Spatial variation in demography and population growth rate: the importance of natal location

Authors

  • J. M. REID,

    1. School of Biological Sciences, Zoology Building, University of Aberdeen, Tillydrone Avenue, Aberdeen AB24 2TZ, UK; Scottish Chough Study Group, Kindrochaid, Bridgend, Isle of Islay, Argyll, PA44 7PT, UK;
    Search for more papers by this author
  • E. M. BIGNAL,

    1. School of Biological Sciences, Zoology Building, University of Aberdeen, Tillydrone Avenue, Aberdeen AB24 2TZ, UK; Scottish Chough Study Group, Kindrochaid, Bridgend, Isle of Islay, Argyll, PA44 7PT, UK;
    Search for more papers by this author
  • S. BIGNAL,

    1. School of Biological Sciences, Zoology Building, University of Aberdeen, Tillydrone Avenue, Aberdeen AB24 2TZ, UK; Scottish Chough Study Group, Kindrochaid, Bridgend, Isle of Islay, Argyll, PA44 7PT, UK;
    Search for more papers by this author
  • D. I. McCRACKEN,

    1. Research Division, Scottish Agricultural College, Auchincruive, Ayr, KA6 5HW, UK; and
    Search for more papers by this author
  • P. MONAGHAN

    1. Institute of Biomedical and Life Sciences, Graham Kerr Building, University of Glasgow, Glasgow, G12 8QQ, UK
    Search for more papers by this author

Jane M. Reid, School of Biological Sciences, Zoology Building, University of Aberdeen, Tillydrone Avenue, Aberdeen AB24 2TZ, UK. E-mail: jane.reid@abdn.ac.uk

Summary

  • 1Understanding the pattern and magnitude of spatial variation in demography and population growth rate (λ) is key to understanding the structure and dynamics of natural populations. However, such spatial variation is challenging to quantify. We use > 20 years of individual life-history data to quantify small- and large-scale spatial variation in demography and λ within a single population of red-billed choughs Pyrrhocorax pyrrhocorax on Islay, Scotland. Critically, we demonstrate a major importance of an individual's natal rather than current location in driving observed spatial variation.
  • 2Breeding success (the number of offspring fledged per breeding attempt) varied among individual chough nest sites but did not vary on a larger spatial scale across Islay.
  • 3The proportion of fledglings observed to survive to recruiting age varied markedly among individual nest sites and also varied more widely across Islay. Spatial capture–mark–recapture models defined two discrete geographical regions where fledgling survival differed significantly: choughs fledged in region ‘BGE’ were more likely to survive than choughs fledged in region ‘CNSW’ as both subadults and adults.
  • 4The asymptotic λ attributable to breeding attempts in region BGE exceeded unity, and exceeded that attributable to breeding attempts in region CNSW. Relatively productive and unproductive regions therefore exist within this population.
  • 5Spatial variation in adult survival was better explained by an individual's natal region than the region where that individual settled to breed. Spatial variation in λ would consequently have remained undetected had survival been measured across resident breeders rather than across individuals fledged in each region. Furthermore, breeding success was a weak predictor of a nest site's estimated productivity of recruits.
  • 6We therefore describe marked spatial variation in demography and λ within a single population of a territorial vertebrate, mediated partly by long-term links between an individual's natal location and its subsequent life-history. Life-long monitoring of individuals of known origin may therefore be necessary to identify accurately subpopulations of intrinsically high and low λ.

Introduction

Spatial variation in demography and population growth rate (λ) is predicted to play a major part in determining the structure and dynamics of populations (Pulliam 1988; Rodenhouse, Sherry & Holmes 1997; Hanski 1999). For example, small-scale variation in territory quality can drive pre-emptive occupation of the most productive territories, thereby determining patterns of habitat occupancy and regulating population size (Sutherland 1996; Rodenhouse et al. 1997). Variation in territory quality can also influence population structure and dynamics at larger spatial scales if territories of similar quality are clustered, creating discrete subpopulations that differ in demography and λ. For example, in source-sink systems, the dynamic of any particular subpopulation will depend on the productivity of, and migration among, surrounding subpopulations rather than solely its own internal demography (Pulliam 1988; Pulliam & Danielson 1991). Therefore, to rationalize and predict the dynamics of natural populations, we require detailed knowledge of the pattern and magnitude of spatial variation in demography and λ (Pulliam & Danielson 1991; Watkinson & Sutherland 1995; Rodenhouse et al. 1997).

While the need to quantify spatial variation is clear, it is less clear how this is best achieved; empiricists face multiple challenges. First, virtually any analysis of spatial variation requires the focal population to be subdivided into spatial units. The most appropriate spatial scale for analysis and the best means of defining subpopulation boundaries are often far from obvious (Amarasekare 1994; Thomas & Kunin 1999). Demographic variation is often compared among distinct habitats that clearly differ in their suitability for the focal species (e.g. Hatchwell, Chamberlain & Perrins 1996; Holmes, Marra & Sherry 1996; Kauffman, Frick & Linthicum 2003; Kreuzer & Huntly 2003). Correspondingly, populations occupying relatively uniform habitat areas are often assumed to constitute single demographic units with little internal structure (e.g. Holmes et al. 1996; Sæther et al. 1999; Reid et al. 2004). This assumption is likely to be simplistic, and population ecologists clearly need to consider the scale and magnitude of demographic variation occurring within individual populations (Coulson et al. 1999; Thomas & Kunin 1999).

