Recruitment to a seabird population depends on environmental factors and on population size


Dr Laurent Crespin, Departamento de Ecología, Pontificia Universidad Católica de Chile, 193 Correo 22 Santiago, Chile. Tel.: +56 2354 2938; Fax: +56 2354 2615; E-mail:


  • 1 A novel capture–mark–recapture (CMR) method was used to build a multistate model of recruitment by young birds to a breeding population of common guillemots Uria aalge on the Isle of May, Scotland. Recruitment of a total of 2757 individually marked guillemots over 17 years was modelled as a process where individuals had to move from an unobservable state at sea, through a nonbreeding state present in the colony, to the breeding state. The probabilities of individuals returning to the colony in a given year, at age 2 and 3–4 years, were positively correlated with an environmental covariate, the winter North Atlantic Oscillation index (WNAO) in the previous years.
  • 2 For 2 year olds, there was a negative relationship with breeding population size, suggesting that density dependence operated in this colony through limitation of food or some other resource.
  • 3 Survival over the first 2 years of life varied with cohort, but was unrelated to the WNAO. Mean survival over this 2-year period was high at 0·576 (95% CI: 0·444; 0·708).
  • 4 This high survival, combined with a low ‘local’ survival after age 5 years of 0·695 (0·654; 0·733) and observations of Isle of May chicks at other colonies, suggests that most surviving chicks return to the natal colony before deciding whether to recruit there or move elsewhere.


Population biologists often contrast the population dynamics of short-lived species, for which the strongest determinant of variation in the population growth rate is recruitment, with long-lived species for which it is adult survival, according to sensitivity analyses of matrix population models (Lebreton & Clobert 1991; Caswell 2001). Because of the low sensitivity of population growth rate to variation in the numbers of young birds joining the population, recruitment should be less strongly affected by natural selection in long-lived species and thus more likely to change with environmental conditions (Porter & Coulson 1987; Gaillard et al. 2000). As a consequence, recruitment in long-lived species is also expected to contribute to density-dependent regulation.

Factors affecting recruitment processes, in particular the mechanisms involved in the transition from the nonbreeding to the breeding stage, in long-lived species are not well understood (Danchin et al. 1991; Cooke & Francis 1993). Two characteristics of the demography of colonial seabirds complicate the study of recruitment: philopatry is often low with many surviving chicks not returning to their natal colony to breed (Harris, Halley & Wanless 1992; Hipfner 2001), and the period of immaturity is long, with individuals remaining unobservable at sea for several years and with pronounced individual variation in the age of return to the colony (Croxall & Rothery 1991; Nur & Sydeman 1999). For these reasons, the investigation of recruitment mechanisms requires long-term studies of marked individuals and complex statistical modelling (e.g. Oro & Pradel 2000; Frederiksen & Bregnballe 2001; Cam et al. 2002).

Chicks of common guillemot (hereafter guillemot) Uria aalge Pont. were individually colour-ringed on the Isle of May, Scotland annually and their presence recorded each breeding season 1983–2001. Young guillemots spend at least one year away from the colony after fledging and, after they return, spend at least another year in the colony before they breed for the first time. We modelled accession of birds to the breeding population as proceeding through three stages (cf. Weimerskirch, Brothers & Jouventin 1997; Nur & Sydeman 1999): (1) nonbreeder at sea (from ringing until first return to the natal colony); (2) nonbreeder at the colony (from return to the colony in the second or subsequent year); and (3) breeder, using a multistate capture–mark–recapture (CMR) approach (Lebreton & Pradel 2002). By considering successive stages in the accession to reproduction, our model is more biologically realistic than current CMR recruitment models, and in that sense approaches multistage models used in epidemiology (Portier & Kopp-Schneider 1991; Heidenreich et al. 2002; Luebeck & Moolgavkar 2002). Our aims were to develop a model to investigate the process of recruitment in a colonial bird, to provide robust estimates of recruitment parameters in the guillemot population and to identify environmental factors affecting the recruitment process.

