SEARCH

SEARCH BY CITATION

Keywords:

  • colonization;
  • colony size;
  • lifetime fitness;
  • natal dispersal distance;
  • survival

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

1. Obtaining empirical evidence of the consequences of dispersal distance on fitness is challenging in wild animals because long-term, unbiased data on reproduction, survival and movement are notoriously difficult to obtain.

2. Lifetime fitness correlates of natal dispersal distance were studied in an isolated population of the facultatively colonial lesser kestrel Falco naumanni (Fleischer) monitored during 8 years at north-eastern Spain, where most birds (83%) dispersed from their natal colony to settle at distances ranging from 112 m to 136·5 km.

3. Neither annual breeding success nor age at recruitment was affected by natal dispersal distance. However, a capture–mark–recapture analysis revealed that survival during the year following recruitment decreased exponentially with dispersal distance, with differences of up to 15% between philopatrics and long-distance dispersers. In subsequent years, it remained similar irrespective of the natal dispersal distance moved. These results did not seem to be biased by long-distance dispersers settling differentially in the periphery of the population (which could emigrate permanently and be considered dead in future occasions) or within-individual consistency in successive dispersal distances, so our results appear to reflect genuine survival differences between dispersal tactics.

4. Average lifetime fledgling production, average lifetime recruitment success and rate-sensitive individual fitness (λind) also decreased with the distance from the natal to the first-breeding colony, indicating that dispersal decisions early in life affecting immediate survival prospects may translate into long-term fitness costs.

5. Both survival and lifetime fitness models including continuous dispersal distances significantly improved the characterization of the effect on fitness compared with models considering dispersal as a discrete process (i.e. dispersal vs. philopatry at a colony level).

6. Long-distance dispersers were more likely to establish new colonies regardless of whether they recruited in the centre or the periphery of the population, revealing their important role in the colonization of unoccupied patches. Individuals experienced a higher probability of mortality in small and newly funded colonies, so lifetime fitness costs of dispersal seem to be explained by recruitment in sites where average quality is low because of high uncertainty in survival prospects.


Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Dispersal is a key parameter in ecological and evolutionary processes. Although the evolution of dispersal has received considerable attention from theoreticians (see reviews in Johnson & Gaines 1990; McPeek & Holt 1992; Clobert et al. 2001), empirical tests of the adaptive significance of this behaviour are comparatively rare. In vertebrates, most empirical approaches have focused on the proximate causes motivating philopatry and dispersal (reviewed in Clobert et al. 2001; Bowler & Benton 2005). Although these studies may be indicative of the selective pressures affecting dispersal, they offer little information on the fitness consequences of adopting different dispersal strategies. Further, most effort dedicated to analysing the correlates of dispersal has treated it as a discrete variable (philopatry vs. dispersal), but at least in some situations an assessment of the costs and benefits of this behaviour as functions of continuous dispersal distance may improve our understanding of both the demographic consequences of movement and the evolution of dispersal strategies (Baker & Rao 2004; Lowe 2010).

The evolutionary consequences and adaptive significance of dispersal in natural populations is usually studied in terms of fitness payoffs, which have two practical difficulties. First, quantifying dispersal itself poses substantial logistical problems (Koenig, van Vuren & Hooge 1996), so the relationship between any fitness component and dispersal may strongly depend on our ability to minimize biases derived from individuals dispersing beyond the limits of the study area (Bélichon, Clobert & Massot 1996; Cooper, Daniels & Walters 2008; Doligez & Pärt 2008). Second, detailed, long-term unbiased data on reproduction and survival are difficult to obtain for free-living animals, with the result that most studies have only concentrated on short-term fitness correlates or on a single fitness component (Bélichon, Clobert & Massot 1996; Doligez & Pärt 2008). For example, annual estimates of breeding success in birds are relatively easy to obtain by directly monitoring nests and broods. In contrast, timing of mortality is usually unknown for natural populations, and the problem of individuals emigrating outside of the study area not only biases dispersal patterns but also underestimates (and probably bias) survival probabilities and lifetime fitness (Baker, Nur & Geupel 1995; Johnston et al. 1997; Zimmerman, Gutiérrez & Lahaye 2007). This problem is exacerbated when individuals show lifetime consistency in dispersal propensity, because if long-distance dispersers tend to disperse a long distance again, they may have a differential probability of emigrating permanently from the study area and be considered dead (Doligez & Pärt 2008). As fitness compensations between fecundity and survival have been reported (Bélichon, Clobert & Massot 1996), accurate measures of fitness should include an analysis of both annual fertility rates and number of reproductive seasons during an individual’s lifetime. The combination of both components can be summarized into the number of fledglings and recruits an individual produces along its life, which are generally assumed to be good proxies of fitness in natural populations (e.g. Newton 1989). However, lifetime fitness estimates such as counts of fledglings or recruits have been criticized because they ignore timing of reproduction. The rate-sensitive estimate of ‘individual fitness’ derived by McGraw & Caswell (1996) solves this problem by incorporating both the timing and number of offspring production, and has been said to be a more accurate estimate of fitness in expanding populations (Brommer, Merila & Kokko 2002).