Second, once appropriate spatial units are defined, the demography of each subpopulation must be rigorously quantified. Many studies use breeding success as the main measure of subpopulation productivity (Sutherland 1996; Sergio & Newton 2003), thereby assuming a primary importance of reproduction in determining overall productivity, or that reproduction and survival are positively correlated (Pulliam & Danielson 1991; Watkinson & Sutherland 1995). These assumptions may not be valid, and overall productivity is better described by an integrated measure such as λ rather than any single demographic rate (Franklin et al. 2000; Caswell 2001). Furthermore, demography and λ need to be measured over multiple seasons in order to distinguish subpopulations that are consistently productive from subpopulations that are productive in some seasons but not others (Dias 1996; Johnson 2004).

Third, the set of individuals used to measure subpopulation demography must be carefully considered. Demography is typically measured across individuals that currently inhabit each focal area (e.g. Hatchwell et al. 1996; Holmes et al. 1996; Murphy 2001). This approach assumes that an individual's life history is primarily determined by its current environment (Pulliam & Danielson 1991; Dias 1996). However, it is clear that life histories can also be permanently influenced by an individual's natal environment (Lindström 1999). Individuals born in high-quality habitats might therefore breed or survive relatively well throughout their lives, irrespective of the habitat in which they subsequently settle (van de Pol et al. 2006). Studies that measure demography and λ across current residents might consequently underestimate spatial variation; for example, because survival or reproduction in unproductive sink areas is enhanced by the presence of phenotypically (or genetically) superior immigrants from neighbouring sources. Rigorous assessment of spatial variation may therefore require demography to be measured across individuals that originate in each subpopulation, rather than across current residents. Studies of spatial variation in demography rarely take this approach (although see van de Pol et al. 2006).

We use > 20 years of individual life-history data to quantify the pattern and magnitude of spatial variation in demography and λ within a single population of red-billed choughs Pyrrhocorax pyrrhocorax (Linnaeus) on Islay, Scotland, and to consider the data and analyses required to detect this variation. First, we focus on a small spatial scale and quantify whether breeding success or fledgling survival varied consistently among individual chough nest sites. Second, we quantify whether breeding success or survival varied on a larger spatial scale across Islay, and define distinct geographical regions within this population that differ in demography and λ. Critically and unusually, we measure life-long demographic variation associated with individuals’ natal location rather than the location where an individual subsequently settled. Finally, we verify whether the spatial variation observed in λ would have been detected had we adopted the common approaches of solely measuring breeding success, or measuring demography across current residents.

Materials and methods

study population

Islay (c. 620 km2) lies 25 km west of the Argyll coast, Scotland, and holds c. 60 breeding pairs of choughs (2002 census, Finney & Jardine 2003). These choughs are resident on Islay and nest in caves and farm buildings. Individual nest sites are re-used over multiple years and generations (Bignal, Bignal & McCracken 1997). The chough's distribution on Islay covers the extent of grassland and pastoral agriculture (Monaghan et al. 1989; Finney & Jardine 2003). Although nest sites are not uniformly distributed across the suitable habitat, there are no clear boundaries that define discrete subpopulations, and no clear expectation that specific areas should be more or less productive for choughs.

Since 1981, the Scottish Chough Study Group (SCSG) has studied the ecology and life histories of choughs on Islay (Bignal et al. 1997; Reid et al. 2004). Pairs breed once each year and defend large territories around their nest site (>> 1 km2). Fledglings from all territories gather to form communal foraging and roosting flocks that inhabit specific traditional locations (Still, Monaghan & Bignal 1986; Bignal et al. 1997). Individuals leave these flocks at recruitment (modal age 3, Reid et al. 2003a) and acquire breeding territories across the island. Breeder sexes have been determined by breeding behaviour and relative size (Tella & Torre 1993). Females and males settle 13·0 ± 1·0 km and 6·0 ± 1·0 km from their natal site on average, and rarely emigrate from the island (SCSG data; Reid et al. 2003a). Surviving breeders occupy the same nest site in > 90% of consecutive years (SCSG data). Hereafter, we use the hierarchical terminology ‘site’ for an individual nest site, ‘area’ for a group of nest sites and ‘region’ for a group of areas.

spatial variation in breeding success

Each year, breeding success (BS, the number of offspring fledged per breeding attempt) was recorded at 50–80% of occupied sites. We used two-way anovas to test whether BS differed among sites or years. Although BS varies with breeder age and identity in choughs (Reid et al. 2003b), we could not include these traits as covariates because many breeders were unringed and of unknown identity. Therefore, to reduce the extent to which among-site variation in BS could reflect individual attempts or breeders, we restricted analyses to sites where BS was recorded in at least five nonconsecutive years (thereby exceeding a chough's modal breeding longevity of 4–5 years, Reid et al. 2003a). We additionally repeated analyses across a more restricted set of sites where BS was recorded in at least 10 years. Site-specific parameter estimates for BS (BSsite) were saved.

We used Mantel tests to assess whether pairwise differences in BSsite increased with the geographical distance between sites and therefore whether BS varied on a large spatial scale across Islay. The probability of obtaining observed correlations between BSsite and distance matrices was estimated by recalculating correlations over 5000 randomizations of the distance matrix (Scheiner & Gurevitch 2001). As Mantel tests may be insensitive to small-scale or nonlinear spatial variation, we used further randomization procedures to test whether BSsite was more similar in nearest neighbouring sites than expected by chance. The identity of each site's nearest neighbour was resampled with replacement from the set of monitored sites. The mean observed between-neighbour difference in BSsite was compared with that calculated over 5000 sets of randomized neighbours.

