1. Adult survival rates strongly affect population growth, but few studies have quantified if and why adult survival differs between breeding habitats. We investigated potential causes of habitat-specific adult survival rates for male and female northern wheatears (Oenanthe oenanthe L.) breeding in Swedish farmland.
2. We used multistate mark–recapture models based on 1263 breeding records between 1993 and 2007 to estimate survival rates based on habitat-type (SHORT vs. TALL ground vegetation) and breeding-success state parameters. We also used breeding-season observations from 2002 to 2007 and an experimental manipulation of ground vegetation height to identify factors influencing adult mortality.
3. Females had lower annual survival than males (0·42 ± 0·02 vs. 0·50 ± 0·02); this difference largely resulted from low female survival in TALL habitats because of higher nest-predation risk and the large proportion of adult females being killed on the nest (>20%) during nest predation events.
4. Among successful breeders, both sexes displayed similar survival rates, but survival was lower for breeders in TALL as compared to SHORT habitats (0·43 ± 0·03 vs. 0·51 ± 0·02). Experimental manipulation of ground vegetation height, controlling for individual and territory quality (n =132), suggested the cost of rearing young to be higher in TALL habitats (survival of successful breeders in TALL vs. SHORT; 0·43 ± 0·11 vs. 0·57 ± 0·05).
5. Detailed observations of food provisioning behaviour during chick rearing revealed a habitat-related difference in parental workload corresponding to the observed habitat differences in adult survival for successful breeders. Adults breeding in TALL habitats were forced to forage further from the nest relative to SHORT-habitat breeders (mean ± SE; 69 ± 10 vs. 21 ± 2 m), which increased the estimated daily workload for adults in TALL vs. SHORT habitats by c. 20%.
6. On-nest predation and parental workload during chick rearing combine to largely explain habitat-specific adult survival rates. The results have implications for our understanding of adult sex ratios, causes of source–sink demography and habitat-specific growth rates. Furthermore, it suggests SHORT field margins and other residual habitat elements to be important for the conservation of farmland passerines breeding in cropland plains.
What causes adult survival rates to vary between habitats is, however, often poorly known. Adult survival probability can be expected to vary with breeding habitat type when habitats vary in their risk of predation (e.g. McLoughlin et al. 2005), parasite load (Gosselink et al. 2007) or food abundance (i.e. increasing risk of starvation; e.g. Doherty & Grubb 2002). Furthermore, between-habitat differences in adult survival rates could be elevated by reproductive trade-offs where the cost of reproduction is higher in certain habitats because of food limitation and/or higher nest predation risk (Martin 1995).
Most of the above-mentioned studies on habitat-specific adult survival rates concern mammals and birds that remain in the same habitat during the breeding and non-breeding seasons. Whether adult survival rates vary with breeding habitat type for migratory species is less clear, because migratory bird species only spend a small part of their life in their breeding habitat and much of the between-year adult mortality probably occurs in other habitats during migration (Sillett & Holmes 2002). Despite this, there is some evidence that adult survival rates also vary with respect to breeding habitats among migratory birds (Murphy 2001; Perlut et al. 2008a). For example, annual adult male survival rates of the migratory northern wheatear (hereafter wheatear; Oenanthe oenanthe L.) varies with breeding habitat type, such that survival rates are lower in habitats characterized by tall and dense as compared to short or sparse ground vegetation (Arlt et al. 2008). The reason for such patterns, and whether similar habitat-specific variation in survival also applies to females is, however, largely unknown.
Here we investigate the potential causes of habitat-specific adult survival rates of male and female wheatears breeding in Swedish farmland. Our long-term study of breeding wheatears is suitable for studying habitat-specific survival rates because: (i) habitats are distinct due to agricultural land-use falling into two broad categories as defined by their field-layer height (i.e. permanently short vs. growing tall; hereafter SHORT or TALL), and (ii) adults breeding in all habitats display high fidelity to the study area allowing survival rates to be estimated with high accuracy (Pärt 2001a, b; Arlt & Pärt 2007; Arlt et al. 2008).
We hypothesized that habitat-specific annual adult survival rates could partly be explained by on-nest predation because nest predation risk is habitat-specific (Pärt 2001a, b) and nest predators may prey upon adults occupying the nest cavity (e.g. Dilks et al. 2003). Additionally, adult survival rates may differ between these two types of habitats because of different parental workloads; wheatears mainly forage on ground and parents breeding in TALL field layers may be forced to fly longer distances to find accessible foraging areas when compared to those breeding in SHORT field layers, potentially increasing energetic demands when feeding nestlings. We estimated local adult survival rates for male and female wheatears by using multistate mark–recapture models (White, Kendall & Barker 2006; Cooch & White 2008) where individuals could move between states relating to breeding habitat type and breeding success between years. We then assessed whether observed habitat-specific rates of apparent survival (hereafter survival) could be linked to predation risk by testing whether survival was linked to breeding success (failed, mainly due to nest predation, vs. successful). The possibility that adults were killed on the nest during nest predation was investigated by detailed searches for adults following nest predation and by examining nest remains. Potential effects of parental workloads on survival were investigated by studying parental food provisioning rates and foraging flight distances during the nestling rearing phase to estimate parental workloads of successfully breeding pairs (see Tremblay et al. 2005). Furthermore, we made use of an experimental manipulation of field layer heights after the establishment of territories to disentangle the effect of individual quality from increased work loads. For this, we used longitudinal data from individuals breeding on the same territory in two consecutive years, but where the field layer height was experimentally manipulated in the second year (see Pärt 2001b).
