Intermittent breeding in the short-tailed shearwater Puffinus tenuirostris

Authors


Dr J.S. Bradley, Biological Sciences, Murdoch University, Western Australia 6150.

Abstract

1. At one colony of short-tailed shearwaters in Bass Strait, Australia, all birds breeding have been recorded individually each year for over 50 years. Among individuals known to be alive and to have bred before, 14% of each sex were not present at their breeding colony, on average, in any one year.

2. A further 15% of males and 13% of females (a significant difference), known to have bred before, were present but not associated with an egg in any one year.

3. Intermittent breeding was associated with pair bond breakdown and with a reduced ability to raise offspring, even when years of absence were allowed for.

4. The frequency of attendance, laying and successful rearing of progeny to fledging increased with age in the early years of breeding. However, this was followed by a decrease, except in the case of laying frequency, which continued to increase throughout life.

5. Our analysis indicated that intermittent breeding in short-tailed shearwaters did not result from individuals implementing trade-offs between the effort required for breeding success and their breeding life span. Rather, we suggest that individuals of higher quality are able to breed more frequently than others without any compensatory reduction in either their annual breeding success or their overall breeding life span.

Introduction

Many interpretations of the life histories of animals include a trade-off between the cost of current investment in offspring and a parent's future survival and reproductive opportunities (Williams 1966). In long-lived species, of which some seabirds are among the best-studied examples, even a small change in the adult survival rate can greatly affect the number of future breeding attempts (Stearns 1992). In long-lived mammals, such as the ungulates, several long-term studies have identified the failure of reproductive females to calve in some seasons (e.g. Clutton-Brock & Albon 1989; Saether 1997), and this has often been associated with high population density and failure to regain body weight after lactation (Clutton-Brock et al. 1997; Saether 1997). Seabirds often live in stochastic environments where in some years the difficulties in producing offspring may be very great (Erikstad et al. 1998). Those seabirds that rear only a single young each year (e.g. all petrels) lack the flexibility to adjust their clutch size at the outset, but their low annual fecundity, coupled with high adult survival, means that any loss of investment in an offspring in difficult years may represent only a small proportion of their total lifetime reproductive output. In situations where they experience or foresee difficulties, it might be expected that some breeding seabirds would occasionally safeguard their own survival by ceasing to breed at any stage between remaining absent from the breeding site and the abandonment of large nestlings (Erikstad et al. 1998). This paper explores this possibility through an examination of intermittent breeding by short-tailed shearwaters (Puffinus tenuirostris Temminck).

Among birds that exhibit fidelity to nesting sites, failure to breed can be manifest in two ways. First, an individual previously recorded with an egg may not be recorded at its breeding site, but be recorded there in a subsequent year. Second, a bird recorded at the breeding site may not be seen to be associated with an egg. The absence from their breeding colony of individuals known to be alive and also known to have bred in earlier years has been documented in several seabirds (Coulson 1984; Harris & Wanless 1994; Calladine & Harris 1997) including annually breeding shearwaters (Brooke 1990; Mougin, Jouanin & Roux. 1997). Procellariform seabirds, such as shearwaters, lay a single egg which is not replaced if lost (Warham 1990). Most species within this order have an annual breeding cycle, but a number of albatrosses have a more protracted breeding period such that completion of chick rearing in one year does not allow sufficient time for them to initiate breeding in the following year.

The causes of intermittent breeding are not understood. It has been suggested that parents exert less reproductive effort, through smaller clutches, than would be required to produce the maximum number of surviving fledglings in a given year (Stearns 1992; Monaghan & Nager 1997). This would enhance the fitness of the parent by increasing the probability that it would survive to the next breeding season. Where habitat quality at the start of the breeding season is poor, the clutch size may be lowered to conserve adult condition, and failure to breed could represent an extension of this mechanism to a zero clutch size. Alternatively, failure to breed in otherwise annually breeding species might result from a strict adherence to a bet-hedging strategy, with intermittent breeding arising from contingent events in the experience of each individual, such as loss of mate, stress caused by storms or starvation, disease or parasite effects. In this second scenario, failure to initiate breeding is not any extrapolation of habitat quality. Of course, these two models represent the ends of a spectrum rather than a dichotomy. The following analysis attempts to establish whether short-tailed shearwaters breed intermittently as a result of bet-hedging or trade-offs. The approach adopted is to examine the distribution of intermittent breeding between years and between individuals, and to estimate correlations between reproductive parameters both between years and within individual life histories.

