Hidden space use behaviors of a nonbreeding migratory bird: The role of environment and social context

Movement behavior strongly mediates species and environment interactions, yet our understanding is constrained by challenges tracking space use at fine spatiotemporal resolutions. Using an automated telemetry array, we quantified variation in and drivers of space use for a nonbreeding population of migratory bird, the American redstart Setophaga ruticilla. We identified three distinct and common behaviors – territoriality, floating, and transience – that were governed primarily by precipitation-driven resource availability. Within seasons, declines in weekly resources increased the prevalence of forays and the area of space utilized. Individuals adopting these alternative space use behaviors were less likely to maintain body condition throughout the nonbreeding season, which is expected to negatively influence fitness and survival. Our study demonstrates that nonbreeding migratory birds exhibit a high degree of plasticity in space use that is driven by both resource availability and the social environment. Ultimately, shifts in the distribution of space use behaviors are likely to occur due to broad-scale climatic changes, which are expected to have important implications on migratory populations.

Methods
Study System

This study was conducted January-May of 2016-2019 on a nonbreeding population of American redstarts at the Font Hill Nature Preserve on the southwest coast of, St. Elizabeth Parish, Jamaica (18° 02’ N, 77° 57’ W, < 5 m above sea level). Redstarts are relatively easy to capture, tag, and monitor, allowing us to not only track their changes in space use within and between seasons but also to investigate the consequences of these behaviors. Resource availability in our Jamaican system can vary widely within and between nonbreeding seasons. Further, though most redstarts are territorial in this population [55], redstarts have been shown to exhibit a diverse mix of space use behaviors that range from territoriality, occupation of home ranges, mixed species flocking, and floating [55–58].

To quantify the movement behavior of individuals, we employed two tracking methodologies: one based on manual hand tracking and another automated telemetry using a network of 5 automated receivers (sensorgnomes). Hand tracking allowed us to follow individuals at a coarse temporal resolution (hourly-daily) but at a precise spatial resolution (< 5 m). The network of automated receivers provided a very fine temporal resolution (detections every ~10 seconds) but the coarse spatial resolution (~300 m detection range of each station; presence/absence). Coupling both approaches allowed us to gather a more complete understanding of both the spatial and temporal variability in movement behavior.

Over the course of the study, we equipped a total of 141 redstarts (N2016 = 23, N2017 = 66, N2018 = 15, N2019 = 37) with a 0.29 g digitally coded VHF radio transmitters (NTQB1-1 & NTQB2-1, Lotek Wireless Inc., Newmarket, ON, Canada) using a modified leg-loop harness [59]. Transmitters operated continuously at a ~ 10.3-second cycle, which resulted in a battery lifespan of approximately ~30 - 45 days. Individuals were captured randomly across the study sites using a passive netting approach to avoid potentially biasing our tagged sample towards territorial individuals that are more susceptible to capture using playback. Further, to avoid the effects of the migratory preparation on individual movement behavior, we only included individuals tagged before April 1 of each year, which is about 4-5 weeks before May 5, the average date of departure for this population [60,61]. Of the 141 redstarts radio-tagged, a final sample of 74 individuals (39 males, 35 females) were ultimately included in this study. Upon capture, we classified individuals into age and sex classes using plumage and feather characteristics [62] and then uniquely marked them with a combination of USGS aluminum band and two-color bands. We measured standard morphometrics, including body mass (g), fat score, pectoral muscle size [63], and tarsus length (mm). We estimated body condition using the scaled mass index following [64], where body mass was scaled by tarsus size and reported as standardized grams. To investigate the potential role that dominance status plays in influencing space use behaviors (territorial vs. floater), we relied on age and sex as well-established proxies of dominance since adult males have been demonstrated to be dominant over females and young males [65,66]. Further, to investigate the potential consequences of space use behaviors on the maintenance of individual condition across the nonbreeding season – an important contributor to an individual’s fitness [25,61,67,68] – we attempted to recapture as many individuals as possible throughout the season (Mean Capture Window = ) to quantify changes in body condition.

Movement behaviors are often tied to changes in resource availability, and to explore the plasticity in space use behaviors, we attempted to link changes in the distribution in individual movement behavior to changing resource availability both between years and within a season at weekly intervals. In this system, changes in food availability mirror changes in habitat quality driven by rainfall [61,69,70] because drought negatively impacts leaf biomass and subsequently arthropod biomass declines [70]. As such, we utilized normalized difference vegetation index (NDVI) as a proxy for resource availability that provided both a seasonal metric of resource availability to facilitate comparisons across years but also captured changes in resource availability within season. We extracted NDVI across the study site from weekly Sentinel-2 multispectral imagery (10 m resolution) sourced using Google Earth Engine [71].

