Using song dialects to reveal migratory patterns of Ruby-crowned Kinglet populations

. Conservation of a migratory species requires knowledge not only of its breeding range, but also of its migratory path and non-breeding range. Except for timing, other aspects of the migration of the Ruby-crowned Kinglet ( Corthylio calendula ) remain largely unstudied, with no published data on migration routes. Breeding populations of this species in the Sierra Nevada and Cascades mountain ranges, as well as those in eastern Canada and the northeastern U.S., have experienced significant declines, whereas Rocky Mountain breeders have increased. Understanding the winter range and migratory pathways used by different breeding populations may be key to explaining these contrasting population trends. Song dialects of the Ruby-crowned Kinglet differ regionally among various breeding populations, and these dialect regions were previously mapped. Because this kinglet sings during spring migration and winter, we obtained archived, non-breeding-season recordings of song and assigned each to one of those regional song dialects. This allowed us to assess the likely winter ranges and migration pathways of different breeding populations. This approach offers some advantages over typical methods of tracking movements. Birds do not need to be captured; one can easily obtain data over large ranges and from many individuals; and it can be applied to species, such as this kinglet, that are too small to permit use of most tracking devices. We were able to assess likely winter range and spring migration routes for populations that breed in the eastern U.S. and Canada, the interior of Alaska, and for the subspecies C c. grinnelli that breeds along the Gulf of Alaska and western British Columbia