Methods

Field protocols

Short-tailed shearwaters nest in burrows around south-eastern Australia, especially on islands in Bass Strait, Tasmania. The single egg laid by a monogamous pair is not replaced if lost or unsuccessful. Almost all eggs are laid between 23 and 28 November each year (Meathrel et al. 1993). In a colony of 100–200 pairs on Fisher Island, in Bass Strait (40°10′S, 148°16′E), all individuals located in all burrows on the island have been recorded each year since 1947, together with the presence of any egg or young. All adults and large young are identified by durable individual leg-bands and a double-banding programme ensures that band loss due to wear does not render individuals unidentifiable (Wooller, Skira & Serventy 1985). Many individuals banded as young on Fisher Island later returned there to breed. These known-age recruits comprise up to half of the breeding population (Serventy & Curry 1984; Wooller et al. 1990)

Non-attendance by an individual known to have been associated with an egg or young on Fisher Island in an earlier year was recorded if that individual was recorded as alive in a later year. In addition, non-attendance was recorded for any birds that were present during the prelaying period but were not subsequently recorded in a burrow; these were very few. Records of non-layers were most unlikely to include many eggs lost shortly after hatching because the extreme synchrony of laying meant that eggs were recorded very soon after they had been laid. Whilst it was feasible that a breeding bird may never have been recorded as present because its foraging trips coincided with burrow checks, this was an unlikely event, again because the synchrony of laying and incubation allowed accurate timing of checks in relation to shift changeover periods. Evidence of such detection failures would lie in absences recorded for an individual whose mate of the previous year was recorded with an egg or nestling, but no partner was recorded for that mate. Of just over 4000 breeding records for the Fisher Island study, only seven records fell into this category; three were with eggs only and four with nestlings. This suggests that the rate of failure to detect breeding birds was extremely low.

Statistical methods

A Monte-Carlo procedure was used to examine whether annual variation in the proportion of individuals in attendance at the colony or with an egg, differed from that expected under a simple model, where the underlying probabilities of attending or laying remained constant over time or exhibited a linear trend. On the basis of the model selected, a theoretical probability of attendance or the presence of an egg was assigned to each year. For each of a thousand replicates in each year the states of each breeding bird (attending or non-attending, etc.) was assigned by a binomial random number generator, using the appropriate probability. Thus, for each year, the likelihood was estimated of the observed numbers in each state.

Logistic regression analysis was performed using the random effects regression option of the EGRETTM logistic regression package (SERC 1990) to fit a logistic binomial regression model.

Results

Breeding by previous partners of absent birds

Overall, the mean proportions of males and females known to have bred before but recorded as absent (excluding their first breeding and their last attendance record) in any one season were 13·7% ± 1·2 and 13·6% ± 1·3; these figures do not differ significantly between the sexes. Similarly, the mean annual rates for attendance without an egg were 17·9% ± 1·8 and 14·5% ± 1·5 and there was a significant difference between the sexes in this case ( inline image = 5·10, P = 0·024).

The rates of absence (males = 30·2%, females = 29·7%) of their mates of the previous year, for individuals themselves absent, were significantly higher (males inline image = 59·8, P < 0·001, females inline image = 56·2, P < 0·001) than the overall rates, indicating that the absence of an individual is not independent of the absence of its mate. However, in most cases the two birds were not absent at the same time (Table 1), which strongly suggests that a substantial proportion of recorded absences reflects a response to the circumstances of an individual rather than the circumstances of its mate. The rate of attendance without eggs (males = 50·6%, females = 38·4%) by the previous mates of absent individuals was also markedly and significantly higher than the overall rates (males inline image = 164·3, P < 0·001, females inline image = 111·8, P < 0·001), but this was entirely a result of the high incidence of attendance by those previous mates without a new partner being present (Table 1). The rate at which previous mates attended with a new partner but without an egg (males = 18·8%, females = 12·7%) did not differ from that in the general population. This indicates that the absence of an individual has a profound impact on the reproductive activity of its mate of the previous year. Often, that mate was also absent or, more frequently, attended but failed to breed with its new partner.

