Marking and behavioural observations

In each study period we selected nests within 1 m from the periphery of the colony with chicks of variable ages (49 nests in December 2005, 33 in February 2006, 14 in December 2006 and 41 in February 2007). Selecting nests in the interior of the colony would cause disproportional amounts of disturbance. Cape Gannets at the fringes are similar in age and quality as birds in the interior (Klages 1994), so no biases in behaviour are expected.

At each selected nest we captured the parents from their nest by manipulating a hooked pole around their neck. Gannets were then measured (bill length to the nearest 0.1 mm, length of flattened wing chord to the nearest mm and body mass to the nearest 25 g) and dyed with picric acid. Our measuring routine differed slightly between the two years. In 2005/06 we measured and dyed the adults of the selected nests in the morning to obtain pre-feeding morning mass. In 2006/07 adults were marked in the afternoon without capture using a marker tied to a stick and we took measurements the next morning if the adults were still present at the nest. Partners were captured and measured within four days after the capture of the first parent, again after they spent the night on the nest to obtain the closest estimate of pre-feeding mass. Each study nest thus had individually recognisable parents; one dyed on the back, the other either had a colour ring or was dyed on the head.

Nest attendance of each bird was recorded by checking nests every hour from sunrise to sunset (05:00 – 20:00, Namibian summer time). Gannets are visual hunters and therefore do not forage at night (Ropert-Coudert et al. 2004). If a different parent was observed attending the nest in the morning than on the previous night, we assumed that the nest relief happened shortly after sunset. We calculated several variables from the observational data: (1) hours away from nest (trip duration), (2) hours on the nest (attendance bout), and (3) average number of foraging trips per day (trip frequency).

The Cape Gannet is a monomorphic species, so sexes could not be distinguished in the field. Therefore we collected some feathers of one bird per nest for later sex determination through DNA analyses (see Fridolfsson & Ellegren 1999 for detailed methods). To verify the sexing method we collected feathers from both parents of 10 nests. All pairs consisted of two sexes.

Chick Growth

We measured bill and wing length of study chicks at 5–7-day intervals. Body mass was measured to the nearest 5 g (<1 kg) or 25 g (>1 kg) with Pesola spring balances. Each date, we weighed chicks at fixed times and in the same sequence. Chicks were fitted with colour rings after 3–4 weeks, depending on tarsus size. Chick age was estimated from the first measurement of each chick, using algorithms derived from 103 known aged chicks from Navarro (1991). If wing length was <40 mm, age was computed by age = -ln ((89.78 - b / 6.15 × b) / 0.086) + 0.5 days, where fa is bill length in mm. For chicks with wings a 40 mm, age was computed by age = 1.395 - ln (ln (588.8 / w) / 0.0264) + 0.5 days, where w is wing length in mm. Age was then calculated by adding the actual number of days that passed since the first measurement. We calculated growth indices (z-scores) from the mass increments of each chick, as departures from average growth (Mullers et al. 2007). This growth index is independent of chick age. For each interval we averaged growth indices for all chicks from the study nests. In total we obtained growth data for 132 chicks (n = 47, 12, 40 and 33 respectively) from which we calculated 434 growth indices.

Survival was calculated as daily survival rate, to correct for differences in length of the observational periods, and expressed as weekly survival rates. Chicks that did not survive were assigned to one of three categories; 1) died without being predated (called starved), 2) taken while at least one parent was present at the nest (predated) or 3) died of weather conditions or predated while no parent was present at the nest (abandonment). We cannot exclude the possibility that ‘starved’ chicks died from other causes (for example from a virus infection).

