Determinants of partial bird migration in the Amazon Basin

Summary

1. Little is known about mechanisms that drive migration of birds at tropical latitudes. Because most migratory bird species in South America have populations that are present year-round, partial migration (in which only some individuals of a given population migrate at the end of the breeding season) is likely to be common, providing an opportunity to assess proximate mechanisms of migration.

2. Two non-mutually exclusive hypotheses explaining intraspecific variation in migratory behaviour were tested in a Tropical Kingbird Tyrannus melancholicus population in the southern Amazon Basin, where a dramatic dry season decrease in the abundance of insect food for kingbirds may promote migration of some individuals.

3. The Dominance hypothesis predicts sub-dominant individuals migrate at the end of the breeding season and dominant individuals do not, whereas the Body Size hypothesis predicts smaller individuals migrate and larger individuals do not.

4. Based on 4 years of data on individually-marked birds, strong support was found for occurrence of partial migration in the study population.

5. In the best model, the largest males (which are typically older and dominant to younger individuals) had the highest probability of migrating. Younger females (which are the smallest individuals in the population) were also more likely to migrate than other kingbirds, except the largest males. Thus, an individual’s probability of migrating was associated with a more complex interaction of size, age and sex than predicted by current hypotheses.

6. These results suggest that determinants of migratory behaviour differ between North temperate and tropical latitudes. Most tests of partial migration theory have been conducted on granivores (e.g. emberizids) or omnivores (e.g. turdids and icterids) at North temperate latitudes, where seasonality is primarily defined by temperature cycles. In tropical South America, however, the most common long-distance migrants are primarily insectivorous (i.e. tyrannids), and seasonality is largely defined by rainfall cycles.

7. We propose a hypothesis based on seasonal food limitation to explain partial migration of primarily insectivorous species in seasonal tropical habitats.

Introduction

The movement of organisms can influence structure and dynamics of populations, communities and ecosystems (Nathan et al. 2008); consequently, the ecology of movement constitutes a diverse and rapidly growing field (e.g., Holden 2006). Yet, understanding how and why animals migrate has been a challenge for researchers, in part because of the interdependency of factors driving migration. Among individuals, what factors underlie the decision to migrate? Partial migration, in which some individuals migrate and others remain sedentary, provides an opportunity to address this question.

Partial migration may involve migrants and residents overwintering together, with migrants leaving to breed elsewhere (e.g., American Dipper, Cinclus mexicanus, Gillis et al. 2008), or migrants and residents breeding together, and migrants leaving to overwinter elsewhere (e.g., House Finches, Carpodacus mexicanus, Belthoff & Gauthreaux 1991). The latter type may be facultative, depending on the condition and previous history of each individual (e.g., Nice 1937; Brackbill 1956; Ogonowski & Conway 2009) or obligate, in which an individual is born a migrant or resident for life (e.g., Terrill & Able 1988; Berthold 1991). Hereafter, we focus solely on facultative partial migration.

In partial migrants, a continuum of ecological tradeoffs determines each individual’s probability of migrating, such that decisions of whether or where to migrate depend on multiple factors including age, sex and physical condition (Ketterson & Nolan 1976; Boyle 2008).

Most workers have approached the question of partial migration from a mechanistic (proximate) perspective, seeking to understand how and why the costs and benefits of migration differ among individuals (Gauthreaux 1982; Adriaensen & Dhondt 1990; Boyle 2008). ‘Migrant’ and ‘resident’ are viewed as alternative strategies, available but not equally beneficial to all members of a population because of individual asymmetries such as body size (Swingland 1983) or because of frequency-dependence (i.e. an individual’s likelihood of migrating depends upon the relative frequencies of dominant and subdominant individuals in the population; Lundberg 1987, 1988).

Several hypotheses have been proposed to explain which individuals within a population are most likely to migrate and how far they travel (reviewed by Cristol, Baker & Carbone 1999; Bell 2005; Boyle 2008). The Arrival Time hypothesis (Ketterson & Nolan 1976) posits that individuals that establish territories in the spring (typically adult males) are less likely to migrate as far as individuals who do not establish territories because a shorter migration distance ensures a more rapid return to the breeding grounds and thereby first access to the best territories. The Dominance hypothesis states that because subordinate individuals are poor competitors relative to dominant individuals, they are most likely to accept the costs associated with migration to avoid competition with dominant individuals when food becomes scarce on or near the breeding grounds (Gauthreaux 1978; Ketterson & Nolan 1979). This hypothesis has received support from several studies (e.g. Lundberg 1985), but has been rejected by others (e.g. Rogers et al. 1989; Boyle 2008). The Body Size hypothesis predicts that individuals of larger body sizes (typically males) are able to withstand colder temperatures and better endure food shortages than smaller individuals. Thus, larger individuals are less likely to migrate than smaller individuals. This hypothesis has also received support from some studies (e.g. House Finches; Belthoff & Gauthreaux 1991) but not from others (Guillemain et al. 2009). Collectively, all three hypotheses were developed and have been tested primarily at North temperate latitudes (but see Boyle 2008). How well do they explain partial migration elsewhere, where seasonality may differ?