Table 1. . The fate of their previous mate for shearwaters absent from their breeding colony and for shearwaters present, but without an egg, whose mates were absent or dead
Status of previous mateAbsent
Total
Proportion malePresent with no egg
Total
Proportion male
Absent but alive2160·54540·69
Dead1910·50670·60
Attends with no mate1720·46  
Attends with new mate; no egg510·47  
Attends with new mate; egg only1490·62  
Attends with new mate; fledgling1330·54  

There was a significant difference between the sexes (Table 1) in the activity of the previous mates of absent birds ( inline image = 10·31, P = 0·006), in that the prior mates of absent males were more likely to be found with an egg or nestling than those of absent females. This indicates that when a female fails to attend, but her previous mate is present, he is less likely to form another bond than the former female partner of an absent male. Such a circumstance might arise if a male tended to wait longer for an absent female, whereas a female accepted an alternative male partner more rapidly.

Table 1 shows the status of the previous mates of individuals absent from their breeding colony or present, but without an egg. The relationship between the sex of the non-breeding individual, its absence or presence, and the status of its previous mate (dead or alive) was explored using loglinear analysis. No significant three-way effects were detected and the only significant two-way effect was that between the sex of the birds and whether they were present or absent ( inline image = 5·30, P = 0·021). That is, an individual present but without an egg and with its previous mate absent was more likely to be male than was an absent bird with an absent mate. This is another measure of the effect considered in the previous paragraph. No further comparisons between these groups could be explored because of the complications introduced by the acquisition of new mates.

This indicates the presence of strong contingent effects upon intermittent breeding. The absence of its mate more than doubles the probability of absence for an individual, or vice versa, and has an even stronger effect on the likelihood of it not having an egg. Nonetheless, a substantial proportion of individuals whose previous partners were absent, bred successfully with new partners, although they sometimes reverted to their old partners in subsequent years.

Correlations in annual variation of reproductive parameters

Substantial annual variation in reproductive success has been observed in the short-tailed shearwater population on Fisher Island (Fig. 1), with both local and more broadscale phenomena implicated. No strong relationships have been detected between measures of annual reproductive success and climatic or oceanographic oscillations (R.D. Wooller & J.S. Bradley, unpublished information). The annual variation in intermittent breeding, measured by the proportions of individuals that had bred there before that were in attendance and that had eggs, was less marked (Fig. 1a). The frequency of catastrophic breeding events was greater at this colony during the 1980s than in previous decades (Fig. 1c), at least in part as a result of local predation by carnivorous rodents. In addition, the proportion of individuals present that were associated with eggs exhibited a fairly steady and significant increase over the period of the study (F1,41 = 29·67, P < 0·001; proportion = year × 0·006 ± 0·001 + 0·716 ± 0·025). While this could reflect some minor changes in the intensity of monitoring over a very long period (Serventy & Curry 1984), the synchrony of laying and lack of predation makes it unlikely that this is the result of changes in sampling resolution and it is more probable that this is a real trend in reproductive performance.

Figure 1.

Annual means (± SE) of three reproductive parameters for the population of short-tailed shearwaters on Fisher Island, Australia, from 1952 to 1994 inclusive. (a) The proportion of individuals that had bred before and were known to be alive (experienced breeders) that were recorded at the colony in that year. (b) The proportion of those experienced breeders present at the colonies that were associated with an egg in that year. (c) The proportion of eggs laid in that year that resulted in fledglings.