Materials and methods

study area and field methods

Field data were collected on the Isle of May (56°11′N, 2°34′W), south-east Scotland, between 1983 and 2001. Each year the total number of individuals present at the colony were counted in early June and the count converted to breeding pairs using data from concurrent population studies (Harris 1989). The breeding population initially declined, but then increased rapidly (Fig. 1). Because guillemots do not return to the colony when 1 year old, the data set contains observations of 17 cohorts hatched between 1983 and 1999. Each season, samples of chicks were ringed with metal and easily read individually numbered colour rings, with engraved numbers 5 mm tall repeated three times (42–266 per year, total n = 2757), and intensive searches were made throughout the colony for returning birds on an almost daily basis. Birds were typically resighted at distances of < 30 m using a × 60 telescope. Often the numbers on both the numbered and colour-rings were read; in the few instances where these did not agree, or where there was any doubt, the records were discarded. If a bird was seen with an egg or a chick, it was recorded as a breeder, otherwise as a nonbreeder. The category of nonbreeders inevitably contained some birds that were actually breeding, particularly older individuals that were often seen only early or late in the season and could have bred unobserved. For further details of fieldwork procedures see Harris et al. (1992).

Figure 1.

The number of pairs of common guillemots on the Isle of May from 1983 to 2001.

the capturemarkrecapture recruitment model

In typical applications of CMR models to estimate survival in bird populations, individuals are recaptured or resighted during a short observation period each year (e.g. Lebreton et al. 1992). In between they either die or survive. In multistate models, this scheme is generalized to allow surviving individuals to move within a finite set of states (Hestbeck, Nichols & Malecki 1991; Lebreton & Pradel 2002). As a consequence, there are three types of parameters: survival probabilities, transition probabilities (between states) and recapture/resighting probabilities. The approach used a generalization of the recruitment model of Clobert et al. (1994), reformulated as a multistate CMR model (Lebreton, Almeras & Pradel 1999). Further generalizations of this model were made by allowing limited age dependence in some parameters after the age of first return to the colony. The identifiability of the resulting models was checked by profile likelihood methods (Gimenez et al. 2005).

Each year, each bird was considered to be in one of three states: nonbreeder absent from the colony (NBA), nonbreeder present at the colony (NBP), or breeder (B). Individuals returned progressively to the colony from age 2 onwards (Birkhead & Hudson 1977; Halley, Harris & Wanless 1995), and chicks at time of ringing were considered to be in state 1 (NBA). A preliminary season at the colony as a nonbreeder seems to be a prerequisite for breeding on the Isle of May, and as a consequence the direct transition from NBA to B was considered to be impossible. Individuals could only be detected at the colony (i.e. in state NBP or B) and the probability of resighting in state NBA was therefore constrained to zero (i.e. this state was unobservable). As is generally the case in recruitment models with unobservable states (Clobert et al. 1994; Lebreton et al. 2003), constraints on the transition and survival probabilities among states are needed to make parameters identifiable. These constraints correspond to biologically reasonable assumptions. First, we assumed that survival after the age of first possible return to the colony at age 2 was the same across the three states. Second, there was no return to the unobservable state, i.e. from NBP or B back to NBA. However, as NBP comprises both young birds not yet having started their reproductive career and previous breeders skipping one or several seasons, both transitions between states NBP and B were considered. The terms ‘juvenile’, ‘immature’ and ‘adult’ refer to birds of age 0–2, 3–4 and 5+ years, respectively. Some further features of the model and biological predictions are as follows.

Resighting probability (P)

The resighting probabilities in states NBP and B were expected to be constant over time, but to differ strongly between states. The resighting probabilities of adults decline slowly after the age of first reproduction (Crespin et al. in press), but because most resightings concerned nonbreeders, this decline is ignored. Nonbreeders are obvious, spending much time on exposed rocks at the edges of the colony. In contrast, rings are much harder to see on breeders, which occur at high densities (M. P. Harris, pers. obs.). Resighting probabilities were therefore predicted to be much higher for state NBP than B, explaining much of the observed individual heterogeneity in P.

Survival probability

The models had different survival probabilities for juveniles (up to 2 years old), immatures (3–4 years old) and adults (5–18 years old). Although a senescent decline in survival of adults has been demonstrated in this population (Crespin et al. in press), data from older birds were sparse here, so senescence was ignored. No birds were seen at age 1, so it was only possible to estimate a compound probability of survival over the first 2 years of life. This juvenile survival probability was allowed to vary over time, i.e. across cohorts. This value represents a ‘local’ survival probability as it is the product of survival probability and the probability of coming back to the study colony. Immature and adult survival probabilities were allowed to vary over time.