Because of climatic and geographic differences between the Northern and Southern Hemispheres, and the wide variety of migration strategies employed by species in the Southern Hemisphere, species that breed in the Southern Hemisphere offer a unique opportunity to better understand the evolutionary ecology of bird migration (Dingle 2008). Approximately two-thirds of Neotropical austral migrants (i.e. species that migrate wholly within South America) have distributions in which individuals are always present in some part of the species’ range (usually towards the Equator), while individuals elsewhere in the range occur only seasonally (Chesser 1994; Stotz et al. 1996; Jahn et al. 2006). Although partial migration is strongly suggested by this pattern, it has not yet been confirmed for any South American species through studies of individually-marked birds.

Tropical Kingbirds Tyrannus melancholicus; hereafter ‘kingbirds’ are completely migratory where they occur south of approximately 18°S latitude in South America, but are present year-round on the continent north of that latitude (Chesser 1995; Fig. 1). The region immediately north of 18°S, the southern edge of the Amazon Basin (Fig. 1), is characterized by pronounced wet and dry periods (October – February and March – September, respectively) that result in strong seasonality of flowering and fruiting (Batalha & Martins 2004), and insect abundance (Pinheiro et al. 2002; Jahn 2009). This seasonal cycle could result in partial migration, with some kingbirds migrating away after the breeding season when drier conditions reduce food availability.

Figure 1.  Distribution of Tropical Kingbird in South America, from Chesser (1995) and Ridgely & Tudor (2009). The gray polygon located mostly north of 18°S latitude represents the area of permanent residency. The dashed line within the gray polygon denotes the Amazon Basin. The white polygon represents the area where the species is found only during the austral summer. The star represents the study site, Caparú Biological Station, from where some kingbirds migrate northwards after breeding (i.e. partial migration in the fall).

We had two goals: (i) determine whether partial migration exists in kingbirds in the southern Amazon Basin and if it does; (ii) test the Dominance and Body Size hypotheses. For the Dominance hypothesis, we predicted that dominant individuals would be less likely to migrate than subordinate individuals. For the Body Size hypothesis, we predicted that: (i) individuals of the smaller sex would be more likely to migrate than individuals of the larger sex; and (ii) within a given sex, smaller individuals would be more likely to migrate than larger individuals. These predictions assume that the rationale for body size-related differences in endurance of cold temperatures and low food availability applies both within and between sexes. We did not test the Arrival Time hypothesis because we were unable to unambiguously determine when male kingbirds established territories in spring.

Given the potential for confounding results because male kingbirds are larger than females and older kingbirds are larger than younger kingbirds, we employed a maximum-likelihood approach, which allowed comparisons of additive and interactive models with multiple covariates among demographic (i.e. age and sex) and morphological (i.e. body size) variables.

Materials and methods

The study was conducted at Caparú Biological Station (CBS), Department of Santa Cruz, eastern Bolivia (14° 49′S, 61° 11′W; 170 m elevation; Fig. 1). Habitat at CBS is primarily grassland with scattered trees (mostly Curatella americana, 4–6 m height). Humid forest surrounds the site, except to the south.

Kingbirds at the site breed September–January (Jahn 2009), and were usually captured by placing a net and a model of a nest predator (Purplish Jay; Cyanocorax cyanomelas) near a nest, or by placing nets near ponds where kingbirds bathed. Nestlings were also ringed. All kingbirds were ringed with uniquely-numbered aluminium rings provided by the Museo de Historia Natural Noel Kempff Mercado, Santa Cruz, Bolivia, as well as with up to three celluloid colour rings in unique colour combinations. Ringing was conducted during most months of the year, and intermittently from October 2004 to July 2007. Data recorded on captured birds included skull pneumatization (measured on a 7-point scale), subcutaneous fat content (on an 8-point scale), presence of juvenile plumage, unflattened wing chord, and body mass (Ralph et al. 1993; Pyle 1997). Mass was used as a metric of body size because it is pertinent to testing the Body Size hypothesis (i.e. that due to a lower surface area to volume ratio, larger individuals can withstand colder conditions better than smaller individuals). Although mass can fluctuate in migratory birds due to fat deposition, we are confident that body mass provides a good estimate of body size in our study population because we rarely encountered individuals with more than trace amounts of fat in the process of ringing during the breeding season. Wing chord (instead of tarsus length) was used as a structural parameter independent of nutrient reserves because the tarsus is very short in kingbirds, such that there is more expected variation in wing length than tarsus length.