spatial variation in survival

Each year, fledglings were individually colour-ringed at a sample of nest sites (Reid et al. 2004). In most years, 40–90 fledglings from 15 to 29 sites were ringed, representing c. 30–55% of choughs fledged on Islay. However, few fledglings were ringed in 1982, 2001 or 2003 (11–22 fledglings from five to 11 sites) due to poor breeding success and/or reduced site access. Over 9000 resightings of colour-ringed individuals have been documented. These data allow the use of capture–mark–recapture (CMR) models to estimate apparent survival (φ) and resighting (p) probabilities (Lebreton et al. 1992; Reid et al. 2003a). However, as survival varies among years in choughs (Reid et al. 2004), there were insufficient data to use CMR approaches to estimate φ and p for choughs fledged from each individual nest site. Such analyses will rarely be feasible for species with low fecundity (see Zimmerman, LaHaye & Gutiérrez 2003). We therefore used a two-stage approach to quantify spatial variation in φ. First, we estimated site-specific fledgling survival as the proportion of colour-ringed fledglings from each site that was observed to reach specific ages. On their own, such ‘observed’ survival rates (S) cannot prove that φ varies among sites, as variation in S could be confounded by among-site variation in p. However, we used initial estimates of S to identify groups of sites (‘areas’) that were similar with respect to location and S. We then used these areas as initial units of comparison within spatially structured CMR models, and rigorously tested whether φ varied among choughs fledged in different areas of Islay while controlling for spatial variation in p. These analyses are described in the following sections.

‘observed’ survival

We used two-way anovas to test whether the proportion (q) of each colour-ringed brood that was observed to reach age 1 (S1) or age 3 (S3, the modal age of recruitment) varied among sites or years. Proportional data were transformed by 2(√n)arcsin(√q), where n is the brood size (Draper & Smith 1981). Analyses covered broods colour-ringed during 1982–2000. We restricted analyses to sites where fledglings were ringed in at least three nonconsecutive years, and repeated analyses across sites where fledglings were ringed in at least 5 years. Site-specific parameter estimates for observed fledgling survival to each age (S1site and S3site) were saved.

We used Mantel tests to assess whether S1site or S3site varied on a large spatial scale across Islay (as for BS). We then used agglomerative cluster analyses, based on z-score standardized Ssite and site coordinates, to identify groups of sites (‘areas’) that were similar with respect to Ssite and geographical location. Agglomerative cluster analysis combines individual observations into progressively larger groupings of similar cases (Scheiner & Gurevitch 2001). As we required groups to provide a basis for spatially structured CMR analyses, we chose the level of aggregation that maximized the number of areas while maintaining sufficient sample sizes within each area to fit CMR models. In practice, final CMR models were robust to the level at which initial areas were defined. Three geographically isolated sites that formed outliers within the cluster analysis were excluded from further analyses.

‘apparent’ survival

We used program mark (White & Burnham 1999) to estimate maximum likelihood φ and p for choughs fledged within each area defined by cluster analysis of Ssite. Individuals that were not included in analyses of S (because fledglings were ringed at their natal site in less than 3 years) were included in CMR analyses if their natal site lay within one of the defined geographical areas. Individuals were noted as observed or not observed during 1 May−1 July each year from 1983 to 2004 (Reid et al. 2003a). φ and p were therefore estimable for the years 1983–1984 and 2002–2003. As Islay's chough population is relatively isolated, spatial and temporal variation in φ is likely to reflect variation in mortality rather than emigration (Reid et al. 2003a).

Previous analyses show that φ differs among first-year (φ1, fledging to age 1), second-year (φ2, age 1–2) and adult (φad, all older ages) choughs, and that φ1, φ2 and p vary among years (Reid et al. 2003a, 2004). However, there were insufficient data to estimate year-specific φ or p for each age-class once data were divided into discrete geographical areas. The initial CMR model therefore included year-specific p and area-specific but year-independent φ1, φ2 and φad. Bootstrap goodness-of-fit tests indicated that this model fitted the data (P = 0·35). To quantify overdispersion (and therefore the degree of model fit), we estimated the median variance inflation factor (c, Cooch & White 1998). For the initial model, median c = 1·03 ± 0·10, suggesting that the model fitted the data (c = 1·00 indicates perfect fit). We then constrained the initial model to identify the most parsimonious grouping of areas into ‘regions’, and test whether areas or regions of Islay differed significantly with respect to φ or p of colour-ringed fledglings. We combined areas for which initial estimates of φ and p were most similar and tested whether the reduced model was better supported at each step. Having defined regions among which φ differed significantly, we introduced among-year variation in φ within each age-class in each region and tested whether statistical support for among-region variation in φ was retained. We used Akaike's Information Criterion (AIC), adjusted for small sample sizes and overdispersion (qAICc, Lebreton et al. 1992; Burnham & Anderson 1998), to identify the most parsimonious model in each candidate set, and likelihood ratio tests to verify whether φ or p differed significantly among years or regions.

Having defined distinct regions of Islay that differed with respect to fledgling survival, we modelled data from choughs that both fledged and settled to breed within final regions to test whether variation in φad and BS was better explained by an individual's natal or settlement region (see Results).

population growth rate

We used stage-structured Leslie matrix projection models to estimate the asymptotic population growth rate (λr) associated with each region of Islay defined by CMR models. We calculated mean λr for each region based on mean demographic rates estimated across all years, and λr for each region for each individual year for which all demographic rates could be estimated. Projection models assumed pre-breeding censuses and birth-pulse dynamics (Caswell 2001), and included second-year, third-year and adult stage-classes (see Reid et al. 2004 for details and parameterization). λr was calculated as the dominant eigenvalue of the projection model: λr = 1·0, 1·0 and < 1·0 indicate stable, increasing and declining populations, respectively (Caswell 2001). This model accurately predicts the observed population trajectory of choughs on Islay, and among-year variation in φ1, φ2 and φad and BS is sufficient to account for observed variation in population size (Reid et al. 2004).