Transition probability

To model the initial phase of recruitment, the transition probability from NBA to NBP was allowed to vary with age. This probability was constrained to zero for birds less than 2 year olds and modelled independently for 2 year olds, 3–4 year olds and 5–18 year olds. We expected the transition probability from B to NBP to be constant over years, because the proportion of nonbreeding individuals among breeders was low and relatively constant (5–10%, Harris & Wanless 1995). However, to allow for some variation in the probability of nonbreeding with age, records for birds aged 5–9 years were treated separately from those aged 10–18 years, as most individuals are likely to have started breeding by age 9 (Harris, Halley & Swann 1994). The reverse transition, from NBP to B, corresponding to the recruitment probability in the strict sense, was assumed to differ between 4 year olds (the youngest age at which guillemots were observed to breed) and older birds. Considering a constant transition probability from some age onwards makes parameters identifiable without having to assume that the transition probability is equal to 1 at some given age (contra Clobert et al. 1994). The cumulative effect of a constant transition probability is that the proportion of individuals having bred approaches 1 asymptotically (Fig. 2).

Figure 2.

Cumulative age-specific percentages of breeders for a constant recruitment probability (transition from NBP to B) equal to 0·04 for age 4 years and to 0·31 from age 5 years onward.

Parameters that varied over time might have done so in response to the pronounced changes in breeding population size over the study period (Fig. 1), and/or variation in environmental conditions. The effects of changes in breeding population size (BP) and in the environment were examined for all parameters that varied over time in the selected model (see next section). Relating variation in seabird demography to particular aspects of the marine environment is difficult (Croxall & Rothery 1991; Harris et al. 1997; Gaston & Smith 2001). The effect of the North Atlantic Oscillation (NAO; Hurrell, Kushnir & Visbeck 2001), which is an integrated measure of environmental conditions in the north-east Atlantic Ocean, was then examined. Winter NAO index (Hurrell 1995; hereafter WNAO) was used because the NAO is primarily a winter phenomenon (Hurrell 1995), most guillemot mortality occurs in winter (Birkhead & Hudson 1977) and WNAO is well correlated with environmental conditions in the North Sea (Ottersen et al. 2001) where most Isle of May guillemots winter (Harris & Swann 2002). Previous studies have shown lagged effects of environmental conditions on seabird demography (Thompson & Ollason 2001; Barbraud & Weimerskirch 2003), so the WNAO index lagged by 1–3 years (WNAO-1 to WNAO-3) was also used as covariates.

goodness of fit and model selection

A typical capture history, over the 19 years encompassed by the study, consisted of marking in state 1 (NBA) and resightings in state 2 (NBP) or state 3 (B), e.g. 0010020002030000000 (with 0 indicating no resighting in that year) for a chick marked in 1985 that was first seen ashore in 1988 and breeding in 1994. The first part of the capture history of each bird (001002 in the example above) provides all the information needed to estimate the survival over the first 2 years of life and the first recruitment parameter (transition from NBA to NBP). These parameters may vary according to age and/or years. A sufficiently general model then has at least as many parameters as pieces of statistical information for this first part of the capture history. As a consequence, there is no information left to assess the goodness of fit (GOF) of the model in this part of the capture history. The second part of the capture history of each individual, obtained by removing the first capture and restricted as a consequence to states 2 and 3 (0000020002030000000 above), contains all the remaining information. In the corresponding part of the model, there was a limited age dependence, allowing survival and transition probabilities to be different between juveniles, immatures and adults. The model was thus a generalization of that of Clobert et al. (1994). Hence an approximate GOF test of the general model consisted of a GOF test of an Arnason–Schwarz model (Arnason 1973) with time dependence and two age classes, applied to the second part of the capture histories (R. Pradel personal communication). Schematically, the GOF test of an Arnason–Schwarz model consists of two components: test3G, which tests the dependence on the previous capture history for individuals released at same time and testM, which tests the dependence on current capture (Pradel, Wintrebert & Gimenez 2003). The GOF test was carried out using program U-CARE (Choquet & Pradel 2002), skipping the subcomponents WBWA and TEST3.SR to account for the age dependence. As the GOF test was significant (see below), a variance inflation factor was calculated (Lebreton et al. 1992). A GOF test on the second part of capture histories neglecting the states was used for comparison; the stratification into states NBP and B was expected to improve the fit by removing heterogeneity.