Kingbirds were sexed using molecular markers (Fridolfsson & Ellegren 1999) and primary feather notch length (Pyle 1997), although the latter was not useful for juveniles. One hundred and sixty individuals were males, 144 were females and 145 could not be sexed with confidence. Kingbirds were aged from skull pneumatization, plumage, and ringing records. For the purpose of ageing, each year began on 1 September, just prior to the start of the kingbird breeding season at CBS (i.e. September–January). We defined kingbirds in their first year of life as hatch-year individuals (HY), and individuals older than this as adults. Because nestlings were obviously still growing when captured, and because we did not have data for some HY individuals, in analyses we assigned them average body mass and wing chord measurements of juveniles of the same sex. Table 1 provides definitions and abbreviations of measurements.

Table 1.   Definitions of model variables

Model notation	Description
Age	Hatch year (HY, an individual in its first year of life; year begins on 1 September of every year) and adult (individual that is not an HY).
Capture	Effect of capture on survival. Different survival for the first month after capture of non-nestlings, which allows for proportion of transients (kingbirds passing through, never to show up again) to be modelled.
Mass	Kingbird body mass (grams). Ten post-fledging kingbirds, for which we did not have body mass data, were assigned the average body mass of kingbirds in the population of the same age and sex. This was also done for nestlings (using the average body mass from juveniles in the population of the same sex), since nestling body mass does not reflect body mass as an HY individual in the fall.
Nestling	Two categories: nestling (recently hatched birds marked while still in the nest) and older. The nestling effect only applies to the first month after capture for survival, but applies indefinitely for spring migration (to account for philopatry).
Season	Two categories: winter is May–September, and non-winter (including migration periods) is October–April.
Sex	Three categories: male, female, unknown.
Wing	Wing chord measured on an unflattened wing. One hundred and six post-fledging kingbirds, for which we did not have wing chord data, were assigned the average wing chord of kingbirds in the population of the same age and sex. This was also done for nestlings (using the average wing chord from juveniles in the population of the same sex), since nestling wing chord does not reflect wing chord as an HY individual in the fall.
Year	A 12-month period, used as a categorical variable for modelling purposes.
Zero	No migration.

To determine presence/absence of individually-marked kingbirds, we divided the ∼700 ha study site into 23 plots and systematically searched for ringed kingbirds approximately every 2 weeks (i.e. ‘secondary sampling occasions’, see below) from February 2005–August 2007, except June–September of 2005. Ringed kingbirds were also searched for on most plots from 28 January–12 February, 2–19 March, and 15–27 June, 2008. When a colour-ringed individual was observed, its location was geo-referenced using a GPS receiver (Garmin GPS 76, Olathe, KS, USA), noting the date, time and ring colours.

To determine dominance, intraspecific agonistic encounters were documented via focal observations on individuals, taking continuous data by following individuals and recording observations into digital voice recorders (Sony ICD-B16, Tokyo, Japan). In a typical agonistic encounter, a kingbird demonstrated agonistic behaviour to another kingbird by using a crouch posture (sensuSmith 1966) while emitting a characteristic, sharp, repetitive call, or did not crouch but emitted the agonistic call. A ‘winner’ in such an encounter was defined as the individual that displaced the other individual from its perch. A ‘loser’ was defined as the individual who was displaced. When colour-ringed kingbirds were observed in such encounters, the colour ring combination, location, time and date were noted.

Migration is often difficult to detect from mark–recapture data because failure to observe a particular individual does not allow one to conclude that it migrated; it could have died, dispersed, or remained at the site, undetected (Kendall, Nichols & Hines 1997; Kendall & Nichols 2000). To tease apart probabilities of survival, dispersal, detection and migration, we used a maximum-likelihood modelling approach. Specifically, we employed the robust design with Huggins closed captures (Pollock 1982; Kendall & Nichols 1995; Kendall, Pollock & Brownie 1995) using R package RMark (Laake & Rexstad 2008; R Development Core Team 2008) as an interface for program mark (White & Burnham 1999). To estimate detectability vs. temporary emigration (i.e. migration), the robust design relies on data collected when the population is assumed closed to immigration or emigration (i.e. secondary sampling occasions) within each primary sampling occasion. Between each primary occasion, the population is allowed to be open. For the purposes of this study, an encounter history was built for each colour-ringed kingbird in the population, using half-month intervals as secondary sampling occasions (n = 61 occasions) and 1-month intervals as primary sampling occasions (n = 30 occasions).