Analyses

We have no sex-specific data on parental behaviour in February 2006. Therefore we first present the averages of the behaviours, body condition and chick growth for all four periods and then look in more detail into the three study periods of which we do know the sexes of the adults. Chick survival was analysed using a logistic regression. Multilevel models were used to take repeated measures and dependent observations into account for chick growth (individual and observation as levels) and parental behaviour (nest, individual and observation as levels). Body mass divided by wing length was used as an index of body condition (Lewis et al. 2006) and regressed on the onset of breeding (Linear regression: chick age r² = 0.08, F_1,193 = 16.6, P < 0.001), after which we used residual body condition. During the three study periods for which we know the sex of each adult, we collected data from 204 different birds, concerning 2482 foraging trips and 2380 nest attendance bouts. Behavioural response variables were lntransformed. Average chick age differed between the study periods (One-way ANOVA F_3,123 = 22.3, P < 0.001, Table 1) and was included as a covariate in statistical models exploring parental behaviour. All multilevel models included sex and study period (n = 3) as factors and chick age as a covariate. To test the relationship between parental behaviours and chick growth or survival, we either averaged (trip duration and attendance bouts) or summed (trip frequency) behaviours for both sexes per nest for the same intervals as the growth increments, so parental behaviour and chick growth could be related to each other for the same period. Included in these models were study period as a factor and chick age and behaviour (attendance bout and trip frequency) as covariates. The models that looked at the association between parental behaviour and chick growth or survival were compared by calculating the Deviance Information Criterion (DIC). This is a generalisation of the Akaike Information Criterion (AIC), derived through Markov Chain Monte Carlo (MCMC) methods, and used to compare multi-level models. Models with the lowest DIC fit best, as with the AIC (Spiegelhalter et al. 2002).

Table 1.

Mean age (SD) of Cape Gannet chicks and mean mass and body condition (mass/wing length) of parents per sex at the beginning of each study period in 2005/06 and 2006/07.

Chick growth and survival

In 2005/06 chick growth increased from 19 December until 9 January, then decreased (Fig. 1). At the end of February growth rates were similar to those of January, but they decreased in the last two weeks of the study period. In 2006/07 chick growth rates in December were within the range of growth the year before (Fig. 1). In February growth rates were less than earlier in the year and decreased even more in the last week, but it was similar to growth in the year before at that time. Chick growth differed significantly between years and weeks (multilevel model: year χ² = 7.1, P = 0.008; week χ² = 57.3, P < 0.001) and decreased during both years (multilevel model: year n.s.; season half χ² = 24.6, P < 0.001; interaction n.s.).

Weekly survival of the study chicks was better in 2005/06 than in 2006/07 (98.8% and 92.6% respectively). In 2005/06 weekly survival of the study chicks was constant and high throughout the breeding season; 98.8% in both study periods. In 2006/07 survival decreased from 97.0% in December to 90.8% in February. Survival tended to be higher early in the breeding season, but not significantly so (98.4% in December and 94.7% in February, logistic regression: year χ² = 15.8, P < 0.001; season half χ² = 3.2, P = 0.072, interaction n.s.). Survival was lower for chicks with slow growth rates and of younger age (logistic regression: growth χ² = 7.2, p = 0.007; chick age χ² = 38.9, P < 0.001). The average age at which chicks died was 25.9 days (SD = 9.2, range 14–56).

Parental body condition

The initial body mass of adults did not differ between years (2005/06: 2580 g, SD 247.0; 2006/07: 2552, SD 186.1) and was lower at the end of 2006/07 (GLM: chick age n.s.; year n.s.; season half F_1,239 = 5.1, P = 0.025; interaction F_1,239 = 6.0, P = 0.015). Wing length was longer in 2006/07 compared to 2005/06, but decreased from early to late 2006/07 (GLM: year F_1,237 = 33.7, P < 0.001; season half F_1,237 = 6.1, P = 0.014; interaction F_1,237 = 4.7, P = 0.031). Body condition (mass/wing length) did not differ between the two breeding seasons, but was significantly lower later in the year. Female body condition was on average 6.45% greater than male body condition (Table 1) and body condition decreased with increasing chick age (GLM: year n.s.; sex F_1,187 = 6.5, P = 0.012; season half F_1,187 = 5.4, P = 0.021; chick age F_1,187 = 11.9, p = 0.001).

Table 2.