Model selection started from the general model described above, with age and year effects in the first part, and year and limited age effects in the second part. All models were fitted by maximum likelihood using program m-surge (Choquet et al. 2004). Supplementary Appendix S1 contains m-surge model notation for all models fitted. Models with lower QAIC (Akaike's Information Criterion corrected for lack of fit) were preferred, as being a good compromise between bias and precision (Burnham & Anderson 1998). The effects of external covariates were also tested using analysis of deviance (anodev, Skalski, Hoffmann & Smith 1993), and the equivalent of a squared correlation coefficient, denoted R2, was calculated based on differences in deviance between models with, and without, time dependence and with the covariate concerned. All parameter estimates are given with 95% confidence intervals.


The numbers of observed transitions between the three states are summarized in Table 1. Because survival was assumed to be equal between the three states, these numbers reflect the product of the probability of resighting and the probability of transition between two states. Thus the low numbers moving into state B reflect the very low detectability in this state (i.e. a low resighting probability for breeders). As a consequence, the near absence of observed transitions from B to B does not imply a transition probability (probability of remaining a breeder) close to zero. On the contrary, the observed number of transitions from B to NBP, which is low relative to NBP to NBP, indicates a low probability of nonbreeding for established breeders and thus a high probability of remaining a breeder.

Table 1.  A summary of numbers of the observed transitions between states in the data (cumulative over several intervals). In the model, birds can move between three states: from the state NBA (nonbreeder absent from the colony, i.e. a nonobservable state) to the state NBP (nonbreeder present at the colony), from NBP to B (breeder at the colony) or from B to NBP. The transitions NBA to B and B to NBA were assumed to be impossible
State of departureNos releasedState of arrival
NBA27570850 7
B  640 37 1

Results of the GOF test are summarized in Table 2. As expected, the stratification into NBP and B improved the fit considerably, with the ratio χ2/d.f. decreasing from 2·286 to 1·526. The strong trap dependence found in the single-state test (inline image, P < 0·0001) was mostly explained by transitions between the two states (test component M ITEC accounts for remaining trap dependence). Nevertheless, although not pronounced, the lack of fit remained significant and was uniform across the various test components, as indicated by the relative constancy of the ratio χ2/d.f. (last column of Table 2) except for the component for immediate trap dependence (M ITEC). Trap dependence is here taken in the statistical sense, as a tendency for animals just sighted to be resighted at the next occasion with a different probability than animals that were not sighted. The ratio χ2/d.f. = 1·526 was used as an estimated variance inflation factor to calculate QAIC and adjust standard errors.

Table 2.  Results of goodness-of-fit tests of the general multistate CMR recruitment model. Test components as in Pradel et al. (2003). For comparison, the overall result of the GOF test of the single-state model is also given. See text for details
Test componentχ2d.f.Pχ2/d.f.
3G SM 51·88043  0·1661·207
M ITEC 27·79911  0·0032·527
M LTEC  8·902 5   0·1131·780
LRT (JMV vs. AS) 16·69710   0·0441·670
Sum105·27869  0·00251·526
Single-state137·135460< 0·000000012·286

First, an appropriate model for resighting and survival probabilities was selected (Table 3). Resighting probabilities could easily be made constant over time (model 1 vs. 2). For the survival probabilities, differences among the three states were not needed (model 2 vs. 3). Removing the year-to-year variation in the adult survival probabilities improved the model considerably (model 3 vs. 4, decrease in QAIC of about 22). The same was true for immatures (model 3 vs. 5, decrease in QAIC of about 24) and for juveniles but less so (model 3 vs. 6, decrease in QAIC of about 5). A better model was obtained by keeping the year-to-year variation only for juvenile survival (model 7). Further attempts to simplify the survival probabilities resulted in considerably worse models (increases in QAIC of at least 20 for models 8–11, Table 3).

Table 3.  Model selection for resighting and survival probabilities. Model selection was performed according to QAIC (Burnham & Anderson 1998). The better the model the lower the QAIC. Models are ranked by QAIC for the sake of simplicity. Notations about effects were taken from Lebreton et al. (1992). At this stage, the most general model was used for transition probabilities. The preferred model at this stage is shown in bold
Age 1–2Age 3–4Age 5–18
  1. + = only additive combinations between covariates in the model; * = a model with full interactions between covariates; t = categorical year effect; c = constancy over time; np = number of identifiable parameters in the model.