INTRODUCTION
Migratory species present challenges to understanding their ecology and any conservation challenges they may face (Faaborg et al. 2010). Whereas most bird research is focused on the breeding season (Marra et al. 2015), migratory birds generally spend most of their year in migration or in the winter range (Newton 2008). Development of effective conservation plans for such species requires knowledge of habitat conditions in both the breeding and non-breeding ranges.
Newer tracking technologies, such as light-level geolocators or satellite tracking, have dramatically improved our ability to learn about bird migration (Stutchbury et al. 2009, Flack et al. 2022. However, these techniques have limitations. They require capturing individuals, usually twice, to attach and recover the devices, and they are limited to tagging animals above a minimum mass (Brlik et al. 2020). Bird banding may also offer glimpses into the migration path or winter range of birds, although encounter rates away from the original banding location, particularly in small passerines, are often low.
For bird species that use different regional song dialects, and also sing during migration or winter, such dialects can allow one to use recordings to assess the non-breeding ranges of birds using these distinct song dialects. This approach offers advantages over other techniques because birds do not need to be captured and large numbers of individuals can be studied from many locations. This approach was used to reveal the migratory strategies of the Puget Sound subspecies of the White-crowned Sparrow (Zonotrichia leucophrys pugetensis; DeWolfe and Baptista 1995) and the Golden-crowned Sparrow (Z. atricapilla; Pandolfino and Douglas 2021), and to determine the source of irruptive individuals of the Black-chinned Sparrow (Spizella atrogularis; . The large and rapidly growing archives of bird recordings offers promise that this approach may have broad applications to many species. The Ruby-crowned Kinglet (Corthylio calendula) breeds from easternmost Canada and the northeastern U.S. across the boreal regions of Canada and the upper Great Lakes and into western Canada and much of Alaska (Swanson et al. 2021). The breeding range extends through most of the montane western U.S., including the Rockies, Cascades, Sierra Nevada, and higher elevations in the Great Basin region. The winter range includes the Pacific coast from southernmost British Columbia through Baja California, lower elevation inland areas of the western U.S., the U.S. Eastern Seaboard from southern New England south, most of the southern half of the eastern U.S., and south through Mexico (Swanson et al. 2021) into Guatemala (Howell and Webb 1995). According to Swanson et al. (2021), no migratory-route information is available. And the Ruby-crowned Kinglet, at six to seven grams (Swanson et al. 2021), is so small that only the largest individuals meet the criteria for tagging with the smallest currently available geolocators.
With the apparent extinction of the Guadalupe Island endemic C. c. obscurus (Quintana-Barrios et al. 2006), there are now two recognized subspecies, nominate C. c. calendula and C. c. grinnelli (Swanson et al. 2021). C. c. grinnelli breeds in coastal forests from the Gulf of Alaska through British Columbia. There is disagreement about the non-breeding movements of C. c. grinnelli. Browning (1979) asserted that this taxon is resident within its breeding range, and Campbell et al. (1997) speculated that C. c. grinnelli makes short latitudinal movements with northern breeders, possibly wintering in southern British Columbia. Others have characterized it as partially migratory with regular winter movements south along the west coast (Phillips 1991), at least into Washington (Wahl et al. 2005) and Oregon (Marshall et al. 2003), and possibly as far south as coastal central California (Dawson 1923). Winter specimens of this taxon have been collected in coastal California as far south as San Diego County (Unitt 2004).
The winter ranges of various populations of the nominate subspecies remain unknown. Some have suggested that these birds may make altitudinal movements from the western mountain breeding areas to nearby lower elevations (Browning 1979, Ryser 1985, Campbell et al. 1997, Wahl et al. 2005, Floyd et al. 2007). Hobson et al. (2014) analyzed stable isotopes from Ruby-crowned Kinglets wintering in Mexico, but the results showed a very wide range of values potentially representing most of the species' breeding range.
Different breeding populations of the Ruby-crowned Kinglet have shown starkly different breeding-population trends since the mid-1960s (Sauer et al. 2020). Birds breeding in the easternmost parts of the range demonstrated significant negative population trends, as have breeders of the Sierra Nevada and Cascades ranges in California, Oregon, and Washington. In the Sierra Nevada of California this species was abundant and widespread in the early 20th century (Dawson 1923, Grinnell andMiller 1944), but is now an uncommon and sparsely distributed breeder (Beedy and Pandolfino 2013), even in areas where the habitat is well-protected and largely unchanged from earlier conditions (Moritz 2007). Indeed, a recent Breeding Bird Atlas in the central Sierra Nevada, where this kinglet was previously common, did not record a single observation of the species (Rose and Rose 2019). In contrast, most of the Rocky Mountain breeding populations from Canada to Arizona and New Mexico showed population gains (Sauer et al. 2020). To understand factors that may be driving these contrasting trends, we need to know the migration pathways and winter ranges of these different regional populations.
The Ruby-crowned Kinglet uses distinctly different versions of their songs (dialects) in different regions within their breeding range, and  recently mapped these dialects throughout this species' breeding range. Because this kinglet also sings during spring migration and winter (Marshall et al. 2003, Wahl et al. 2005, Swanson et al. 2021, we chose to use non-breeding season recordings to assess winter ranges and migration pathways of various populations by assigning one of the regional song dialects to each individual recorded. Because spring migration tends to be generally direct in most birds species studied (Newton 2008, Nilsson et al. 2013, La Sorte et al. 2016, we hypothesized that spring migration of the Ruby-crowned Kinglet would be direct.