Manual Hand Tracking
Following the initial capture and tagging, we used a standardized protocol of localizing each individual daily through a mix of triangulation and homing localization techniques using an SRX-800 (Lotek Wireless Inc., Newmarket, ON, Canada) and a 3-element yagi. Each day, we searched for every active tag (once per round) for approximately 5-10 minutes near the capture location or the last location the individual was detected. Upon detection, individuals were either localized via homing (identified visually via color bands), and a GPS location taken (<1 m) or triangulated by a single observer taking multiple bearings to the tag within a 3-minute period. Because of the close distances (< 50 m), triangulated positions were relatively precise (± 5m; LOAS software, Ecological Software Solutions). If an individual was not detected during the initial 5-10 minute search of that round, nor opportunistically while traversing the study site between tag detections, we expanded our search to include all areas of the study site that were not originally traversed for approximately 60 minutes. Because the nanotags are digitally encoded and transmitted on a single frequency, we were able to scan the 166.380 mHz continuously while tracking, enabling us to locate individuals as we moved through the study site. Therefore, this protocol allowed us to confidently assess whether an individual was still alive, died, or departed the study site (~ 200 ha). Of the 74, 11 individuals were likely transients as they were only tracked a few times before ultimately relocating away from the study site and were excluded from further analysis. Following this protocol each season daily from ~ 6 am - 1 pm, which resulted in an average of 26.5 locations per individual (range = 5 - 108) for the 63 individuals included in subsequent analyses

We quantified space use and the variability in movement behavior in two complimentary ways using this temporally coarse but spatially precise hand-tracking data. First, to quantify an individual’s space use area, each individual’s ‘home range’ size was calculated as the area of the minimum convex polygon. While any estimate of home-range area is sensitive to sample size [72], our aims with this approach was to identify large deviations in home range area that amounted from individuals utilizing different space use tactics and not in relatively minute differences between individuals employing the same tactic. Further, as opposed to alternative home range estimators such as KDEs, MCPs typically underestimate home ranges at smaller sample sizes thus only likely underestimating home range estimates for floaters that already occupy areas that are orders of magnitude larger than average territory size. As such, MCP approach allowed us to compare the scale of space use in a continuous way across all individuals irrespective of their space use behavior because differences in home range area varied by orders of magnitude between territorial (< 0.5 ha), floater (> 2-10 ha), and transient individuals (> 10 ha). Second, , to draw comparisons between previous work and evaluate outcomes of space use behavior on individual performance (detailed below), we used net square displacement (NSD) to broadly categorize individuals into sedentary (‘territorial’) and alternative space use behaviors (‘floaters’ and ‘transients’). NSD is a frequently used movement metric that captures the scale and breadth of an individual’s trajectory as it measures the square of the Euclidean distance between subsequent locations. Distinct patterns in NSD time-series are expected from specific movement strategies with asymptotic patterns in NSD associated with a sedentary behavior (territory or home range) and NSD increasing over time representative of nomadic movements (floaters and transients). Although many studies have used non-linear parametric models to assign individuals to discrete movement strategies [73–75], patterns in net square displacement over time were distinct enough to visually assign individuals in this study (see Appendix, Figure S2).

Automated Telemetry

To track individuals at a fine temporal resolution across the study site, we utilized an automated radio tracking system that consisted of 5 Sensorgnome receivers (Appendix, Figure S1).

All tagged individuals were initially caught within the detection range of any given receiver ensuring that most individuals but because individuals often moved from their point of capture we only included individuals that were detected consistently for at least 5 days to accurately capture space use behaviors. Of the 63 individuals captured as part of this study, 36 were detected frequently enough across the array to include in this foray analysis.

Each receiver was equipped with four horizontally polarized omnidirectional antennas positioned 9 meters high on a galvanized steel mast. These receivers continuously collected incoming signals from any nearby transmitter and logged the tag ID, timestamp, signal strength, and antenna port for each detection. Data collected by the automated telemetry system was uploaded to the Motus Wildlife Tracking System network for preliminary processing [46] and we used the R packages Motus [76] & tidyverse [77] to download, filter, and analyze the data. Although detections occurred approximately every 10 seconds, we smoothed the detection data at one-minute intervals by calculating the median signal strength from a series of consecutive detections for each receiver and then converted this into a detection history (1s and 0s).