Nonetheless, although a significant linear trend existed in the proportion of birds with eggs, no significant trend could be found in the proportion of breeding birds known to be alive that were in attendance. However, examination of the binomial standard errors (Fig. 1) indicates strongly that annual fluctuation in this variable was greater than that expected from simple binomial sampling variation. This was established by calculating the overall proportion in attendance, and then estimating the probability for each year associated with the test that the observed proportion for each year differed, under the appropriate binomial distribution, from the overall proportion. The test probabilities were then used to generate the standard metastatistical deviance difference statistic (Sokal & Rohlf 1995). A similar procedure was performed on the annual proportion of individual with eggs, to see whether the annual variation could be explained purely by the trend line combined with sampling error. In this case, the approach differed only in that the expected proportion for each year was generated from the regression. The annual variation was significantly different from that expected under the binomial model both for the proportion of breeding birds known to be alive that were in attendance (G78 = 191·2, P = 0·000) and for the proportion with eggs (G84 = 226·3, P = 0·000). The existence of a trend in the proportion with eggs, in concert with large non-binomial fluctuations in both variables, thus indicate the existence of real and substantial annual effects upon intermittent breeding.

If intermittent breeding represents a strategy to safeguard physical fitness in ‘poor’ years, then one might expect a correlation between reproductive failure as a result of widespread, climate-related causes and the proportion of individuals failing to breed. Such a relationship would imply that prebreeding conditions permitted predictions of subsequent resource availability during the incubation and chick-rearing phases. Selecting the years between 1955 (when the total number of birds with a known breeding history rose above 50) and 1984 (after which predation led to breeding failures in some years), a significant serial correlation existed between the proportion of individuals with eggs and the proportion of those eggs that resulted in fledglings (breeding success) (r = 0·42; P = 0·022). None of the correlations between the proportion of individuals present, the proportion of these individuals with eggs, and breeding success (the proportions of eggs producing fledglings), were significant (Fig. 2). In addition, the effect of the previous season's reproductive success was correlated with the proportions present, but again not significantly. The existence of a significant correlation between the proportion of individuals that produce an egg and successful fledging of the ensuing offspring would appear to imply that some relationship does exist between resource availability around the time of laying and its availability later during chick-rearing. Poor correlations between the proportion of individuals in attendance and either egg production or reproductive success suggest that non-attendance is a separate phenomenon from laying failure, and that these two aspects of intermittent breeding represent distinct responses.

Figure 2.

Scatter plots of annual reproductive parameters (see Fig. 1 for details) between 1955 and 1984. Previous fledglings per egg indicates the proportion of eggs that resulted in fledglings on the previous occasion, whereas fledgling per egg represented the same measure for the year concerned. A line of best fit indicates the existence of a significant correlation.

Variation in reproductive parameters with increasing age from first breeding

This analysis used those individuals for whom the year in which they first bred was known and, in order to avoid bias in comparisons, excluded their first and last recorded breeding year because it was not possible to be absent on the last breeding attempt. The proportion of individuals present at the colony rose with increasing breeding age, defined as the years elapsed since first associated with an egg or young (sensuWooller et al. 1990), for approximately the first 10 breeding years (Fig. 3). After this, there was some evidence that the probability of an individual being present declined with breeding age. This change in attendance patterns closely parallels that shown by breeding success, as measured by the proportion of eggs that produced fledglings (Wooller et al. 1989, 1990). The proportion of individuals present associated with eggs, however, appeared to increase consistently throughout life, with no sign of a downturn.

Figure 3.

The means (± SE) of three reproductive parameters (see Fig. 1 for details) in relation to breeding age (the number of years elapsed since an individual bred for the first time).

These changes were tested using logistic regression. For the three dependent variables, individual identity was fitted as a random factor and year as a fixed factor. The sex of an individual and whether it survived for 12 years or not, were also included as factors. Age at first-breeding was fitted as a polynomial up to the cubic term. Non-significant terms were omitted from the fit.

The proportion of individuals in attendance required a linear and quadratic term, the proportion with eggs a linear term only, and the proportion of eggs resulting in fledglings (breeding success) required the full cubic term. The function shape for attendance and breeding success up to maximum breeding age was a convex curve with a single maximum value (Table 2). This indicates that these two variables increased with breeding age in younger individuals, but, having achieved a maximum, declined in older birds. No such decline was apparent for egg production (Table 2). The sex of an individual was not a significant factor except in the case of egg production (Wald's χ2 = 4·29, P = 0·038). However, the breeding age by sex interaction term was not significant, indicating that the rate of increase of egg production with age did not differ between the sexes.

