Miniaturized light-level geolocators provide novel insight into the migration ecology of an endangered songbird

Abstract

Understanding songbird movements throughout their annual cycles is critical to inform conservation planning decisions and management priorities for migratory species. From 2017 to 2020, we attached miniaturized light-level geolocators to 309 male golden-cheeked warblers (Setophaga chrysoparia; hereafter warbler) at five study sites located across the species’ breeding range in Texas. From 2018 to 2021, we recovered data from 61 geolocators and used these data to determine the locations and timing of warbler movements during their non-breeding seasons and to quantify migratory connectivity and migration networks for the species. Mean duration of fall migration (37 d ±19 d) was longer than spring migration (27 d ±14 d), and, on average, second-year warblers initiated fall migration 10 d ±13 d earlier than after-second-year warblers. Migration initiation dates during spring, as well as duration and distance of fall and spring migrations, were similar for both age classes. Overall, 80% of warblers migrated along the Sierra Madre Oriental and Mexico’s Gulf coast between their breeding grounds in Texas and wintering grounds in southern Mexico and Central America during fall and spring; 20% migrated across or eastward around the Gulf of Mexico, primarily in fall. Overall, stopover duration varied along the warbler’s migration route (range = 3–61 d), with the longest stopovers in Texas and along the coast of Tamaulipas and Veracruz in Mexico. The highest percentages of stationary overwintering locations occurred in Chiapas, Mexico (27%), Guatemala (25%), and Honduras (17%), and many warblers used more than one location during winter. Some warblers used areas outside the predicted winter range; one extralimital location was confirmed during field-based surveys in Oaxaca, Mexico. Our results suggested that warblers exhibit weak migratory connectivity and likely disperse widely across the winter range. Given the narrow migration pathway used by most warblers and confined winter range in montane forests, continued and increased conservation along the warbler’s migration route and on the wintering grounds could help ensure population persistence and simultaneously protect co-occurring species.

Supplementary information

The online version contains supplementary material available at 10.1186/s40462-026-00626-0.

Background

Over the last century, 60% of Neotropical migratory songbirds have experienced population declines [1–3] and Rosenberg et al. [4] estimated a cumulative net loss of 2.5 billion native migratory birds in North America since 1970. The drivers of avian population declines are species-specific and there is growing evidence that events during non-breeding periods contribute to their declines [1, 5, 6]. Unfortunately, we still lack data to describe winter habitat associations and migratory behavior for most species, including the timing, location, and duration of migration, and the extent to which individuals from a specific breeding location retain geographic spacing during the non-breeding period (and vice versa; i.e., migratory connectivity) [7–9]. Such information is necessary to develop and implement more effective conservation strategies that address threats across the full annual cycles of migratory birds.

The miniaturization of light-level geolocators (hereafter geolocators), which record ambient light data that can be used to estimate the latitude and longitude of an individual’s location, has been particularly useful for examining migratory movements of small-bodied species ( < 12 g [10, 11]. For example, Augustine et al. [12] found evidence that Canada warblers (n = 10; 9–13 g; Wilsonia canadensis) from the southern extent of their breeding range in North America spent the non-breeding season in northern Columbia and may follow a pattern of counterclockwise loop migration, which could minimize flight distance and time to the breeding grounds. With repeated tracking, Stanley et al. [13] discovered that wood thrushes (n = 10; ~43 g; Hylocichla mustelina) exhibited consistency in the timing of migration but plasticity in their individual migration routes, suggesting that their potential ability to adjust migration timing in response to climate change may be limited. Similarly, geolocator data from three aquatic warblers (~12 g; Acrocephalus paludicola) revealed a previously unknown migration route and identified needed protections for previously unknown stopover sites that may be critical for the survival of this globally threatened species [14]. Geolocators have inherent limitations and cannot fully resolve the complexities of migration data collection. For example, geolocators must be retrieved to download data recorded by the device’s light sensor and they do not have the same spatial resolution as global positioning systems (GPS) and satellite transmitters (location accuracy of geolocators typically ranges from 12 to 200 km; e.g. [15]). However, they often represent the only option for researchers to study the migration ecology of small birds, appear to have minimal impacts on the vital rates of birds < 100 g [16], and have become an important tool for exploring the long-distance movements of birds.

According to Web of Science, well over 200 peer-reviewed papers have been published using geolocator data to describe migration routes, overwintering locations, and sex- and age-based differences in migratory strategies of birds since Stutchbury et al. [17] used the devices to track long-distance movements of wood thrushes and purple martins (42–65 g; Progne subis). Data collected to quantify migratory connectivity, or the degree to which breeding and non-breeding populations of a migratory species are linked [9, 18, 19], have been especially useful for identifying threats throughout species’ annual life cycles. Strong migratory connectivity suggests that individuals from a breeding site migrate to the same location on their wintering grounds (and vice versa), whereas weak migratory connectivity suggests that individuals from the same breeding site disperse to several locations on their wintering grounds (or vice versa) [19, 20]. For species with strong migratory connectivity, localized threats on the wintering grounds could have more profound effects on demographics than they would experience by those with weak migratory connectivity [7, 19]. The strong migratory connectivity found by Hallworth et al. [21] for ovenbirds (n = 28; ~19 g; Seiurus aurocapilla) illustrated this concept; the authors of this study found that eastern and western ovenbird populations were completely separated throughout their annual cycle. The origin maps developed by Hallworth et al. [21] could help inform conservation efforts or predict the influence of disturbances on specific populations.

While geolocator technology has been enthusiastically adopted by avian ecologists, we still lack migration data for most species, especially very small songbirds ( < 12 g). Until recently, the smallest geolocators weighed > 0.5 g and the total weight of all attachments (geolocator, harness, bands) is typically restricted to < 5% of an individual’s body weight. In addition, it is difficult to obtain adequate sample sizes from multiple geographically distinct study sites due to low return rates, low retrieval rates, battery failures, and cost [11]. These challenges are evident in the peer-reviewed literature. For example, Cresswell and Patchett [22] analyzed published tracking data from landbirds to assess potential biases in measuring migratory connectivity. They aggregated data by flyway and species, including only species with more than two sampled populations and at least four individuals per population. The resulting dataset included an average of 24.8 individuals per tracked population pair. However, some studies are able to collect data from much larger samples, for example, Sharp et al. [23] used geolocator data from 112 painted buntings (~16 g; Passerina ciris) monitored at 11 breeding sites in the U.S. to describe migratory connectivity of disjunct eastern and interior populations.

Until recently, few studies had examined migratory connectivity in the smallest wood-warblers (Family Parulidae) across geographically distinct breeding locations. However, available research reveals substantial variation among species. Hermit warblers (S. occidentalis; ~12 g; n = 22; 6 sites [24]); and blue-winged warblers (V. cyanoptera; ~9 g; n = ~25; 3 breeding regions [25]); exhibited low connectivity, whereas worm-eating warblers (Helmitheros vermivorum; 12–14 g; n = 21; 4 sites [26]), black-throated green warblers (S. caerulescens; 9–11 g; n = 13; 2 sites [27]), and cerulean warblers (S. cerulea; 8–10 g; n = 26; 14 sites [28]); showed moderate connectivity. Golden-winged warblers (Vermivora chrysoptera; ~10 g; n = ~59; 2 breeding regions [25]); had high connectivity, while blackpoll warblers (Setophaga striata; ~12 g; n = 47; 11 sites [29]); exhibited low to moderate connectivity depending on time of season. This variation underscores the need for species-specific migratory connectivity research to inform comprehensive conservation planning for warblers.