To model migration, the robust design estimates two independent movement parameters: γ’, the probability of being unavailable for subsequent capture given that the individual was unavailable for initial capture (i.e. staying off the study site), and γ”, the probability of being unavailable for subsequent capture given that the individual was available for initial capture (i.e. moving off the study site) (Kendall & Nichols 1995; Cooch & White 2007). Fall migration (i.e. temporary emigration of some individuals from the breeding grounds) is defined as γ_t”, the probability of an individual migrating from the site in month t in fall (hereafter, ‘fall migration’, when partial migration would occur), given that it was on the site in month t-1. The probability of returning to the site in month t in the spring (hereafter, ‘spring migration’), given that it was away from the site in month t-1 is defined as 1-γ_t’. The fall migration season was defined as late January – early May (γ_t” fixed at 0 for all other months), the spring migration season was defined as October – November (γ_t’ fixed at 1 for all other months), winter as late May – September, and summer as December – early January. Model selection and inference were conducted using second-order Akaike’s Information Criterion (AIC_c; Anderson, Burnham & White 1994; Burnham & Anderson 2002).

To reduce the number of possible models, the effects of covariates on parameters other than fall migration (hereafter referred to as non-focus parameters) were modelled first. A general model for fall migration (γ”_{age + sex + mass}) was used to sequentially test the effects of covariates on the following non-focus parameters: (i) survival (fixed, age, capture, and by season), including the effect of migratory kingbirds not breeding at the site (i.e. transients; these are known to exist, Jahn et al. 2010b); (ii) detectability (fixed, by season, and by sampling effort); and (iii) spring migration (fixed, by year, and by age). To keep the number of models manageable, only models within two AIC_c units of the top model were included in the next stage of the analysis, and any models that did not estimate all parameters were discarded.

The best survival model, which was used in all subsequent models, was an additive effect of nestling, capture and season (Table 1). The best detectability model, which was also used in all subsequent models, was an additive effect of sampling effort (person-hours spent searching for colour-ringed individuals) and season. The two best spring migration models, which were included in subsequent modelling, had an effect of returns by ringed nestlings and an effect of nestling returns with an additive year effect. These models for the non-focus parameters were used in all modelling of fall migration (γ”).

Because the study period spanned four migration seasons, a year effect on the probability of partial migration in the population was also included and tested. Since strong support was found for an effect of year, an additive effect of year was included in all subsequent models of fall migration. The presence of partial migration was tested by comparing the fit of models with or without fall and spring migration. We used age as a proxy for dominance in the population (see below), and evaluated the Dominance hypothesis by testing the singular, additive and interactive effects of age and sex on fall migration. The Body Size hypothesis was tested using the singular, interactive and additive effects of body mass, wing chord, age and sex on fall migration. Competing models were defined as those within ten AIC_c units of the top model (Burnham & Anderson 2002).

To check whether kingbirds that disappeared from the study site in the fall were not migrating but simply moving to nearby areas or different habitats, we placed radio-transmitters on kingbirds and searched for them within a 12-km radius during the 2007 fall migration and subsequent winter (Jahn et al. 2010a). We found no evidence of local movement; individuals either stayed on site (i.e. within 1 km from the study site plots) or disappeared (A.E. Jahn, unpublished data). To ensure that migration was not confounded with permanent dispersal, spring migration for non-nestlings was fixed at 1 for a model (i.e. the model assumed that birds returned the following spring if alive) and nestlings were allowed to return or not (because nestlings in the original modelling had a lower probability of returning in the spring). This model had very similar estimates of fall migration as our original spring migration model, suggesting that few birds recorded as migrants were permanently dispersing.

In case the robust design assumption of closure (i.e. no additions or removals from the population) between secondary sampling occasions was violated (potentially biassing estimates of movement if movement happened at a similar temporal scale as sampling; Kendall 1999), we attempted to estimate migration rates using multistratum models (Brownie et al. 1993). This permitted a relaxation of the assumption of closure within months. However, mark was not able to estimate all parameters in these models. Nevertheless, preliminary results from multistratum models were generally consistent with results of the robust design, such that we have confidence in the detected differences in migration between groups.