Table 2. . Terms in the logistic polynomial regressions of breeding age on attendance, egg production and breeding success. All regressions were fitted with individual identity as a random factor and year as a fixed factor. Insignificant terms in the polynomial are omitted. In the case of attendance and breeding success, the polynomials produced convex functions with single maxima, indicating that the dependent variable increased in frequency with breeding age in younger individuals and decreased in older individuals
Dependent variableTerms in
breeding age
CoefficientStandard
error
P for Wald’s testAge for
maximum value (years)
AttendanceLinear0·08160·02920·00510·4
Quadratic−0·00390·00130·003 
Egg productionLinear0·03410·01080·002 
Breeding successLinear0·23280·0364< 0·0016·6
Quadratic−0·01500·0039< 0·001 
Cubic0·00030·00010·016 

The initial increases in all three variables might result from a selective effect if individuals with higher rates of absence or failure to lay or rear an offspring also had a lower survival rate, and thus were progressively eliminated from the population. Indeed, Bradley et al. (1989) demonstrated a positive correlation between reproductive output and survival in younger breeders. In order to exclude the possibility that differential survival accounted for the trends observed, individuals were categorized on the basis of surviving for at least 12 years, and the survival factor fitted to the model as a main effect and an interaction term with breeding age. In no case were these terms significant, indicating that the fitted breeding age functions reflected age-dependent changes in reproductive variables rather than differential survival.

Thus, the probability of attendance by an individual increased by about 3% per year until around the eleventh year, then declined approximately 9% thereafter. Breeding success increased about 7% until around the seventh year and decreased approximately 9% thereafter. The proportion of individuals in attendance that were associated with an egg increased approximately 3% per year throughout their breeding lives. The question remains as to why both reproductive success and attendance should decrease markedly in older birds while egg production continues to increase, perhaps even at a faster rate.

Correlations in reproductive parameters within individual life histories

If a failure to attend their breeding colony or to lay an egg represents a strategic withdrawal of effort by an individual, then one might expect that non-breeding would produce advantages elsewhere. At one extreme, if a sufficient proportion of the life history was measured, one might discover that individuals associated with non-breeding events did not produce fewer viable offspring than those that attended and laid consistently, because the success of non-breeders on subsequent breeding occasions was enhanced.

However, when absences are examined in a group of individuals that survived at least 12 years after starting to breed, the cumulative total of fledglings produced was reduced (F = 46·8, P < 0·001) with increasing absences (Fig. 4). A single absence may not produce any difference from the reproductive success of birds with no absences (using Hochberg's GT2 a posteriori comparison between means, zero does not differ from one absence at P = 0·05, but both differ from all other levels of absence). Nevertheless, a decrease in reproductive output by individuals exhibiting a substantial number of absences is inevitable, because the seasons available to them to produce offspring are thereby much reduced. If absence represents a risk-minimization strategy to conserve resources for future breeding attempts, one might expect that birds with absences would have a higher success in egg production, fledgling production or both, in those years when they do attempt to breed, thereby compensating for their reduced number of attempts. In order to examine this we calculated correlations (Table 3) between the proportion of years available that individuals were absent (excluding the first year of breeding, in which, by definition, all individuals were present with an egg), the proportion of years present in which an egg was produced (again excluding the first), and the proportion of those eggs which gave rise to fledgling.

Figure 4.

The mean (± SE) cumulative reproductive success (circles represent number of eggs, squares numbers of fledglings) in relation to the number of years absent, for individuals that survived 12 years from commencement of breeding. The results of the first breeding occasion are omitted because, by definition, at first-breeding birds must be associated with an egg.

Table 3. . Correlations and tests of significance between four parameters of reproduction during the first 12 years of an individual's breeding career
VariableProportion of
breeding years
available in which
absent
Proportion of years
present at the colony
in which associated
with an egg
Proportion of eggs
that resulted in
fledglings
Proportion of years
present at colony
in which recorded
with a mate
Proportion of years present at
the colony in which associated
with an egg
−0·28
P = 0·0003

  
Proportion of eggs that
resulted in fledglings
−0·26
P = 0·0006
0·04

P = 0·6462
 
Proportion of years present
at colony in which recorded
with a mate
−0·30
P = 0·0001
0·70
P = 0·0000
0·18
P = 0·0182