In most animals in which dispersal is an active behaviour, the frequency distributions of movements are highly leptokurtic, i.e. have a positive kurtosis with many individuals remaining close to the point of origin and a long tail reflecting relatively few individuals that move long distances (Shields 1982; Johnson & Gaines 1990; Paradis et al. 1998). This pattern is widespread even in long-distance migratory birds, generating the testable prediction that philopatry, i.e. settling in or close to the natal area, is in general adaptive. Large-scale dispersal should be selected against because it preserves co-adapted genomes (Shields 1982) and involves a loss of familiarity with resources, predators and conspecifics (e.g. Clobert et al. 2001). Conversely, however, theory also predicts that dispersal may be advantageous for avoiding competition among kin (Hamilton & May 1977), prevent inbreeding (Greenwood 1980), and escaping from degrading or poor-quality environments (Travis & Dytham 1999). Multiple selective opposing pressures may indeed affect dispersal behaviours at different spatio-temporal scales, the balance of which should theoretically select for prevailing dispersal strategies (Dobson & Jones 1985; Clobert et al. 2001). To date, most studies investigating lifetime fitness correlates of dispersal in the wild have found that dispersal is under negative selection (Verhulst & van Eck 1996; Bensch et al. 1998; Wheelwright & Mauck 1998; Forero, Donázar & Hiraldo 2002; Hansson, Bensch & Hasselquist 2004; Pasinelli, Schiegg & Walters 2004; Robbins & Robbins 2005; Pärn et al. 2009) although an opposite pattern (Nilsson 1989; Wauters, Matthysen & Dhondt 1994; MacColl & Hatchwell 2004) or no differences between dispersal tactics (Marr, Keller & Arcese 2002) have also been reported. Unfortunately, potential biases associated with fitness estimates are frequently ignored, and hence, differences in fitness are usually open to alternative interpretations (Doligez & Pärt 2008).

Here we examined a large data set on demography and dispersal in a population of a colonial bird, the lesser kestrel Falco naumanni, to investigate the fitness correlates of continuous natal dispersal distances. Our studied population at the Ebro valley, north-eastern Spain, constitutes one of the rare cases in which dispersal of individuals beyond the limits of the study area is negligible (see Serrano et al. 2001, 2003, 2004, 2005; Serrano & Tella 2003; Serrano, Carrete & Tella 2008). Briefly, we intensively monitored a discrete, geographically isolated population occupying a large area (10 000 km2) and encompassing colonies separated by maximum distances (210 km) exceeding by far the maximum natal dispersal distances recorded in this population (136 km, Serrano et al. 2003). Only one case of natal dispersal beyond the limits of the study area was detected in spite of intensive monitoring of birds by other research teams in the neighbouring populations of Spain and France (M. Alberdi, C. Gutiérrez, and L. Brun, pers. comm.). This, besides detailed individual-based demographic information, alleviates the problem of birds dispersing outside the study area and permits calculating reliable fitness estimates, making this population ideal to study the relationship between lifetime performance and dispersal. We concentrated on natal dispersal, defined as the movement of an individual from its birthplace to the place where it settles to breed for the first time, because it usually involves the longest movements of individuals along their lives, and in most species determines the general area in which they will live during their reproductive careers (e.g. Greenwood 1980). As dispersal distances are strongly right-skewed in our study model (Serrano et al. 2003), and individuals settled in or close to their natal colony more frequently than expected under a random settlement scenario (Serrano, Carrete & Tella 2008), we hypothesize that dispersing long distances should be under negative selection. Male lesser kestrels tend to disperse less and shorter distances than females (Serrano et al. 2003), which is a common pattern in birds (Greenwood 1980), and thus, we expect a stronger negative selection for long-dispersal distances in males. We concentrated on the fitness correlates of dispersal distance after birds settled for the first time as breeders because natal dispersal distance is, by definition, unknown before recruitment. Given that fitness depends on both survivorship and reproductive output, we first examined the relationship between dispersal and both fitness components separately. Then, we tested whether the balance between both fitness components agrees with three robust metrics of lifetime fitness in birds: lifetime fledgling production (LFP), lifetime recruitment success (LRS) and rate-sensitive individual fitness (λind).

Materials and methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Study species and study population

Lesser kestrels are small (c. 150 g), hole-nesting threatened falcons that prey mostly on invertebrates. They are migratory birds, and both sexes are easily distinguishable by plumage features in the field. At arrival from the wintering quarters, males choose a nest-hole and defend it vigorously while displaying to attract a female, but are otherwise nonterritorial. Males feed females before laying, which occurs in early May. Lesser kestrels typically recruit into the breeding population when 1 or 2 years old and once they start breeding they attempt to reproduce every year. They are predominantly monogamous and single-brooded falcons.

The study site is located at the Ebro valley, north-eastern Spain, and covered c. 10 000 km2. There, an isolated population of lesser kestrels has been the subject of intensive monitoring from 1993 to 2000. The kestrels breed under small tiled roofs (c. 50 m2) of mostly separated and abandoned farmhouses surrounded by traditionally dry-farmed cereal crops (Jovani et al. 2008). During the study period, these buildings attracted both solitary pairs and colonies of 2–43 pairs. The number of colonies in the study area has increased from 52 in 1993 to 181 in 2000 (see Jovani et al. 2008 for details on the population growth process).