Results

Of 449 ringed kingbirds, 210 were re-sighted at least once during the study period. One hundred and thirty-eight kingbirds were ringed as HY individuals, 50 of which were re-sighted at least once in a subsequent year (i.e. after 1 September). Sixty-one kingbirds were ringed as nestlings. Kingbirds ringed as adults were re-sighted an average of 2·2 times (3·9 SD; range = 1–26).

Unambiguous support was found for partial migration (vs. no migration) in the population (Table 2). Eight models failed to estimate all parameters and were discarded; these included models in which sex, age, and mass were fully interactive. Model selection identified eight competing models (Table 3). Of these, the best model was an additive effect of age and mass, and an interaction of each of these with sex. The only difference between the top model and the second highest model was an additive effect of year on spring migration (Table 3). These two models were by far the strongest, as reflected by ΔAIC_c weight, accounting for approximately 60% of the combined weight of all models (Table 3). These models describe a weak relationship between body mass and migration for HY kingbirds of both sexes, for adult females and for kingbirds of unknown sex (Fig. 2). The most notable patterns are: (i) a striking increase in the probability of migrating with larger body mass in adult males, with the heaviest males having the highest probability of migrating (almost 80%) relative to all other demographic classes; and (ii) a higher migration probability by HY females relative to other males or females, except the heaviest males (Fig. 2). To ensure that HY individuals assigned an average body mass for their age and sex class were not driving the patterns found in the best model, we excluded those HY individuals from the model. Doing so did not alter the conclusion that body mass influences migration probability in males but not in females.

Table 2.   Model comparison results for probability of partial migration in the population, using second-order Akaike’s Information Criterion (AIC_c). Variables listed are: number of parameters (K), Akaike differences (ΔAIC_c), and the Akaike weight (w_i). Model effects are: γ” = fall migration, γ’ = spring migration, γ”_zero = no fall migration, γ’_zero = no spring migration. For all models, survival was modelled with an additive effect of capture, nestling and season. Detectability was modelled with an additive effect of season and sampling effort. See Table 1 for description of variables

Model	K	ΔAICc	wi
γ”year,γ’nestling	13	0·00	0·666
γ”year,γ’nestling + year	15	1·38	0·334
γ”zero, γ’zero	7	147·38	0·000

Table 3.   Model selection results for the best models (i.e. ΔAIC_c < 10) of γ”, fall migration (γ’ represents spring migration). Variables listed are: number of parameters (K), Akaike differences (ΔAIC_c), and the Akaike weight (w_i). For all models, survival was modelled with an additive effect of capture, nestling and season. Detectability was modelled with an additive effect of season and sampling effort. See Table 1 for description of variables

Model	K	ΔAICc	wi
γ”sex × (age + mass) + year,γ’nestling + year	23	0·00	0·306
γ”sex × (age + mass) + year,γ’nestling	21	0·07	0·296
γ”age × sex + mass + year,γ’nestling	19	1·65	0·134
γ”age × sex + year,γ’nestling	18	1·73	0·129
γ”age × sex + year,γ’nestling + year	20	2·42	0·091
γ”sex × mass + year,γ’nestling	18	5·64	0·018
γ”sex × mass + year,γ’nestling + year	20	6·16	0·014
γ”sex × wing + year,γ’nestling	18	9·53	0·003

Figure 2.  Monthly probability of fall migration (γ”) for 2007, which is representative of all years of the study, as a function of mass, age and sex (‘unknown’ represents individuals of unknown sex) in a population of Tropical Kingbirds. The migration probabilities are from the minimum (best) Akaike’s Information Criterion model (γ”_{sex × (age + mass) + year,}γ’_{nestling + year}). Error bars represent standard error. See Table 1 for definitions of terms.

The third highest model is very similar to the two top models, except that sex is interactive with age, not with mass. In the fourth and fifth highest models – the only models without a body size component – females tend to be more likely to migrate in their first year of life (HY) than as adults, while the reverse is true in males (Fig. 3). In the sixth and seventh highest models, larger males and smaller females have a higher probability of migrating. In the lowest model, males with longer wings and females with shorter wings have higher migration probabilities.

Figure 3.  Monthly probability of fall migration (γ”) as a function of age and sex in a population of Tropical Kingbirds, by year from the highest Akaike’s Information Criterion model that does not incorporate body mass (γ”_{age × sex + year,}γ’_nestling). Individuals of unknown sex are not shown but had the same pattern as that of females. Error bars represent standard error. See Table 1 for definitions of terms.