Multilevel mixed-models testing the main effects on parental behaviours of Cape Gannets in 2005/06 and 2006/07. Coefficients (B) indicate direction of effects, relative to the reference category. Weeks coincide with growth interval of the chicks.

Figure 1.

Mean standardised Cape Gannet chick growth and standard errors for each interval (4–7 days). The lines represent the mean growth per study period.

Parental behaviour

As we do not know the sexes of the parents in February 2006, we first compared the average behaviour of all birds with the means of February 2007. Foraging trips were 6 h shorter in February 2006 compared to February 2007 (28.2 and 34.3 h respectively, multilevel model: χ² = 8.5, P = 0.003) and birds stayed 5 h shorter on the nest (16.3 and 21.4 h respectively, χ² = 20.1, P < 0.001).

Figure 2.

Sex-specific means and standard errors of A) trip duration, B) attendance bouts, and C) trip frequency of Cape Gannets for two study years. No sex-specific data were available for the second half 2005/06.

In the three study periods with parents of known sex, both trip duration and attendance bouts decreased with increasing chick age, so nest visitation rates increased when chicks became older (Table 2). Females made longer trips and shorter attendance bouts than males and consequently males had higher nest visitation rates (Fig. 2, Table 2). In February 2007 (the year for which we have sex-specific data), trip durations of female Gannets were elevated by 59% compared to the mean of December 2006 (December 2006: 27.0, SD 10.4; February 2007: 43.6, SD 38.9) and were 17.5 h longer than male trip duration (males: 26.1, SD 26.2). Male trip duration increased to 33.9 h in the last week. The sex differences in trip frequency were not constant between weeks. Male Gannets increased their trip frequency faster with increasing chick age than females (multilevel model: sex x weeks χ² = 10.4, P = 0.001; sex x chick age χ² = 20.7, P = 0.008).

Chick growth in relation to parental behaviour and body condition

Parental trip duration was highly correlated with trip frequency (Pearson r = -0.617, P < 0.001) and we therefore only looked at the association between chick growth and trip frequency or attendance bouts. Attendance bouts and trip durations were positively correlated between the sexes (bivariate correlations: attendance bout r = 0.540, P < 0.001; trip duration r = 0.334, P < 0.001).

The model with trip frequencies separated per sex showed that paternal trip frequency was associated with chick growth, but maternal trip frequency was not (multilevel model: maternal trip frequency B (SE) = 0.24 (0.17), χ² = 2.0, P = 0.155; paternal trip frequency B (SE) = 0.44 (0.17), χ² = 6.8, P = 0.009, DIC 756.6). The model that best explained the association between chick growth and trip frequency combined the trip frequency for both sexes: chick growth was positively associated with the number of parental nest visits (Table 3, DIC 755.7). Variation in attendance bouts was not associated with variation in chick growth in any of the models (Table 3).

Chick growth in the first week after measuring parental body condition was positively associated with residual maternal body condition (Linear regression: residual maternal body condition F_1,86 = 4.3, P = 0.042). When study period was included in the model the association between chick growth and maternal residual body condition was not significant (GLM: residual maternal body condition n.s.; study period F_2,84 = 3.5, P = 0.035). Both parental and paternal residual body condition were not associated with chick growth, neither in any of the models with residual body mass instead of residual body condition.

Table 3.

Multilevel mixed-models testing the main effects on Cape Gannet chick growth for behaviour of both parents combined in 2005/06 and 2006/07. Parental behaviours are attendance bout (h), and trip frequency (per d). Coefficients (B) indicate direction of effects, relative to the intercept. Study periods indicate the three periods in which the sex of the parents is known in 2005/06 and 2006/07.

Table 4.

Multilevel mixed-models testing the main effects on Cape Gannet chick survival for A) behaviour of both parents combined and, B) behaviour per sex. Parental behaviours are either attendance bouts (h) or trip frequencies (per d). Coefficients (B) indicate direction of effects, relative to the intercept. Study periods indicate the three periods in which the sex of the parents is known in 2005/06 and 2006/07.