The golden-cheeked warbler (Setophaga chrysoparia; hereafter warbler) is a small (8–12 g) endangered Nearctic-Neotropical migrant that breeds exclusively in Ashe juniper (Juniperus ashei) woodlands in central Texas and overwinters in montane evergreen cloud forests, pine-oak forests, pine forests, oak forests, and shrubland in southern Mexico and Central America [30, 31]. The U.S. Fish and Wildlife Service (USFWS) listed the warbler as endangered under an emergency decision in 1990 citing “ongoing and imminent habitat destruction” as the greatest risk to the species [32]. The primary sources of habitat loss and fragmentation on the warbler’s breeding grounds are removal of Ashe juniper for livestock grazing and urban development, and the primary threats to warbler habitat on their wintering grounds are deforestation for agriculture (cattle ranching and coffee plantations), firewood collection, and wildfires [33–36].

The warbler’s breeding habitat is well defined (e.g. [37, 38]), and research conducted across the warbler’s breeding range estimated the population size at 263,339 males (95% CI: 223,927–302,620) [39] and 217,444 males (95% CI = 153,917–311,965) [38], respectively. In addition, there have been significant efforts to find and study warblers on their wintering grounds (e.g., Pronatura and Defensores de al Naturaleza). However, there are few location records for warblers during migration. Further, researchers have banded thousands of warblers on the breeding grounds, but no banded warblers have been detected on the species’ wintering grounds or during migration. While most warblers are thought to migrate north and south through mountains in eastern Mexico, some researchers have proposed that warblers have a protracted migratory flight or that the species migrates over the Gulf of Mexico [40, 41]. Understanding the migration ecology of warblers is essential for conservation planning, as it could allow researchers and land managers to identify important habitats across the species’ full life cycle, assess potential threats along the species’ migratory routes, and ensure habitat protection efforts are targeted where they are most needed.

From 2017 to 2020, we deployed 309 geolocators on male warblers at five study sites across their breeding range in Texas to examine the species’ movements during non-breeding seasons. We used data from 61 geolocators we retrieved to describe the timing, location, and duration of warbler migration and the wintering period, and to quantify migratory connectivity and migration networks for the species. At 8 to 12 g, the warbler is one of the few very small ( < 12 g) songbird species used in a multi-year geolocator study. Warblers have high annual return rates (24–67% [42, 43]), strong site fidelity [44, 45], and high recapture rates ( > 75%; J. Macey, unpublished data), which made the species an excellent candidate for geolocator research.

Given the warbler’s narrow breeding range [37, 38] and constrained wintering distribution [46], opportunities for spatial segregation once birds reach the winter grounds are limited [47]. Therefore, we predicted weak migratory connectivity for warblers, likely reinforced by mixing during migration and redistribution across shared stopover sites, as observed in similar species (e.g., blackpoll and hermit warblers [24, 29]). We anticipated that most individuals would follow a direct north–south corridor through the montane highlands of eastern Mexico during migration, with only limited Gulf crossings. Because of the species’ compressed longitudinal breeding range, parallel (e.g. [28]), leapfrog (e.g. [48]), and chain-like (e.g. [49]), connectivity patterns seemed unlikely, though we could not completely rule them out. Understanding connectivity strength and structure, along with non-breeding movements, could enable targeted conservation actions that support long-term population viability for this species.

Methods

Study area

We deployed geolocators at five study sites in Texas, USA: Dinosaur Valley State Park (DVSP), Fort Hood (FH), Balcones Canyonlands National Wildlife Refuge (BCNWR), Joint Base San Antonio-Camp Bullis (CB), and Kerr Wildlife Management Area (KWMA) (Fig. 1). Study sites were separated by 51 to 110 km across the breeding range in the Edwards Plateau and Cross Timbers ecoregions [38, 50]. Study sites ranged from 617 to 86,994 ha and contained 437 to 19,977 ha of warbler breeding habitat (Table 1). Elevations ranged from 152 to 661 m (Table 1). Based on 30-year averages (1991–2020), minimum and maximum monthly temperatures ranged from 2˚ to 36 °C; and annual precipitation ranged from 81 to 95 cm (Table 1).

Fig. 1.

Study sites where we fitted male golden-cheeked warblers (Setophaga chrysoparia) with geolocators to study their migration ecology (2017–2021) shown in relation to the species’ breeding range [38] shaded in gray)

Table 1.

Characteristics of the five study sites in Texas we used to examine the migration ecology of male golden-cheeked warblers (Setophaga chrysoparia) from 2017 to 2021

Study Sitea	Area Total (ha)	Warbler Habitat (ha)	County	Latitude, Longitude	Elevationg (masl)	Low (c)f	High (c)f	Annual Precipitation (cm)f
DVSP	617b	437b	Somervell	32˚15’11.7“N, 97˚49’07“W	207–259	−1.67	36.11	94.77
FH	86,994c	19,977c	Bell, Coryell	31˚19‘50“N, 97˚74’13“W	152–349	2.28	35.89	88.27
BCNWR	11,128d	4,653d	Burnet, Travis	30˚37‘21“N, 98˚04’06“W	212–455	1.67	34.44	83.82
CB	11,285e	3,431e	Bexar	29˚62‘51“N, 98˚57’00“W	314–455	5.00	35.56	81.79
KWMA	2,628b	1,027b	Kerr	30˚06‘33“N, 99˚50’66“W	594–661	1.11	34.44	81.28

^aStudy site acronyms: DVSP = Dinosaur Valley State Park; FH = Fort Hood, BCNWR = Balcones Canyonlands National Wildlife Refuge; CB = Joint Base San Antonio-Camp Bullis; KWMA = Kerr Wildlife Management Area

^bhttp://www.tpwd.state.tx.us

^c[43]

^dS. Rowin, U.S. Fish and Wildlife Service, personal communication

^e[51]

^f[52]

^g[53]

Field methods

We deployed geolocators on warblers from March to June of 2017 to 2020. Beginning in early March of each field season, we searched from sunrise to ~ 1300 daily for warblers on our study sites. When we detected a warbler by sight, we used a GPS unit to record the bird’s latitude and longitude, and we noted the warbler’s sex and the color combination of any bands present on the bird’s legs. We did not use this data for analysis purposes, but rather to relocate individuals during subsequent monitoring and capture attempts. Banding and monitoring are also components of long-term warbler research projects conducted at Fort Hood and Camp Bullis. Once behavioral observations indicated that warblers had established their territories (e.g., males consistently found in the same area, localized patterns of singing activity, aggressive or defensive behaviors toward conspecifics in the area), and thus, were unlikely to move long distances from our study sites, we used standard mist-netting techniques with playback of conspecific songs to capture them [54]. We banded warblers with a U.S. Geological Survey (USGS) aluminum leg band and two or three plastic color leg bands. We recorded the USGS band number, color combination, date, time, body mass to the nearest 0.01 g, age, and we used GPS units to identify the Universal Transverse Mercator (UTM) coordinates of all banding locations. We aged individuals as hatch year (HY), second-year (SY), or after-second-year (ASY), and sexed each bird according to Pyle [55] and Peak and Lusk [56, 57].

We attached geolocators (Migrate Technology Intigeo P30Z11-7 Dip with a 5-mm light stalk; 0.41 g with elastic harness using Stretch Magic brand elastic cord) to male warblers at all study sites using a modified version of the leg-loop harness [58, 59]. We only attached geolocators to males to increase recapture rates, and to birds with a minimum weight of 9.2 g, so that the weight of all the attachments (USGS band, color bands, and geolocator) did not exceed 5.5% of the birds’ body weight [59, 60]; USFWS #TE023643-11, #TE32917C-1, and #TE082496-0; USGS #21999 and #24126; TPWD #SPR-0409-079, #SPR-0417-097; and Louisiana State University AgCenter Animal Care and Use Protocols #2018-11 and #A2021-10). We also recorded locations for tagged individuals throughout the breeding season to assist with relocation efforts for the following breeding seasons and to examine the potential impacts of geolocators on warblers for a concurrent study [61]. From 2018 to 2021, we returned to each study site to recapture warblers with geolocators using the same capture techniques described above so that we could remove the geolocators and retrieve the stored data.