Materials and methods
The wheatear is an insectivorous, ground-foraging bird frequently inhabiting open farmland habitats, and usually nests in cavities within stone piles. Wheatears prefer to forage in sparse or short vegetation (e.g. grazed grasslands; Tye 1992; Pärt 2001a, b) where prey availability is higher (Tye 1992). Reproductive success is higher in habitats with short as compared to tall field layers, probably a result of increased food availability, with these short field layers also characterized by a lower risk of nest predation (Pärt 2001a). The most important nest predators, i.e. weasels (Mustela nivalis), stoats (Mustela erminea) and adders (Vipera berus), largely prefer hunting in tall field layers where their main prey (i.e. voles) tend to be most abundant (Pärt & Wretenberg 2002; Macdonald, Tew & Todd 2004). The wheatear is one of many farmland bird species which has shown a marked population decline during the last three decades (Wretenberg et al. 2006). Possible reasons for this decline include the loss of semi-natural pastures and nesting sites: especially stone piles and stonewalls.
Our study area (60 km2) is a heterogeneous agricultural landscape situated southeast of Uppsala in southern central Sweden (59°50′ N, 17°50′ E). From 1993 to 2007, all territories previously occupied or potentially suitable for wheatears (229 territory sites, annual number of pairs: mean ± SD, 144 ± 23; range 120–180) were regularly monitored throughout the breeding season (at least every fifth day from mid-April to the end of June). Detailed data were collected from a central and intensively studied 40 km2-part of the total study area, with the surrounding area monitored for adult dispersal from these central territories (see below). Each territory was categorized as belonging to one of the following six habitat types characterized by different land use: (i) farmyards including bare ground, mowed lawns and gardens; (ii) pastures grazed by cattle or sheep; (iii) pastures grazed by horses; (iv) spring-sown crop fields; (v) autumn-sown crop fields; and (vi) ungrazed pastures and other unmanaged grassland habitats. The first three habitat types were generally characterized by a field layer kept permanently shorter than 5 cm and grouped together as short field layer habitat (hereafter SHORT). The latter three habitat types were characterized by a field layer which was often short at the time of territory establishment, but grew to 15 cm or more during late incubation and nestling care, and grouped together as tall field layer habitat (hereafter TALL; see also Pärt 2001a; Arlt & Pärt 2007; Arlt et al. 2008). When a nest site was located at the border of these two habitat categories, the territory was classified as SHORT because the pair’s territory was almost exclusively contained within the short habitat (>90% of the foraging trips were made in the short patch; T. Pärt & D. Arlt, unpublished data).
Wheatears built their nests in cavities; in our study area mainly within stone piles (c. 80% in stone piles and stone walls) or under roof tiles of farm buildings. Egg laying started in early May, incubation by the female lasted for about 13 days, brooding by the female but not the male was commonly noted during the nestling stage, with nestlings spending c. 15 days in the nest before fledging (T. Pärt & D. Arlt, unpublished data). A breeding attempt was judged to be successful when we observed fledglings or heard intense warning calls of the parents at or after the predicted time of fledging (Pärt 2001a; Arlt & Pärt 2007). Nest failures were mostly due to predation (Pärt 2001a). Each year we uniquely colour-ringed chicks from c. 90% of all successful nests, as well as a proportion of unmarked adults, so that, on average, 57% of breeding adults were marked by the end of the breeding season. Adults were aged as first-year breeders or older based on plumage characteristics (Pärt 2001a).
Adult survival analyses
Survival rate estimates often are biased due to the difficulty of separating mortality from permanent emigration (Doligez & Pärt 2008). To minimize the potentially confounding effect of breeding dispersal (permanent emigration) on survival, we estimated adult survival only for birds first detected breeding in the 40-km2 central area, with subsequent surveys including the entire (60 km2) study area. This means that all adults dispersing within 2–4 km from the outer border of the central area between years could be detected (only 19 of 723 birds shifted from the central to the outer area during the study and their inclusion or removal from analyses did not influence model ranking or survival estimates). Our survival estimates are likely to reflect true survival rates because adult wheatears in our study area only disperse short distances between years [median ≤ 350 m, 90% quantile < 1500 m; measures of adult breeding dispersal calculated from a small core area (8 km2) with a buffer zone of 6–8 km for resighting; Arlt & Pärt 2008; Arlt et al. 2008]. Furthermore, outside the 2 km buffer zone (in the northern part of the study area) there was little suitable breeding habitat for wheatears for at least 2 km. Thus, from our dispersal data we estimate that the number of permanently emigrating adults was negligible. Moreover, breeding dispersal patterns did not differ between habitat types enabling us to compare habitat-specific survival rates (Arlt et al. 2008).