Most lesser kestrels in the study area (83%) dispersed from their natal colony to settle as first-time breeders in another colony (median dispersal distance = 7225 m, mean = 15442·3 m, range 112–136 500 m, N = 751 individuals, Serrano et al. 2003), although they tended to settle in the surroundings of their birth site (Serrano, Carrete & Tella 2008). Once established as breeders, 72% of lesser kestrels showed fidelity to their previous year breeding colonies, and the birds that changed colony between consecutive years usually moved very short distances (median dispersal distance = 1600 m, mean = 4030·4 m, range = 100–65 220 m, N = 486 subsequent breeding attempts) compared with the size of the study area (Serrano et al. 2001).

Field procedures

From 1993 to 1999, 4901 fledglings captured in their nests were marked with a numbered metal ring and a plastic colour ring engraved with a unique two-digit alphanumeric code that could be read with spotting scopes. We estimate that more than 90% of the fledglings were marked annually. Each year, regular surveys were carried out to locate buildings occupied by kestrels, both in previously known subpopulations and in appropriate areas where the species was not breeding in previous years. Once a single pair or a colony was located, we proceeded to identify marked birds, and to assign all observed individuals to their nests, which were mapped on detailed schemes of the roofs. Intensive observations of marked kestrels were mostly made during the prelaying and laying periods (March-middle May), but capture–recapture histories were also completed during late May and June when we surveyed the colonies directly to confirm reproduction, record breeding parameters and capture brooding adults. Some birds visited two or more colonies along the mating period before finally settling as first breeders (Serrano & Tella 2007). We thus defined as first breeders those birds for which their first reproduction attempt was confirmed and their natal dispersal distance as the straight-line distance from the natal colony to that where they definitively settled. Colony size was defined as the final number of established pairs initiating the reproduction. Further details about field procedures can be found in the study by Serrano & Tella (2003, 2007) and Serrano et al. (2005).

Measuring dispersal and fitness

Dispersal was defined in two ways: (i) Euclidean distance between an individual’s natal colony and its first breeding colony, assigning zero values to birds which were philopatric at the colony scale and (ii) dispersal status as a discrete event (philopatry or dispersal at a colony scale).

From 1994 to 2000, we obtained data of dispersal distance, colony and age of recruitment from 962 individuals. We also determined the fate and number of fledglings raised from 1308 nests for which the natal dispersal distance of at least one of the parents was known. To estimate survival rates, individual histories were constructed by considering that individuals were marked when they were reencountered (recaptured or resighted) for the first time as breeders. Because using this methodology survival probabilities cannot be estimated for birds recruited in the last year of study, live encounter histories from 750 individuals (years 1993–1999) were used to analyse the relationship between this parameter and natal dispersal distance.

For the analyses of LFP, LRS, and individual fitness λind, we only used birds with known reproductive success for every year during their reproductive careers. We only considered individuals not seen during at least 2 years prior to the end of this study (year 2000), which had a very high probability of being dead (probability of resighting at least once a live individual over 2 years was 0·97 for males and 0·95 for females, respectively, see Results). Following these criteria, data on LFP (the sum of all fledglings raised by an individual during its lifetime) were available for 359 individuals, while LRS (number of breeding offspring) could be calculated for 340 individuals, because a few offspring were marked only with one metal ring, and thus their potential recruitment could have gone unnoticed. Individual fitness λind was calculated for 359 individuals by constructing individual projection matrices with annual breeding success and survival (McGraw & Caswell 1996). The individual fitness of each bird was estimated as the dominant eigenvalue of its matrix. As demographic parameters can be affected by large-scale geographical variations in the environment, four major, geographically discrete subpopulations were considered in all statistical models as group effects: (i) West Ebro Valley, (ii) Bujaraloz, (iii) Ventas and (iv) South Ebro Valley.

Statistical analyses

Fecundity, age at first breeding and phenotypic traits

We used a Poisson generalized linear mixed model with a log-link function to test whether dispersal distance was a good predictor of the number of fledglings produced annually. In this model, dispersal distance, sex and individual age together with all first-order interactions were fitted as fixed effects, while individual identity, year and subpopulation of recruitment were fitted as random terms. The natural logarithm of age was fitted in these models to account for nonlinear relationships. This initial model was overparameterized and had problems of convergence. A comparison of the full fixed-effects model with and without the subpopulation random term yielded a nonsignificant effect of the random effect (χ2 = 0·03, d.f. = 1, P = 0·86), so we removed it from the initial model. This new model had not further problems and was used as the starting model.

To examine age at first breeding in relation to dispersal distance, a Poisson generalized linear mixed model with a log-link function was also used. Natal dispersal distance and sex were included as fixed effects, while cohort and subpopulation of recruitment were fitted as random terms. Data on wing length (a good indicator of body size in this species) and body mass at recruitment were used to determine whether dispersal distance was associated with phenotypic traits, which could be related to individual competitive abilities at first settlement (see Serrano & Tella 2007). Natal phenotypic traits and conditions (i.e. nest, hatching date, brood hierarchy, and body condition at fledging) were not considered because they were previously shown to be unrelated to natal dispersal (Serrano et al. 2003). As data on dispersal distance are highly skewed, we fitted a negative binomial mixed model in which wing length, body mass, sex and time elapsed because individual laying date (which is related to changes in body mass) was included as independent fixed effect (see also Serrano & Tella 2007), and subpopulation and year of recruitment as random terms.

Data on lifetime breeding fecundity are also typically highly skewed, with most individuals producing zero or few fledglings and recruits. Therefore, we also used a negative binomial model with a log-link function to determine whether dispersal distance correlates with LFP and LRS. Sex and its interaction with dispersal distance were also included as independent fixed effects, and subpopulation identity was included as a random term. The relationship between individual fitness λind and natal dispersal distance was analysed by employing the same model structure with a normal distribution of errors and the identity link function, previous transformation logeind + 0·5).