Geolocator analysis

We used the R package BAStag [62] to import raw light data from each geolocator, convert the.lux file to “TAGS” format (https://www.nceas.ucsb.edu), and to detect and validate twilights. Consistent with previous studies on other forest-dwelling songbirds, we used a light threshold of 1.5 to define separation of day and night [63]. We visually inspected each twilight event and removed false twilights that occurred due to shading of the light’s sensor (i.e., quick transitions from light to dark or vice versa) [63]. We then used the R package FLightR [64] to estimate latitude and longitude coordinates from the processed light data. FLightR used a hidden Markov chain model that combined an animal movement model with an observational model of light measurements to estimate locations based on position information from both current and adjacent twilights [64]. Prior to running each model, we used deployment coordinates and the maximum time we could confirm that each warbler was in a specific location on their breeding grounds to calibrate our analysis (i.e., establish the relationship between observed and expected light levels for each device). We defined the spatial extent for our analyses as 500 km beyond the known range of the warbler, set the maximum distance a warbler could travel per day to 750 km [64], and programmed the animal movement model so that individuals could not be stationary over water. However, we did not prohibit over-water crossings in the analysis because we could not rule out the possibility of alternate migration routes across the Gulf of Mexico (hereafter Gulf migrations). Model outputs included location means and medians with their associated credible intervals, arrival and departure dates, and stopover durations [64, 65]. We used two methods to calculate accuracy of our geolocator data. We first calculated differences in latitudinal and longitudinal distances between each individual’s capture point and the individual’s mean geolocated coordinates during the stationary breeding period (see migratory connectivity and migration networks section below). Second, we calculated the mean and standard deviation (SD) estimated distances from the upper and lower credible intervals (CI) for latitude and longitude stationary periods per period (winter, migration).

We used spring and fall departure dates, as well as evidence of short- and long-distance flights to help us define breeding, migration, and wintering periods. We considered long-distance movements as sustained flights > 100 km and chose this distance due to the accuracy limitations of geolocators. We used the output from FlightR to define stationary periods (i.e., migration stopovers and winter stationary periods) as locations sustained for ≥3 days. We considered fall and spring migration to be the periods when individuals made sustained long-distance flights toward their wintering or breeding grounds, respectively. We were unable to categorize the fall and spring migration periods for all warblers based on dates alone because the timing of these movements varied by individual. Thus, we defined the wintering period for each individual as the date of their last long distance, sustained flight during fall migration to the date of each individual’s first long distance, sustained flight toward the breeding grounds in spring.

We used the mean latitude and longitude and the associated CIs to identify an individual’s locations during each period. We used the output from FLightR to calculate distances traveled between points identified as breeding and wintering locations, measure durations, and convert arrivals and departures to Julian date for each individual. We calculated means for each of these variables and compared them among years, locations, and age classes using analysis of variance (ANOVA [66]); or t-tests [67] with α = 0.05. For ANOVAs, we used Tukey’s honestly significant difference (HSD; α = 0.05 [68]); test to determine which means were significantly different from each other. We calculated SDs and Cohen’s d (d) to measure effect size. We examined if there were correlations between Julian departure and arrival dates for migration using the Pearson Correlation Coefficient.

We used estimated fall and spring migration locations and wintering stationary locations (weighted by duration) to run separate kernel density estimates (KDEs) for each season with ArcGIS Pro 2.9.12 (ESRI, Redlands, CA, USA). We calculated KDEs using a 75 km search radius and output results in raster images with 1 km resolutions. Next, we used the raster calculator tool to estimate KDEs by percent [69]; specifically, we divided the raster cells by the highest KDE recorded for each period, multiplied the results by 100, and binned the percentages (i.e., 25–49%, 50–74%, 75–89%, 90–100%) for visualization purposes. As such, KDE percentages represent the proportion of geolocated points estimated to be within each pixel compared to the pixel with the highest estimated point density. To complement this analysis, we also visualized estimated fall and spring migration locations and wintering stationary locations with their 95% CIs. We provided additional maps in the Supplementary Materials that visualize these data based on overlap of 75% utilization distributions (UDs) for each bird and an overlap of convex hulls created from the 95% CIs around each geolocated point (Figs. S.6 and S.7)

Migratory connectivity and migration networks

We estimated migratory connectivity of geolocated warblers between breeding and wintering locations using the Mantel correlation coefficient (rM) and MC [7]. Specifically, rM quantifies the correlation between distances of individuals during the breeding and wintering periods to examine whether individuals remain similarly close or distant from one another during the wintering period compared to the breeding period [20]. Similarly, MC estimates migratory connectivity between groupings of birds on the breeding and non-breeding grounds (referred to as ‘populations’) and accounts for uneven sampling effort and differing transition probabilities across populations [7]. We chose to examine rM because warblers have a limited breeding range and showed little genetic structure on their breeding grounds [70], making it difficult to delineate breeding regions that represent distinct populations (see [7]). However, we suspected that rM would be biased toward low connectivity because rM uses the correlation of distances between individual birds in its calculation. We sampled relatively few sites across the entirety of the breeding range, and the non-breeding grounds were much larger than the breeding grounds (see [7, 20, 22]). As such, we also estimated MC and grouped birds based on conservation regions in the breeding and non-breeding grounds (hereafter “regions”). The MC estimate thus examines whether birds within each breeding region overwinter in the same non-breeding regions (i.e., high migratory connectivity) or across different non-breeding regions (i.e., low migratory connectivity [7]). For our MC analysis, we used Hatfield et al. [71] proposed recovery regions for warbler’s breeding grounds, which included 3 regions (i.e., North, Central, and South) delineated by Texas county boundaries (see Fig. 5 in the results). We used DeSaix [72] conservation regions, which combined information on Level I ecoregions and political boundaries to represent biomes that are often governed or managed by similar entities to group warblers on the wintering grounds [73]; see Fig. 5 in the results).

Fig. 5.

Delineated breeding range regions [71] with corresponding utilization distribution centroids in overwintering conservation regions [72] for all golden-cheeked warblers (Setophaga chrysoparia) geolocated during our study (2017–2021)

We conducted our migratory connectivity analyses in program R (V. 4.4.2; R Core Team 2024) using packages sf [74], MigConnectivity [75], geosphere [76], ggspatial [77], terra [78], tidyterra [79], tidyverse [80], and mignette [72]. For both rM and MC, we defined our origin points as the capture locations in the breeding grounds and our target points as the centroid of each bird’s 75% UD in the wintering grounds, which we calculated using the R package adehabitHR [81]. Almost half of the birds (49%, n = 30 out of 61) had more than one 75% UD, so in these cases we used the 75% UD in which the bird spent the longest duration. We were unable to calculate the 75% UD for two birds due to error (location points were too close), so we instead substituted the centroid for the 50% UD. The centroids of six birds were located in water, preventing regional assignment. However, their UDs were located entirely or almost entirely (i.e., > 90%) within a single conservation region. As such, we moved these points < 25 km to the nearest terrestrial location for the purposes of the MC analysis.

We calculated the mean, standard deviation, and range of distances between origin points in the breeding grounds and between target points in the wintering grounds to describe the spread of birds during each period [82]. We also subset target points by Hatfield region origin and similarly calculated the mean, standard deviation, and range of distances between target points for each region. Since geolocators cannot record the exact locations of individuals, we then estimated location error in the wintering grounds in accordance with Cohen et al. [7] for later use in the rM and MC calculations. We determined the difference in latitudinal and longitudinal distances between each individual’s capture point with their geolocated coordinates during the breeding period and took the mean of these values, which resulted in a dataset with one latitude error and one longitude error per bird. For these calculations, we excluded birds with geolocators that began capturing points after June (n = 18), as some birds began moving outside of their breeding locations during July and we assumed that points collected through June represented a stationary period. Next, we created intercept-only linear models using the latitude and longitude error as dependent variables, which allowed us to estimate bias (i.e., the intercept coefficient) for latitude and longitude, as well as the variance-covariance matrix between latitude and longitude error [7]. We then used the bias and variance-covariance matrix as an estimate of location error for the wintering grounds; we acknowledge that breeding period location error may not be representative of wintering period location error, but we could not estimate wintering period location error without capturing and geolocating birds on the wintering grounds (see [7]). We calculated rM with the estMantel function with 1,000 bootstraps using all origin points, target points, and location bias and variance-covariance matrix [20]. We examined the rM estimate and 95% confidence interval (α = 0.05) and defined strong migratory connectivity as rM > 0.5 and weak migratory connectivity as rM ≤0.5 [47].