Proportion of years present
at colony in which mate of
previous year was retained
−0·46
P = 0·0000
0·45
P = 0·0000
0·33
P = 0·0000
0·38
P = 0·0000

There was a significant decrease, in those individuals with a higher proportion of absences, in the proportion of years in which an individual was in attendance during which eggs were produced, and of eggs producing fledglings. This strongly suggests that intermittent breeding is characteristic of birds with a lower competence in incubation or chick-rearing, and that birds prone to absences are also prone to laying failures. However, there was no significant correlation between the production of eggs and the proportion of those eggs resulting in fledglings. Individuals that were associated with most eggs were not necessarily those that were most efficient in rearing their young. With one exception, the correlations in Table 3 do not differ significantly between the sexes. The exception is the correlation between the proportion of absences and the proportion of eggs producing fledglings. In males, this correlation is not significant (r = −0·10, P = 0·366), whereas in females it is highly significant (r = −0·41, P < 0·0001). This may again indicate that the factors determining attendance in males and females differ. A male whose previous partner is not in a condition to lay may attend the colony either because he is unaware of her condition or because there is the possibility of another pairing. A female unready to lay may simply not attend. Under these circumstances, the significant correlation in the female would be generated by poor condition, possibly arising from foraging difficulties.

The question remains as to whether this reduced efficiency is related to individuals or, rather, is characteristic of the pair bonds in which they find themselves. Those individuals absent more often were also less likely to be recorded with a mate, and with the same mate, during those years that they attended the colony (Table 3). Similarly, those individuals more likely to be associated with eggs during their attendances were more likely to be recorded with a mate, and with the same mate, as were individuals producing more fledglings per egg. Poor reproductive performance may be associated with poor condition in the mate, leading to higher mortality, and we have previously demonstrated a negative correlation between breeding success and mortality in younger breeders (Bradley et al. 1989). Alternatively, we have demonstrated that divorce is negatively correlated with breeding success and that pair bonds terminating in divorce have a higher rate of absences than those that terminate through the death of a participant (Bradley et al. 1990). Thus, intermittent breeding appears to be associated with a reduced ability to acquire a mate and raise offspring. However, it is not clear how far pair bond breakdown is a result of absence and non-laying, rather than their cause.

Discussion

The general pattern of reproductive performance in birds is generally considered to increase with age during the first reproductive years, to reach a plateau in middle age, sometimes followed by a decline in old age (Forslund & Pärt 1995). However, because failure to instigate a breeding attempt by an animal that has bred previously is difficult to measure in the field, this profile has been largely developed from analyses of age trends in clutch size and breeding success. Nonetheless, evidence of age-related trends in intermittency has been found in several seabird species (Mougin et al. 1997).

In two species of large gulls Larus argentatus Pontoppidan and L. fuscus L. studied over two consecutive years, 32–41% of all individuals that had bred before and were recorded alive in a later year, were not associated with a nest, egg or young in any one year (Calladine & Harris 1997). Indeed, half of these individuals failed to breed in both years and intermittently breeding gulls were 18% less successful than others. As in short-tailed shearwaters, intermittent breeding in these gulls appeared to be associated with differences in the quality of individuals (Calladine & Harris 1997). In common guillemots [Uria aalge (Pontoppidan)], 5–10% of individuals did not breed in any year. These individuals often failed to breed repeatedly and had both a lower breeding success and lower survival rate than more regular breeders, indicating that they too were poorer quality individuals (Harris & Wanless 1995).

Kittiwake gulls (Rissa tridactyla L.) also demonstrate a high incidence of non-breeding, particularly among individuals with less breeding experience (Wooller & Coulson 1977). Cam et al. (1998) found that breeding and non-breeding kittiwakes were equally likely to be recaptured, and detected no differences between males and females. However, non-breeders had lower survival rates than breeders, their survival varying from year to year in parallel. In addition, non-breeding individuals were more likely to fail to breed in the following year than breeding individuals. They concluded that non-breeding by individuals that had previously bred tended to involve lower quality individuals.