Hence, model 7 was used as a starting point for modelling the transition probabilities (Table 4). There was no evidence for year-to-year variation in the transition B to NBP (model 7 vs. 12, decrease in QAIC of about 26), nor in the transition NBP to B (model 7 vs. 15, decrease in QAIC of about 23). Indeed, removing the year-to-year variation from both these transitions led to a much lower QAIC (model 16, Table 4). Further, the removal of the year-to-year variation in the transition from NBA to NBP for adults (age 5–18, model 16 vs. 20) led to a further decrease in QAIC of about 9.

Table 4.  Model selection for transition probabilities. Conventions are as Table 3 with, for the probabilities of transition from NBA to NBP, the effects included for the three age classes (2, 3–4, 5–18 years) being shown separated by commas. The preferred model at this stage is shown in bold
ModelP and survivalTransitionDeviancenpQAIC
13Model 7c,t,tAge*tAge*t8742·161045936·81
14Model 7t,c,cAge*tAge*t8756·81925922·41
7See Table 3t,t,tAge*tAge*t8648·401205907·37
18Model 7c,t,tAgeAge8810·46565885·56
15Model 7t,t,tAgeAge*t8695·05935883·94
12Model 7t,t,tAge*tAge8671·98995880·82
17Model 7t,c,cAgeAge8831·99445875·67
22Model 7t,t,ccAge8790·98575874·78
19Model 7t,c,tAgeAge8789·12575873·58
16Model 7t,t,tAgeAge8716·28725855·85
21Model 7t,t,cAgec8749·00575847·29
20Model 7t,t,cAgeAge8742·19585844·83

Model 20 was thus the starting point to consider the effect of covariates on parameters where year-to-year variation was apparent (Table 5). At this stage, including WNAO or BP in the juvenile survival led to a poorer fit (models 23 and 24 vs. model 20, increase in QAIC of more than 30). Generally, the inclusion of only one variable (BP or only one lag of WNAO) in the juvenile and immature transitions from NBA to NBP did not lead to a reduced QAIC. For instance, including BP for the juvenile (age 2) and immature (age 3–4) transitions led to an increase in QAIC (model 20 vs. 25 and 28). However, both juvenile and immature transitions seemed best modelled with the inclusion of WNAO with lags 0–2 (model 20 vs. 26 and 29 and see Appendix S1). In contrast, including BP in the model led to a trivial increase in QAIC for juveniles but a large increase for immatures (models 25 vs. 20 and models 28 vs. 20). Including both covariates BP and WNAO in an additive way decreased the QAIC score of the model (models 27 and 30 vs. model 20). At this stage, however, the inclusion of BP for the immature transition was questionable (model 29 vs. 30, difference of 1·34 units of QAIC). Given the similarity of the structure of the transitions for juveniles and immatures, a parameter indicating additivity between juveniles and immatures (denoted a) was added in the transitions and this improved the fit considerably (model 32 vs. model 20, a decrease of about 10 units of QAIC). Adding the interaction between this additive parameter and BP improved the fit further (model 32 vs. model 33, a decrease of about 12 units of QAIC), indicating different slopes of the relationship with BP. The slope of the relationship between BP and the immature transition was not significant [β(BP) = −0·365 (−1·849; 1·118)] and removing the influence of BP on the immature transition led to a better model with one parameter less (model 34, QAIC = 5821·46). The addition of interactions between a and WNAO, WNAO-1 and/or WNAO-2, did not lead to any improvement (models 35 and 36 vs. model 34). Given that the inclusion of neither of the covariates for juvenile survival improved the fit further (models 37 and 38 vs. model 34), model 34 was the preferred model.

Table 5.  Testing the effect of covariates on survival of juveniles and transition probabilities to the nonbreeder state present in the colony. The preferred model is shown in bold
ModelSurvival age 1–2NBA→NBP age 2NBA→NBP age 3–4DeviancenpQAIC
  1. np = number of identifiable parameters in the model; t = categorical year effect; c = constancy over time; a = additive parameter between juvenile and immature transitions on logit scale; WNAO, WNAO-1, WNAO-2 = North Atlantic Oscillation index with a lag of, respectively, 0, 1 or 2 years; WNAO(0 to x) = all lags until x are used in the model; BP = breeding population size; BP(juv) = breeding population size is used to model only the juvenile transition.