For our MC analysis, we estimated proportional abundance within each breeding region using abundance estimates from Mueller et al. [38]. We then estimated the distance between breeding regions and between non-breeding regions using the centroid of each region; because some regions were irregularly shaped or disconnected (see Fig. 5 in the results), we estimated the centroid that fell within each region using the st_point_on_surface function. We estimated transition probabilities for each region using the estTransition function with 1,000 bootstraps, which included all origin points (including those that we had to move to be within a region), target points, regions, and location bias and variance-covariance matrix calculated above. We then estimated MC with the estStrength function and 1,000 bootstraps, which used the regions, distances between breeding regions and between non-breeding regions, breeding region proportional abundance, and transitional probabilities. Again, we examined the MC estimate and 95% confidence interval (α = 0.05) and defined strong migratory connectivity as MC > 0.5 and weak migratory connectivity as MC ≤ 0.5 [47].

Finally, we created a migratory network representing the proportion of birds in each breeding region predicted to overwinter in each non-breeding region. In addition to the aforementioned packages, we also used the R packages ggalluvial [83], rjags [84], and ebirdst [85] for the migratory network analysis and we visualized and exported figures using ggplot2 [86], ggrepel [87], and ggpubr [88]. We first created a frequency table that represented the number of birds that occurred within each pair of breeding and non-breeding regions [72, 73]. We then used seasonal mean relative abundance (3 km) from the 2022 eBird Status and Trends data [89] to estimate relative abundance within each region, using the breeding period abundance for breeding regions and the non-breeding period abundance for non-breeding regions. The eBird Status and Trends data often underestimates abundance due to factors like sampling bias, though estimates likely retain the relative proportion of abundance across regions [89]. As such, we chose to use these data as they were the only estimates of warbler abundance across the entirety of the non-breeding range and the migratory network function converts abundance estimates into a percentage [72]. For consistency, we also used eBird trends data to estimate relative abundance in breeding regions for this analysis. We used the function run_network_model, which requires the aforementioned frequency table and relative abundance data, to create a migratory network. We examined model fit (α = 0.05), then visualized predicted proportions and 95% confidence intervals to determine differences in the proportion of birds predicted to breed and overwinter in each combination of regions.

Results

We attached 309 geolocators to male warblers across our study sites and recovered 80 (26%) of the devices (Table 2). We retrieved fewer geolocators from Camp Bullis than the other study sites, but return rates at Camp Bullis were consistent with previous estimates [43, 51] and broader research on warblers (reviewed by [42]). Further, during a concurrent study at Fort Hood (formerly Fort Cavazos) and Camp Bullis, we found little evidence that geolocators impacted warbler return rates or breeding behavior [61]. However, return rates of SY warblers with geolocators were lower than SY warblers without geolocators at Camp Bullis, and individuals in both age classes were challenging to recapture [61]. Overall, three males returned to the breeding grounds without their geolocators, and we were unable to use the data recorded by 19 (24%) of the geolocators that we recovered due to poor or corrupt data. Thus, our final analyses included data from 61 males (28 SY [46%], 33 ASY [54%]). Fifty-five of the 61 geolocators (90%) recorded light data from the date of attachment through the date of recovery ( = 346 d ±49.68 d) and 6 geolocators (10%) recorded light data from the date of attachment through at least 01 December ( = 251 d ±74.85 d). See Table S.1 in the supplemental material for error estimates by study site and Table S.2 in the supplemental material for the estimated distances for the stationary periods (migration and winter) from the estimated means to the upper and lower credible intervals (latitudes and longitudes) along with SDs per period.

Table 2.

Number of light-level geolocators attached to and retrieved from male golden-cheeked warblers (Setophaga chrysoparia) per study site and year (2017–2021) at Dinosaur Valley State Park (DVSP), Fort Hood (FH), Balcones Canyonlands National Wildlife Refuge (BCNWR), Joint Base San Antonio-Camp Bullis (CB), and Kerr Wildlife Management Area (KWMA)

Location	2017–2018		2018–2019		2019–2020		2020–2021		All years
	Deployed	Retrieved	Deployed	Retrieved	Deployed	Retrieved	Deployed	Retrieved	Deployed	Retrieved
DVSP	15	5 a	0	0	13	4	11	2	39	11
FH	25	12	19	6	28	7	18	6	90	31
BCNWR	20	7 a	0	1	20	3 a	20	7	60	18
CB	20	1	20	2	20	0	0	0	60	3
KWMA	20	2	0	0	20	5	20	10	60	17
Total	100	27	39	9	101	19	69	25	309	80

^aNumber does not include one warbler captured without the geolocator we attached to the individual during the previous breeding season

Fall and spring migration

Warblers initiated fall migration over a two-month period from early July to early September (Fig. 2; Table 3). On average, warblers arrived on their wintering grounds 37 d ±19 d after they left the breeding grounds; spent 167 d ±23 d on their wintering grounds; initiated spring migration over a two-month period from mid-January to mid-March; and arrived on the breeding grounds 27 d ±14 d after they left the wintering grounds (Fig. 2; Table 3). The average distance from their capture location to their first stationary wintering location was 1,910 km ±269 km (range: 1,083–2,491 km; n = 61) and the average distance from their last overwintering location to their capture location on the breeding grounds was 1,932 km ±274 km (range: 1,428–2,705 km; n = 56). Eighty percent of warblers (n = 49) migrated over land along the Sierra Madre Oriental during both migrations, whereas 20% (n = 12; 1 to 6 individuals per year) migrated over the Gulf during one migration (fall [n = 11], spring [n = 1]; Fig. 3; Table 3). Of the Gulf migrants, 50% were SY warblers and 50% were ASY warblers.

Fig. 2.

Kernel density estimates for fall (top left) and spring (top right) golden-cheeked warbler (Setophaga chrysoparia; n = 61 light level geolocators) migration locations that occurred along the north/south land route along with migration stopover locations, durations, and 95% credible intervals for fall (bottom left) and spring (bottom right). Data were collected from 2017 to 2021 and shown in relation to the breeding [38] and winter [46] ranges

Table 3.

Mean (± SD, range) of initiation/arrival dates and migration durations for male golden-cheeked warblers (Setophaga chrysoparia) based on data we obtained from 61 light-level geolocators (2017–2021). We summarized these metrics for all individuals (overall), individuals that only migrated over land (land), and individuals with trans-gulf migrations (Gulf)

Overall	Fall migration (n = 61)	Spring migration (n = 55)
Initiation date	6 Aug ±13 d, 1 Jul–05 Sep	26 Feb ±11 d, 19 Jan–19 Mar
Duration	37 d ±19 d, 5–106 d	27 d ±14 d, 5–66 d
Arrival date	11 Sep ±19 d, 13 Aug–27 Oct	24 Mar ±11 d, 04 Mar–24 Apr
Land	Fall migration (n = 49)	Spring migration (n = 46)
Initiation date	04 Aug ±12 d, 01 Jul–05 Sep	27 Feb ±11 d, 19 Jan–19 Mar
Duration	36 d ±17 d, 12–106 d	26 d ±15 d, 5–66 d
Arrival date	09 Sep ±18 d, 13 Aug–27 Oct	23 Mar ±11 d, 04 Mar–24 Apr
Gulf	Fall migration (n = 12)	Spring migration (n = 9)
Initiation date	14 Aug ±13 d, 18 Jul–03 Sep	23 Feb ±9 d, 05 Feb–05 Mar
Duration	40 d ±25 d, 5–91 d	30 d ±10 d, 20–51 d
Arrival date	23 Sep ±20 d, 24 Aug–18 Oct	25 Mar ±13d, 12 Mar–14 Apr

Abbreviation: d = days

Fig. 3.