We used Monte Carlo simulations to estimate confidence limits for λr (Alvarez-Buylla & Slatkin 1991; Wisdom, Mills & Doak 2000); 5000 projection matrices were independently parameterized by randomly drawing demographic rates from lognormal (BS) and beta (φ) distributions with mean and variance matching observed values. As observed demographic variance includes both biological and sampling variance, this approach will overestimate variance in λr. To check whether inflated confidence intervals affected conclusions, we repeated simulations using half the observed variance for BS and φ (Wisdom et al. 2000). Analyses were run in SPSS (v14·0), Excel and R (v1·8·1). Means are presented ± 1 SE. Partial effect sizes (η2) are reported for among-site variation in BS and S.

Results

breeding success

There were 54 sites where BS was recorded in ≥ 5 nonconsecutive years during 1981–2004 (n = 624 observations, median of 10 observations per site covering 15 ± 1 year on average). BS averaged 2·0 ± 0·1 fledglings per attempt (CV = 0·12) and varied among sites and years (site F53,547 = 1·6, P = 0·005, η2 = 0·14, year F23,547 = 1·7, P = 0·022). BS also varied among 33 sites where BS was recorded in ≥ 10 years (site F32,441 = 2·0, P = 0·001, η2 = 0·13, year F23,441 = 1·7, P = 0·025, n = 497 observations). Across all 54 sites, predicted BS varied from 1·0 to 3·0 fledglings per attempt (assuming an average year).

Pairwise differences in BSsite did not increase with the geographical distance between sites (Fig. 1a, Mantel test, r = 0·05, P > 0·25). Therefore, there was no evidence that BS varied on a large spatial scale across Islay. However, BSsite was more similar in nearest neighbouring sites than expected by chance (randomization test, P = 0·018).

Figure 1.

Spatial pattern of site-specific parameter estimates for (a) breeding success (BSsite), (b) ‘observed’ survival to age 3 (S3site) and (c) productivity of recruits (Rsite) for chough nest sites on Islay. Darker markers indicate sites with higher BSsite, S3site or Rsite. To ensure that nest site locations cannot be deduced, site coordinates have been nonlinearly transformed. Figures therefore provide a qualitative overview of the pattern of spatial variation.

‘observed’ survival

There were 42 sites where fledglings were colour-ringed in ≥ 3 nonconsecutive years during 1982–2000 (n = 876 individuals from 308 broods, median of six to seven broods per site covering 11 ± 1 year on average). Overall, 36% and 21% of fledglings were observed to reach ages 1 and 3, respectively. The proportion of fledglings that was observed to reach each age varied among sites and years (S1: site F41,248 = 2·6, P < 0·001, η2 = 0·30, year F18,248 = 4·0, P < 0·001; S3: site F41,248 = 1·9, P = 0·002, η2 = 0·24, year F18,248 = 3·2, P < 0·001). Observed survival also varied among 31 sites where fledglings were colour-ringed in ≥ 5 years (S1: site F30,218 = 2·6, P < 0·001, η2 = 0·27, year F18,218 = 3·5, P < 0·001; S3: site F30,218 = 1·9, P = 0·007, η2 = 0·21, year F18,218 = 2·7, P < 0·001).

Pairwise differences in S1site and S3site increased with the geographical distance between sites (Fig. 1b, Mantel tests, r > 0·25, P = 0·001), suggesting that φ and/or p varied on a large spatial scale across Islay. Cluster analyses identified seven geographically discrete groups of sites that were similar with respect to Ssite and location (Fig. 2, hereafter ‘areas’ Ballygrant, Central, Gruinart, North, East Rhinns, South Rhinns and West Rhinns; B, C, G, N, E, S and W, respectively). As S1site and S3site were tightly correlated across sites (r = 0·87, n = 42, P < 0·001), clusters derived from each were similar. We used these seven areas as initial groups within CMR models, to rigorously test whether φ or p varied among choughs fledged in different areas or regions of Islay.

Figure 2.

Seven groups of chough nest sites that were similar with respect to location and ‘observed’ fledgling survival (S) to age 3. Hatched and open areas comprise regions BGE and CNSW, respectively (see Results).

‘apparent’ survival

CMR models used encounter histories of 883 colour-ringed fledglings from 68 sites within the seven initial areas (n = 113, 117, 167, 131, 38, 134 and 183 individuals from 7, 9, 13, 14, 7, 7 and 11 sites in areas B, C, G, N, E, S and W, respectively). Resighting probability (p) varied among years, from 0·44 in 1997 to 0·96 in 1998 (mean 0·73 ± 0·03, inline image = 150·7, P < 0·0001), but did not vary among age-classes or among choughs fledged in different areas. The most parsimonious model therefore included year-specific but not age- or area-specific p (Table 1a). This model fitted the data (Bootstrap GOF, P = 0·21, median c = 1·04 ± 0·11) and was used to test for spatial variation in φ.