Although adult females were more likely than males to shift breeding site between years, this difference was minor (as most birds shifted within a distance of two territories) and breeding dispersal distance did not differ between the sexes (Arlt & Pärt 2008). Following breeding failure, males, rather than females, dispersed slightly further; this increased dispersal distance in males was mainly caused by a larger fraction dispersing between 500 and 1500 m, whereas there was no difference in the proportion of males and females dispersing >1500 m (D. Arlt & T. Pärt, unpublished data). Thus, we are confident that sex, habitat and breeding status did not bias our survival estimate comparisons (see below).
We used multistate mark–recapture models in program mark (Version 5.1; White & Burnham 1999) to estimate adult survival probabilities based on 723 adults of known age and breeding history (336 males, 387 females), where survival was determined from sighting records of ringed individuals during the intensive breeding season monitoring program (for more details see Pärt 2001a, b; Arlt & Pärt 2007, 2008; Arlt et al. 2008). We used a multistate model framework because it permits the estimation of survival probabilities specific to states which may change for individuals from one survey (i.e. year) to the next and, thus, cannot be incorporated within a traditional Cormack–Jolly–Seber (CJS) encounter history (White et al. 2006; Cooch & White 2008). In this study, individuals were assigned to one of four states from which they could shift between years: breeding success or failure in a SHORT or TALL habitat (a total of 1263 breeding records; 619 for females, 644 for males). This allowed adult survival estimates to be calculated based on combinations of the state variables (habitat and breeding success) and utilizing variables associated with the individual (sex and age), e.g. survival probability to year t + 1 of a young female (individual variables) successfully breeding in TALL habitat (state variables) in year t (for more details, see Appendix S1, Supporting Information).
These four variables were included in models because they are likely to be related to adult survival: (i) sex, female mortality rates in bird populations are generally higher than for males (Donald 2007); (ii) age, young birds may have lower survival rates compared to older birds because of factors associated with individual quality or experience (Forslund & Pärt 1995); (iii) breeding success, successful chick rearing and adult survival probability may exhibit a positive relationship (e.g. as a proxy for adult quality or adult predation risk; Forslund & Pärt 1995; Dilks et al. 2003) or a negative relationship (e.g. if parents trade-off reproduction against survival; Forslund & Pärt 1995); and (iv) habitat category, tall field layer habitats exhibit negative population growth rates compared to those in short field layers (Arlt et al. 2008). Also, the state variables are likely to be linked to reproductive effort and, thus, survival differences between the various state-variable categories help to highlight differences in reproductive costs for these states (as measured by post-reproduction survival probability in year t to year t +1; Nichols et al. 1994; White et al. 2006; Hadley, Rotella & Garrott 2007). Our expectation was that if habitat-specific reproductive costs were related to parental workload, then habitat-specific differences in adult survival would be more pronounced for successful than failed breeders. This is because these habitat-specific workload differences are more likely to be evident during chick rearing, and that many failed breeders would not pay this cost.
There was no evidence of overdispersion in the data (500-sample median ĉ goodness-of-fit test for the non-time-dependent global model = 1·0 ± 0·01); thus for each analysis, we compared candidate models using Akaike’s information criterion with a second-order correction for sample size (AICc). AIC weights (wi) were used to determine the relative strength of support for each model, model averaging used when reporting parameter estimates, and AIC relative-importance weights used to rank the importance of each variable (Burnham & Anderson 2002; Cooch & White 2008).
To determine the model structure with the greatest support, we fit a series of factors to each parameter (Φ, p, ψ), while holding the remaining parameters constant (for a similar procedure, see McElligott, Altwegg & Hayden 2002; Hadley et al. 2007). First, we examined support for various structures of the resighting parameter (±sex, habitat, year) when the survival and transition structures were held constant with relatively complex structures (i.e. Φsex, habitat, breeding success and ψsex, habitat, breeding success; Table S1). The subscripts denote the parameters which were allowed to vary independently during the estimation procedure (Cooch & White 2008); thus Φsex, habitat, breeding success will produce eight separate estimates for survival based on the possible combinations of sex (male and female), habitat (SHORT and TALL) and breeding success (success and failed). There was strong support for a sex difference in resighting probability P [mean ± SE: male P =0·98 ± 0·01 (range from all years 0·93–1·0, median = 1·0); female P =0·89 ± 0·03 (range 0·75–1·0, median = 0·86)] with habitat, breeding success and year effects having virtually no support [Table S1; relative-importance weights (Burnham & Anderson 2002) for each variable: sex (0·92) >> habitat (0·13) ≈ breeding success (0·11) ≈ year (0·0)]. This sex-varying resighting parameter (psex) was then used for all subsequent analyses on the full data set. Second, we looked at support for various constraints on the transition parameter (ψ ± sex, habitat, breeding success), with the other parameters held constant (i.e. psex and Φsex, habitat, breeding success). In this study, the transition structure of ψhabitat had overwhelming support [Appendix S1, Table S2; relative-importance weights for each variable: habitat (1·0) >> sex (0·15) ≈ breeding success (0·01) ≈ year (0)]; thus, ψhabitat was used for all survival parameter modelling (Table 1).