Dominance hypothesis

We were unable to assign a dominance score (number of wins divided by the total number of interactions; Poisbleau, Jenouvrier & Fritz 2006) to individual kingbirds because most ringed kingbirds that were observed in agonistic encounters with other kingbirds were seen in only one encounter. However, numerous lines of evidence strongly suggest that adults are dominant over HY kingbirds. First, HY individuals were never observed to win an agonistic encounter with other kingbirds. Second, adults were significantly more likely to win (n = 17) than lose (n = 6) in intraspecific encounters (χ² = 5·261, P = 0·023). We do not know the identity of the opponents of these kingbirds, since they were usually not ringed; presumably some were HY individuals. Third, adults of both sexes are significantly larger than HY individuals (Fig. 4; see below), giving them a size advantage in an encounter with an HY individual. Finally, previous studies of intraspecific dominance in passerines have generally found older birds to be dominant over younger individuals (Arcese & Smith 1985; Zink & Watt 1987; Piper & Wiley 1989; Keys & Rothstein 1991). We therefore use age as a proxy for dominance, with older kingbirds dominant over younger kingbirds.

Figure 4.  Boxplot of body mass of Tropical Kingbirds from the study population, by age and sex, excluding nestlings. Numbers in parentheses on x-axis labels represent sample size. Circles represent outliers and stars represent extremes. See Table 1 for definitions of terms.

The Dominance hypothesis was not strongly supported. The top three models indicated that the largest males, which are also generally older and therefore dominant over smaller males (Fig. 4), had the highest probability of migrating (Table 3 and Fig. 2). However, the top three models indicated that younger females (Fig. 4) had a consistently higher migration probability than older females (Table 3 and Fig. 2), as predicted. Likewise, the fourth-highest model indicated that younger females had a notably higher migration probability than older females (Table 3 and Fig. 3). Thus, effects of dominance on probability of migration appear to apply only to females; dominance status was not a consistent predictor of partial migration.

Body Size hypothesis

Adult males (n = 129) had significantly greater body mass than adult females (n = 118; t = 3·047; P = 0·003) and HY males (n = 15; t = 9·890; P < 0·0001), and adult females had significantly greater body mass than HY females (n = 11; t = 8·032; P < 0·0001; Fig. 4). Overall, adult males had the greatest body mass in the population (Fig. 4). Adult males (n = 105) also had significantly longer wings than adult females (n = 83; t = 12·945; P < 0·0001) and HY males (n = 10; t = 5·773; P < 0·0001), and adult females had significantly longer wings than HY females (n = 11; t = 4·790; P < 0·0001).

Given that in the strongest models, males with the highest body mass had the highest migration probability (Table 3 and Fig. 2), we found no general support for the prediction that individuals of the smaller sex are more likely to migrate than individuals of the larger sex. Additionally, the two best models indicated a general increase in migration probability for heavier individuals of both sexes, which contradict the prediction that within a sex, smaller individuals are more likely to migrate than larger individuals. The only support for this prediction and the Body Size hypothesis was that in the two top models, younger (smaller) females had an overall higher migration probability than older females (Fig. 2).

Discussion

These results provide unambiguous support for partial migration of kingbirds in Bolivia, but provide only limited support for the Dominance and Body Size hypotheses. In females, smaller, less dominant HY individuals were more likely to migrate than adults. In males, however, the opposite pattern was observed – older, larger, more dominant individuals were more likely to migrate than HY males. More generally, adult males were more likely to migrate than any other group.

Previous research has also failed to support the Dominance and Body Size hypotheses. Rogers et al. (1989) found little support for the Dominance hypothesis in Dark-eyed Juncos Junco hyemalis because individuals that migrated further in fall were not necessarily subdominant to those that overwintered closer to the breeding grounds. Likewise, in Costa Rica, Boyle (2008) found that older male White-ruffed Manakins Corapipo altera were more likely to migrate than younger, likely subdominant males. Studying differential migration of Teal Anas crecca, Guillemain et al. (2009) failed to find support for the Body Size hypothesis.

Although the destination of migratory kingbirds at CBS remains unknown, it is most likely deeper into the Amazon Basin. Several lines of evidence support this supposition. First, the population of kingbirds declines significantly at CBS in the non-breeding season (Jahn 2009), while north of the study site near Manaus, Brazil, abundance of kingbirds in the same season almost triples (A.E. Jahn, unpublished data). Second, kingbirds that breed further south of the study site likely migrate into the Amazon Basin, north of the study site, to spend the non-breeding season (Jahn et al. 2010b). Third, rainfall is highest in north-central South America during the southern Amazon Basin’s dry season (Garreaud et al. 2009), and the abundance of the kingbird’s insect prey is positively related to rainfall (Jahn 2009). Indeed, central and northern regions of the Amazon Basin are less seasonal in terms of rainfall than southern regions (Espinoza et al. 2009).