Models were implemented in sas v.9.2 (Cary, NC, USA). Model selection was based on Wald’s F significance tests for fixed effects. We started from full models including all main effects and first-order interactions and sequentially removed all nonsignificant fixed terms beginning with the interactions. Random terms were retained in all analyses. The extra-dispersion parameter was examined to assess the goodness-of-fit of each model (range 0·58–1·64) and corrected when necessary. In addition to linear effects, nonlinear relationships between fitness estimates and dispersal distance were tested by incorporating quadratic and logarithmic effects in the models. Moreover, generalized additive models (GAMs, Hastie & Tibshirani 1990) were fitted to the data to explore more flexible nonlinear relationships between dispersal and fitness. A cubic smoothing spline with four degrees of freedom was used.

Survival probabilities

Apparent survival (φ) was modelled following capture–mark–recapture basic methods for open populations, in which return rates were corrected for recapture (p) probabilities (Lebreton et al. 1992). First, program U-Care (Choquet et al. 2005) was used to test the goodness-of-fit of our global model. Program MARK (White & Burnham 1999) was then used to select the models and estimate the parameters on a logit scale. Our general starting model was the time-dependent Cormack–Jolly–Seber (CJS) model (Lebreton et al. 1992) with sex and subpopulation differences in both apparent survival and encounter probabilities. Following Lebreton et al. (1992), this starting model was denoted φt*s*subpt*s*sub, where subscripts t, s and sub denoted time-, sex- and subpopulation-specific effects, respectively. Less parameterized and biologically reasonable versions of the general model, including additive effects, were constructed and compared using the Akaike Information Criterion adjusted for small sample sizes (AICc), and derived AICc weights (Burnham & Anderson 2002). Apart from time, subpopulation, and sex effects, dispersal distance was included in the structure of survival models, and recapture was additionally constrained to vary with the distance from the colony of recruitment to the population centroid (calculated in the year of recruitment). Dispersal distance and distance to the population centroid were treated as individual covariates in MARK, and both linear and logarithmic effects were tested. Models differing in <2 AICc points were considered equivalent (Lebreton et al. 1992).

Testing potential biases in fitness

Long-distance dispersers could be settling differentially in the borders of the study area, with the subsequent risk of leaving it definitively and thus being considered dead (Doligez & Pärt 2008). We tested this possibility by regressing dispersal distance against distance from the colony of recruitment to the population centroid in the year of recruitment with a Gaussian mixed model in which year of recruitment was introduced as a random term. In addition, dispersal distances and distances to the centroid were treated as individual covariates in the structure of recapture models to assess and account for spatial variability in sampling effort (see Survival probabilities above). Even though long-distance dispersers were not settling differentially in the borders of the study area, true survival could be underestimated if these individuals had a high probability of dispersing a long distance again, that is, if there was individual consistency in dispersal distance. Within-individual nonrandom dispersal was tested by correlating natal and first breeding dispersal distance with a negative binomial model.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Annual productivity and age at first breeding

Annual production of fledglings did not vary with natal dispersal distance (F1,430 = 0·35, P = 0·55, Table 1) or its interaction with sex (F1,430 = 0·08, P = 0·77) or loge(age) (F1,430 = 0·41, P = 0·52). This variable neither differed between sexes (F1,430 = 2·54, P = 0·11), although there was a strong positive relationship between individual loge(age) and annual breeding performance (Estimate ± SE = 0·51 ± 0·08, F1,430 = 68·97, P < 0·0001).

Table 1.   Effect sizes of the relationship with natal dispersal distance of variables that showed a nonsignificant statistical effect for the population of lesser kestrels of the Ebro valley
 EstimateSEMalesFemales
MeanSDNMeanSDN
  1. The estimate and the standard error of each effect are shown, together with the mean, the standard deviation and the sample size of each variable.

Annual fledgling production2·09E–062·56E−061·491·615841·581·59724
Age at first breeding7·16E–071·81E−061·880·724421·430·63520
Body size−0·00330·0209522919·5322327·6160
Body mass0·012030·02064231·77·328231·66·896
First breeding dispersal distance0·0180·01151·483·05763·077·98125

After controlling for cohort and subpopulation of recruitment effects, age at first breeding was not correlated with dispersal distance (F1,940 = 0·74, P = 0·39) or its interaction with sex (F1,940 = 0·11, P = 0·74), although males recruited as breeders later than females (Estimate ± SE for males = 0·23 ± 0·06, F1,940 = 14·47, P = 0·0002).

Analysis of the number of fledgling only in the year of recruitment yielded qualitatively identical results (results not shown).

Survival

Our survival data fitted the general CJS model correctly (TEST2 + TEST3: χ2 = 43·497, d.f. = 69, P = 0·99). However, the directional Z-test for transience was significant (Z = 1·972, one-tailed P = 0·024), indicating that there were differences in the probability of being later reencountered between individuals resighted for the first time as breeders and birds with more breeding experience (Pradel et al. 1997). After suppressing the first encounter occasion, this effect was not longer significant (Z = 0·298, one-tailed P = 0·383), so we constructed models in which survival for the first time interval (denoted φ′ in model structure) was separately estimated (i.e. an ‘age’ effect, see Pradel et al. 1997). Further, the general model inline image fitted the data better than the same model with three ‘age’ classes (ΔAICc = 8·41), so model selection began with a (φ′, φ) parameterization.