Table 1. Candidate set of multistate mark–recapture models to determine the best covariate structure for the survival parameter (Φ); various effects were modelled for Φ while the resighting and transition parameters were fixed at the covariate structures with the greatest support (psex and ψhabitat respectively; see text for more details)
Factors included in the models were sex, age, breeding success (BS), habitat category (habitat) and year; subscripts define which parameters were estimated for that model. The BS × habitat parameter is not a true interaction term, but instead represents a model where survival rates are the same for both habitats when nesting attempts fail, but differ when nesting is successful (see text for more details). ΔAICc = difference in AICc relative to the best model, wi = AICc weight of the model, K = number of parameters in the model, Dev = model deviance.
Φsex, BS, age
Φsex, BS, habitat
Φsex, BS×habitat, age
Φsex, BS, habitat, age
ΦBS, habitat, age
Φsex, habitat, age
Φsex, BS×habitat, year
When modelling survival we considered not only combinations of the parameters of interest (±habitat, breeding success, sex and age), but also an additional constraint applied to the habitat and breeding success estimates via a manipulation of the PIMs. For this we set the PIM structure so that separate survival estimates would be generated for the two habitat types when breeding was successful, but only a single estimate would be generated when breeding failed regardless of habitat type (a form of habitat and breeding success interaction). We did this based on our prediction that if reproductive costs differed between habitats because of differences in parental workload, then this would show as a survival cost primarily when breeding attempts were successful because they had to provide for offspring throughout the chick rearing stage. Thus, the ‘interaction’ model (Φhabitat × breeding success) has fewer parameters than the general model (Φhabitat, breeding success) because it assumes habitat-specific adult survival rates are the same when nesting attempts fail, whereas the general model assumes survival rates differ between habitat types and breeding outcomes (and thus has one additional parameter to estimate). We also considered an ‘age’ variable (first-year breeder or older) in models which contained habitat because there was a higher proportion of first-year breeders in TALL habitat compared to SHORT (TALL vs. SHORT; males: 39% vs. 26%, χ2 = 12·2, P <0·001; females: 48% vs. 35%, χ2 = 11·7, P <0·001). Comparing the strength of support for survival models with and without the ‘age’ variable allowed us to assess whether differences in survival between habitat categories were confounded by the effects of individual age. Survival estimates were generally modelled without year effects because the inclusion of ‘year’ returned over-parameterized models and analyses showed that including this term had no AIC support over models which did not include a year term (ΔAICc of the best-supported model in Table 1 when including a year term = 72).
Predation of adults
Most nest failures in our study area result from nest predation (85% of 146; Pärt 2001a); however, the likelihood of adults being killed at the time of depredation was unknown. Thus, between 2002 and 2006, we collected information from depredated nests to identify the fate of the focal adult pair. The character of nest disturbance was noted, as were the remains of any chicks or adults within or near the nest cavity. The territory was searched at the time of nest failure to determine which of the adults were known to be alive at the time the nest was found depredated. In addition, territories were regularly monitored during the remainder of the breeding season and in the post-breeding season. This further increased the probability of re-sighting any surviving adults in the study area as adult wheatears generally remain in the vicinity of their breeding territory in the post-breeding season to perform a flight- and body-feather moult (Arlt & Pärt 2008).
Parental workload observations
To estimate the relative cost associated with provisioning for offspring in TALL vs. SHORT habitats, we monitored nest feeding visits and foraging by adult wheatears in 31 territories (15 TALL and 16 SHORT) between 25 May and 15 June 2007. TALL territories were characterized by the nest being located in a rock pile (n =12) or under the roof of a small shed (n =3), surrounded by a uniform, dense, cereal crop field layer (>50 cm high) which isolated it from the nearest foraging habitat with short field layer by 20–110 m (mean ± SD: 44 ± 31 m). In SHORT territories, the nest was located under a rock or in a rock pile within or near a cattle or horse pasture, with continuous short field layer habitat being available in at least a 180° arc from the nest. Within each type of habitat, we randomly selected territories to be monitored and the selected territories thus covering the natural variation of field layer structure within each habitat type. To minimize confounds between habitat comparisons, we matched TALL and SHORT territories as much as possible with regard to age and number of chicks in the nest and undertook a 2-h observation at each matched TALL and SHORT pair (13 pairs = 26 territories) at the same time on the same day. Observers alternated between TALL and SHORT territories to minimize observer bias accruing within a particular habitat category. For the remaining five territories, we were unable to match observations to the same time, but all observations were made in similar weather conditions (i.e. warm, dry and little wind). All observations were made between 08.00 and 16.00 h when chicks were between 7 and 10 days old (the age of chicks was estimated by comparing them to photographs of known-age nestlings). Observations were comparable because adult feeding rates show little diurnal fluctuation during this period (Low et al. 2008).
