The Within-Population Formation of the Upper Range Limit in a Songbird


 The formation of the upper distributional range limit of species at mountain slopes is often based on environmental gradients resulting in changing demographic rates towards high elevations. However, we still lack an empiric understanding of how the interplay of demographic parameters forms the upper range limit in highly mobile species. Here, we study apparent survival and within-study area dispersal over a 700 m elevational gradient in barn swallows (Hirundo rustica) by using 15 years of capture-mark-recapture data. Annual apparent survival of adult breeding birds decreased while breeding dispersal probability of adult females, but not males increased towards the upper range limit. Individuals at high elevations dispersed to farms situated at lower elevations than would be expected by random dispersal. These results suggest higher turn-over rates of breeding individuals at high elevations, an elevational increase in immigration and thus, within-population source-sink dynamics between low and high elevations. The formation of the upper range limit therefore is based on preference for low-elevation breeding sites and immigration to high elevations. Thus, shifts of the upper range limit are not only affected by changes in the quality of high-elevation habitats but also by factors affecting the number of immigrants produced at low elevations.


Introduction
All species show limitations in their distribution and thus, form distributional range limits 1,2 . Generally, the distributional range of a species is the consequence of spatial variation in demographic rates i.e.
reproductive output, survival, emigration and immigration 3,4 . The variations in demographic rates in turn are based on spatial variation in biotic and abiotic factors 2,4-6 . Within this framework, theoretical studies showed that in situations of xed environmental gradients, range limits may be additionally affected by differential dispersal patterns 3,7,8 .
Strong climatic and environmental gradients are typically found along mountain slopes 9 . Populations inhabiting high elevations are expected to evolve life-histories different from populations at low elevations as an adaptation to mountainous environments 1,10 . Recent reviews on life-history changes in relation to elevation revealed that high-elevation populations show consistently lower fecundity than lowelevation populations, but this productivity decrease is only balanced by higher adult survival rates in some cases 11,12 . A possible reason for this paradox might be that the increased fecundity in populations at low elevations is realized by a smaller fraction of the reproductively mature individuals due to intraspeci c competition 12 . Alternatively, juvenile survival may be increased in high elevation compared to low elevation populations due to increased parental care or offspring body condition 13,14 .
Adaptations to mountainous environments require su cient genetic isolation from low elevation populations 15,16 . Alternatively, populations at the upper range limit can be sink populations maintained only by immigration from lower elevations thereby preventing adaptation 17,18 . Theoretical considerations 3,8,10 and transplant experiments 18 suggest that in species with high dispersal ability, range limits are shifted upwards beyond conditions supporting sustainable populations. Such species establish sink populations at the upper range limit producing source-sink dynamics over the elevational gradient 3,10-12 .
Upper range limits normally occur within populations, in particular in highly mobile species such as birds where populations cover large extents of elevations associated with environmental gradients 19 , connected by high rates of dispersal between high and low elevations. Knowledge of within-population elevational gradients of demographic rates rather than that of between-population differences in demography at varying elevations can help to understand the mechanisms underlying the formation of the upper limit of population distribution. It also contributes to the understanding of the demographic mechanisms resulting in the upward shifts of range limits observed in some mobile species due to climate change in mountainous areas 17,[20][21][22][23][24] .
In birds, we still lack a comprehensive understanding of how distributional ranges are limited at increasing elevations and what is the role of spatial dynamics in the formation of the upper range limit.
To understand these processes, researchers called for studies investigating several demographic rates across elevational gradients at the same time in long-term studies 4,25,26 . Decreased fecundity or reproductive output at the upper range limit suggests source-sink effects within populations 17,18 .
However, also adult mortality or emigration rates could increase with elevation resulting in higher turnover rates of individuals and increased spatial dynamics from and to the upper range limit.
Reduced reproductive output is not enough to form an upper range limit of a population if individuals disperse randomly within the elevational gradient. At least a preference for (better) breeding sites at lower elevations is additionally required. In such situations, we expect downwards directed within-population dispersal (natal or breeding dispersal) and thus, either lower recruitment rates (natal dispersal) or higher turn-over rates of individuals (breeding dispersal) at high compared to low elevations. Here, we study the spatial variation in apparent survival and within-study area dispersal in an Alpine population of barn swallows (Hirundo rustica) in relation to the elevation of nest sites by using a long-term capture-markrecapture data set. The study area includes potential breeding sites at elevations that exceed the current upper distributional range limit of the species by far. In a recent study, we showed that reproductive effort in terms of fecundity and multibroodedness is increased, but that the annual reproductive output is decreased at high elevations 27 . While the decrease in annual reproductive output is assumed also for this study, increased reproductive effort might result in reduced survival due to reproductive costs. However, since reduced reproductive output has been shown to increase breeding dispersal rates in barn swallows 28 , we expect higher within-population dispersal rates downwards at high elevations. Our results contribute to the understanding of spatial processes in mountainous gradients restricting elevational distributions of birds.