Cory's shearwater [Calonectris diomedea (Scopoli)], in which 10–16% of the population fails to breed, also shows a decline in the frequency and duration of non-breeding with increasing breeding experience (Mougin et al. 1997). Intermittent breeding did not appear to be related to prior breeding success in this species but 45% of birds absent had mates similarly absent (compared with 10% overall). After their return, such absent birds were more prone to divorce (Mougin et al. 1997). About 9% of experienced great skuas (Catharacta skua Brünnich) and 6% of parasitic jaegers (Stercorarius parasiticus L.) failed to breed, mainly because of loss of a mate (Catry et al. 1998). Non-breeding by great skuas was most frequent in the youngest and oldest members of the breeding population (Catry et al. 1998).

A substantial number of possible explanations for an increase in reproductive performance with age have been suggested but, as Forslund & Pärt (1995) point out, these generally fall into one of three categories: progressive disappearance of poor performers; age-related improvements in competence; and optimization of residual reproductive effort. The effects of the first cause can be excluded, as has been done for this study, by measuring reproductive parameters in birds known to have survived beyond a specific age.

In short-tailed shearwaters, the downturn in attendance and breeding success, but not in egg production, at older breeding ages suggests that different processes (or mix of processes) operate for different aspects of reproduction. Significant effects linked to breeding age indicate the existence of real qualitative differences between individuals in reproductive performance. Attributes modified by age may also differ intrinsically and repeatably between individuals of the same breeding age because of consistent differences in their reproductive qualities or physical fitness.

The different responses between egg production and attendance imply that these are very distinct phenomena, rather than non-laying just being another form of absence. The decline in attendance and breeding success may reflect the reduced ability of older individuals to cope with physical challenges (Clutton-Brock 1988). The lower physical fitness of older birds constrains their ability to forage, to rear a chick or to attain the body condition necessary to commence reproduction. In addition, because a substantial correlation occurs between the ages of breeding birds (Bradley, Wooller & Skira 1995), mate condition or death will also moderate attendance.

Alternatively, more experienced individuals may be less prepared to continue at these stages in the breeding cycle if their lack of reserves warrant it. However, although individuals with sufficient reserves may be in attendance at their breeding colony, other processes may then interfere with laying; for instance, accelerating rates of egg production are consistent with the residual reproductive value hypothesis (Stearns 1992; Forslund & Pärt 1995). Laying may be the stage at which many strategic decisions are made. That is, sufficient information may exist at this point to evaluate the forthcoming season. Younger individuals, that are more likely to have taken the wrong option, would then be able correct it (the ‘cold feet’ hypothesis), whereas older individuals, which are more likely to have guessed correctly, would continue with their choice.

The downturn in breeding success occurs at a younger breeding age than that for attendance and may stem from the determination by an individual that its condition is too low to continue breeding without detriment. Galbraith et al. (1999) found that the efficiency in use of time and energy by terns was low in the youngest breeding males, rose to a peak around 12 years of age, then appeared to decline in the oldest breeding males. They suggested that the improvement with age was a result of increased efficiency rather than increased effort (Galbraith et al. 1999).

A similar increase in efficiency resulting from cumulative experience may underlie the continual increase in egg production with breeding age by short-tailed shearwaters, although we cannot exclude the possibility of increased effort later in life in accord with the residual reproductive value hypothesis. Whatever the reasons for the continual increase throughout life, and despite potentially substantial reproductive costs prior to laying (Jönsson, Tuomi & Järemo 1998), the egg still represents a considerable and irrevocable commitment to a breeding attempt each year.

Several studies of long-lived seabirds have demonstrated that reproductive output is impacted by variation in the provisioning rate (Croxall & Rothery 1991). If an individual was in a position to process predictive information about such aspects of a breeding season as resource availability and required foraging effort then, in some years, the balance of probabilities may determine that the optimal strategy for that individual would be a zero clutch. However, it has also been suggested that some individuals may employ bet-hedging, with less than maximal reproductive effort in a year when resources are more available than average in order that they have some residual physical fitness to support a better-than-expected breeding effort in a year when resources are less available (Slatkin 1974; Stearns 1992; Saether, Ringsby & Røskaft 1996). Reproductive output would thereby be dragged towards the average, rather than directly reflecting annual variation in resources.