For each nest visit by an adult during an observation period, observers recorded: (i) sex; (ii) time of arrival; (iii) load size; and (iv) distance from the nest to the first feed after leaving the nest. Observers used the unaided eye or binoculars to follow movements of adults within the territory, and a 20–60× magnification telescope to observe prey load size when birds arrived at the nest site. Load size was a relative estimate on a rising 3-point scale (1 = prey items occupying less than half the length of the beak, 2 = prey items occupying more than half the length of the beak, and 3 = prey items occupying the length of the beak with prey body-parts extending beyond the outer margins of the beak), with these estimates being consistent between observers relative to habitat category and the sex of the bird (F3,67 = 1·26, P =0·29). At the end of the observation period, distances to foraging locations in the territory were measured using a GPS. We also used data on daily feeding rates collected for an additional 11 nests. We fitted data-loggers into the entrance holes and recorded all movements of adults into and out of the nest chamber for the following 48 h (for more details, see Low et al. 2008).
Manipulation of field layer height
Because survival rates relating to field layer height may be confounded by factors such as individual quality or unaccounted-for differences between territories, we used an experimental approach similar to that used by Pärt (2001b) to disentangle the effects of territory and individual quality from habitat-linked survival estimates. For this, we used longitudinal data of individuals breeding in the same territory for two consecutive years (years 1 and 2), but where the habitat was manipulated in some territories in year 2 due to changed management regimes (i.e. grazed in year 1 and grazed vs. not grazed in year 2). We only used birds breeding in SHORT territories in year 1 because these were grazed in late autumn, which produced the preferred short grazed vegetation structure also when individuals selected and established territories in year 2. The decision by farmers to change management regimes was made after the wheatears had chosen their breeding territories in year 2; thus, all territories displayed the same habitat characteristics at the time of establishment in both years. However, a subset of these territories were not grazed in year 2, causing the ground vegetation to grow TALL during the nestling rearing stage whereas the other group remained SHORT. These between-year changes were unpredictable because of the multitude of factors on which decisions were based (Söderström, Pärt & Linnarsson 2001), meaning that the manipulations of field layer height can be viewed as random with respect to territory quality (Pärt 2001b). To test whether habitat type affected adult survival when putting effort into chick rearing we analysed only nests which were successful in year 2. From this subset, survival from years 2 to 3 could then be estimated and compared for two groups: those which experienced the same conditions in years 1 and 2 (n =111; SHORT and SHORT) and those which changed category from SHORT in year 1 to TALL in year 2 (n =21). This protocol allowed us to control for individual and territory quality effects (for more details, see Pärt 2001b) and remove on-nest predation confounds. We used a CJS analysis in program mark and compared four models with different constraints on the survival parameter – constant survival, habitat differences, sex differences, and habitat and sex differences.
Because all measurements during feeding observations (e.g. load size, distance travelled) were taken multiple times for each bird, means for each bird and each territory were used for analyses. Mean distances were log transformed prior to analysis to normalize the skewed distribution; means were used because the calculation of flight costs required total daily foraging flight distances (mean distance multiplied by number of visits). Data were incomplete for some nests; thus, sample sizes may vary between particular analyses. For parental workload estimates, comparisons between the male and female at one site could not be made because the male did not attend the nest during observations; thus this nest was removed for male–female comparisons. Analyses were performed in jmp (version 7.0; SAS Institute Inc., Cary, NC, USA). Means are expressed ±SE unless otherwise indicated.
The five models with AICc support (Σwi > 0·99) included an effect of sex and breeding success on adult survival probability [Table 1; relative-importance weights for each variable: breeding success (1·0) ≈ sex (0·98) > habitat (0·66) > age (0·20) ≈ year (0)]. In general, female survival was slightly lower than male survival, and survival was markedly lower for females whose nests failed as compared to those successfully fledging young (any association between male survival and breeding success was marginal; Fig. 1). The best supported model, with more than twice the support of the next-best model (wi = 0·52 vs. 0·21; Table 1), also included habitat category. This model indicates that the effect of nesting outcome on adult survival also depends on breeding habitat. Birds breeding in TALL habitat territories had lower survival than those in SHORT habitat territories, but only for pairs where chicks were successfully fledged; for failed pairs, there was no difference in survival between the two habitat categories (Table 1; Fig. 1). However, the difference in female survival between habitats (TALL vs. SHORT; 0·35 ± 0·03 vs. 0·45 ± 0·03) was not entirely the result of habitat-specific survival differences of successful breeders; lower female survival in TALL habitats also resulted from a higher probability of breeding failure in this habitat category (breeding failure in TALL vs. SHORT: 39% of 784 vs. 24% of 1162; χ2 = 51·8, P <0·0001), and its corresponding low female survival. Of the seven models which included an age variable, none of these had greater support than the equivalent model which did not include this term (see Table 1). Thus, there is scant support for the hypothesis that differences in survival between habitat classes were a function of the differing habitat-specific age structure (see also the field layer height experiment below). State transitions from year t to year t +1 were influenced by habitat, but not breeding success or sex in year t (Table S2 and see Methods). In general, birds tended to move from TALL to SHORT habitats between years [80% of birds stayed in SHORT habitats from year t to year t +1; whereas only 50% of birds remained in TALL habitats from year t to year t +1 (for detailed state-transition estimates, see Appendix S1)], which suggests that experienced birds have a preference for SHORT habitats.