The six top-ranked models (Table 2) indicated that recapture probabilities depended on the interaction between sex and the natural logarithm of natal dispersal distance for first breeders and was sex dependent for older birds. The best survival model (model 1, Table 2) indicated that φ′ was a function of the natural logarithm of dispersal distance, while φ varied between years. Alternatively, the statistically equivalent second ranked model (model 2) assumed an additive effect of sex and the natural logarithm of natal dispersal distance on φ′, and again a time-dependent effect on φ. According to AICc weights, models in which annual survival after the first reproductive attempt was a function of the natural logarithm of natal dispersal distance were 4·3 times more supported by data than models with dispersal distance untransformed (Σwi = 0·649 and 0·149, respectively), and 3·2 times than models unconstrained by any of the two individual covariates (Σwi = 0·201). Finally, models with dispersal status (philopatry vs. dispersal) received less support than models of continuous dispersal distance. The highest-ranked model considering dispersal as a discrete event (model 10, Table 2) was eight times less supported than the top-ranked one.

Table 2.   Models of annual survival in the year following recruitment (transient effect, φ′), and in subsequent years (φ) for lesser kestrels at the Ebro valley
ModelAICcΔAICcwiKDeviance
  1. Recapture rates (p) modelled separately for first breeders and more experienced breeders were indicated with subscripts fb and ad, respectively. Other subscripts denote effects modelled as constant (.), sex-specific (s), time-specific (t), different for philopatrics and dispersers at a colony scale (distatus) and constrained by natal dispersal distance (dis) or its natural logarithm (lnd). Symbol ‘*’ represents interaction between effects and symbol ‘+’ additive effects. AICc values, differences in AICc with respect to the top-ranked model (ΔAICc), model weight (wi), number of estimable parameters (K) and model deviance are also shown. Models are ranked by AICc values. Only the 15 top-ranked models from the 160 biologically sensible ones are shown.

 1inline image2163·8500·15279132137·5597
 2inline image2165·131·28130·08051142136·7961
 3inline image2166·232·37890·0465192148·0858
 4inline image2166·322·46710·0445132140·0268
 5φ′.φtpfb(s*lnd), ad(s)2166·782·9340·03523122142·5354
 6inline image2167·023·1740·03125152136·6406
 7inline image2167·713·86180·02216102147·5368
 8inline image2167·864·01430·02053142139·5291
 9inline image2168·014·15550·01913102147·8305
10inline image2168·064·20970·0186292149·9166
11inline image2168·124·2730·01804102147·948
12inline image2168·194·3370·01747102148·012
13inline image2168·234·37890·01711102148·0539
14inline image2168·244·38540·0170582152·121
15inline image2168·464·60710·01526152138·0737

A model averaging procedure of the two top-ranked models (i.e. those within two AICc points in Table 2) indicated that reencounter probabilities the year after the first breeding attempt depended on dispersal distances, but in a different way for males and females. The reencounter probability of males that recruited in their natal colony was 0·84 ± 0·06 (Estimate ± SE), but decreased to 0·80 ± 0·06 for birds dispersing 100 km. In females, however, reencounter probabilities were very similar (0·78 ± 0·03) in all the range of natal dispersal distances. In more experienced individuals, average reencounter estimates were 0·92 ± 0·04 and 0·79 ± 0·04 for males and females, respectively. Under these averaged models, survival probability in the year following recruitment decreased linearly with the natural logarithm of dispersal distance and was slightly lower for males than for females (Fig. 1). Probability of a bird surviving 1 year immediately after recruitment was around 0·70 ± 0·05 for birds settling to breed in their natal colony, while for birds dispersing 100 km, this value decreased to 0·54 ± 0·04 for males and 0·56 ± 0·04 for females (Fig. 1). Adult survival probabilities after the first occasion varied annually between 0·53 ± 0·05 and 0·78 ± 0·09.

image

Figure 1.  Relationship between natal dispersal distance and apparent survival probabilities of lesser kestrels during the year following recruitment, as resulted from model averaging over the two top-ranked models (Table 2). Mean values and 95% confidence intervals are shown for males (solid lines) and females (dashed lines).

Download figure to PowerPoint

Lifetime performance and individual fitness

Lifetime fledgling production in lesser kestrels decreased with increasing natal dispersal distance (F1,354 = 7·40, P = 0·0068, Fig. 2), but did not vary with sex or its interaction with dispersal distance (sex: F1,352 = 1·00, P = 0·32; sex × distance: F1,352 = 2·56, P = 0·11). A GAM model with four degrees of freedom did not improve the fit of the data (linear effect: t = −4·39, d.f. = 1, P < 0·0001; nonlinear effect: χ2 = 5·9, d.f. = 3, P = 0·12). The number of recruits an individual produced along its life (LRS) followed a similar, but weaker trend (F1,335 = 4·16, P = 0·042, see Fig. 2). There was no difference between sexes, nor when the interaction between sex and distance was incorporated into the model (sex: F1,333 = 0·50, P = 48; sex × distance: F1,333 = 1·53, P = 0·22). The nonparametric part of a GAM model evidenced little support for a nonlinear relationship (linear effect: t = −2·47, d.f. = 1, P = 0·014; nonlinear effect: χ2 = 3·2, d.f. = 3, P = 0·36). Individual fitness λind also correlated significantly with dispersal distance (GLM, F1,354 = 4·65, P = 0·0318). Again, sex or its interaction with dispersal distance did not affect λind (Sex: F1,352 = 1·28, P = 0·26; Sex × Distance: F1,352 = 0·87, P = 0·35). The corresponding GAM model supported a linear rather than a nonlinear relationship between dispersal distance and individual fitness (linear effect: t = −2·13, d.f. = 1, P = 0·037; nonlinear effect: χ2 = 1·7, d.f. = 3, P = 0·63). When dispersal was characterized as a discrete process, lifetime fitness was unrelated to dispersal status in all models (LFP: F1,354 = 2·07, P = 0·15; LRS: F1,335 = 0·33, P = 0·57; λind = F1,354 = 2·16, P = 0·14).