Table 1.  Capture–mark–recapture models testing for temporal and spatial variation in resighting (p) and apparent survival probabilities (φ). 1, 2 and ‘ad’ indicate first-year, second-year and adult age-classes. y indicates full-year dependence in p and φ. Superscripts ‘all’ and ‘BGE,CNSW’ indicate parameters split by all seven initial areas, and by regions BGE and CNSW. A subset of tested models is shown. The most parsimonious model in each section is indicated in bold
 ModelqAICcParametersDeviance
(a) Temporal and spatial variation in p
1φ(1all,2all,adall)p(y)3651·4421814·3
2φ(1all,2all,adall)p(1,2,ad)3759·0241959·1
3φ(1all,2all,adall)p()3760·8221965·0
4φ(1all,2all,adall)p(all)3766·8291956·7
(b) Spatial variation in φ
1φ(1all,2all,adall)p(y)3651·4421814·3
5φ(1,2,ad)p(y)3674·7241874·9
6φ(1BGE,CNSW.,2BGE,CNSW.,adBGE,CNSW.)p(y)3634·1271828·1
7φ(1BGE,CNSW.,2,adBGE,CNSW.)p(y)3633·1261829·1
8φ(1BGE,CNSW.,2,ad)p(y)3638·0251836·0
9φ(1,2,adBGE,CNSW.)p(y)3669·2251867·2
10φ(1BGE,CNSW.,2,adBGE,CNSW.)p(yBGE,CNSW.)3642·8471795·2
(c) Temporal variation in φ
7φ(1BGE,CNSW.,2,adBGE,CNSW.)p(y)3633·1261829·1
11φ(1yBGE,CNSW.,2,adBGE,CNSW.)p(y)3574·0611696·8
12φ(1BGE.,1yCNSW.,2,adBGE,CNSW.)p(y)3566·2441724·9
13φ(1BGE.,1yCNSW.,2y,adBGE,CNSW.)p(y)3558·3621678·9
14φ(1BGE.,1yCNSW.,2y,adyBGE,CNSW.)p(y)3594·9971639·4
15φ(1BGE.,1yCNSW.,2y,ady)p(y)3579·4781665·6
16φ(1y,2,adBGE,CNSW.)p(y)3648·1431808·9
17φ(1BGE.,1yCNSW.,2y,adBGE,CNSW.)p(yBGE,CNSW.)3578·7831654·0

A model where φ differed among choughs fledged in all seven areas of Islay was better supported than a model where φ was uniform across all areas (Table 1b). However, there was greatest support for models with intermediate spatial structuring. The most parsimonious model defined two groups of areas (‘regions’), comprising B, G and E, and C, N, S and W, respectively (hereafter regions ‘BGE’ and ‘CNSW’, Table 1b). Apparent survival averaged higher for choughs fledged in BGE than choughs fledged in CNSW across all three age-classes. φ1 and φad were significantly higher for choughs fledged in BGE (φ1: BGE 0·61 ± 0·04, CNSW 0·38 ± 0·05, inline image = 110·1, P < 0·0001; φad: BGE 0·84 ± 0·02, CNSW 0·77 ± 0·02, inline image = 6·1, P = 0·014). The difference in φ2 was not significant (BGE 0·69 ± 0·04, CNSW 0·65 ± 0·04, inline image = 1·5, P = 0·21), and a model with no spatial variation in φ2 was marginally better supported (Table 1b). There was no support for models where p differed between choughs fledged in BGE and CNSW (Table 1b).

First-year survival varied among years in choughs fledged in CNSW (Table 1c, inline image = 106·6, P < 0·0001). There was most support for models where φ1 was constant across years for choughs fledged in BGE (Table 1c), although a likelihood ratio test suggested that φ1 did vary among years (inline image = 28·2, P = 0·04). After pooling choughs fledged in all areas, φ2 varied among years (Table 1c, inline image = 46·0, P < 0·001). There was no support for models that included among-year variation in φad for choughs fledged in either BGE or CNSW (Table 1c). The difference in φ1 between choughs fledged in BGE and CNSW was still strongly supported when among-year variation was included (Table 1c). Again, there was no support for models where p differed between choughs fledged in BGE and CNSW (Table 1c). The final model therefore included region-specific φ1 with among-year variation in CNSW, year-specific but region-independent φ2 and region-specific but year-independent φad. This model fitted the data (Bootstrap GOF P = 0·24, median c = 1·01 ± 0·12).

The most parsimonious CMR model therefore identified two distinct regions of Islay that differed significantly with respect to life-long survival of choughs fledged in these regions (Fig. 2). φ1 and φad averaged higher for choughs fledged in BGE than choughs fledged in CNSW and tended to vary less among years (Fig. 3, CVs: φ1: BGE 0·28, CNSW 0·55; φad: BGE 0·09, CNSW 0·16). Among-year variation in φ1 and φad was not tightly correlated across choughs fledged in BGE and CNSW (Fig. 3).

Figure 3.

Year-specific estimates of (a) first-year (φ1), and (b) adult (φad) survival for choughs fledged in region BGE (open symbols, dashed line) and region CNSW (filled symbols, thin line), and associated standard errors. Population-wide estimates of φ1 and φad are also shown (thick line, after Reid et al. 2004). There were insufficient data to estimate φ1 for BGE in 2001. Among-year variation in survival was not tightly correlated across BGE and CNSW (φ1: r = −0·05, n = 19, P > 0·7, φad: r = 0·17, n = 18, P > 0·5).

breeding success and population growth rate

To estimate λr for the BGE and CNSW regions defined by CMR analyses, we first quantified BS for each region. Across all observed attempts, BS averaged marginally higher in BGE than CNSW (2·1 ± 0·1 and 1·9 ± 0·1 fledglings per attempt, respectively, region F1,627 = 3·8, P = 0·05; year F23,627 = 1·5, P = 0·06). Among-year variation in mean BS was not tightly correlated across BGE and CNSW (r = 0·22, n = 24 years, P = 0·30, based on means of 10·6 ± 0·5 and 16·6 ± 0·8 attempts per year in BGE and CNSW, respectively).

The mean asymptotic λr attributable to breeding attempts made in BGE (λBGE) and CNSW (λCNSW) was 1·12 and 0·95, respectively. Simulations indicated that λBGE exceeded 1·0 (95% CL 1·00–1·25, or 1·06–1·18 with 50% variance), that λCNSW did not differ from 1·0 (95% CL 0·81–1·08, or 0·88–1·02 with 50% variance) and that λBGE exceeded λCNSW.