Predation of adult females
Of the failed nests where female survival could be determined (i.e. females with known identity; n =52), only 19% (n = 10) of females returned to breed in the following year (Table 2; cf. 22% as estimated from the survival model for failed nests across all years in Table 1; Fig.1). In 10 cases (19%), the recovery of feathers, rings or a carcass provided direct evidence of the female having been killed on the nest. While the identification of the predator in these cases was not definitive, it was most likely a small predatory mammal (e.g. stoat or weasel) based on the state of the nest (intact nest cup; domestic cat, fox and badger destroy the nest cup by dragging/digging it out of the nest cavity) and recovered feathers and/or carcass remains. In an additional 12 cases (23%), the female was missing from the territory directly following nest depredation and was not observed during subsequent surveys. None of these females returned to breed in the following year, in contrast to nests where the female was observed following nest failure (0% vs. 33% return rate respectively; χ2 = 5·2, P =0·022; Table 2). These figures show that at least 19%, but perhaps up to 42%, of females were killed at the time of nest failure. Similar data on male survival (n =47 nests) suggest that male predation at the time of nest failure is rare (Table 2).
Table 2. Observations of 52 females and 47 males from depredated nests, between 2002 and 2006, based on observations of the adult in the breeding territory immediately after nest failure and during the post-breeding census
Numbers in parentheses indicate those in each group which returned to breed in the following year (i.e. apparent survival).
Direct evidence of adult predation
Not seen after nest failure
Seen after nest failure
The probability of on-nest predation should be related to the time adults are present at the nest cavity. As only females incubate the eggs, nest predation during incubation should result in a higher risk of concurrent adult female predation when compared to nest predation during chick rearing. We tested this by expanding the analysis to all failed nests for which we knew the identity of at least one of the adults (and, hence, could determine that bird’s survival) and had accurate data on nesting stage at the time of failure (n =159). Female, but not male, survival was significantly lower for nests which failed during incubation when compared to nests which failed during chick rearing (surviving females per nest: 5 of 40 vs. 19 of 65, respectively, χ2 = 3·9, P =0·047; males 25 of 51 vs. 28 of 60, respectively, χ2 = 0·06, P =0·80).
Females were observed delivering food to the nest at a higher rate than males (paired t-test, t29 = 2·50, P =0·017), but with similar load sizes (t29 = 0·26, P =0·79) and travelling similar distances from the nest to the first feed (t29 = 0·79, P =0·43; see Fig. 2.). The total rate of nest visits did not differ between habitat types (t30 = 0·6, P =0·62), but in TALL habitat mean load size was significantly smaller (t30 = 2·3, P =0·027) and average distances travelled much longer (t30 = 3·85, P <0·001) when compared to SHORT habitat (Fig. 2).
For birds in SHORT habitat territories, the frequency of foraging trips declined exponentially as a function of distance from the nest; 50% of trips were within 10 m and 90% within 50 m, with virtually no movements beyond 100 m. For birds in TALL territories, the pattern of foraging distances differed, with distances mainly reflecting the spatial access to short field layer foraging areas: 40% of foraging trips were within 10 m of the nest entrance (i.e. on the habitat island containing the nest) but more than 25% covered distances >100 m, the latter reflecting movements over the tall field layer matrix to find short field layer foraging areas.
Field layer height experiment
The manipulation experiment provided additional evidence that successfully raising a brood in a TALL habitat incurred greater costs when compared to SHORT habitats (survival estimates from Φhabitat model: TALL = 0·43 ± 0·11, SHORT = 0·57 ± 0·05), with this effect similar for both sexes (see below). Because only a small number of territories changed from SHORT to TALL between years compared to those where it remained SHORT (n =21 vs. 111), the CJS analysis could not differentiate between models where survival varied between habitats or was constant [strength of support for candidate models Φ (ΔAICc, wi) = Φconstant (0, 0·41), Φhabitat (0·3, 0·36), Φsex (1·6, 0·19), Φsex, habitat (4·4, 0·04)]; thus we also report model-averaged survival estimates from the four models above [for habitat: TALL = 0·49 ± 0·07, SHORT = 0·55 ± 0·04; for habitat and sex: TALL = 0·49 ± 0·07 (female), 0·50 ± 0·07 (male), SHORT = 0·55 ± 0·05 (female), 0·56 ± 0·04 (male)]. These habitat-related survival differences for successful breeders are similar to those reported from the multistate analysis on the entire data set (see Fig. 1).
Annual survival rates of adult wheatears differed significantly between the two categories of breeding habitats investigated, with survival being lower in habitats characterized by tall as compared to short field layers. Furthermore, females displayed lower survival rates than males, primarily because of a remarkably low female survival from nests that were depredated. Our results suggest two major causes for the difference in adult survival between habitats: (i) on-nest predation and (ii) increased parental workloads.