image

Figure 2.  Relationship between natal dispersal distance and (a) lifetime fledgling production, (b) lifetime recruitment success and (c) individual fitness (λind) in lesser kestrels.

Download figure to PowerPoint

Dispersal distance, colony selection and phenotypic traits

Our results did not seem to be affected by large-scale environmental variations because covariance parameter estimates of the subpopulation random term was in all models very close to zero. However, birds dispersing long distances could have settled in colonies of lower quality, i.e. recruited in empty buildings or in small colonies where survival prospects are lower (see Serrano et al. 2005). To evaluate this possibility, we tested whether the probability of funding a new colony or joining a very small one (i.e. settling in a building with 0–2 pairs, see Serrano et al. 2005) was a function of natal dispersal distance. A binomial mixed model with year and subpopulation of recruitment as random terms showed that this probability increased with dispersal distance (estimate ± SE 0·23 ± 0·05; F1,939 = 50·52, P < 0·0001), with a similar effect in both sexes (sex: F1,939 = 0·04, P = 0·85; sex × distance: F1,939 = 0·01, P = 0·91). Indeed, when first-breeding colony size was introduced together with dispersal distance in lifetime fitness models, results showed that the effect of colony size was more important (LFP: distance: F1,353 = 5·45, P = 0·02; colony size: F1,353 = 10·70, P = 0·0012; λind: distance: F1,353 = 3·22, P = 0·07; colony size: F1,353 = 10·13, P = 0·0016) or both effects became nonsignificant (LRS: distance: F1,334 = 3·74, P = 0·054; colony size: F1,334 = 0·82, P = 0·36). Natal dispersal distance was neither correlated with body size (F1,172 = 0·00, P = 0·98, Table 1) nor with body mass at the time of recruitment (F1,105 = 0·08, P = 0·77, Table 1), after taking into account sex and time-elapsed-laying effects in females.

Potential biases in fitness estimates

The best model including distance from the colony of recruitment to the centroid of the population in the recapture structure (inline image) received little support (AICc = 2168·58, ΔAICc = 4·73, wi = 0·0143). Moreover, there was a negative relationship between dispersal distance and distance to the population centroid (distance: estimate ± SE = −0·09 ± 0·04; F1,951 = 9·81, P = 0·0018; sex: F1,951 = 1·30, P = 0·25; sex × distance: F1,951 = 0·31, P = 0·58), indicating that long-distance dispersers tend to get closer to the centre of the population rather than settling in the borders of the study area. First breeding dispersal distance did not correlate with natal dispersal distance, when either all individuals (natal dispersal distance: F1,197 = 0·01, P = 0·93; sex: F1,197 = 0·04, P = 0·85; natal dispersal distance × sex: F1,197 = 0·05, P = 0·82, see Table 1) or only breeding dispersers were analysed (natal dispersal distance: F1,91 = 0·01, P = 0·92; sex: F1,191 = 0·02, P = 0·88; natal dispersal distance × sex: F1,91 = 0·2, P = 0·66). This suggests low within-individual consistency in dispersal patterns. Finally, recapture probabilities were not dependent on breeding success during the first breeding attempt (results not shown).

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

In vertebrates, current evidence suggests that natal dispersal could be penalized in terms of fitness, although some researches have reported opposite patterns (see references in Introduction). In the present study, natal dispersal distance correlated negatively with the number of fledglings and recruits an individual produced through lifetime, which suggests that recruiting in or close to the natal colony is currently under positive selection at the Ebro valley. Indeed, lesser kestrels settled in their natal colony more frequently than expected by chance (Serrano, Carrete & Tella 2008), and dispersal distances were strongly biased towards the point of origin (Serrano et al. 2003). Although our results only include fitness differences once the settlement process has been successfully completed, all studies to date have found equal or higher rates of mortality in dispersing individuals relative to residents during the movement stage (e.g. Small, Holzwart & Rusch 1993; Larsen & Boutin 1994; Alberts & Altmann 1995; Devillard & Bray 2009; Johnson et al. 2009), so any fitness compensations, i.e. costs after settlement balanced by survival benefits during the movement stage, are highly unexpected.

The separate analysis of each fitness component indicated that annual survival probabilities decreased exponentially with the distance moved from the natal colony, while annual production of fledglings or age at recruitment did not vary with dispersal distance. Further, the differential effect of dispersal distance on survival seemed to be more pronounced in males, in agreement with hypotheses predicting more benefits of remaining near the natal site for this sex (e.g. Greenwood 1980), but the strength of the apparent difference was small. Once birds survived their first breeding year, the probability of survival was similar irrespective of the natal dispersal distance they moved.