As there were insufficient data to estimate year-specific φ2 for choughs fledged in BGE and CNSW separately, we estimated year-specific λBGE and λCNSW under two assumptions. First, we used year-specific φ2 measured across the whole population. This approach assumes that φ2 did not differ between BGE and CNSW and is likely to underestimate regional variation in λ. Second, we used mean φ2 estimated for each region across all years. This approach may better indicate the magnitude of the difference between λBGE and λCNSW but may underestimate among-year variation. Under both assumptions, λBGE equalled or exceeded 1·0 in 16 of 17 years for which all parameters could be estimated, and λCNSW equalled or exceeded 1·0 in 6 of 18 years (Fig. 4). λBGE exceeded λCNSW in 16 of 17 years. λBGE and λCNSW did not increase or decline over time (r < 0·15, P > 0·4, n = 17 and 18 years, respectively) but were positively correlated across years (r = 0·53, n = 17, P = 0·03, Fig. 4). λCNSW tended to vary more than λBGE among years (CVs: 0·17 and 0·11 assuming year-specific φ2, or 0·15 and 0·09 assuming region-specific φ2), although this difference was not significant (Levene's test, P > 0·17).

Figure 4.

Year-specific estimates of λr for region BGE (open symbols) and region CNSW (filled symbols) assuming year-specific but region-independent φ2 (solid lines) and region-specific but year-independent φ2 (dashed lines). Population-wide λ is also shown (thick line, after Reid et al. 2004). The horizontal line indicates λ = 1·0.

effects of natal and settlement region

We have shown that choughs fledged in BGE had higher φad than choughs fledged in CNSW (Fig. 3b). This link between natal location and φad could arise because choughs typically settle to breed in their natal region (and therefore experience their natal environment during their adult life), or because φad is permanently influenced by an individual's natal location irrespective of where it settles. To distinguish these possibilities, we tested whether φad differed among choughs fledged in BGE and CNSW while accounting for the region in which they settled. In total, 176 colour-ringed choughs fledged in BGE or CNWS were known to settle in these regions, comprising four residence categories: fledged BGE-settled BGE (n = 37), fledged BGE-settled CNSW (n = 57), fledged CNSW-settled CNSW (n = 56) and fledged CNSW-settled BGE (n = 26). An initial CMR model with year-specific p and category-specific φad fitted the encounter histories of these individuals (Bootstrap GOF P = 0·32, median c = 1·06 ± 0·12). p varied among years, but not among the four residence categories, or among choughs that fledged or settled in BGE vs. CNSW (Table 2a). A model where φad differed among all four categories was better supported than a model where φad was uniform across all categories (Table 2b). However, there was most support for models with intermediate structure. Consistent with analyses of the full data set (Table 1), φad averaged higher for choughs fledged in BGE than choughs fledged in CNSW, irrespective of where they settled (Tables 2b and 3, inline image = 4·1, P = 0·04). In contrast, φad did not differ between choughs that settled in BGE and CNSW, irrespective of where they had fledged (Tables 2b and 3; inline image = 2·0, P = 0·16). φad therefore varied more with a chough's natal region than the region where it settled, indicating a long-term effect of natal location on adult survival. However, the most parsimonious model suggested that φad averaged higher for choughs that either fledged or settled in BGE than for choughs that both fledged and settled in CNSW (Tables 2b and 3; inline image = 10·3, P = 0·001). p did not differ between these groups (Table 2). The final model fitted the data (Bootstrap GOF P = 0·20, median c = 1·03 ± 0·09).

Table 2.  Capture–mark–recapture models testing for variation in resighting (p) and apparent adult survival probabilities (φad) among four categories of choughs that fledged and/or settled in regions BGE or CNSW (Table 3). y indicates full-year dependence in p and φ. Superscripts ‘all’, ‘f’ and ‘s’, respectively, indicate parameters split by all four residence categories, and by choughs that fledged or settled in each region. The most parsimonious model in each section is indicated in bold
 ModelqAICcParametersDeviance
(a) Variation in p
1φ(adall)p(yfBGE,fCNSW)1629·9491111·0
2φ(adall)p(ysBGE,sCNSW)1629·0491110·1
3φ(adall)p(y)1606·8281133·4
4φ(adall)p(all)1700·8111262·8
5φ(adall)p()1697·7 81265·8
(b) Variation in φ
3φ(adall)p(y)1608·9281133·4
6φ(ad)p(y)1610·9251143·8
7φ(adfBGE,fCNSW)p(y)1606·8261139·7
8φ(adsBGE,sCNSW)p(y)1611·0261141·8
9φ(adf or sBGE,f & sCNSW)p(y)1602·6261133·5
10φ(adf or sBGE,f & sCNSW)p(ysBGE,sCNSW)1624·8471110·3
11φ(adf or sBGE,f & sCNSW)p(yf or sBGE,f & sCNSW)1619·5471105·0
12φ(ady)p(y)1624·7421121·2
Table 3.  Mean estimated apparent survival probabilities (φad), breeding success (BS) and λ for (a) four categories of male and female choughs that fledged (f) and/or settled (s) in regions BGE or CNSW, and (b) all choughs that fledged or settled in each region. nad, nmobs, nfobs, nm and nf indicate the number of individuals included in survival analysis, observations of breeding males and females, and individual males and females, respectively
 FledgedSettlednadφadnmobs (nm)Male BSnfobs (nf)Female BSλ (95% CI)
(a)BGEBGE 370·82 ± 0·03 78 (21)2·1 ± 0·2 41 (12)2·2 ± 0·21·11 (1·01–1·22)
BGECNSW 570·83 ± 0·02 72 (18)2·1 ± 0·2 71 (22)2·0 ± 0·21·02 (0·93–1·10)
CNSWCNSW 560·71 ± 0·04 44 (15)1·5 ± 0·2 61 (19)1·9 ± 0·20·88 (0·78–0·98)
CNSWBGE 260·84 ± 0·03 22 (8)1·9 ± 0·3 16 (7)1·8 ± 0·41·10 (1·00–1·20)
(b)BGE  940·83 ± 0·02150 (39)2·1 ± 0·1112 (34)2·1 ± 0·11·11 (1·02–1·20)
CNSW  820·76 ± 0·04 66 (23)1·7 ± 0·2 77 (26)1·9 ± 0·20·94 (0·83–1·05)
BGE 630·83 ± 0·02100 (29)2·1 ± 0·1 57 (19)2·1 ± 0·21·04 (0·93–1·14)
CNSW1130·79 ± 0·03116 (33)1·9 ± 0·1132 (41)2·0 ± 0·10·99 (0·89–1·10)