Estimated Gulf migration routes based on stationary locations (by duration), along with 95% credible intervals, and first stationary location post migration of male golden-cheeked warblers (Setophaga chrysoparia; n = 12) based on data from 61 light-level geolocators (2017–2021). Lines do not identify exact migration routes but rather represent connections between the mean estimated stationary locations

Of the fall Gulf migrants, four departed toward the wintering grounds from the coast of Texas, two from Louisiana, two from Alabama, and three from Florida (Fig. 3). The one spring Gulf migrant we identified departed from the Yucatán Peninsula and arrived in Louisiana before traveling west to the warbler’s breeding grounds (Fig. 3). All of the Gulf migrants that we identified were located at four study sites during the breeding season (KWMA 31% [n = 5 of 16]; DVSP 29% [n = 2 of 7]; FH 17% [n = 4 of 23]; BCNWR 7% [n = 1 of 14]). We had insufficient and uneven sample sizes to make statistical comparisons between warblers that migrated over land and warblers that completed Gulf migrations. However, departure dates for fall and spring migration, durations of fall and spring migration, and fall migration distances were similar between the two groups (Table 3).

Fall migration took ten days longer ( = 37 d ±19 d, range: 5 – 106 d) than spring migration ( = 27 d ±14 d, range: 5 – 66 d) (t₅₄ = 3.27, p < 0.01, d = 0.62) (Table 3). We found no significant correlations between fall and spring Julian departure dates (r = −0.05, p = 0.74), fall and spring arrival dates (r = 0.16, p = 0.24), or spring departure and arrival dates (r = 0.21, p = 0.13). However, we did find a weak and positive correlation between fall departure and winter arrival dates (r = 0.36, p < 0.01). We found no statistically significant differences between initiation dates for fall or spring migration, the durations of fall or spring migrations, or the distances of fall or spring migrations across study sites (Fig. S.3, Table 4) or years (Fig. S.4, Table 5). When we compared these metrics for SY and ASY warblers, we found no statistically significant differences between initiation dates for spring migration, the durations for fall or spring migration, or the distances of fall or spring migration (Fig. S.5, Table 6). However, we did find an age-based difference in fall departure dates (t₅₉ = 9.95, p < 0.01; d = 0.82) whereby SY males initiated fall migration 10 days before ASY males (SY:  = 01 Aug ±13 d, range: 1 Jul–3 Sep; ASY:  = 11 Aug ±12 d, range: 21 Jul–05 Sep) (Fig. S.5, Table 6).

Table 4.

Summary of means, standard deviations, ranges, and results from our analysis of variance tests comparing fall and spring migration initiation dates (Julian date), fall and spring migration durations (days [d]), and fall and spring migration distances (km) across study sites for male golden-cheeked warblers (Setophaga chrysoparia) based on data from 61 light-level geolocators (2017–2021). Our study sites included Dinosaur Valley State Park (DVSP), Fort Hood (FH), Balcones Canyonlands National Wildlife Refuge (BCNWR), Joint Base San Antonio-Camp Bullis (CB), and Kerr Wildlife Management Area (KWMA) in Texas

	Study Site
	DVSP (n = 7)	FH (n = 23)	BCNWR (n = 14)	CB (n = 1)	KWMA (n = 16)	Results
Fall initiation	07 Aug ±7 d	04 Aug ±15 d	06 Aug ±14 d	29 Jul	11 Aug ±11 d	F4,56 = 0.80
	29 Jul–17 Aug	01 Jul–03 Sep	18 Jul–05 Sep	-	29 Jul–01 Sep	p = 0.53
Fall duration	26 d ±9 d	38 d ±21 d	40 d ±20 d	32 d	40 d ±20 d	F4,56 = 0.66
	14–39 d	15–106 d	16–91 d	-	5–73 d	p = 0.63
Fall distance	1,824 km ±394 km	1,954 km ±261 km	1,930 km ±280 km	1,774 km	1,846 km ±289 km	F4,56 = 0.54
	1,083–2,254 km	1,350–2,391 km	1,451 –2,491 km	-	1,460 –2,278 km	p = 0.70
Spring initiation	03 Mar ±8 d	27 Mar ±13 d	23 Feb ±12 d	27 Feb	25 Feb ±10 d	F4,51 = 0.66
	18 Feb–10 Mar	19 Jan–17 Mar	26 Jan–11 Mar		05 Feb–19 Mar	p = 0.63
Spring duration	27 d ±12 d	26 d ±16 d	31 d ±14 d	16 d	25 d ±13 d	F4,50 = 0.55
	13–43 d	5–66 d	10–60 d	-	9–63 d	p = 0.70
Spring distance	1,952 km ±305 km	1,899 km ±237 km	1,912 km ±327 km	1,602 km	2,004 km ±268 km	F4,50 = 0.70
	1,627 –2,596 km	1,428 –2,380 km	1,604 –2,705 km	-	1,607 –2,440 km	p = 0.59

Table 5.

Summary of means, standard deviations, ranges, and results from our analysis of variance tests comparing fall and spring migration initiation dates (Julian date), fall and spring migration durations (days [d]), and fall and spring migration distances (km) across years for male golden-cheeked warblers (Setophaga chrysoparia) based on data from 61 light-level geolocators (2017–2021). Our study sites included Dinosaur Valley State Park, Fort Hood, Balcones Canyonlands National Wildlife Refuge, Joint Base San Antonio-Camp Bullis, and Kerr Wildlife Management Area in Texas

	Deployment Year
	2017 (n = 16)	2018 (n = 5)	2019 (n = 16)	2020 (n = 24)	Results
Fall initiation	07 Aug ±14 d	02 Aug ±8 d	09 Aug ±15 d	05 Aug ±13 d	F3, 57 = 0.34
	21 Jul–05 Sep	26 Jul–14 Aug	01 Jul–01 Sep	19 Jul–31 Aug	p = 0.80
Fall duration	31 d ±12 d	37 d ±10 d	40 d ±30 d	39 d ±15 d	F3,57 = 0.73
	16–62 d	29–54 d	5–106 d	8–78 d	p = 0.54
Fall distance	1,824 km ±319 km	1,930 km ±211 km	1,942 km ±303 km	1,922 km ±269 km	F3,57 = 0.55
	1,083 –2,278 km	1,721 –2,258 km	1,350 –2,491 km	1,460 –2,385 km	p = 0.65
Spring initiation	28 Feb ±9 d	24 Feb ±3 d	03 Mar ±10 d	23 Feb ±13 d	F3,52 = 2.36
	16 Feb–14 Mar	21 Feb–27 Feb	18 Feb–19 Mar	19 Jan–10 Mar	p = 0.08
Spring duration	27 d ±16 d	21 d ±9 d	27 d ±13 d	28 d ±15 d	F3,51 = 0.28
	9–63 d	14–34 d	5–51 d	6–66 d	p = 0.84
Spring distance	1,882 km ±243 km	1,950 km ±345 km	1,929 km ±238 km	1,962 km ±316 km	F3,51 = 0.24
	1,428 –2,372 km	1,602 –2,380 km	1,627 –2,596 km	1,604 –2,705 km	p = 0.87

Table 6.