Despite nest predation being identified as an important influence on reproductive output, predation of adults on the nest has often been neglected as a potentially important factor affecting individual breeding behaviour, life histories and population growth (e.g. Martin 1995; Evans 2004). One reason is that many studies focus on avian nest predators (e.g. corvids) that may rarely kill adults, or species that nest in open areas where adults may readily escape predators. Furthermore, unless specifically searched for, adult predation may go unnoticed because many nest predators remove carcasses for consumption elsewhere (Moorhouse et al. 2003). For cavity-nesting species, however, potential predators (e.g. small predatory mammals and snakes) are efficient killers when the adult bird is unable to escape the predator by virtue of being ‘trapped’ within the nest cavity (Dilks et al. 2003; Moorhouse et al. 2003). Comparative studies have shown that non-excavator cavity-nesting species have higher annual mortality rates than open nesters (Martin & Li 1992; Martin 1995), although the timing of this mortality is generally unknown. Detailed studies of nest predation of cavity-nesting species, however, frequently report substantial effects of on-nest predation on adult survival rates (Lundberg & Alatalo 1992; Dilks et al. 2003; Moorhouse et al. 2003). Increased on-nest predation risk of females may explain the male-biased adult sex ratios reported for many species (Donald 2007). In the current study, on-nest predation of adult females occurred during c. 20–40% of all nest predation events, which explains a large proportion of the difference in annual survival estimates between females successfully fledging young (46·5%) and those failing (22%). On-nest predation can also explain a large part of the observed habitat-specific difference in female survival rate, because predation risk was higher in TALL habitats, and probably resulted from the preference of hunting TALL habitat for the most common nest predators (see Methods). Our study design, which utilized a buffer zone around the study area to detect dispersing individuals, in conjunction with data showing that between-year breeding dispersal was: (i) short in distance; (ii) similar for males and females; and (iii) unbiased with regard to breeding success and habitat type (see Methods), strongly suggest that our results are robust to the potentially confounding effects of permanent emigration (e.g. Doligez & Pärt 2008).
Reproductive effort and adult survival
Both male and female wheatears dramatically increased their work-load when feeding nestlings in TALL as compared to SHORT habitats because of longer foraging flights. Similar increases in foraging flight distances have been found in other studies as a result of locally limited availability of preferred foraging habitat or food availability (Bruun & Smith 2003; Tremblay et al. 2005).
Based on the foraging behaviour data, we could roughly estimate the difference in energetic cost between SHORT and TALL habitats when feeding young. Previous data based on data loggers suggest parents make 502 ± 41 nest visits per day (n =11 nests; Low et al. 2008). Based on parental workload observations (females visited more often than males: 57% vs. 43% respectively; Fig. 2) and assuming a temporally constant ratio of female to male visits we expect, on average, 286 of these visits have been from the female and 216 from the male. Assuming that distances travelled to foraging sites remain constant throughout the day (TALL vs. SHORT: 69 ± 10 vs. 21 ± 2 m, i.e. birds in TALL territories travel, on average, 2 × 48 m = 96 m further per foraging trip), this equates to females and males in TALL habitats flying an additional 27·46 and 20·74 km day−1, respectively, compared to adults in SHORT habitats. Assuming a flight speed of 9 ms−1 (Blake & Chan 2006), female and male wheatears in TALL habitats spent an extra 50 and 38 min flying each day respectively. This additional flight activity carries a high energy cost [33 kJ h−1 or 20 × basal metabolic rate (BMR); Moreno 1989; Nudds & Bryant 2000], which, if compared to a bird perching or preening during this time (2 BMR; Goldstein 1988), would mean that foraging in TALL habitats costs for females and males an additional 25 and 19 kJ day−1 respectively (Nudds & Bryant 2000). These figures of birds breeding in TALL habitats correspond to an additional 20–28% of daily energy expenditure as compared to northern wheatears feeding nestlings in SHORT habitats (87 kJ day−1, Moreno 1989; 95 kJ day−1, Tatner 1990), i.e. a dramatic increase in reproductive effort in absolute terms.
Admittedly, our estimates are based on behavioural data with several assumptions on flight costs and flight speed (Nudds & Bryant 2000), but it is likely that the assumed parameters would be similar for birds in the two types of habitats. Because food provisioning of nestlings is the predominant behaviour at this stage of life, relative differences in daily flight distances (and time) are likely to reflect corresponding differences in daily energy expenditure. Other studies have shown that variation in feeding rates is positively linked to metabolic rates in the field (e.g. Moreno 1989; Pärt, Gustafsson & Moreno 1992; Nilsson 2002), but our results point to the importance of also considering the effort expended per nest visit when estimating workload because average flight distances may differ markedly between habitats and despite similar feeding rates (see also Grieco 2002; Stauss, Burkhardt & Tomiuk 2005). Although some other studies present behavioural data to suggest that energy expenditure may differ between breeding habitats (e.g. Stauss et al. 2005), few studies have estimated the relative difference in terms of daily energy expenditure (but see O’Halloran et al. 1990; Ens et al. 1992).
The increased workload and the decreased survival of wheatears breeding in TALL as compared to SHORT habitats suggest a habitat-specific difference in reproductive costs. Such phenotypic correlations, however, are potentially confounded by individual quality (van Noorwijk & de Jong 1986) and relationships based on absolute work loads therefore rarely show reproductive costs (Lindén & Møller 1989). Thus, to show a reproductive cost one has to break-up the potential positive correlation between absolute workload and individual quality by experimental manipulation of reproductive effort (Lindén & Møller 1989; Stearns 1992). However, reproductive costs may be realized in unmanipulated situations when low-quality individuals have a higher absolute workload than high-quality individuals or when individuals make a terminal investment (see Clutton-Brock 1984; Pärt et al. 1992). By manipulating field layer height after territory establishment in an individual’s second year when breeding in the same SHORT territory, we were able to show the direct effects of field layer height on annual adult survival rates. Although the sample size of manipulated sites was small, the manipulation from SHORT to TALL field layers in year 2 suggested a similar reduced subsequent survival of successfully breeding pairs as shown by our cross-sectional observational data. Nevertheless, differences in individual quality may still contribute to our non-experimental differences in survival within habitats, which would explain the greater survival of successful vs. failed males in SHORT habitats. In TALL habitats, such a difference could be masked by the strong negative effect of increased parental workload of successful breeders.