Similarly, we tested whether the (small) difference in mean BS between attempts made in BGE and CNSW was better explained by a breeder's natal or settling region. In males, BS declined with age and varied with natal region but not settling region; males fledged in CNSW had lower BS irrespective of where they settled (n = 216 attempts made by 58 individual males: male identity P > 0·1; year F18,195 = 1·5, P = 0·09; age F1,195 = 4·5, P = 0·035; natal region F1,195 = 5·8, P = 0·017; settling region F1,194 = 0·1, P = 0·71, Table 3). In females, BS varied with age, but not natal or settling region (n = 189 attempts made by 58 individual females: female identity P > 0·1; year F21,165 = 1·0, P = 0·49; age F1,165 = 13·3, P < 0·001; age2 F1,165 = 14·1, P < 0·001; natal region F1,164 = 0·6, P = 0·44; settling region F1,164 = 0·2, P = 0·68, Table 3).

We then tested whether the significant spatial variation observed in λ (Fig. 4) would have been detected had we measured demography and λr across individuals that were resident in BGE and CNSW rather than individuals fledged there. We estimated mean BS across all breeding attempts observed in each region, and φad across resident breeders (Table 3). As choughs fledged from all sites spend their pre-breeding years in communal flocks (Still et al. 1986), we made the apparently reasonable assumption that φ1 and φ2 did not vary among choughs fledged in different regions. λBGE and λCNSW were 1·04 and 0·99, respectively, under these assumptions, and did not differ significantly from 1·0 or from each other (Table 3). Therefore, the marked spatial variation in λ that exists within this population was not detectable when demography was measured across current residents of BGE and CNSW.

productivity of recruits

Finally, we tested the common assumption that BS predicts a site's overall productivity (in terms of the number of recruits contributed to the breeding population). We estimated each site's average annual productivity of recruits (Rsite), where Rsite = BSsite·S3site, and investigated to what extent BSsite predicted Rsite. As p did not vary among choughs fledged across regions of Islay, S1site and S3site can validly be used to compare fledgling survival among sites. Although S will underestimate φ, comparisons indicate that the discrepancy is relatively small (S1 = 0·36 vs. φ1 = 0·43; S3 = 0·21 vs. φ3 = 0·22, see Reid et al. 2004).

Across 42 nest sites, BSsite was weakly correlated with S1site (r = 0·28, P = 0·07) but not correlated with S3site (r = 0·14, P = 0·39, Fig. 5a). Rsite varied markedly among sites, from 0·05 to 1·46 recruits per cohort (mean 0·41 ± 0·06, assuming an average year). Pairwise differences in Rsite increased significantly with the geographical distance between sites (Mantel test, r = 0·28, P < 0·001, Fig. 1c). S3site and BSsite explained 88% and 12% of variation in Rsite, respectively. Breeding success was therefore a weak predictor of an individual nest site's estimated contribution of recruits to the breeding population (Fig. 5b).

Figure 5.

Relationships between site-specific parameter estimates for breeding success (BSsite) and (a) ‘observed’ survival to age 3 (S3site), and (b) estimated productivity of recruits (Rsite).

Discussion

spatial variation in breeding success, survival andλ

Across 24 years, chough breeding success varied significantly among individual nest sites. As our analyses only included sites where breeding was monitored in multiple years, among-site variation is unlikely to solely reflect the success of individual attempts or breeders. However, as high-quality individuals are expected to acquire high-quality sites, among-site variation in BS may be exacerbated by correlated variation in breeder quality (Sutherland 1996; Franklin et al. 2000; Sergio & Newton 2003). There was no evidence that BS varied on a large spatial scale across Islay (although neighbouring sites showed similar BS). Variation in BS may therefore reflect local variation in site or territory quality.

The proportion of fledglings observed to survive to ages 1 and 3 varied among individual nest sites and also varied markedly across Islay. A key assumption underlying the interpretation of these patterns is that p did not differ among choughs fledged from different sites. This assumption is reasonable as choughs fledged from all sites spend their pre-breeding years in communal flocks and do not necessarily settle near their natal site (Still et al. 1986; Bignal et al. 1997). Indeed, CMR analyses confirmed that p did not vary across Islay, and that choughs fledged in BGE were more likely to survive than choughs fledged in CNSW as both first-years and adults. Owing to this clear spatial variation in φ and slight correlated variation in BS, the mean asymptotic λ attributable to breeding attempts made in BGE exceeded that attributable to attempts made in CNSW. Furthermore, as φ and λr tended to vary more among years in CNSW than BGE, λr may overestimate the true stochastic growth rate to a greater extent in CNSW than BGE (although this effect may be small, Lande 1988; Reid et al. 2004).