Chick survival in relation to parental behaviour and body condition

Chick survival was reduced when parents visited the nests less frequently (Fig. 3, Table 4A). Chick survival was significantly associated with maternal trip frequencies (B (SE) = 2.34 (1.0), χ₂ = 5.51, P = 0.019), but not with paternal trip frequency (B (SE) = 1.49 (0.9), χ² = 2.59, P = 0.107). Despite the positive correlation between the trip frequencies of both sexes, the model with only maternal trip frequency showed the lowest DIC (0.43 vs. 0.58 for the parents combined).

The best model explaining variation in chick survival from variation in attendance bouts was the model with the behaviour of the sexes separately; increased attendance bouts of males coincided with lower chick survival (Table 4B, DIC 0.74), but there was no association with maternal attendance bouts. The estimates of the sexes differed in direction, indicating different consequences of parental behaviour for chick survival. Also when tested in separate models, paternal attendance bouts were associated with chick survival (multilevel model: B (SE) = -2.44 (1.1), χ² = 5.40, P = 0.020), whereas maternal bouts were not (B (SE) = 0.46 (0.65), χ² = 0.51, P = 0.477).

Figure 3.

Survival of Cape Gannets as a function of trip frequency by logistic modelling. The top panel gives the frequency distribution for successful and unsuccessful nests as fraction of total per category.

Survival of chicks was not associated with either parental or paternal residual body condition, but tended to be reduced if maternal residual body condition was reduced as well (Logistic regression: residual maternal body condition: χ² = 3.8, P = 0.051). None of the models that included study period showed any significant association between chick survival and residual body condition and residual body mass of parents combined or separated.

Table 5.

Weekly survival rates of Cape Gannet chicks on Ichaboe during 2005/06 and 2006/07. The three causes of death of our study chicks and their absolute and relative contribution to the number of mortalities in each study period are indicated.

Nest absences and predation

We could not statistically demonstrate that the level of parental non-attendance was associated with chick survival. We do have indications, however, that attendance is important for chick survival. In February 2007 chicks were left alone by the parents after which they either died of cold temperatures (4 cases) or were predated by Kelp Gulls (7; Table 5). Chicks younger than 30 days that were left alone were usually killed by Kelp Gulls within a few hours. For two ‘predated’ chicks we do not know whether they were abandoned and then predated, or were predated and then the attending parent left. Chicks that did not survive were left unattended for the first time when they were on average 10 days younger than surviving chicks (21.2 and 31.6 d respectively, One-Way ANOVA: F_1,52 = 32.2, P < 0.001).

Chick growth and survival

Unfortunately, no data on food availability in the Namibian waters were available, so we could only derive indirect estimates of changes in food availability. When foraging conditions become poorer, trip durations and foraging ranges of breeding animals increase concomitantly (Boyd et al. 1994, Suryan et al. 2002, Kowalczyk et al. 2006). The longer foraging trips in 2006/07 compared to 2005/06 would suggest lower food availability around Ichaboe in 2006/07. This interpretation is consistent with the delayed onset of breeding by both Cape Gannets and Cape Cormorants in 2006/07.

The longer foraging trips in 2006/07 were associated with slower chick growth and decreased survival in 2006/07 compared to 2005/06. Within both years chick growth decreased during the breeding season. In the first year the reduced growth was not associated with decreased survival rates, which remained high throughout the year (98.8%). In February 2007 however, the weekly survival rates decreased with 8%, which was associated with slower chick growth, but also with reduced provisioning rates of the parents. The decreased growth and survival rates of the chicks in concert with the increased foraging trip duration of the parents in 2006/07 suggest adverse changes in the environment like food shortage during which parents could not sustain the energy demands of their offspring.

