# Data for: Faster growth and larger size at crèche onset are associated with higher offspring survival in Adélie Penguins

## Cite this dataset

Jennings, Scott (2023). Data for: Faster growth and larger size at crèche onset are associated with higher offspring survival in Adélie Penguins [Dataset]. Dryad. https://doi.org/10.5061/dryad.d51c5b07d

## Abstract

We conducted the first assessment of Adélie Penguin chick survival that accounts for imperfect resighting. We found that when chicks are larger in size when they enter the crèche stage (the period when both parents forage at the same time and chicks are left relatively unprotected), they have a higher probability of survival to fledging. We investigated the relationships between growth, crèche-timing, and chick survival during one typical year and one year of reduced food availability. Chicks that hatched earlier in the season entered the crèche stage older, and chicks that both grew faster and crèched older entered the crèche at a larger size. These relationships were stronger in the year of reduced food availability. Thus, parents increased their chicks’ chance of fledging if they provided sufficient food for faster growth rates and/or extended the length of the brood-guarding period. Early nest initiation (i.e., early hatching) provided parents with the opportunity to extend the guard period and increase chick survival. However, to extend the guard stage successfully, they must provide larger meals and maintain higher chick growth rates, even if just one parent at a time is foraging, which previous work has shown is not possible for all individuals. We show that the factors governing trade-offs in chick-rearing behavior of Adélie Penguin parents may vary in accord with environmental conditions, a result from which we can better understand species’ adaptations to environmental changes.

## Methods

**Study system**

This study was conducted on Cape Crozier, Ross Island, Antarctica (77°27’15.00”S, 169°13’45.00”E) during the summers of 2012–13 and 2013–14 (hereafter 2012 and 2013, respectively). Cape Crozier is the largest Adélie Penguin colony in the southern Ross Sea and one of the largest for the species (Lynch and LaRue 2014). It is surrounded by hundreds of nesting South Polar Skuas (*S. maccormicki*), with most of the colony within skua foraging territories (Wilson et al. 2016). Our study included 43 chicks in 2012 and 69 in 2013 (112 total). Across both years, 84 chicks survived to the crèche stage and could be used to model crèching size and age and survival during that period. The mean crèching age was 21.3 days (SE = 0.46, range 15-26, n = 33) in 2012, and 18.9 days (SE = 0.41, range 10-25, n = 51) in 2013.

Across the entire colony, not just study chicks, we observed that substantially more chicks died from apparent starvation in 2013 than in 2012. Although the average amount of food delivered to chicks per day was similar between the two years, in 2013 there was a longer interval between food deliveries, indicating that parents required more time foraging to provision their chicks (Jennings et al. 2021). Chick fledging mass in 2013 (2741 ± SD 483 g, n = 110) was the lowest recorded between 2000–2016, but in 2012 (2948 ± 626 g, n = 206), it was approximately average for that time period (Ainley et al. 2018). Similar patterns in chick mortality and fledging size were observed at Cape Crozier during 2001–2005 when large icebergs made ocean access more difficult and increased the duration of foraging trips, leading to lower chick feeding rate (Ballard et al. 2010, Dugger et al. 2014). During the present study, in 2013, chick carcasses accumulated at a rate to indicate the supply of carrion was apparently greater than the skua demand, and most of these carcasses had empty stomachs and no sign of predation or scavenging, neither of which was not observed in 2012. Although we did not collect data to quantify annual variation in skua population size or predation pressure during our study, we did not observe any evidence of differences between 2012 and 2013. Nor did frequency of high wind events or temperatures differ between the two years (Supplementary Material Figure S1). Thus, chick-raising challenges of 2013 apparently resulted from food availability and not predation pressure or weather.

Beyond provisioning and brood guarding, other Adélie Penguin parental behaviors and characteristics can influence chick growth rates and outcomes. Older and more experienced parents start breeding earlier in the spring (Ainley et al. 1983) and are generally more successful (Taylor 1962, Lescroël et al. 2009, Kappes et al. 2021), although parent age is a poor predictor of chick growth rate for this species (Jennings et al. 2021). However, earlier nesting allows chicks and parents more time to gain weight before the onset of molt and winter weather (Ainley 2002, Chapman et al. 2011). Nests on the colony edge, adjacent to skua territories, experience higher predation risk and are guarded longer (Ainley et al. 1983, Davis and Mccaffrey 1986). First-hatched chicks and those without siblings grow faster than second-hatched ones (Ainley 2002, Jennings et al. 2021), and male chicks average faster growth than females (Jennings et al. 2016).