Summary of means, standard deviations, ranges, and results from t-tests comparing migration durations (days [d]) and fall and spring migration distances (km) between second-year (SY) and after-second-year (ASY) male golden-cheeked warblers (Setophaga chrysoparia) based on data from 61 light-level geolocators (2017–2021). Our study sites included Dinosaur Valley State Park, Fort Hood, Balcones Canyonlands National Wildlife Refuge, Joint Base San Antonio-Camp Bullis, and Kerr Wildlife Management Area in Texas

	SY (n = 28)	ASY (n = 33)	Results
Fall initiation	01 Aug ±13 d	11 Aug ±12 d	t59 = 9.95
	01 Jul–03 Sep	21 Jul–05 Sep	p < 0.01
Fall duration	39 d ±21 d	35 d ±17 d	t59 = 0.70
	8–106 d	5–73 d	p = 0.40
Fall distance	1,943 km ±343 km	1,867 km ±225 km	t59 = 1.07
	1,083 –2,491 km	1,451 –2,278 km	p = 0.30
Spring initiation	29 Feb ±12 d	25 Feb ±11 d	t54 = 1.44
	26 Jan–17 Mar	19 Jan–19 Mar	p = 0.24
Spring duration	26 d ±13 d	27 d ±15 d	t53 = 0.10
	5–66 d	9–60 d	p = 0.76
Spring distance	1,922 km ±255 km	1,940 km ±294 km	t53 = 0.61
	1,607 –2,596 km	1,428 –2,705 km	p = 0.81

Migration stopovers

Warblers made 0–5 stopovers ( = 1.64 stopovers ±1.27 stopovers) and were stationary for 3–54 d per stopover ( = 12.49 d ±11.70 d) during fall migration (Figs. 2, 3). Warblers also made 0–5 stopovers ( = 1.50 stopovers ±1.13 stopovers) during spring migration and were stationary for 3–61 d per stopover ( = 10.63 d ±11.32 d) (Figs. 2, 3). However, 10% (n = 6 of 61) of the fall migrants and two percent (n = 1 of 55) of the spring migrants had no stopovers during migration. We found no significant differences in the number of stopovers (t₅₅ = 0.98, p = 0.33; d = 0.03) or the durations per stopover (t₈₃ = 0.89, p = 0.38; d = 0.16) between fall and spring migration.

Wintering period

All warblers arrived on their wintering grounds between mid-August and mid-October and, on average, remained there for 167 d ±23 d (range: 105–204 d; Fig. 4). The mean number of overwintering locations per individual was 1.51 ± 0.76 (range: 1–4 locations), which we defined by the number of 75% UDs per individual that were > 50 km apart. We found no significant difference in the number of overwintering locations for SY ( = 1.46 ± 0.65, range: 1–3 locations) and ASY ( = 1.55 ± 0.85, range: 1–4 locations) warblers (t₅₅ = 0.43, p = 0.67; d = 0.12). Our KDE (density/duration) showed a pronounced use of habitat within the warbler’s estimated winter range [46] in southern Mexico and Central America (Fig. 4). The KDE illustrated three distinct areas of high density warbler use, with the largest and highest density area located in central to northern Chiapas, Mexico with a portion in Tabasco, Mexico (Fig. 4). Two smaller areas with high density warbler use were in Guatemala and in El Salvador-Honduras (Fig. 4). The highest percentages of individual warbler stationary periods were in Chiapas, Mexico (27%); Guatemala (25%); and Honduras (17%; Fig. 4). See Figures S.6 and S.7 for additional wintering illustrations.

Fig. 4.

Kernel density estimates (density/duration) for stationary wintering locations (left), and stationary periods (by duration) along with credible intervals (right) of male golden-cheeked warblers (Setophaga chrysoparia) based on data from 61 light-level geolocators (2017–2021)

Migratory connectivity and migration networks

For the 43 birds with geolocated points on the breeding grounds before July, the estimated mean latitude error was 19.8 km ±16.2 km (range: 0.7–94.2 km) and estimated mean longitude error was 29.4 km ±45.1 km (range: 0.4–209.9 km). The mean distance between capture locations on the breeding grounds was 121 km ±89 km (range: 0–294 km), while the mean distance between non-breeding centroids was 463 km ±272 km (range: 6–1550 km). The mean distance between non-breeding centroids for individuals from the north Hatfield region (DVSP and FH) was 432 km ±259 km (range: 20–1,484 km); the mean distance between non-breeding centroids for individuals from the central Hatfield region (BCNWR and CB) was 500 km ±310 km (range: 49–1,221 km); and the mean distance between non-breeding centroids for individuals from the south Hatfield region (KWMA) was 453 km ±256 km (range: 6–1,147 km). The estimated rM was 0.08 (95% confidence interval: −0.02–0.21), suggesting low migratory connectivity between individuals (i.e., individuals that were closer together on the breeding grounds were not similarly close on the wintering grounds). Similarly, the estimated MC was 0.13 (95% confidence interval: 0.01–0.30), suggesting low migratory connectivity between individuals from the same breeding regions in non-breeding regions (i.e., individuals from the same Hatfield region did not consistently overwinter in the same non-breeding conservation region; Fig. 5).

Simulations from our migratory network model showed a similar fit to our observed data (Bayesian P-value = 0.41). Our migratory network models predicted that the mean proportion of warblers breeding in the South region and overwintering in Highland Central America was 2 to 40 times greater than the proportions for other combinations of breeding and non-breeding regions (Fig. 6). Similarly, the mean predicted proportion of warblers breeding in the Central region and overwintering in Highland Central America was 3 to 17 times greater than the proportion of birds breeding in the Central and North regions and overwintering in other regions (Fig. 6). Further, the mean predicted proportion of birds breeding in the Central region and overwintering in Highland Central America was 7 and 12 times greater than the proportion of birds breeding in the South region and overwintering in Atlantic Lowland Mexico and Southwest Mexico, respectively (Fig. 6). The mean predicted proportion of birds breeding in the North region and overwintering in Highland Central America was, on average, 6 times greater than the proportion of birds breeding in the Central and North regions and overwintering in Southwest Mexico (Fig. 6). Similarly, the mean predicted proportion of birds breeding in the South region and overwintering in Lowland Central America was, on average, 6 times greater than the proportion of birds breeding in the Central and North regions and overwintering in Southwest Mexico (Fig. 6). The mean predicted proportion of birds breeding in the North region and overwintering in Atlantic Lowland Mexico was, on average, 4 times greater than the proportion of birds breeding in the Central and North regions and overwintering in Southwest Mexico (Fig. 6). All other predicted proportions had 95% confidence intervals that overlapped (Fig. 6).

Fig. 6.

Means and 95% confidence intervals (top) and alluvial plot (bottom) showing the predicted proportion of golden-cheeked warblers (Setophaga chrysoparia) that breed in each Hatfield region (i.e., South [S], Central [C], and North [N] [71]); and overwinter in each non-breeding conservation region (i.e., Highland Central America [HCA], Lowland Central America [LCA], Atlantic Lowland Mexico [ALM], and Southwest Mexico [SWM] [72]); from a migration network model created with the “mignette” R package [72, 73]. Values of n indicate the number of birds assigned to each region. In the alluvial plot, the width of each box represents the proportion of the total estimated population that occurs at each region and the width of the colored lines represents the proportion of the population that is estimated to migrate to and overwinter in each non-breeding region

Discussion

Our research provided novel insights into the movement ecology of male warblers throughout their annual cycle. Overall, 80% of warblers migrated along the Sierra Madre Oriental between their breeding grounds in Texas and wintering grounds in southern Mexico and Central America, but 20% took an alternate route over the Gulf of Mexico, primarily in fall. We identified several areas along the warblers’ migration routes and on their wintering grounds that were used by most birds we tracked and could be focal areas for conservation of migratory and overwintering habitat for this species. We also found locations outside the estimated winter range where warblers occurred; one estimated location was confirmed by biologists during field surveys in Oaxaca, Mexico (E. Molina, personal communication). Finally, we found that warblers exhibit weak migratory connectivity. Thus, while specific breeding populations do not appear to be limited to specific locations on the wintering grounds, increased conservation efforts along their narrow migratory pathways and of the montane forests used by most warblers in winter could help ensure population persistence.