Sex-specific foraging behaviour

In Cape Gannets the parental behaviours of both sexes were positively correlated, but averaged over two years, males made considerably shorter foraging trips (21.1 h) and visited the nest more often (0.56 visits per day) than females (29.7 h trips; 0.48 visits per day). In the closely related Northern Gannets Morus bassanus the sexes do not differ in trip duration, but females make longer and deeper dives and spend more time at the sea surface than males (Lewis et al. 2002). The differences in foraging behaviour may suggest different roles for the sexes. Longer foraging trips of female Gannets may indicate that they explore different feeding grounds (Weimerskirch 1993, Gonzalez-Solis et al. 2000, Weimerskirch et al. 2006), thereby possibly reducing intra-specific competition at the feeding sites near the colony (e.g. Hunter 1983).

During the presumed deterioration of the food situation, female Gannets increased their trip duration by 59%, several weeks before the males increased theirs. The lower provisioning rates of both parents affected chick growth, but only the reduced trip frequencies of mothers were associated with chick survival. Foraging strategies should maximize the chances of finding local sites of high food availability in order to minimize the energetic costs of foraging (Parrish et al. 1998). If good quality prey shifts to more distant sites, females possibly attempt to follow such prey and increase trip duration. Males responded by staying longer at the nest with the chick, hence the association between male attendance bouts and chick survival. Short intervals between consecutive feedings and the regularity of food deliveries are important in optimizing seabird chick growth (Schaffner 1990, Weimerskirch et al. 2000, Gray & Hamer 2001). Only in nests where females managed to keep their trip frequency high, the offspring managed to survive. The evolutionary significance of sex-differences in foraging behaviour in monomorphic species like the Cape Gannet remains a puzzle (Lewis et al. 2002) and we need at-sea data to look into this more specifically.

Figure 4.

Diagram showing the interactions between food availability, parental behaviour and chick growth and survival of Cape Gannets and the dynamics with other seabird species. We have no data on food availability (hence the dashed arrows), but there is strong evidence for its importance in the system.

Parental behaviour, chick growth and survival

We suggest the following scenario for the two breeding seasons in which we collected data (see Fig. 4). Under favourable conditions in 2005/06 (shorter foraging trips and higher chick growth rates and survival) both sexes increased provisioning rates when the chick became older to sustain increasing chick energy demands. When conditions were less favourable in 2006/07, female Gannets almost doubled their trip duration. Consequently males had to stay longer at the nest when chicks were at an age that they still needed brooding for thermoregulation or protection against predators. As a result the total provisioning rate to the chick declined and chicks showed reduced growth rates and survival. Cape Cormorants normally breed in large numbers on Ichaboe Island (Crawford et al. 2007b). Kelp Gulls mainly predate Cape Cormorant eggs, paying little attention to Gannets. In 2006/07 the Cape Cormorants delayed breeding, most likely due to decreased food availability. This led to Kelp Gulls increasing their predation pressure on the Gannets. Cape Gannet parents with young chicks had two options; either stay at the nest to protect their chick, which then had increased chances to starve, or leave their chicks to increase provisioning rates, giving Kelp Gulls the opportunity to predate their offspring. From our data it appeared that Cape Gannets left their chicks to increase foraging effort, apparently not to jeopardize their own survival (Williams 1966). This mechanism is also found in Arctic Skuas Stercorarius parasiticus (Davis et al. 2005); low food availability forced skuas to increase their foraging effort and decrease attendance rates, which in turn increased the predation pressure by Great Skuas Catharacta skua.

Final remarks

Cape Gannets breeding in the Benguela ecosystem are limited in their foraging behaviour by food abundance and it was suggested that Gannets are foraging at their energetic limits (Lewis et al. 2006, Pichegru et al. 2007). We found that parental body condition did not differ between years, indicating that even when food conditions were poor parents remained at similar body mass as in favourable years. In this way, allocation decisions in poor years favoured the survival of parents above increased foraging effort for the offspring. This parental response seemed the underlying cause of the reduced reproductive output of the whole breeding colony in poor foraging years. Conservation measures in the breeding colonies, by actively keeping Kelp Gulls away from Gannet nests, would therefore be insufficient to enhance the breeding success of Cape Gannets. If local food availability does not improve, the decline in Cape Gannet population numbers is unlikely to reverse.