Results
Recapture probability and apparent survival Recapture probability was lower for rst year birds than for older birds (Table 1). For older birds, recapture probability was essentially independent of elevation, whereas for rst year birds, it decreased with increasing elevation. Apparent survival was lower for females compared to males only at low elevations (Table 1, Fig. 1). At high elevations apparent survival of males and females was similar due to a steeper decline of survival values with elevation for males compared to females. Apparent survival of juveniles (i.e. recruitment) was low and independent of elevation. The proportion of males among the not identi ed individuals was estimated to be 45%.

Dispersal Probability
Within-study area dispersal probabilities were lower for adults compared to juveniles ( Table 2, Fig. 2, Fig. 3). 91% of juveniles with recaptures dispersed from their natal farm (N = 43; 4 males returned to their natal farm), whereas only 17% of the adult breeding birds (N = 83) changed the breeding site from one to the next year. In adult females, dispersal probability clearly increased with increasing elevation, while in adult males and in juveniles such a relationship was not present ( Table 2, Fig. 3). Adult males and females did not differ in dispersal probabilities at low elevations, but adult females nearly reached the high dispersal rates of juveniles at the upper limit of the elevational range (Fig. 3).

Dispersal Distances And Corrected Elevational Shift Of Individuals
The data set for the analyses of within study area dispersal distances and corrected elevational shifts included 56 occasions of dispersal events with known start and end point (12 females, 5 males, 39 juveniles; 51 individuals from 26 farms). Breeding dispersal distances (adults) were smaller than natal dispersal distances (juveniles; breeding dispersal distance = 1.25 km, SD = 2.12 km, n = 17; natal dispersal distance = 4.12 km, SD = 3.56 km, n = 39; Table 3, Fig. 2). Credible intervals of estimated correlations between dispersal distance and elevation all included both medium to strong negative as well as medium to strong positive correlations. Consequently, we refrain from drawing conclusion from this case study. However, barn swallows of high elevations dispersed to farms situated at lower elevations than the average farm within the range of dispersal as seen in the negative correlation between corrected elevational shift and elevation (Table 3, Fig. 4). This pattern was similar in all age and sex classes. However, in juveniles it was most pronounced.