Saether et al. (1996) propose a classification of avian life histories based on the quality of the breeding and survival (non-breeding) habitat. They distinguished high reproductive species (breeding habitat good, non-breeding poor) with large clutch sizes and relatively high adult mortality, bet-hedging species (both habitats good, but breeding varying stochastically) with a (generally) long life and relatively large clutch size, and survivor species (breeding habitat poor, non-breeding good) with a long life but small clutch size. Saether et al. (1996) considered procellariform seabirds the archetypal group of survivor species. However, the distinction between survivor species and bet-hedgers seems somewhat artificial. The quality of the breeding habitat for bet-hedgers must vary stochastically, because that provides the basis for the underlying fitness advantage of the strategy. There is evidence in shearwaters of very large annual variation in reproductive success (Wooller, Bradley & Croxall 1992), implying large variation in the breeding habitat. The production of a single egg, without replacement, implies a constancy of clutch size in which reproductive effort is strongly constrained to a (low) constant value, rather than allowed to reflect annual variation in habitat quality. Similarly the very narrow and annually invariant laying period recorded across the range of short-tailed shearwaters (Meathrel et al. 1993) is consistent with strong constraint to an annual average. Therefore, it can be argued that the survivor species category may contain within it the extreme of the bet-hedging range, namely those species that have most prolonged their lives and most constrained their reproductive effort against habitat variation.

If the single egg of petrels is an outcome of this process, then intermittent breeding should be fairly infrequent and invariant in this group. Most breeding individuals would be expected to attempt to breed in most years, but to regulate their foraging effort in accord with the quality of the year. A single-egg clutch allows little flexibility in response (one or zero) from year to year, and the establishment of such a small, fixed clutch size must also be accompanied by a high survival rate for there to be a positive innate capacity for increase. This longer life span, however, means that the average breeding bird will sample a number of breeding seasons, so that the ‘average’ performance should be a well-defined entity. The pelagic habitat exploited by wide-ranging seabirds appears highly stochastic, especially in the southern oceans, where irregularities are driven by Antarctic weather systems. In the case of a bird with as high a survival rate (93%) as the short-tailed shearwater (Bradley et al. 1989), it is surprising that no years have been recorded when all, or at least the majority of birds, fail to attempt to breed (Serventy & Curry 1984), as has been documented in other seabirds (Hatch 1989; Erikstad et al. 1998).

Under this model, failure to breed in an otherwise annually breeding species stems from strict adherence to a bet-hedging strategy, with intermittent breeding resulting from contingent events in the experience of each individual, such as physical challenges, food shortages or loss of mate. It does not arise from predictive, prebreeding signals of habitat quality, which alert individuals to a high probability of reproductive failure. Thus, a year in which conditions were so bad that they precluded a substantial proportion of the population from attempting to breed would be very rare indeed, and would be unlikely to occur except over very long periods, such as centuries. Individuals would not make early predictions of the quality of the season because non-breeding is not a ‘normal’ option. However, they might well discontinue breeding at a later stage if their condition threatened to fall to a point too low to allow them to hold reserves for the following season's breeding attempt. Under these circumstances, attendance and laying should be relatively invariant, and contingent from year to year, irrespective of habitat quality, while breeding success should be highly variable.

In conclusion, 14% of short-tailed shearwaters known to be alive and to have bred before, were absent from their breeding colony in any one year and a similar proportion were present, but without an egg. This intermittent breeding was associated with a reduced ability to acquire a mate and raise offspring. Our results do not indicate that intermittent breeding in this species stems from a trade-off between the effort required for reproductive success and breeding life span. Rather, we suggest that individuals of higher quality are able to breed more frequently than others without any compensatory reduction in either their annual breeding success or their overall breeding life span. The lower annual variation in intermittent breeding compared to greater annual variation in reproductive success and the invariant breeding phenology is consistent with the occurrence of bet-hedging as a component of this species reproductive strategy.

Acknowledgements

We thank all our colleagues for their continuing support of this work which was funded by Australian Research Council grants A1901054 and 1153. We are especially grateful to Ian James for advice concerning logistic regression.

Received 23 January 1999; revision received 14 December 1999

Ancillary