coasts of Virginia and Massachusetts [12,13,14].
Capture and tagging
We tracked movements of piping plovers and common terns using uniquely coded VHF transmitters (“nanotags”, Lotek Wireless, Ontario, Canada) at focal breeding sites in southern New England. We were interested in tracking breeding-season and fall migratory departure movements of these species along the U.S. Atlantic Coast and adjacent OCS waters from Cape Cod, Massachusetts in the north to Back Bay, Virginia to the south. As of May 2022, this area contains 25 BOEM Renewable Energy Commercial Lease Areas and one Research Lease Area [6].
We tagged piping plovers between 2015 and 2017 in two principal breeding locations: southeastern Massachusetts (Monomoy Island and South Beach), and the southern coast of Rhode Island (Fig. 1). Monomoy and South Beach account for ~ 9% of the Massachusetts population of piping plovers, while focal sites in Rhode Island supported at least one third of nesting pairs monitored in the state [14]. From 9 May through 27 June, we captured adult plovers at their nest sites during the egg incubation period using funnel traps. A small number (n = 10) of piping plovers were tagged in the Bahamas during March 2017 with the aim of tracking northbound migratory movements. Tagging sites were in the northwestern Bahamas on North Andros (Young Sound and Kamalame Cay) and the Joulter Cays. Plovers were captured in diurnal foraging areas using drop nets. All piping plovers were fitted with Lotek NTQB-4-2 (1.1 g; 12 × 8 × 8 mm; < 3% body mass) nanotags with 16.5-cm antennas. Tags were attached by clipping a small area of feathers from the interscapular region and gluing the tag to the feather stubble and skin with a cyanoacrylate gel adhesive [8].
Between 2014 and 2017, we tagged common terns at several breeding colonies across the same southern New England range (Fig. 1): at Great Gull Island, NY, in eastern Long Island Sound; at Monomoy Island (2014); and at three colonies in Buzzard’s Bay, Massachusetts (Bird, Ram, and Penikese Islands; 2016–2017). From 9 June to 12 July, staff at nesting colonies used walk-in treadle traps to capture adult common terns at their nests, within approximately 1–5 days of their eggs hatching. Common terns were fitted with waterproofed Lotek NTQB-4-2 nanotags with 1-mm tubes at the front and back of the transmitter, bringing the total tag weight to 1.5 g. The transmitter and attachment materials weighed < 2% of the body mass of tagged terns. We attached nanotags to the dorsal inter-scapular region using cyanoacrylate adhesive and two sutures (Prolene: 45-cm length, 4.0, BB taper point needle, catalog # 8581H) inserted subcutaneously and secured to the end-tubes of the transmitter [14].
All transmitters were programmed to continuously transmit on a shared frequency of 166.380 MHz from activation through the end of battery life. Times between transmissions (burst intervals) were specific to each transmitter and ranged from 4 to 6 s. All transmitters were uniquely identifiable on the shared frequency using a combination of an encoded tag ID and known burst intervals in coordination with the Motus Wildlife Tracking System [9]. The expected life of the NTQB-4-2 nanotags ranged from 146 days (4 s burst interval) to 187 days (6 s burst interval). For each tracked individual, we tallied the total number of transmissions received and calculated the duration of transmission by subtracting the deployment date from the date of last transmission received.
Automated telemetry stations
We obtained signals of tagged birds using an array of land-based automated radio telemetry stations (i.e., stations that automatically scan for organisms with uniquely-coded VHF transmitters and record detections; hereafter, stations) throughout the study area (Additional file 1: Table S1). Typical station specifications consisted of a 12.2-m tall radio antenna mast supporting 4–6, 9-element (3.3 m) Yagi antennas mounted in a radial configuration at 60-degree intervals. Antennas were connected to ports on a receiving unit (Lotek SRX-600 or 800, Lotek Wireless, Ontario, Canada) via coaxial cable (TWS-200). Each receiving station was operated 24 h per day using one 140-W solar panel and two 12-V deep- cycle batteries. When tagged birds were within detection range, the receivers automatically recorded transmitter ID number, date, time stamp, antenna (defined by monitoring station and bearing), and signal strength value of each detection. All raw data and metadata were uploaded to the Motus Wildlife Tracking System for processing [9]. We used the ‘motus’ package in R to download and filter detection data using default criteria [15].
Data analysis
Bootstrapped inclusion values
We began by using detection radius estimates and active dates provided on the Motus website [16] to combine any receiving stations that were placed in identical locations and eliminate any stations or station groups that were not active during the full study period. We then used a bootstrapping approach to assess how well different subsamples of the population represented occurrence patterns of un-sampled individuals [17, 18]. This involved repeatedly selecting random subsamples of n individuals from the overall sample N and calculating the inclusion value. We defined the inclusion value as the proportion of stations used by the subsampled population that were also used by the remaining individuals not included in the subsample. We conducted all analyses in R [19].
For each sample size n up to a maximum of N − 1, we repeated the process 50 times and calculated the mean (µ) and standard deviation of the inclusion value for a given sample size (In) over the 50 random subsamples. We then fitted a non-linear model to the means of the bootstrapped samples of the form µIn = an/1 + bn, where a and b represent numeric coefficients. We estimated parameter values via non-linear least squares estimation, with starting values of a = 1 and b = 0.1. We used the modeled relationship between the inclusion value and sample size to identify sample sizes required to achieve inclusion at three different levels—80%, 90%, and 95%—that approximately correspond to confidence intervals commonly used to detect significant effects. After fitting a non-linear function to the data, we calculated the representativeness for any given sample size (Rn) by dividing the projected inclusion value for that subsample (In) by the asymptote of the fitted non-linear function (Imax): Rn = In/Imax. We also calculated the representativeness of each capture site as the asymptote for that capture site alone, divided by the asymptote across the full population (R(capture site) = Imax(capture site)/Imax(population)).
We calculated representativeness across different combinations of tagging sites and years by constraining our random draws to specific number of tagging sites (1–2) or years (1–3), then comparing the inclusion values from each combination of sites and years with the asymptotic inclusion value Imax for the overall sample. Since the Bahamas were only sampled in one year for piping plovers, we removed individuals captured at this site from analysis of site-year effects. Similarly, since common terns were only captured at one site (GGI) in 2015, we removed this year from analysis of site-year effects but retained all years with two distinct capture sites.
Finally, for piping plovers, we also assessed the detection probabilities of detecting at individual receiving stations by calculating the probability that each used station would be included in the subset of stations used by a bootstrapped subsample of n individuals (Tn). Since common tern sampling was designed to detect individuals during post-breeding staging but not during migration [20], we expected station characteristics to play a limited role in station-specific detection rates for this species; we therefore did not assess this metric for common terns.
Modeling
We fit generalized additive models (GAMs) with residual maximum likelihood [21] using the mgcv R package [22] to determine the effects of dividing any given sample size across multiple tagging sites or years on both inclusion values and their variability. We modeled either the mean or the standard deviation of the raw bootstrapped inclusion values (response) as a function of number of transmitters (smoothed predictor), number of tagging sites (categorical predictor), number of years (categorical predictor), average duration of transmission (i.e., number of days between tag deployment and last detection; smoothed predictor), and average number of detections per individual (smoothed predictor). We identified a predictor as significant if its 95% confidence interval (CI) did not overlap zero. To assess factors affecting inclusion probabilities for used stations, we modeled means and standard deviations of station-specific inclusion probabilities Tn as a function of number of transmitters, distance to tagging site, number of receiving antennas, antenna altitude, and number of detections at the station (smoothed predictors).