Discussion
The long-term mark-recapture study in a small Alpine population of barn swallows revealed clear demographic patterns over a 700 m elevational gradient. First, annual apparent survival of adult breeding birds decreased with increasing elevation towards the upper range limit. Second, breeding dispersal probability of adult females, but not males increased strongly towards the upper range limit. And third, adult and juvenile barn swallows at high elevations dispersed to farms situated at lower elevations than expected by chance. By considering more than one demographic parameter at the elevational range limit 4,29 , we show for a highly mobile passerine bird that not only reproduction and survival is reduced at the upper range limit, but that also breeding dispersal probability is increased and dispersal is directed downwards. Thus, this study provides evidence for a higher turn-over rate of breeding individuals and increased spatial dynamics at the upper range limit.
Unfortunately, we lack reliable long-term data on reproductive output in our Alpine study population of barn swallows. However, in a recent study over 13 Swiss barn swallow populations including our study population we show that though fecundity is increased at high elevations, nestling survival is considerably reduced and start of breeding delayed 27 . A delayed start of breeding is shown to result in a decrease in both, the annual number of successful broods and the number of edglings in successful broods 30 , and thus, in a reduced annual reproductive output 31 . Moreover, since the activity of aerial insects, the main food of barn swallows, strongly depends on temperature 32 , we suggest that spells of cold weather have stronger effects on the reproductive output at high elevations than in lowlands 33,34 .
We therefore have good evidence that barn swallows breeding at the upper range limit in the Swiss Alps experience reduced reproductive output.
As expected, within-study area dispersal probability was high for juveniles (natal dispersal) and low for adult birds (breeding dispersal), con rming that adult barn swallows are highly faithful to their breeding site 28,35−38 . However, this was only the case at low elevations: within-study area dispersal probability of females strongly increased at elevations approaching the upper range limit. A likely underlying mechanism at least partly responsible for this pattern is the decline in reproductive success at high elevations shown to provoke increased dispersal probabilities of females 28 . In contrast, male dispersal probability within the study area was independent of the elevation of the breeding site. These results suggest that the environmental gradient towards high elevations negatively affecting reproduction results in a spatial gradient of female breeding dispersal and in increased turn-over rates of females at the upper range limit.
Dispersal at high elevations was directed downwards. Thus, barn swallows preferably selected breeding sites at lower elevations either due to climatic or other environmental gradients changing with elevation.
Since in this study all nest sites at both low and high elevations were located in the preferred cowsheds hosting cattle 37,39 , small-scale quality of nest sites can be excluded as a reason for the observed pattern. Settlement decisions towards low elevation might be affected by an increased availability of patches with high density of aerial insects 32,40,41 32,40,41 or by the prolonged daily and seasonal duration of high insect activity due to temperature gradients 32 . We suppose that these nest site preferences are not only the reason for directed downwards dispersal, but also prevent settlements at farm buildings with cattle at even higher elevations. The preference for breeding sites at lower elevations suggests that immigration of juvenile birds into the study area rst occurs at low elevation until a critical breeding density is reached.
Later arriving individuals, often individuals of lower quality 37 , then start to select less preferred breeding sites at elevations over 1000 m 42 .
At low elevations, apparent survival showed the well-known sex-and age-speci c patterns of small passerines in continuous habitats. While apparent survival of juveniles (i.e. recruitment) was considerably lower than that of adults also due to reduced rst-year survival and higher rates of natal than breeding dispersal [43][44][45] , males showed higher apparent survival than females 28 . The latter can be explained by higher dispersal rates out of the study area by females than by males after brood loss or reduced reproductive success 28 . However, at high elevations, apparent survival of adult breeding birds declined for individuals of both sexes. This pattern can arise due to either increased breeding dispersal out of the study area or reduced true survival at high elevations.
The increased within-study area dispersal rates of females at high elevations suggest that part of the female decline in apparent survival is due to increased dispersal out of the study area. However, the elevation-independent breeding dispersal probability of males does not t to this explanation for the male decline in apparent survival and higher male dispersal rate outside of the study area but not within the study area seems unlikely. This suggests higher mortality at high elevations. A decline in true survival in both sexes could be due to higher reproductive efforts at higher elevations 27 potentially bearing higher reproductive costs, or because low quality individuals that were outcompeted in the lowlands settle at high elevation. Consistently, delayed start of breeding is shown to be associated with lower annual survival 30 . The sex speci c difference then might be due to the fact that males arrive earlier at breeding sites 37,38 and therefore are more prone to adverse weather conditions in early spring [46][47][48] .
The demographic gradients in combination with the downwards directed dispersal shown in this study revealed that the population covering an elevational gradient of 700 m shows characteristics of sourcesink dynamics resulting in a dispersal-extended upper range limit 18,49 . Similar to source-sink dynamics between distinct populations or patches 29,49,50 , dispersal allows the section of the population at the range limit to persist although it could not persist in the absence of dispersal. The dispersing and dead breeding birds at high elevations must be replaced to maintain population size at the upper range limit. Since recruitment rates of juveniles remained unchanged and low, only immigration can maintain the number of breeding pairs at high elevations. This is also the case in study areas of continuous barn swallow populations at low elevations 28,30,45 . However, immigration at the upper range limit in this study must be considerably higher than at low elevations, but immigrating birds may come from further away.
Increased immigration at high elevations can have several consequences. First, the location of the range limit does not only depend on the environmental gradients, but also on factors affecting the immigration rate to high elevations, i.e. density-dependent effects at low elevations 17,18 . Thus, in years after low reproductive output or annual survival i.e. in years with growth rates λ <= 1 at low elevations, we expect low numbers of immigrants to high elevations. After several years of such conditions, we predict a descending upper range limit. In contrast, several years of λ > 1 at low elevations might result in a rise of the upper range limit extending the limit even more upwards to elevations with low nest site preference and low reproductive output. Second, the within-population elevational source-sink dynamics is likely to result in spatial structuring of the population by sorting individuals with different traits to different Page 9/19 elevations: late arriving immigrants are more likely to end up at high elevations than early arriving immigrants. As immigrants are predominantly rst-year breeders 37 , we expect an altered age-structure with higher proportion of rst-year breeders at high than low elevations. Moreover, late arriving individuals are often of low quality or in bad body condition 37,46,51 . As rst-year breeders and individuals of low quality and body condition show reduced reproductive success and survival 37 the accumulation of rstyear breeders and individuals of low quality at high elevations will further reinforce the demographic gradients towards the upper range limit. Thus, environmental gradients at mountain slopes in combination with within-population source-sink effects leading to spatial structuring can result in steep gradients of demographic rates.
In conclusion, this study provides evidence that the formation of the upper range limit of barns swallows is based on two mechanisms: preference for low-elevation breeding sites and the immigration to high elevations associated with source-sink effects. We therefore suggest that within-population elevational range shifts of barn swallows and other mobile vertebrates can occur due to factors affecting both habitat selection and immigration to high elevations. The occurrence and speed of the expected shift of the upper range limit depends not only on the improvement of high-elevation habitats due to climate change (i.e. changes affecting environmental gradients 21,52 ), but also on the effects of environmental changes (climate and land-use change) on reproduction and survival of the population sections living at low elevations. One of the reasons for the high variation in shift directions and the smaller upwards shifts than expected from regional increase in temperatures in Alpine bird species 20,23,53 might therefore be the declining population sizes of many bird species at lower elevations.