Potential biases in dispersal and fitness

An important caveat in studies of dispersal is that results can be biased by individuals travelling beyond the limits of the study area to never return. We tried to minimize this bias by monitoring an isolated population in a large area, and the observed frequency distribution of dispersal distances suggests that we obtained a reasonable figure of the actual dispersal kernel. Despite we embraced colonies separated by more than 200 km, only <5% of the birds dispersed more than 50 km to breed for the first time (Serrano et al. 2003), which strongly suggests that the number of birds recruiting outside the study area should be very low. Further, there is no reason to expect abrupt changes in the relationship between dispersal and fitness if these birds would have been incorporated in the analyses. At the very least, our sampling situation included the vast majority of dispersal events in marked individuals and a great variance in dispersal distances, a situation in which relevant ecological and evolutionary effects may be expected. Another potential source of bias in survival estimates is that the complement of survival in CJS models includes both losses by mortality and permanent emigration, so it is important to consider whether our results could be explained by long-distance dispersers emigrating definitively from the study area the year after recruitment. This possibility seems unlikely because permanent emigration of adult birds, i.e. breeding dispersal, from our study area has never been documented. This is not surprising, because breeding dispersal is of much lesser magnitude than natal dispersal in this species (only 28% of adults changed colony between consecutive years and dispersers moved a median distance of 1600 m, Serrano et al. 2001). Moreover, we have tried to assess and minimize potential biases in fitness estimates motivated by subsequent breeding dispersal by (i) testing for spatial biases in recruitment with respect to dispersal distance, (ii) analysing the effect of dispersal distance and the border of the study area on reencounter probabilities and (iii) testing the existence of within-individual nonrandom dispersal. Long-distance dispersers tend to recruit differentially in the centre of the population, probably as a result of conspecifics attraction (Serrano & Tella 2003), and reencounter probabilities did not vary with the distance to the population centroid. So, differences in apparent survival are not explained by either an asymmetric settlement of long-distance dispersers or a low recapture effort in the periphery of the population. Recapture probabilities of first breeders, however, decreased with dispersal distance, an unexpected result that is probably explained by the lower recapture rates in small colonies (Serrano et al. 2005), where long-distance dispersers tend to recruit. The effect size of the relationship was however weak, only evident for males, and controlled for in survival analyses. Finally, there was no relationship between natal- and first-breeding dispersal distance, which suggests that long-distance dispersers were not leaving the study area differentially in subsequent years. Thus, this is one among the very few studies in which the costs of dispersal are hardly open to alternative explanations of biased fitness estimates (Doligez & Pärt 2008). Apparent survival must be therefore very close to true survival, and asymmetries in fitness between individuals dispersing different distances in our system seem to be mainly explained by immediate survival prospects.

Dispersal as a continuous rather than a discrete process

One important aspect of our approach is that costs and benefits of dispersal have been mainly studied in the past by considering dichotomous definitions of dispersal rather than the continuous distance moved between sites (Baker & Rao 2004; Lowe 2010). Indeed, the majority of studies based of lifetime fitness estimates have compared philopatric and immigrant individuals (e.g. Verhulst & van Eck 1996; Bensch et al. 1998; Wheelwright & Mauck 1998; Marr, Keller & Arcese 2002; Hansson, Bensch & Hasselquist 2004; Pärn et al. 2009). A discrete characterization of dispersal implies by definition that dispersers are attributed to a unique category, while they may constitute a heterogeneous group of individuals in terms of the distance they travelled, the quality of the habitat where they were born, and even in their genetic composition (Hansson, Bensch & Hasselquist 2004). Our results indicate that at least in some situations, where dispersal is not constrained by the spatial scale considered and breeding habitat is homogeneously distributed, a continuous rather than a discrete characterization of dispersal may be more informative.

Why dispersal entails fitness costs?

By relying on individuals settling in 211 different colonies, here we have shown that the probability of recruiting solitarily or in a small-sized colony, where individuals have the highest probability of predator-induced adult mortality and nesting failure (Serrano et al. 2005), increases with dispersal distance. The fitness costs of dispersal in our population seem therefore mediated by the quality of the habitat where individuals settled to breed for the first time, and in fact, colony size was in general a better predictor of lifetime fitness than dispersal distance. Empirical studies of vertebrates have attributed lifetime fitness costs of natal dispersal to loss of familiarity with the natal area or reduced mating success (Bensch et al. 1998; Wheelwright & Mauck 1998; Forero, Donázar & Hiraldo 2002; Hansson, Bensch & Hasselquist 2004; Pärn et al. 2009). Loss of familiarity may be important in our study system, because survival probability decreased exponentially with dispersal distance (Fig. 1). However, no study has so far suggested a consistent lifetime fitness cost of dispersal distance linked to uncertainty in habitat quality at recruitment.

Why some individuals disperse so much, and to low-quality colonies?