Fall and spring migration

Historically, warblers were thought to migrate over land between the Sierra Madre Oriental and the Gulf of Mexico, with no evidence of trans-Gulf migration [30]. Most warblers included in our study did indeed migrate north/south over land as expected, but 20% (1 to 6 individuals per year) migrated over the Gulf during one migration, primarily in fall. We do not know the true locations of individuals outside our study sites, and the accuracy and precision of geolocator data depends on many factors (e.g., time of year, bird behavior, environmental conditions) [63], but the general routes we identified were supported by abrupt changes in median longitudinal and latitudinal estimates following stationary periods. These abrupt changes occurred consecutively over time for individual warblers but were asynchronous across warblers that took the Gulf routes.

Golden-cheeked warblers are difficult to detect during non-breeding seasons. However, there are some published reports and eBird records for warblers outside the southern/northern land route during migration, including St. Croix, Virgin Islands (23 Nov 1939 to 8 Jan 1940 [90]), Pinellas County, FL (24 Aug 1964 [91]), Galveston Island, TX (Aug 1977 [92]), South Padre Island, TX (09 Mar 2012, 06 Jul 2014, 17 Jul 2020 [93]), and Aransas NWR, TX (Jul 1999 [94]). The alternate paths we identified do not appear to be the predominate routes taken by warblers during migration and may represent anomalies, but at least some warblers appeared to take a Gulf route during our study. Why warblers might take these alternate paths remains unclear. We found no evidence that the differences we observed were related to age, departure date, or location on the breeding grounds. Other factors such as prevailing winds [95], extreme weather events [96], resource availability along the migration route [97], body condition [98], or risk of predation [97] could influence warbler migration routes, but elucidating the mechanisms that drive differences in the migratory paths of warblers was beyond the scope of our current research and would require additional study.

Migration timing

Similar to other studies that used geolocators to explore the migratory behavior of songbirds (e.g. [17, 99, 100]), our results indicated that the length of spring migration was shorter than fall migration for warblers. However, the departure and arrival dates for fall and spring migration both spanned approximately two months. In a review of approximately 100 species of New World land birds, Faaborg et al. [101] indicated that winter departure and breeding arrival windows can span weeks to months for birds, and emphasized that food availability, weather, and social cues can cause variation in the length of migration. Contrary to our results, La Sorte et al. [102] found that short‑distance migrants tend to have broader, more plastic windows (weeks to months) while long‑distance migrants are more constrained (often 1 to 3 weeks). We also found limited evidence that fall departure and arrival dates were related. This finding is similar to a review by Schmalijohann [103] of 21 songbird species and a review by McKinnon et al. [104] of 18 Neotropical migrants species which found the start of migration was significantly and positively correlated with arrival timing, which could be related to a specific time needed for travel (i.e., departure date determines arrival date; e.g. [103]), and having an adequate time needed for fuel deposition rates to reach the wintering grounds (e.g. [105]).

The timing of migration is heavily influenced by photoperiod [106], but many factors can influence migration initiation dates and durations. For example, molt is energetically expensive [107]. Warblers molt between June and August, and resource availability on the warblers’ breeding grounds prior to fall migration may influence when molt occurs. Molt timing could vary across locations and years, thus influencing departure dates and the duration of fall migration [108–110]. In addition, weather patterns may be more variable in fall than spring, meaning warblers may need to adjust their migration routes or timing to avoid large storms, and research suggests that migratory songbirds could take advantage of tailwinds in spring that are not available in fall [41, 111].

There are also fewer time constraints during fall migration compared to spring migration. In spring, birds must reach the breeding grounds quickly so that they have sufficient time to establish their territories, pair, and fledge young; early arrival on the breeding grounds may reflect strong selection pressures [112–114] and may result in higher pairing and fledging success for warblers, as previous studies found that warbler daily nest survival decreases as the breeding season progresses (e.g. [112]). Warblers arriving earlier on the breeding grounds may also have more time to re-nest following nest failure or to double-brood after a previous nest fledged. In addition, resource availability along their migration routes may be greater in spring than fall, reducing the need for extended stopovers during spring. Food abundance on the wintering grounds can also advance or delay migration [115] and there is evidence of seasonal differences in hypothalamic gene expression that cause songbirds to gain more fat and body mass during spring compared to fall [116]. Though we cannot comment on these specific factors, we did find that age may influence initiation dates for fall migration, as SY males started fall migration ~10 d earlier than ASY males, and SY males tend to arrive on the breeding grounds 1–2 weeks later than ASY males (J. Macey unpublished data). Age-based differences in the timing of migration are relatively common for songbirds and could reflect variation in experience, body condition, or social dominance behaviors [117–119].

Migration stopovers

Stopover sites are critical for migratory songbirds as these locations allow individuals to rest or refuel en route to their breeding or wintering grounds, and the condition of stopover sites can impact survival or breeding success [101, 120]. Geolocators have revealed unexpectedly long stopovers ( > 7 days) for some species in the fall (e.g. [104, 121]), spring (e.g. [104, 122]), and during both migrations (e.g. [104, 123]). Many of these long-duration stopovers exceed the amount of time needed for refueling based on energetic models alone [124]. However, the relationship between stopover duration and energetic needs is complicated and can be influenced by fuel deposition rates, distances between stopovers, predation risk, and wind conditions, among other factors [97].

We found that warblers spent much of their migratory periods at stopover sites; 48% of the stopovers lasted ≥ 8 days and 18% of stopovers lasted for ≥ 20 days at a single location. In addition, fall stopover locations appeared to be more spread out than spring stopover locations, possibly indicating that in the spring males may be more prone to make a direct flight to reach the breeding grounds sooner [113, 114]. However, as previously stated, our stopover locations do not represent exact locations but rather identify areas that warblers likely used to obtain food resources, rest, or reduce their risk of mortality due to predation. As such, following geolocator studies up with field-based surveys to pinpoint where warblers occur represents a valuable opportunity to improve our understanding of the functions and relative importance of stopover sites. Others could use this information to protect areas that are critical to the survival of warblers during the non-breeding season [125].

Wintering

Few warblers remained in one location throughout the wintering period, which has also been documented for other songbirds [126–128] and may be associated with changes in weather [129], resources [130–132], age [133, 134], or interactions with co-occurring species [135] on a bird’s wintering grounds. However, based on our KDE and wintering stationary locations, most warblers overwintered within or in close proximity to the predicted warbler winter range [46] in southern Mexico and Central America. Our analyses indicated that locations of particular importance to warblers during the winter according to both our KDE and stationary periods (Fig. 2) included central to northern Chiapas and eastern Tabasco, Mexico. Several additional areas had moderate use by warblers, including southern Chiapas, south-eastern Guatemala, and the borders of south-western Honduras and El Salvador, with most of the higher density areas in or near the delineated winter range.

We also identified overwintering locations that were outside the known winter range (Fig. 5). However, locations outside of the delineated wintering range may be attributed to aforementioned geolocator accuracy limitations or could be related to warblers’ migration speed slowing down when they get closer to the wintering grounds, thus erroneously estimating that warblers were in their wintering period. Regardless, as most overwintering locations occurred within a relatively small spatial extent, warblers in these areas may be at higher risk from catastrophic weather events, wildfire, or land use change. Last, we did not find evidence that warblers segregated on their wintering range by age class, and while sex-based segregation occurs on the wintering grounds for some species [119, 136], we did not explore this phenomenon in warblers.