It is clearly valuable to quantify spatial variation in demography and λ. However, it is often less clear how such variation should be measured (Amarasekare 1994; Thomas & Kunin 1999). Any analysis requires the landscape to be subdivided at some spatial scale, and any subdivision will be to some extent arbitrary (Thomas & Kunin 1999). Relevant subpopulation boundaries are particularly difficult to define in relatively continuous landscapes where habitat quality is difficult to quantify (Coulson et al. 1999; Franklin et al. 2000). We approached this problem for choughs on Islay by identifying discrete geographical regions where survival was relatively uniform, and then comparing demography and λ among these regions. Regions were therefore defined by survival models rather than relying on a priori assumptions about the spatial pattern of demographic variation. We did not attempt post-hoc optimization of region boundaries because data were not available from all nest sites. However, we demonstrate that discrete geographical regions of relatively high and low λ can be identified within Islay's chough population. Although habitat varies across Islay, there was no clear expectation that λ would vary markedly in space, or means of predicting more or less productive areas. Furthermore, the spatial variation observed in λ was not detected by measuring breeding success, and became evident only by measuring long-term survival of fledglings of known origin; BS was a weak predictor of a site's estimated productivity of recruits. This situation conflicts with many theoretical and empirical studies where site or habitat quality is assessed in terms of reproduction, with little consideration of variation in offspring or adult survival (Watkinson & Sutherland 1995; Sergio & Newton 2003). Closer focus on survival is desirable, particularly in long-lived species where λ is typically most sensitive to variation in survival (Franklin et al. 2000; Sæther & Bakke 2000).

Year-specific estimates of BS, φ1 and φad were not tightly correlated across BGE and CNSW, suggesting that the incidence or impact of key drivers of demographic variation may vary across Islay. However, λBGE was correlated with λCNSW across years (Fig. 4). This correlation arose because different demographic rates were correlated across regions: BS and φad in BGE were correlated with φ1 in CNSW (r ≥ 0·33). Ringsby et al. (2002) noted that environmental covariation does not necessarily lead to demographic covariation across subpopulations. Our data show that covariation in λr does not necessarily imply covariation in any individual demographic rate, and that an absence of direct demographic covariation does not necessarily imply an absence of covariation in λr.

effects of natal and settlements region

Spatial variation in demography is typically measured across current residents of each focal area (Pulliam 1988; Thomas & Kunin 1999). This approach assumes that an individual's life history is primarily shaped by its current environment. Instead, we measured demographic variation associated with an individual's natal rather than current location. Variation in φad was better explained by an individual's natal region than the region where that individual settled to breed. Therefore, across choughs that settled to breed in CNSW, low φad among individuals fledged locally was ameliorated by higher φad of individuals that had immigrated from BGE. Evidence of links between an individual's natal area and subsequent breeding success was weaker, partly because BS varied relatively little between regions. However, males fledged in BGE bred more successfully than males fledged in CNSW irrespective of where they settled. These patterns could reflect life-long effects of natal conditions on survival and reproduction (as previously suggested in choughs, Reid et al. 2003a), genetic variation (if choughs breeding in BGE are of heritably high quality), or be exacerbated because choughs fledged in BGE acquire the best territories within CNSW. Irrespective of the cause, the link between natal region and subsequent life history meant that large-scale spatial variation in λ would have gone undetected had demography been measured across current residents of each geographical region.

Natural populations are unlikely to behave as strictly defined ‘sources’ and ‘sinks’ as these categories occupy extreme positions on the continuum of possible demographic variation (Thomas & Kunin 1999). However, it is valuable to identify populations that approach these extremes in order to rationalize observed dynamics of habitat occupancy and plan effective management (Pulliam 1988; Dias 1996; Murphy 2001). Sources and sinks, however, are difficult to identify from observed demographic variation. Uncertainties arise where not all demographic rates can be measured, or measured over multiple seasons (Watkinson & Sutherland 1995; Thomas & Kunin 1999). Furthermore, patterns of habitat-specific density dependence must be described in order to distinguish sinks from pseudo-sinks, where intrinsically positive growth is depressed by the presence of immigrants (Watkinson & Sutherland 1995). We show that long-term covariation between natal area and subsequent life history presents an additional problem. Given such effects, the presence of immigrants could obscure the intrinsic demographic characteristics of any focal area. The magnitude of this effect will depend on the relative influence of natal vs. current location on life history, and on the rate at which breeders disperse rather than recruit to their natal area. Under extreme conditions, ‘pseudo-sources’ could arise, where the presence of high-quality immigrants causes an intrinsically negative growth rate to appear positive when demography is measured across current residents. The population dynamic consequences of delayed life-history effects have been discussed in the context of single, spatially homogeneous populations (Beckerman et al. 2002; Lindström & Kokko 2002). Their implications for spatial dynamics require further consideration. In such cases, the definition of sources and sinks as areas where local production of new individuals exceeds local losses, or vice versa (Dias 1996; Johnson 2004), may be too simplistic.

None the less, the BGE and CNSW regions of Islay show tendencies towards a source-sink system (Pulliam 1988). BGE consistently showed a demographic surplus while λCNSW showed a (marginal) deficit on average. Furthermore, BGE exported more fledglings to CNSW than vice-versa (Table 3). Choughs are of conservation concern in Europe and protected areas have been designated on Islay. The existence of previously undescribed regions of higher and lower λ should be considered as designations are reviewed (bearing in mind that areas of lower λ may still facilitate population persistence, Murphy 2001). Further studies are required to determine whether observed spatial variation reflects chough density, foraging conditions, predation or other factors that could themselves be the focus of management, and to investigate the role of spatial variation in regulating this isolated population (Rodenhouse et al. 1997).

Acknowledgements

We thank the numerous people who collected data during the long-term study and Earthwatch, Merial Animal Health, NCC, RSPB, SEERAD, SNH and WWF-UK for funding. Islay farmers generously allowed access to nest sites, and visits were licensed by NCC and SNH. Tim Coulson, Morten Frederiksen, Dan Haydon and Bernt-Erik Sæther commented on manuscript drafts. JMR acknowledges support from Jesus College (Cambridge) and the Royal Society.

Ancillary