As lesser kestrels can easily assess colony quality by cueing on the number of previously settled conspecifics (Serrano et al. 2004), the question would remain as to why the probability of settling in a low-quality colony increases with dispersal distance. A number of studies have found that dispersers and residents differ in a variety of phenotypic traits (e.g. Clobert et al. 2001), so we may hypothesize that differences in colony size at recruitment may be caused by low competitive phenotypes dispersing in response to intensified local competition, with poor competitors being obligated to travel long distances and ending up in a small, low-quality colony. Indeed, several theoretical models assuming differences in competitive abilities predicted the skewed distribution of dispersal distances in animals to be functions of competition for suitable sites (e.g. Murray 1967; Waser 1985). Accordingly, many first-breeding kestrels recruited in colonies smaller than those at which they first tried to settle owing to agonistic interactions with resident adults (Serrano & Tella 2007). Further, natal philopatry to large colonies, where probability of predation is low, but competition with conspecifics is high, correlated negatively with arrival date from the wintering quarters, indicating that only high competitive individuals can settle in crowded environments (Serrano et al. 2003). Dispersing and resident kestrels were previously shown to do not differ in body size or condition in their year of birth (Serrano et al. 2003), and here, we have found no relationship between dispersal distance and these two variables measured after settlement. Yet, these are only crude estimates of individual quality, and other aspects of the phenotype could be implicated in competitive abilities. This scenario, however, does not explain why some individuals move as far as they do because the availability of both vacant sites (i.e. buildings with a surplus of adequate nest-holes, Forero et al. 1996) and optimal foraging habitats (Tella et al. 1998) seemed to be high all over the study area (thus allowing a rapid population and range expansion, Jovani et al. 2008). Therefore, virtually all dispersing individuals had opportunities to settle in an empty or small-sized colony by moving a short distance. Moreover, spontaneous movements seemed to be frequent in our studied system (Serrano & Tella 2007), so induced dispersal alone cannot satisfactorily explain the relationship between dispersal distance, colony size and fitness.

Recent advances in the study of animal personalities and dispersal help propose alternative scenarios. A growing body of evidence from a variety of taxa indicates that dispersal is not random with respect to phenotypic variation in behavioural traits, i.e. dispersers and nondispersers consistently differ in dispersal-related traits such as boldness, aggressiveness and sociality (Fraser et al. 2001; Dingemanse et al. 2003; Cote & Clobert 2007). Dispersers could be behaviourally better skilled than other individuals to colonizing empty patches, while residents would be more suited to social environments. For example, Duckworth & Badyaev (2007) found that aggressive Western Bluebirds able to exclude heterospecifics were also more prone to disperse than less aggressive, resident phenotypes, and postulated that coupling of aggression and dispersal has a key role on patch colonization and range expansion. In great tits, slow-exploring, short-distance dispersing individuals were better at coping with defeat in interactions with conspecifics, enabling them to remain in socially stressful environments (Verbeek et al. 1999; Dingemanse et al. 2003). The negative relationship between dispersal distance and colony size reported here indicates that dispersers are more likely to establish new colonies, supporting the interpretation that individuals moving long distances could be more adapted to new environments and thus to colonizing empty sites (see also Cote et al. 2010). Estimates of heritability in breeding colony size were high for lesser kestrels, which suggest within-individual consistency in social predisposition (Serrano & Tella 2007). In this scenario, the average fitness correlates of dispersal found here would strongly depend on the prevailing environmental characteristics of our population, i.e. high predation rates in small-sized colonies (Serrano et al. 2005), and the maintenance of good colonizers (i.e. long-distance dispersing, asocial phenotypes) would be dependent on spatiotemporal variations in these selective pressures (Duckworth 2008).

Implications and further prospects

Overall, our results may be important to understand colonization success, expansion range and rescue effects, especially in a context of habitat fragmentation and global change. In spatially structured populations, these processes involve ‘jumps’ between isolated patches which are included in the tail of the dispersal kernel (Veit & Lewis 1996), so a negative relationship between dispersal and fitness may contribute to explain slow rates of population spread in advancing fronts of populations, as well as metapopulation declines in spite of significant connectivity. From an evolutionary perspective, realized gene flow between distant populations should be lower than as suggested by the movement of individuals (Verhulst & van Eck 1996; Bensch et al. 1998), with pivotal consequences for phenotypic divergence. Skewed dispersal distances and increased costs of dispersal, however, only translated into a slight neutral genetic differentiation in our population, probably because a very small number of immigrants per generation may be enough to maintain gene flow (Alcaide et al. 2009). Importantly, different phenotypes may differ in dispersal tactics depending on the characteristics of the habitat (Verhulst, Perrins & Riddington 1997). Although in our study the relationship between dispersal and fitness probably arises from habitat quality at recruitment, it is not clear whether (and how) this is mediated by phenotypic traits of individuals (Doligez & Pärt 2008). This point deserves further attention, given that phenotype-dependent dispersal in variable social and environmental circumstances may change our view of the adaptive evolution of dispersal and its ecological implications in animals (Duckworth 2008; Clobert et al. 2009).

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

We are grateful to E. Ursúa, A. Gajón, R. López, J.M. Grande, I. Luque, R. Jovani, A. Giráldez, F.J. Moreno, M.G. Forero, J.A. Donázar, F. Hiraldo and Y. Menor for their help with the field work. We would like to thank Jaime Potti, Dani Oro and three anonymous reviewers for their lucid commentaries and suggestions. During writing, D. Serrano was supported by a Ramón y Cajal contract from the Spanish Ministry of Science and Technology. Financial support during field work was partially provided by collaborative projects with Gobierno de Aragón (1994–1995 and 2000), SEO/BirdLife (1998–1999) and DGICYT and DGES Projects PB93-0040 and PB96-0834.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References