An implicit assumption of all theories on partial migration is that the non-breeding season represents a period of limited resources and/or harsh climatic conditions, such that some individuals of a population must temporarily emigrate (e.g. Gauthreaux 1978, 1982; Lundberg 1985; Smith & Nilsson 1987; Boyle 2008). At issue is why adverse environmental conditions affect individuals differently. At our site, there is a dramatic decline in the abundance of kingbird prey in the non-breeding (dry) season relative to the breeding season (Jahn 2009). The abundance of kingbirds’ most important prey type, coleopterans, was more than twice as high in the wet season than in the dry season, a pattern also reflected in other prey types (Jahn 2009). This precipitous decline in food availability in the dry/non-breeding season is noteworthy because food is generally a limiting resource for birds (Lack 1954; Martin 1987), especially in the non-breeding season (Alerstam & Hogstedt 1982; Lovette & Holmes 1995; Brown & Sherry 2006). Kingbirds at our study site did not switch diets to non-arthropod prey during the non-breeding season; aerial sallies for arthropod prey (Fitzpatrick 1980; Cintra 1997; Gabriel & Pizo 2005) were by far the most common foraging manoeuvre (i.e. >85% of foraging manoeuvres, A.E. Jahn, unpublished data) throughout the year.

We suggest two reasons that previous studies have not found that larger individuals are significantly more likely to migrate from partially migratory populations. First, most research on partial migration has focused on granivorous (e.g. Song Sparrow, Melospiza melodia; Nice 1937; House Finch, C. mexicanus; Belthoff & Gauthreaux 1991) or omnivorous species (e.g. European Robin, Erithacus rubecola; Biebach 1983; Blackbird, Turdus merula; Lundberg 1985; Blue Tit, Cyanistes caeruleus; Smith & Nilsson 1987; Nilsson et al. 2006; Nilsson 2007). During the winter, these species often forage in flocks on resources that are patchily distributed; access to those resources within a flock is generally determined by competitive ability (Lundberg 1985; Smith & Nilsson 1987). Thus, large, dominant individuals in these or similar species would acquire more food (Piper 1997; Leary, Sullivan & Hillgarth 1999) and subordinate individuals would be forced to migrate away (Lundberg 1985; Smith & Nilsson 1987).

In contrast, Tropical Kingbirds never forage in flocks (except rarely for fruit; A.E. Jahn, personal observation), and likely do not deplete the insect food resources around the perches from which they forage (Fitzpatrick 1981). Thus, a kingbird’s ability to compete directly with conspecifics for available food during the non-breeding season may be relatively unimportant. Effects of dominance on partial migration in this species may instead be driven by competition for space or perches for foraging. If the abundance of flying insects becomes too low to sustain large-bodied adult males during the non-breeding season, they will likely migrate away, leaving behind smaller individuals with lower energetic requirements. In contrast, because seeds are generally available during winter at temperate latitudes of North America and Europe, large individuals of granivorous and omnivorous species can access them if they can successfully compete with flock members.

Second, winter temperatures at north temperate latitudes, where most studies on partial migration have occurred, are generally much lower than at our tropical study site at 15°S. In north temperate latitudes, variation in temperature is a good predictor of migratory tendency, whereas in the milder and drier Southern Hemisphere, migration is more tightly correlated with variation in precipitation (Chesser 1994; Dingle 2008). Indeed, our study site is characterized by high variation in precipitation, with an average daily rainfall of 6·3 mm (SD ± 13·6) from mid-September to mid-February and 2·9 mm (± 9·3) during the rest of the year (data from October 2004 to August 2007). Variation in temperature at our site, on the other hand, is low. The average daily temperature from mid-September to mid-February (the wet season) is 27·5 °C (± 13·8), whereas the average daily temperature during the rest of the year is 26·4 °C (± 9·2). Minimum daily temperatures in the non-breeding season at our site range from 10 to 15 °C (A.E. Jahn, unpublished data), while at north temperate latitudes where most research on partial migration has been conducted (i.e. >35°N latitude), they are regularly well below 0 °C. The ability of a bird to remain at northern latitudes may primarily depend on its ability to withstand cold temperatures, in which case a large body size would be desirable (due to a lower ratio of surface area to volume), if the bird were able to maintain a high food intake rate. For an insectivorous kingbird that depends on flying insects for food and that inhabits an area of highly seasonal rainfall, the relatively mild temperatures and greatly reduced food availability during the non-breeding season means that the greater food demands associated with a large body size may outweigh the advantage of lower heat loss associated with a lower surface-to-volume ratio.