METHODS
We used Raven Pro software (KLYCCB 2022) to visually examine all song recordings available from xeno-canto (https://www.xenocanto.org) and the Mark Robbins/Macaulay Library (https:// www.macaulaylibrary.org) of Ruby-crowned Kinglets outside the breeding season (August through mid-May). We compared the songs of each individual recorded to the 43 dialect song types (hereafter "song types") previously identified by . Following , the song types were based on the type of repeated phrases used in the third part (Unit C) of the song (Fig. 1). As these authors demonstrated, these phrases are visually distinct and can be unambiguously assigned to a type based on visual inspection of spectrograms (Appendix 1), and each individual appears to have only a single song type in their repertoire. When multiple recordings at the same location on the same day were available, we assumed that all those recordings were of the same individual. The assignment of a given song type to a recording was done in a blind fashion. That is, the recordings were coded such that the authors did not know the location of the recording when assigning a song type, to avoid any possibility of unintended bias. Because many of the songs from the nonbreeding period were subsong (a poorly formed vocalization given by birds still perfecting their song), presumably from young birds, we only assigned song types when the song clearly matched one of the types identified by . Of the 1110 recordings available from xeno-canto and the Macaulay Library (through 30 April 2022), 165 were of too poor quality to be used. Another 186 recordings were eliminated as possible duplicate recordings from the same individual (recorded on the same day and location). We eliminated 286 recordings of subsong that could not be unequivocally assigned to an adult song type. We used the same numbering system for these types as .
We were able to assign a song type to 473 individuals (Appendix 2). We grouped these recordings by breeding region based on song types found exclusively or predominantly within that region by , and we used the geographic regions identified by those authors: east, north central, northern Rockies, central Rockies, southern Rockies, west, interior Alaska, and the range of the subspecies C. c. grinnelli (Fig. 2).
We also obtained data on all encounters of banded Ruby-crowned Kinglets from the Bird Banding Laboratory archives (Celis-Murillo et al. 2021). Our sample used only data from encounters at least 500 km from the banding location (Appendix 3). We chose 500 km to eliminate or reduce examples of within-season movement as opposed to migratory movement.  . Ranges (breeding shaded yellow, winter shaded blue) based on Fink et al. (2022).
We grouped the recordings and banding data into the following time frames: fall migration (August through October); winter (November through March); spring migration (April); late-spring migration (first half of May); and breeding (mid-May through July; Fig. 3). Table 1 shows proportions of recordings or banding encounters by non-breeding period. We used the time frames noted above based on eBird movement data as displayed in Swanson et al. (2021). Whereas this schedule may not apply to all populations precisely, it does represent movement timing for the bulk of the population. For example, those data show that, whereas a few individuals may begin moving north during the last week of March,  spring migration appears to be largely restricted to April and early May. These dates are also consistent with earlier spring migration of males, which likely account for all song recordings we used, as demonstrated by Swanson et al. (1999).

Fig. 4. The locations of individuals using song dialects of the east region (A) and the north-central region (B)
. Blue dots represent recordings made from November through March; light green arrows, recordings from April; dark green arrows, recordings from the first half of May; brown arrows, recordings from August through October; red dots, recordings from mid-May through July (Pandolfino and Douglas 2022, but only including song types found in this study). Ranges (breeding shaded yellow, winter shaded blue) based on Fink et al. (2022).

General observations
We found no identifiable song types from August or September, two from October, one from November, none from December or January, and four from February. Nearly all the songs recorded in those months appeared to be subsong and could not be assigned to any song type. We found many individuals giving recognizable song types by March (98), April (315), and early May (54).

East region
Our sample included eight different song types from 241 individuals that were associated exclusively with the east breeding region (types 1, 5, 13, 21, 30, 31, 34, and 40). Song type 21 was the most common, accounting for 60% of the eastern song types in our sample. This was also the most common type found by  in the breeding season. The seasonal distribution of these song types (A, Fig. 4) showed the winter individuals in the southeastern U.S. (from Mississippi eastward) and along the Eastern Seaboard. During spring migration (April), most birds were found along the east coast of the U.S., with most on or near the breeding range by early May.

North-central region
We found four north-central song types (9, 26, 28, and 38) from 124 individuals. As with the east region, the most common song type in the breeding-season samples from  was the most common (77%) in our sample (type 9). Some winter-season individuals using those song types overlapped with the winter range of the east-region birds in the southeast and along the eastern U.S. coast (B, Fig. 4), but most were found farther west from Mississippi into Texas, with one bird recorded in Mexico. One winter individual in western Nebraska was recorded in late March (23 March) and was likely an early-spring migrant. Birds giving these song types showed spring-migration distribution across a wide swath of the eastern U.S. from Iowa to the east coast. However, birds using these song types in spring were found, on average, significantly farther west than the east-region birds (average spring migration longitude = -85.6° ± 6.6° for north-central song types, vs. -76.1° ± 4.3° for east types). As with the east-region breeders, most late-spring migrants were on or near the breeding range.