Given our array configuration (Appendix, Figure S1), we were able to unambiguously determine that detections across different receivers of the same individual were indicative of relatively long-distance forays. First, these omni-directional antennas have reduced detection range compared to the typical directional yagis conventionally used. Based on the average distance at which an individual with known position was not detected, each receiver had an approximate detection range of ~300 m when individuals were moving at or below the canopy level. Further, the spatial configuration of receivers was such that they were distributed over an a large area () and separated by distances between , which are orders of magnitude larger than the average territory size of a redstart (; [78]). Taken together, sequential detections across different towers were indicative of exploratory forays outside of the territory or home range of an individual. Conventional hand-tracking approaches do not offer the temporal resolution needed to effectively capture these relatively rare or infrequent movements and therefore underrepresent the diversity of movement behaviors. As such, quantifying median daily forays enabled us to quantify the extent to which individuals (territorial or floaters) explored the study site on a continuous scale which allowed us to estimate the effect that seasonally variable resource availability, such as NDVI, had on the extent and prevalence of these movement behaviors.

Forays can be quantified in two ways using the automated telemetry array: movements between one receiver and another (Transitions) or movements out of range of one receiver and back (Recursions). Transitions are unambiguous because, with this array configuration, departures from one receiver’s detection range into another’s reflect true, relatively large-scale, movements. Recursions, however, are more ambiguous. They can either represent the movement of individuals temporarily out of range of the receiver but still within its home range or territory, or they can represent forays outside of their respective home range but in areas not covered by our array (Appendix, Figure S1). Given their ambiguity, in this manuscript, we chose to exclude recursions and therefore our quantification of daily forays represents a conservative estimate.

Evaluating the Consequences of Space Use Tactics on Nonbreeding Season Condition

Nonbreeding season condition – defined as how well an individual maintains or improves their body condition throughout the nonbreeding season – is a key trait that underlies the overall performance of an individual and is inherently tied to fitness in this species through a seasonal interaction on departure time [61,69]. However, body condition is an inherently multifaceted trait that is difficult to measure and rarely captured with single metrics. Because we were primarily interested in the relative mass and muscle an individual accumulates or maintains throughout the season, we keyed in on the change in scaled mass index (~size corrected mass, g) [64] and change in pectoral muscle size (mm)[63,79] as response variables in linear models that included space use tactic (territorial vs. floaters) as well as age, sex, and habitat type to account for their confounding effects. Of the 63 individuals tracked in this study, we recaptured 46 at least twice within a season (> 10 days apart), which allowed us to assess how space use strategy influenced changes in nonbreeding season condition. Key positive indicators of condition were positive changes in scaled mass index (increases in body mass) and increases in muscle size. Poor performers, on the other hand, would expect to lose or maintain mass throughout the season and decrease in muscle size as expected from a previous food-reduction experiment in this study system [25].

Data Analysis

We fit general linear models that included age class (second year and after second year) and sex (male and female) to investigate how home range area of an individual, its space use tactic (territorial vs floater), and its median daily forays was influenced by its dominance status. We also included year as a fixed effect to assess how the average space use behavior of the population differed among years. With the home range area models, we opted to weight the variance by the number of locations used to estimate the MCP to account for the fact that home range area is sensitive to the number of locations used to estimate the MCP. As such observations that included more locations were given more weight than observations with fewer locations. While the home range area and forays were modelled using normal distributions, individual space use tactic was modeled using a logistic distribution. We initially fit all three models (home range area, space use tactic, and forays) with two-way interactions between habitat and age and sex but found that those interactions were non-significant. As such, we dropped those terms and report only results of the additive models.

We explored within-year changes in space use behavior (mean daily forays & mean daily displacement) averaged across a week using linear mixed models that included individual as a random effect and weekly NDVI at the study site level as the predictor. We included age class, sex, and habitat type (dry scrub vs. wet mangrove) along with their respective interactions with NDVI to explore how an individual’s space use response to changes in environmental conditions varied by dominance status and habitat type. Because both mean daily forays and mean daily displacement were quite variable and spanned orders of magnitude, we log transformed these variables. In the case of mean daily forays which included 0’s, we added 1 to each value. All analyses were run in R and mixed models were fit using the lme4 package [80]. We assessed significance of all parameters at the level and where appropriate used likelihood ratio tests on reduced model varieties for models that included random effects.