31tWNAO(0–2) + BPWNAO(0–2) + BP9098·279316024·175
37BPVAO(0–2) + BP(jus) + aVAO(0–2) + BP(jus) + a8915·72175876·54
38WNAO(0–2)VAO(0–2) + BP(jus) + aVAO(0–2) + BP(jus) + a8893·72195866·13
30ttWNAO(0–2) + BP8773·537475843·34
32tWNAO(0–2) + BP + aWNAO(0–2) + BP + a8807·262325835·469
27tBP + WNAO(0–2)t8751·98475829·24
35tWNAO(0–2)*a + BP(juv)WNAO(0–2)*a + BP(juv)8783·79355826·09
33tWNAO(0–2) + BP + a + BP.aWNAO(0–2) + BP + a + BP.a8785·65335823·31
36tWNAO(0–1) + BP(juv) + a*a.WNAO-2WNAO(0–1) + BP(juv) + a*a.WNAO-28784·43335822·50
34tWNAO(0–2) + BP(juv) + aWNAO(0–2) + BP(juv) + a8785·88325821·46

All parameters were identifiable in the preferred model, confirming that the inclusion of limited age dependence did not jeopardise the estimation of parameters. Both juvenile and immature transitions from NBA to NBP increased when the WNAO index was high during the previous years [β(WNAO) = 0·119 (0·057; 0·182) β(WNAO-1) = 0·095 (0·031; 0·159) and β(WNAO-2) = 0·330 (0·236; 0·424), see Fig. 3]. The juvenile transition decreased at high population size (BP) [β(BP) = −4·495 (−6·197; −2·794), see Fig. 3]. The combination of WNAO, WNAO-1, WNAO-2 and BP explained 95% of the deviance for the juvenile and immature transitions (Table 6). In contrast, none of the covariates used explained a significant proportion of the variation in survival over the first 2 years. As predicted, the resighting probability in state NBP [0·786 (0·724; 0·838)] was higher than in state B [0·155 (0·103; 0·225)], and the transition probability from B to NBP was relatively low compared with the probability of remaining a breeder [transition B to NBP: 0·440 (0·294; 0·596) for 5–9 year olds and 0·178 (0·065; 0·402) for 10–18 year olds]. The transition probability from NBP to B was lower for 4 year olds [0·036 (0·009; 0·135)] than for older birds [0·312 (0·243; 0·390)]. The mean transition probability from NBA to NBP was 0·104 (0·057; 0·151) for juveniles. This average was calculated only over the first 13 cohorts because the last four estimates of transition were considered suspect as they were an order of magnitude lower and their standard errors suspiciously small (range of estimates: 0·0214–0·2309 vs. 0·0044–0·0087). This was likely due to many individuals marked as chicks in the last years of the study not having yet returned to the colony to breed. Over these 13 years, the mean transition probability from NBA to NBP was 0·449 (0·343; 0·554) for immatures. For adults the estimate was 0·260 (0·113; 0·491).

Figure 3.

Transition probabilities from nonbreeder absent (NBA) to nonbreeder present (NBP). (a) Juvenile and immature transitions plotted against number of breeding pairs at the colony (BP); (b) the juvenile transition plotted against the winter WNAO index; (c) WNAO index 1 year before (WNAO-1); and (d) the winter WNAO index 2 years before (WNAO-2). Point estimates are from a model not containing any of the covariates, but allowing year-specific transitions (model 20, Table 4). Only results from the first 13 cohorts are shown (see text).

Table 6.  Results of ANODEV tests for the covariates used: the number of breeding pairs (BP), the winter North Atlantic Oscillation Index with different lags (WNAO, WNAO-1 and WNAO-2)
ParameterCovariateFPR2 (%)
Juvenile survivalBP 1·63   0·22 9·80
WNAO(0–2) 1·45   0·2725·10
Juvenile and immature transitionBP16·26   0·00150·40
WNAO 0·93   0·35 5·50
WNAO(0–1) 1·97   0·1720·85
WNAO(0–2) 6·10   0·00756·67
WNAO(0–2) + BP69·94< 0·000195·56

Juvenile survival for the first 13 cohorts over the first 2 years of life varied from year to year (Fig. 4), with the mean being 0·576 (0·444; 0·708). This was likely a conservative estimate since the four discarded estimates were all higher than this average (range of estimates: 0·83–1·00). Annual immature survival was estimated at 0·863 (0·805; 0·906), and adult survival (age 5 years onwards) at 0·695 (0·654; 0·733). This low adult survival is discussed below.