Interior Alaska
We found 40 individuals giving song types used by interior Alaska breeders (types 15, 20, 22, and 27). Of those, 33 were of song type 27, also the most common type found by . Most of the winter recordings were from California, with a few from Arizona and Washington, and all spring migrants were close to the west coasts of California, Oregon, Washington, and British Columbia (A, Fig. 5).

Range of C. c. grinnelli
Kinglets in the breeding range of this subspecies used two song types (18 and 38) almost exclusively . We found 10 winter-season birds giving these song types  (B). Blue dots represent recordings made from November through March; light green arrows, recordings from April; dark green arrows, recordings from the first half of May; brown arrows, recordings from August through October; red dots, recordings from mid-May through July (from Pandolfino and Douglas 2022, but only including song types found in this study). Ranges (Breeding shaded yellow, Winter shaded blue) based on Fink et al. (2022).
from the southern tip of Vancouver Island south through Washington, Oregon, and into California (B, Fig. 5). Spring migrants were recorded along the Pacific coast from Washington, through British Columbia, and into Alaska.

Interior west
This area includes four of the regions identified by  encompassing the Rocky Mountains from northern Canada to Arizona, and the Cascades and Sierra Nevada ranges from Washington to California. We found too few recordings of interior-west song types (42 recordings in our sample for all four regions) to derive useful information about winter ranges or migration. For example, our winter recordings of song types from the northern, central, and southern Rockies, and the west regions, numbered just 2, 3, 6, and 1, respectively. In addition, some of the song types from those regions were widespread through more than one breeding region . The few spring-migration examples suggest a fairly direct northward migration, but characterization of migratory pathways is not possible without more information about the winter range of these populations.

Banding data
Banding data provided information about the movements of 46 different individuals (Fig. 6). Unfortunately, we found only a single bird for which there was information about both the winter and breeding locations. That individual was banded 16 July in western Quebec and was encountered the following winter (22 January) in the panhandle of Florida. In addition, three other individuals banded in the first week of May, most likely on or near the breeding range, were encountered in winter. Two of those birds were banded on 6 May and 9 May in southern Ontario and encountered in winter in North Carolina and eastern Texas, respectively, within the winter range inferred from our songdialect work. The other was banded on 6 May in the Sierra Nevada of California and encountered 20 January in northern Mexico. Kinglets wintering in the southeastern U.S. all remained in the two regions expected from the dialect data, and all the birds banded in the west were encountered in the west. Even the individual showing the longest migration (banded in early spring on the Louisiana coast) bred in central Alberta, within the expected north-central region (longest line in Fig. 6). Thus, all these encounters with banded birds were consistent with the migration paths suggested by our song dialect analyses.

Vagrants
Among the 1066 recordings used in this study and by , we assumed that eight were vagrants (Table  2): that is, birds that had wandered well beyond the expected breeding or migration areas based on the song types used. Two of these were birds recorded in the west, singing song types from eastern breeding regions, and six were recorded in the east, singing western-type song dialects.  Fink et al. (2022).

Broader application of the method
To our knowledge, this approach to studying bird movements has only been applied in three prior studies (DeWolfe and Baptista 1995, Douglas 2021, Pandolfino et al. 2022). The large and growing archives of bird song should allow this method to be applied to many other species. This offers the advantages of birds not needing to be captured, and many individuals across a broad range being able to be sampled very economically. Application of this method requires that the species show distinct regional variation in song and that individuals sing during the non-breeding season. It is also very helpful if the species has only one, or very few, song types in its repertoire. Because nearly all such species studied appear to maintain this repertoire throughout their lives (Marler andSlabbekoorn 2004, Catchpole andSlater 2008), one can be reasonably confident that a song type encountered outside the breeding range will match the one used in that range. One must also be aware that females of many species sing. However, in most cases, as with the Ruby-crowned Kinglet (Pieplow 2017), their song is easily distinguished from that of the male. For example, female song of the Ruby-crowned Kinglet lacks Unit C entirely (Pieplow 2017). However, Pieplow (2017) notes singing mainly from February through August, and our analyses of the available recordings (Pandolfino and Douglas, personal observation) showed only a few examples of poorly developed song during fall, with progressively larger proportions of fully crystallized adult song from March through May. By late May through July, all birds recorded were singing fully developed songs.