Migratory connectivity

We found low migratory connectivity in warblers, both when using individual locations (rM) and when grouping individuals into conservation regions (MC). To date, the results from similar research have been mixed; some songbirds have exhibited strong migratory connectivity (e.g. [21, 137–139]), whereas others exhibited weak migratory connectivity (e.g. [47, 139, 140]). Our results aligned with Finch et al. [47], who found that long-distance migrants with relatively small wintering ranges (~1,000 km or less) tended to have low migratory connectivity. Though Finch et al. [47] noted that small wintering ranges can erroneously cause calculations of migratory connectivity to be low, even if there is evidence of inter-population separation and low within-population spread, visual assessments of our data (Fig. 2) and average spread between non-breeding points ( > 400 km for all points and points separated by Hatfield region origin) align with our finding of low migratory connectivity for golden-cheeked warblers (Fig. 2). As previously stated, though most warbler wintering habitat is found in a relatively small area of ~1,000 km from Chiapas, Mexico to Nicaragua [46], we estimated non-breeding locations from Veracruz, Mexico to Costa Rica (~1,500 km; Fig. 2). Even considering a potentially larger wintering range size, our findings remain consistent with previous research on long-distance migrants [47] due to our observed within-population spread ( > 400 km).

Low migratory connectivity may represent an adaptation to uncertain conditions on the wintering grounds, such as climate variability [47]. While populations with strong migratory connectivity and individuals with high winter site fidelity may be better adapted to local conditions [19], events on the wintering grounds can have a stronger effect on migration phenology [141], survival [101, 142], and reproductive success of migratory songbirds [143]. Conversely, populations with weak migratory connectivity that diffuse across their winter range may be more resilient to local variation in resource availability. Similarly, the effects of habitat loss or natural disturbances on the population’s breeding or wintering grounds may be less pronounced for species that exhibit weak migratory connectivity compared to species that exhibit strong migratory connectivity [144, 145]. In addition, populations with weak migratory connectivity may exhibit genetic variation in their migratory behavior (e.g., direction and timing) that allows them to adapt more quickly to changes on their breeding or wintering grounds [19].

Further, though our migratory network model predicted that the largest proportion of warblers breed in the South Hatfield region and overwinter in Highland Central America, our model also predicted that large proportions of birds from both the Central and North Hatfield regions overwinter in Highland Central America. Highland Central America encompasses large portions of the Sierra Madre de Chiapas mountain range, with elevations in the region at an average of 1,200 masl and ranging up to 4,200 masl, aligning with known non-breeding habitat (montane oak-pine forests) for the warbler [146–148]. Our migratory network model also predicted that a non-trivial proportion of birds from all Hatfield regions overwintered in three additional conservation regions generally representative of different average elevations and biomes [73]. However, all three regions included portions of mountainous areas that were > 1,000 masl, so these regions may also include patches of montane pine-oak forest that warblers used during the non-breeding season, though our geolocators also estimated overwintering locations outside of montane environments for several individuals (see above). It is therefore important to note that conservation regions created by DeSaix et al. [73] do not represent the exact boundaries of ecoregions, geologic features, or political entities. Thus, though our analyses suggested that warblers migrated to several regions during the non-breeding period, the actual range of wintering habitat is smaller than that covered by these conservation regions and locations used by warblers may not be representative of the land cover, elevations, and vegetation types that predominate each individual region. Further, though we detected birds in Southwest Mexico, the proportion of the entire warbler population predicted to migrate to this region was < 0.05, suggesting that this region may contain limited non-breeding habitat for the warbler.

Migratory connectivity estimations can be affected by sample size and the distance between capture locations on the breeding grounds (see [22]). We mitigated potential biases in our conclusions to the extent possible by reporting three metrics: two to describe migratory connectivity (rM and MC) and one metric for non-breeding population spread (i.e., distance between non-breeding locations, separated by Hatfield region origin), which is less affected by sampling bias [22]. We were unable to directly estimate geolocator error on the non-breeding grounds and we used 75% UD centroids to represent each birds’ wintering location, so it is possible that error associated with these factors affected our results for migratory connectivity metrics, as well as our assignment of birds to non-breeding conservation regions. However, even considering potential geolocator error and limitations to conservation region boundaries, warblers from the same locations on the breeding grounds had estimated wintering locations that were spread across Central America, suggesting that localized threats on the non-breeding grounds could have fewer negative effects on birds from a single breeding region [47]. In contrast, generalized threats like widespread habitat loss in the non-breeding grounds could cause severe declines in the warbler population across the breeding grounds [47]. Given the limited extent of potential habitat on their wintering grounds (~104,000 km² [46]), habitat losses forecasted to occur under the most optimistic climate change scenario (~43,000 km² [46]), and uncertainties in how songbirds could adapt (e.g., physiologically, morphologically, behaviorally) to increased conservation along the warbler’s migration route and on their wintering grounds, continued and increased research on connections between the breeding and wintering grounds is warranted to ensure population persistence. Future studies could consider using recently developed barometric geolocators (see [149]), integrating data from geolocators or other tracking devices with complementary techniques (e.g., isotopes, genetics), or using methods that allow for more robust analyses (e.g., analyses that account for survivorship bias [150]), all of which could further advance our understanding of the migratory ecology of warblers and other species.

Conclusions

Our research filled a critical data gap for the warbler and provided information that could help narrow sampling periods and locations for field-based surveys in non-breeding habitat, explore how other factors (e.g., weather, climate) might influence the species’ migratory behavior, and identify non-breeding habitat in need of protection. Such knowledge is necessary to develop a comprehensive conservation strategy for the species and could increase the impact of conservation actions by organizations and partners who work to protect migratory birds (e.g., Partners in Flight, Working Lands for Wildlife, Americas Flyways Initiative, Migratory Bird Joint Venture). Future research to understand the migratory ecology of female warblers and to determine whether warblers exhibit sexual segregation on their wintering grounds could also help target conservation efforts where they are needed most.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

ANOVA

Analysis of Variance

Apr

April

ASY

After-second-year

Aug

August

BCNWR

Balcones Canyonlands National Wildlife Refuge

Bldg.

Building

C

Celsius

CA

California

CB

Joint Base San Antonio-Camp Bullis

CI

Confidence interval

cm

Centimeter

d

Day

DSVSP

Dinosaur Valley State Park

e.g.

Exempli gratia

et al.

Et alia

Feb

February

FH

Fort Hood

Fig.

Figure

FL

Florida

g

Gram

GPS

Global positioning system

ha

Hectare

HSD

Honestly significant difference

HY

Hatch year

i.e.

Id est

Jan

January

Jul

July

km

Kilometer

KWMA

Kerr Wildlife Management Area

LA

Louisiana

LCL

Lower confidence limits

Mar

March

masl

Meters above seas level

mm

Millimeter

MO

Missouri

n

Number

Oct

October

OK

Oklahoma

P

Probability

r

Correlation coefficient

R

Range

rM

Mantel’s correlation coefficient

SD

Standard deviation

Sep

September

SY

Second-year

TX

Texas

UD

Utilization distribution

UCL

Upper confidence limits

USA

United States of America

USGS

United States Geological Survey

USFWS

United States Fish and Wildlife Service

UTM

Universal Transverse Mercator

V.

Version

Mean

Author contributions

JNM, AML, and JMK wrote the manuscript with contributions from KC, SR, JMM, SMC, MRC, NG, NMR, MDG, DF, and RL; JNM, AML, and JMK conducted the analysis; JNM and AML obtained funding; JNM, KC, SR, JMM, SMC, MRC, NG, NMR, MDG, DF, and AML conducted field research.

Funding

This project was funded by Department of Defense Strategic Environmental Research and Development Program (SERDP) (W9126G-18-2-0025) and Fort Hood.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request and approval from the corresponding state or federal agency that manages the study sites where we conducted this work.

Declarations

Ethics approval

We conducted all field activities in accordance with the following permits: USFWS #TE023643-11, #TE32917C-1, and #TE082496-0; USGS #21999 and #24126; TPWD #SPR-0409–079, #SPR-0417-097; and Louisiana State University AgCenter Animal Care and Use Protocols #2018-11 and #A2021-10.

Consent for publication

Not applicable.

Competing interest

The authors declare no competing interests.