Figure 4.

Estimates of juvenile survival. These estimates of survival encompass the first 2 years of life as no common guillemot is known to have returned to its natal colony before 2 years old. Estimates are from the preferred model (model 34, Table 5). Only the 13 first cohorts are plotted (see text for details). Dashed lines represent the 95% confidence interval limits.


validity of the model and parameter estimates

This model represents, as far as we are aware, the first published attempt at modelling recruitment to the breeding population as a three-stage process using multistate CMR techniques. It is still preliminary, with several biologically relevant aspects not included due to lack of quantitative data, but most of the parameter estimates seem reasonable. In accordance with predictions, resighting probability was estimated to be much higher for nonbreeders than for breeders, reflecting the marked difference in the ease with which the two categories can be observed. The estimated transition probabilities from NBP to B are consistent with the limited information on age at first breeding. Harris et al. (1994) found that recruitment to two Scottish populations took place over several years, with some birds apparently recruiting when up to 15 years old and the cumulative age-specific percentages of breeders calculated from the transitions show indeed such a protracted recruitment process (Fig. 2). The estimated transition probabilities from B to NPB are high compared with the 5–10% recorded as nonbreeders by Harris & Wanless (1995). However, their results were based on intense daily observations in a small study area throughout the season. In contrast, birds in this study, which covered the entire colony, were often only recorded once or twice during a season, thereby substantially increasing the likelihood of misclassifying a breeder as a nonbreeder.

Adult survival at 0·695 was clearly underestimated. This estimate was based on all birds 5 years or older and, because most birds observed were nonbreeders (cf. Table 1), this group contributes strongly to this estimate of local survival. A separate study in the same colony over the same period of birds ringed as breeders indicated that annual survival was about 0·95, in accordance with previous findings on the Isle of May and elsewhere (Hatchwell & Birkhead 1991; Sydeman 1993; Harris et al. 2000; Crespin et al. in press). The proportion of transients was significantly higher among nonbreeders than among breeders in the data set (0·42 vs. 0·20, one-sided Fisher's exact test P = 0·0018), indicating substantial permanent emigration among nonbreeders. The low local survival of (mainly prebreeding) adults, combined with the high estimate of juvenile survival (see below), suggests that most or all surviving chicks returned to the Isle of May during at least 1 year before deciding whether to recruit there or disperse to another colony. Empirical support for this comes from the sightings at other colonies of 35 Isle of May colour-ringed chicks. Of these, 24 (69%) were also seen on the Isle of May, a figure within the 95% CI of the expected 28 given the average resighting probability of 0·786 for birds in the state NBP. Such large-scale visiting of the natal colony before emigrating has also been recorded in the south polar skua Catharacta maccormkii, herring gull Larus argentatus and common tern Sterna hirundo (Ainley, Ribic & Woods 1990; Vercruijsse 1999; Dittmann, Zinsmeister & Becker 2005). This emigration after initial return to the colony but prior to recruitment is likely to have resulted in a negative bias in estimated adult survival.

Our results suggest that a substantial proportion, perhaps as many as 25%, of immature guillemots that return to the Isle of May later move to other colonies. This agrees well with an earlier assessment of 25–30% of surviving chicks from the Isle of May emigrating to other colonies (Harris, Halley & Wanless 1996). The estimate of annual survival of 3–4 years old immatures (0·863) is difficult to compare with previous estimates but appears biologically realistic, whereas the juvenile survival estimate for the first 2 years of life seems high (see below). The transition estimates from NBA to NBP are the first published, and it is thus difficult to evaluate whether they are reasonable. A further biological complication is that resighting probabilities of nonbreeders would be expected to increase with age, as the amount of time immature guillemots spend in the colony increases from age 2 up to age 5 years (Halley et al. 1995). However, it is unclear if, or how, this would affect parameter estimates.