We propose a food limitation hypothesis: in regions where seasonality is defined by wet-dry cycles, a threshold of insect food availability exists, below which individuals with larger bodies and therefore higher total energetic demands must migrate to wetter regions to find sufficient food. This may be more often the case for species breeding in the Southern Hemisphere, where rainfall is generally a more important driver of migration than in the Northern Hemisphere (Dingle 2008). At CBS, we hypothesize that the largest male kingbirds are at a greater risk of starvation due to food limitation during the dry season than are smaller HY males and adult females. Smaller individuals have higher energetic demands than adults on a per-gram basis because their larger surface-to-volume ratio results in higher mass-specific metabolic rates (Calder 1974); however, their smaller size means their total energy demands are lower than those of larger conspecifics. Larger kingbirds at CBS are not likely to benefit from a low surface-to-volume ratio during the non-breeding season because, as previously mentioned, temperatures at that time are mild.

That the relationship between the probability of migration and age in females is the opposite of that in males suggests that different factors affect the decision to migrate in females and males. Reproductive strategies may primarily affect the female’s decision and if, as in European Robins, Erithacus rubecula, migratory female kingbirds run the risk of later pair formation than non-migratory females (Harper 1985), it may be beneficial for adult females not to migrate. For HY females, on the other hand, the potential costs in terms of competition with dominant conspecifics may outweigh any benefits of staying at the site during the dry season.

For HY male kingbirds, staying at the site during their first non-breeding season may offer a chance to familiarize themselves with potential territories and males with whom they may compete for mates in the breeding season. Indeed, site familiarity has been shown to be positively correlated with dominance in the White-throated Sparrow (Zonotrichia albicollis; Snell-Rood & Cristol 2005), and prior residency has been shown to increase the chances of being dominant in Dark-eyed Juncos at the beginning of the next breeding season (Cristol, Nolan & Ketterson 1990; Holberton, Hanano & Able 1990). Additionally, the smaller body size of an HY male relative to that of older males may permit it to survive at the site during the dry season, since its caloric requirements would be relatively low.

Large adult male kingbirds may enjoy the best of both worlds: by migrating away, they avoid a food-limited time of year at the site, and since they likely have already established their dominance, they may be able to re-assert their position in the social hierarchy upon returning the following breeding season. That dominance can be established during a brief period early in life and can play a major role determining an individual’s dominance status throughout its life has been shown in the White-throated Sparrow (Piper & Wiley 1989; Piper 1995). Subordinates recognize dominants, such that dominant individuals remain dominant (Wiley et al. 1999).

Similar to the results of Boyle’s (2008) work on partial migration of White-ruffed Manakins, we conclude that: (i) the decision by kingbirds to migrate is condition- and sex-dependent. Although migration in our study population could also be partially genetically controlled (e.g. European Robin, Biebach 1983), the possibility remains unexplored. (ii) Leading hypotheses only partially explain partial migration of kingbirds. Taken together, these studies from Central and South America make it clear that current theory on partial migration cannot be reliably generalized to other migratory systems.

A more general message from our study has less to do with the migratory behaviour of kingbirds than with differences in how animals respond to seasonality at temperate vs. tropical latitudes. Theories on the effects of environmental variation on organisms developed and tested at north temperate latitudes need independent evaluation elsewhere. Given the potential of climate change to differentially impact animal populations across latitudes (Deutsch et al. 2008), a geographically broader research perspective is fundamental to appreciate the challenges to movement and survival facing migratory species across the planet.

Acknowledgements

We are grateful to J. Rozenman and G. Weise for their hospitality and ongoing support of research at Caparú Biological Station. We thank E. Bruna, D. Green, S. Robinson, B. Sandercock, K. Sieving, D. Steadman and three anonymous reviewers for many helpful comments. A. Ozgul, K. Pollock and G. White provided advice on analytical procedures. Many people assisted with field and lab work, especially: A. Alcoba, J. Andrews, E. Chiang, J. Cocke, S. Estevez, B. Flores, F. Hilarion, R. Horn, J. Johnson, P. Justiniano, J. Ledezma, T. Mack, S. Ouly, J. Prather, S. Prospero, J. Rozencranz, A.M. Saavedra, M. Saldias, M. Simon, and J. Vidoz. Funding was provided by the American Ornithologists’ Union, the National Science Foundation (OISE-0313429, 0612025), Optics for the Tropics, School of Natural Resources and Environment – University of Florida, Southeast Alliance for Graduate Education and the Professoriate, Western Bird Banding Association, and the Wilson Ornithological Society. This research was conducted with permission of the Dirección General de Biodiversidad of Bolivia.