Migration routes and winter ranges
Our approach does not reveal migration routes of individuals but provides an overview of winter ranges and migration routes for subpopulations. For example, individual birds may make significant intra-winter movements within the winter range and we did not assume that the location of any individual bird in our sample was stable throughout the winter period. Ruby-crowned Kinglets using the song dialects of the two easternmost regions (east and north central) appeared to take a direct spring-migration route from their winter range to the breeding range. Although we had fewer data points for breeders from interior Alaska and members of the C. c. grinnelli subspecies, those spring migrations also appeared to be direct with spring migrants found mainly along the west coast.
Those results are consistent with the general pattern of most migratory birds for which spring migration is faster and more direct than fall migration (Newton 2008, Nilsson et al. 2013, La Sorte et al. 2016. This difference is believed to be driven by the importance of arriving on the breeding grounds early enough to find and occupy the best territories (Myers 1981, Gauthreaux 1982, Francis and Cooke 1986, Chandler and Mulvihill 1990, Newton 2008. Female Ruby-crowned Kinglets may give a short version of song lacking the last part (Unit C; Pieplow 2017), so we assumed that all the singing individuals in our sample were males. Male Ruby-crowned Kinglets migrate earlier than females (a common observation among most passerines studied; Newton 2008), allowing them to establish breeding territories before females arrive (Swanson et al. 1999). Males also appear to winter farther north than females (Fairfield andShirokoff 1978, Swanson et al. 1999); thus, the winter ranges inferred from our song data may be skewed northward.
Based on the winter locations of birds singing the dialects of C. c. grinnelli, some members of this taxon migrate well south of the breeding range, which is contra Browning (1979) andCampbell et al. (1997). We found these individuals at the southern tip of Vancouver Island, British Columbia, and south into Washington, Oregon, and northwestern California. Of course, females wintering farther south would not have been found by our method, and the winter range of this subspecies may well extend through the Marine West Coast Forest ecoregion in California.
The migration routes and wintering ranges of birds breeding in the western mountain ranges (Rockies, Sierra Nevada-Cascades) remain largely unknown due to the small number of recordings we found from birds using those dialects and the widespread nature of some of those dialects. It seems likely that many or most of those western montane breeders may winter in Mexico. However, we have information on only two individuals wintering in Mexico: one in Mexico City (south-central Mexico) giving a song type exclusively associated with breeders in the north-central region, and another banded in the central Sierra Nevada in May and encountered in January in northwestern Mexico.
Our data included only one migrating bird west of the Mississippi River using an east-region song; therefore, it seems unlikely that significant numbers of east-region breeders winter in Mexico. Stable isotope analysis of samples from Ruby-crowned Kinglets wintering in Mexico (Hobson et al. 2014) were consistent with this assumption, suggesting most birds may have bred in the northern Rockies from Canada into Wyoming. However, the full range of values reported in that study was broad enough to encompass nearly the entire breeding range.
Migratory connectivity, the extent to which birds breeding in one area also share a wintering area, is not possible to assess from our data. Whereas there may be groups of individual Ruby-crowned Kinglets that have a high degree of migratory connectivity, the relatively widespread nature of the song dialects we observed suggests only a very diffuse sort of connectivity. For example, song type 21, for which we have the most examples, is found in the breeding range from Newfoundland, south into upstate New York, and in winter along the U.S. Eastern Seaboard from Connecticut to Florida and west into eastern Louisiana.
Our samples included only three song types for which we had robust numbers (> 10) of observations from both the breeding range and the wintering range. Therefore, our data were insufficient to assess whether the Ruby-crowned Kinglet uses a chain-migration strategy (in which the most northern breeders also winter the farthest north), or leapfrog migrations (in which the most northern breeders winter the farthest south). And the lack of data from Mexico, where many Ruby-crowned Kinglets winter, makes any estimate of typical wintering latitudes suspect.