Life-history theory predicts that short-lived species, such as the wheatear, should be more likely to trade their own survival for that of their current reproduction because the future prospects of reproduction are small (Charlesworth 1994). Despite this, previous experimental studies suggest that the cost of reproduction is most frequently paid as a reduced fecundity, especially among short-lived species (Lindén & Møller 1989; Golet, Irons & Estes 1998). However, the relative absence of survival costs in these studies can be partly explained by methodological issues, such as small sample sizes and poor manipulation of reproductive effort (Golet et al. 1998; for more potential problems, see also Lindén & Møller 1989). Thus, given our estimates on workload are at least roughly showing the relative difference in parental effort between parents breeding in these two contrasting habitat types, the observed reduction in adult survival probability is not unlikely (for relationships between increased workload and decreased survival, see also Bryant & Tatner 1991; Daan, Deerenberg & Dijkstra 1996).
Given that migratory species only spend a few months in their breeding habitat and that mortality probably mostly occurs during migration (Sillett & Holmes 2002), it may be surprising to find a link between parental workload and annual adult survival rate such as in our study. However, experimental manipulations of reproductive effort of migratory species show that an increased effort when feeding young may have long-lasting effects on future reproduction (e.g. annual survival or future reproductive output; see review in Golet et al. 1998; see also Gustafsson & Pärt 1990). A possible mechanism for such an effect may be immune suppression resulting in a reduced resistance to parasites (Gustafsson et al. 1994). Clearly, such effects on the immune system have a potential to reduce the condition of individuals and potentially increase the probability of dying during the stressful conditions of migration to and from wintering quarters.
Optimal life-history decision models implicitly assume that phenotypic reproductive costs may differ between habitats (Charlesworth 1994), but it has rarely been studied explicitly. Furthermore, virtually no study has investigated whether habitat-specific demography is partly driven by habitat-specific costs of reproduction: one exception being the study of habitat-specific demography of the migratory kingbird (Tyrannus tyrannus; Murphy 2001). However in contrast to our study, habitat-specific effects on adult kingbird survival rates resulted in a pattern where there was an increased survival in habitats with low breeding success (Murphy 2001). Perlut et al. (2008b), on the other hand, found that bobolinks (Dolichonyx oryzivorus) and savanna sparrows (Passerculus sandwichensis) displayed a simultaneous reduction in reproductive success and adult survival rate in poor habitats (i.e. similar to this study), but these patterns were not explicitly discussed in terms of habitat-specific costs of reproduction.
Our results provide additional insight into three well-studied areas of ecology. First, the habitat-specific survival rates of female wheatears add to the often overlooked notion that on-nest predation may be a significant source of adult mortality (e.g. Moorhouse et al. 2003) and that it could partly explain general patterns of sex-biased survival and skewed adult sex ratios (see Donald 2007), especially in relation to nesting type (e.g. cavity vs. open; cf. Martin 1995). Second, a previous study on habitat-specific population growth rates based on male wheatear data (Arlt et al. 2008) showed that low population growth in TALL habitats (i.e. winter and spring crops) was to a large extent driven by low adult survival rates. The current study helps to explain this pattern by suggesting that increased workloads when feeding nestlings, and thus habitat-specific costs of reproduction, could be an important factor in determining why some habitats may act as sinks. We expect the links between population growth rates and habitat-specific reproductive costs to be more common than the literature suggest; it is currently neglected as a potential factor in source-sink demography. Third, the increased parental flight distances and its links to survival in habitats with TALL field layers contributes to the recent discussion about farmland bird conservation: in particular, the importance of habitat elements with short field layers in crop field landscapes (Bradbury et al. 2000; Vickery, Carter & Fuller 2002; Atkinson et al. 2005). Wheatears share many characteristics (foraging in sparse/short field layers and nesting on ground) of other declining farmland passerines and could be viewed as a good model species for understanding observed declines of other species. Our results linking crop field structure to population demography, therefore, suggests that more focus should be given to the study of the effects of field margins and other habitat elements with short field layers (e.g. tractor paths, residual habitat elements) on farmland bird biodiversity.
We are grateful to the many people who helped us with the collection of field data and to the farmers in our study area for their support. Thanks also to Alex Hartman, Jane Reid and two anonymous referees for their constructive comments on the text. The study was partly funded by the Swedish Research Council (to T.P. and M.L.), Helge Ax:son Johnson’s Foundation, Royal Swedish Academy of Sciences, Alvin’s Foundation, and Swedish Ornithological Society (to D.A.).