Study species
The barn swallow is a migrant passerine normally breeding in agricultural landscapes 37 . In the Swiss Alps barn swallows occur regularly from the lowland up to around 1300 m a.s.l. However, the highest broods were observed at the elevation of 1900 m a.s.l. 24 . In high elevations, breeding sites are restricted to farm buildings inhabited by cattle because they provide increased food resources and enhanced thermal conditions 27 . Thus, the species' dependence on speci c human structures is particularly high at the upper distribution of the species. However, cattle stables and cow sheds in our study area occur also at higher elevations than the distributional range limit of the species. Consequently, it is not the availability of stables that determines the upper range limit for this species at our study site.

Study area and bird ringing
The ). In 2004 the study area was enlarged to the nal size. The abundance of farm buildings declined with increasing elevations. Therefore we mapped all accessible stables within our study sites in order to estimate an expected value of random dispersal to all available nest sites independent of elevation.
During the breeding season, stables and barns of the farms were regularly controlled for detecting the barn swallow broods. Juveniles in accessible nests were ringed at the age of 5 -15 days. Adults were caught during their rearing periods, usually in the late evening when they rested in or close to the nest, using a hand net, or they were caught with mist nets mounted at the entrances of the buildings. Reproductive output (and ringing of juveniles) was only assessed at the fraction of accessible nests and therefore data on reproductive output were not available in su cient quality in this study.

Mark-recapture analysis
To analyse how apparent survival correlates with elevation and age we used a Cormack-Jolly-Seber type of model [54][55][56][57] . In our model, we included linear predictors for apparent survival and for recapture probabilities using the logit-link function. Survival was modelled dependent on age (two classes: rst year and older), elevation of breeding (adults) or edging (juveniles) site and the interaction age x elevation. We further included a variable that indicated whether the breeding or edging site was in the centre or at the edge of the study area (binomial: edge vs. centre) to account for the fact that individuals being born or breeding at the edge of the study area have a higher chance to leave the study area from one year to the next. In addition, we included year and the farm of origin (breeding or edging site) as normally distributed random variable in the linear predictor. As predictors for recapture probability we also used age, elevation of the site and their interaction as xed predictors and year as random effect.

Dispersal analysis
Analysis of dispersal probability was restricted to birds with at least one recapture. Analyses of dispersal distances and elevational shifts within the study area were restricted to birds with at least one dispersal event within the study area. Elevational shift could be biased by the fact that birds of high elevations had higher probability to descend because they had less opportunity to climb (and vice versa). Consequently, birds at high elevations may have dispersed to lower elevations also when they dispersed randomly to the available farms. Therefore, we calculated the mean elevation of the available farms within the radius of the dispersal distance of the respective bird. Thus, a bird dispersed to a farm at higher or lower elevation than the average farm within the range of dispersal. This elevational difference was denoted corrected elevational shift.
Each recapture occasion represented a data point. Thus, individuals with more than one recapture were measured repeatedly. Additionally, dispersal characteristics could be in uenced by characteristics of the farm or the year. We therefore included farm, year, and also individual, as random factors in linear mixed models to account for these correlations. For analysing effects on dispersal distances and on corrected elevational shifts the normal distribution was assumed, whereas for analysing the effects on dispersal probability the binomial error distribution and logit-link function was assumed. The models were tted to the data using the functions lmer and glmer from the package lme4 58 in the program R 59 . The xed part of the models included age (two classes) and, where possible, the sex of the bird (dispersal probability: three levels: female, male, juvenile; dispersal distances and corrected elevational shift: two levels: adult, juveniles), the elevation of the farm before the dispersal event as a scaled covariate and the interactions of sex and age with elevation. We quanti ed uncertainty of the parameter estimates using Bayesian methods as implemented in the function sim from the package"arm" 60 . Thereby, at prior distributions were assumed for each model parameter, and the posterior distributions were described using Monte Carlo simulation. The 2.5 % and 97.5 % quantiles of 2000 simulated values were used as the limits of the 95 % credible intervals (CrI).