juvenile survival of guillemots

Many of the published estimates of survival of juvenile or immature seabirds are imprecise being based on return rates of chicks from ringing until recruitment age with no allowance made for changes in resighting probability (Nur & Sydeman 1999). Although it is difficult to compare estimates obtained using different methods, our mean estimate of 0·576, for local survival over the first 2 years of life seems high. Combined with our estimate of immature survival, this gives a compound local survival until age 4 years of 0·429. A calculation of mean survival to age 4 for the same 13 cohorts using the method of Rothery (1983), which assumes that all birds becoming breeders at the colony have come back at age 4 was 17% lower at 0·358, showing the improvements resulting from using a more up-to-date and realistic modelling framework.

In common guillemots, Birkhead & Hudson (1977) reported values of survival to the fifth year ranging from 0·270 to 0·411, and Hatchwell & Birkhead (1991) showed that the higher of these estimates was more consistent with observed population growth. Survival to breeding age has been estimated for four alcids: 40% for Brünnich's guillemot Uria lomvia and razorbill Alca torda, 33–52% for Atlantic puffin Fratercula arctica and 36% for black guillemot Cepphus grylle (Harris & Wanless 1991; Gaston et al. 1994; Lyngs 1994; Frederiksen 1999). Thus overall our model appears to give reasonable estimates of survival.

factors affecting return to the colony

In the preferred model, transition probabilities from NBA to NBP were much higher for immatures (mean 0·449) than for juveniles (mean 0·104). Both juvenile and immature transition probabilities were influenced by an additive combination of the winter WNAO index in the same and the two previous years (covariates WNAO, WNAO-1 and WNAO-2). For juveniles, there was also a negative relationship between the number of breeding pairs and the transition, i.e. the more breeding pairs, the lower the probability of coming back to the colony to enter the pool of prebreeders on the Isle of May (Fig. 3). This negative relationship suggests increasing competition for some resource when population size is high. Shortage of good nest-sites can be a limiting resource as the deaths of large numbers of breeders can be followed by an increase in recruitment of young breeders (Potts, Coulson & Deans 1980; Porter & Coulson 1987). Competition for a limited number of high-quality sites has also been implicated as leading to an observed decline in mean breeding success for guillemots on the Isle of May (Kokko, Harris & Wanless 2004). However, younger prebreeding guillemots on the Isle of May do not hold sites (Halley et al. 1995) so this explanation cannot apply to our results. Perhaps, there may be competition for food between breeders and prebreeders (Coulson, Duncan & Thomas 1982). Some other studies have found that recruitment increased with breeding population size (Oro & Pradel 2000; Frederiksen & Bregnballe 2001), suggesting that under some circumstances conspecific attraction can be more important than competition for food or nest sites.

Ocean climate has been shown to affect the population dynamics of seabirds in the north-east Atlantic (Thompson & Ollason 2001; Durant, Anker-Nilssen & Stenseth 2003; Frederiksen et al. 2004b). The positive association between the probability of a guillemot returning in a given year and the WNAO index is intriguing. When the WNAO index is high, winters in north-west Europe tend to be mild, wet and windy (Hurrell 1995). Breeding on the Isle of May tends to be earlier following winters when NAO is high (Frederiksen et al. 2004a) so overall such conditions appear good for guillemots. The mechanism behind these associations is unclear, partly because the distribution, ecology and diet of guillemots during the nonbreeding season are imperfectly known. We have not identified any environmental covariates of juvenile survival, probably because our estimates covered a 2-year interval, whereas the values of covariates encompassed only 1 year. However, future studies involving lightweight data loggers (Wilson et al. 2002) combined with the use of fine-scale environmental covariates, may allow identification of proximate factors affecting survival and recruitment of prebreeding guillemots.


We thank Duncan Halley and many other fieldworkers for their contribution to this data set, the UK Joint Nature Conservation Committee for financial support and Scottish Natural Heritage for access to the Isle of May. Emmanuelle Cam and an anonymous reviewer improved the manuscript with their comments. For part of the study M.P.H. was supported by a Leverhulme Emeritus Fellowship and L.C. by the program FONDAP-FONDECYT 1501-001. L.C. and J.-D.L. acknowledge a grant from the international program of scientific collaboration between CNRS and CONICYT (2002).