Vagrancy
Vagrancy, when a bird is observed well away from its expected breeding, wintering, or migration range, is observed among nearly all migratory species (Newton 2008). Such birds are most often found in spring or fall and are usually young birds making their first migration from the breeding area or from their wintering range (De Sante 1983a, De Sante 1983b, Newton 2008). Many of these birds are believed to have an anomalous sense of orientation in which they choose to migrate on a trajectory different from that of most members of that species. This misorientation can vary among individuals from a few degrees to 180 degrees, resulting in migration in the opposite direction from normal (De Sante 1983b, Newton 2008).
Ruby-crowned Kinglets have been observed as vagrants as far east as Scotland (BOURC 2022) and as far west as Attu Island at the western end of the Aleutian Chain (Dunn and Alderfer 2011). Our use of mapped song dialects allowed us to detect so-called cryptic vagrant Ruby-crowned Kinglets that would never have been noted as being out of range. Seven of the vagrants were recorded within regions and on dates when the species is common. Thus, only the detection of unexpected song dialects revealed that they were outside the expected range. The exception was a bird recorded 18 May in Baja California, Mexico, a date when Rubycrowned Kinglets are not expected in that area. In each case, these birds were using a dialect otherwise unrecorded in that area. The fact that more vagrants were detected in the east than the west may reflect the higher concentration of birders in the east, though our sample size is too small to draw any conclusions.

Priorities for future research
The paucity of recordings from Mexico leaves an important conservation question unanswered. The starkly contradictory trends among birds breeding in the Rocky Mountains versus those breeding in the Sierra Nevada-Cascades range may be related to the conditions in the wintering ranges of those birds. A concerted effort to record songs of the Ruby-crowned Kinglet in Mexico in March could yield important information, especially about any difference in wintering ranges among populations. Additional recordings may help reveal details about migratory strategy (e.g., chain versus leapfrog) or connectivity, as well.
Other methods could complement our results and provide more information about movements of individual birds. Stable isotope analyses of feathers can produce information about the molting locations of birds (Hobson 1999), and Ruby-crowned Kinglets are believed to molt on the breeding grounds (Swanson et al. 2021). However, the wide range of isotope values from Rubycrowned Kinglets reported by Hobson et al. (2014) revealed only very general information about breeding ranges. Tagging Rubycrowned Kinglets with MOTUS (https://www.motus.org) transmitters, which are light enough to use with kinglets, could reveal much more about movements of individuals, at least in areas with a high density of receivers.

Acknowledgments:
This sort of research would not be possible without the archives of sound recordings maintained by xeno-canto (https://www.xenocanto.org), the Mark Robbins Macaulay Library (https://www. macaulaylibrary.org) and the many recordists who contribute their recordings to those archives. Appendix 2 includes a complete list of those recordists. Comments and suggestions from two reviewers allowed us to improve the focus and clarity of the manuscript. We are also grateful for the banding data made available by the U.S. Geological Survey's Bird Banding Laboratory (https://www.usgs. gov/labs/bird-banding-laboratory) and the efforts of bird banders throughout North America. This material uses data from the eBird Status and Trends Project (https://ebird.org/science/status-andtrends) at the Cornell Lab of Ornithology, eBird.org. Any opinions, findings, and conclusions

Data Availability:
All data will be made fully available. In fact, all the raw data are in the Appendices and all recordings are available to the public.