**Data collection**

We systematically selected nests to represent a range of parent ages and nest position (edge vs. interior). We checked nests every 1–3 days during incubation and chick-rearing to determine hatch day, the first day of crèche stage, and chick fate. On 5-day intervals from 10 days old through the end of the chick-rearing period (50–55 days old; Ainley and Schlatter 1972, Chapman et al. 2011), we measured mass (to nearest 25 g), and lengths of flipper and tibiotarsus (to 1 mm; Jennings et al. 2016). We individually marked chicks with a T-bar fish tag (Floy Tags Inc., USA) attached to the loose skin on the nape to facilitate individual identification without recapture. Tags were light grey to match the color of juvenile plumage and were removed just before fledging. We determined chick sex molecularly from feather samples (Fridolfsson and Ellegren 1999).

We estimated growth rate for each morphological measurement as the slope coefficient of a regression of size on age, fitted to data for the period of linear growth (full details in Jennings et al. 2021). Linear growth lasted until 40 days old (mass and flipper) or 35 days old (tibiotarsus). Because chicks crèched when 10–26 days old, our growth rate estimate reflected parental behavior during the guard stage and the beginning of the crèche stage. Adélie Penguin chick growth rate declines and in some cases reverses in the final week or so before fledging (Ainley and Schlatter 1972), and this could be important to overall chick outcomes. However, we only used growth during the linear phase because 1) slope estimates for linear models fitted during this period provided a straightforward and intuitive way to compare growth rate among chicks, and 2) growth during this period was more pertinent to our parameter of interest (crèching size) than growth late in the crèche stage.

We only recaptured chicks every 5 days for measurement. If crèching happened on a scheduled measurement day then crèching size was measured directly, otherwise, it was calculated using the linear model described above (Jennings et al. 2021). We checked chick fate every 2-3 days, identifying their tag with binoculars from approximately 5 m. We searched for crèched chicks in a radius of ≤25 m around their respective nest. On each attempted resighting we searched until we found the chick (alive or dead) or for 15 minutes, whichever came first. If a chick was not resighted, we returned and repeated these methods for 4–5 days. If a chick was not detected by day 5, searches were discontinued, and we assumed it had died and was scavenged. Toward the end of the chick-rearing period, as chicks neared fledging age and spent more time closer to beaches, we also conducted several systematic searches for marked chicks along beaches and heavily trafficked routes between nests and beaches.

**Analysis**

We used prior research and knowledge of the study system to develop an *a priori* model set containing variables to test each of our specific predictions while accounting for the effects of two additional variables we thought might have important effects on our response variables. These extra variables were chick sex and relative hatch date (calculated as the difference in days between the hatch date of each chick and the mean hatch date each year). Our sample size limited the number of variables we could include in our models, so we did not include separate variables for parent age, experience, or quality, nor for nest position (see Lescroel et al. 2009). However, the systematic selection of nests from a mix of parent ages and of interior vs. edge nest positions ensured our data reflected the average population response across both these factors. In addition, we accounted for any remaining variation in parental quality by including relative hatch date as a covariate and we used relative rather than absolute hatch date to account for any population level differences in nesting phenology between the two years.

We used an information theoretic approach to weigh relative evidence for each model in each model set. We used Akaike Information Criteria (AIC) values corrected for small sample size (AIC* _{c}*) and for extra binomial variation (QAIC

*) to judge the relative support for each candidate model by comparing its value to that of the model with the lowest value ( ΔAIC*

_{c}*or ΔQAIC*

_{c}*; Burnham and Anderson 2002). We used the “build-up” strategy advocated by Morin et al. (2020) to evaluate variable importance while limiting the number of models considered. We first fitted candidate models with additive combinations of year, sex, and relative hatch date and we included the year * sex interaction based on previously established relationships between provisioning and growth rates (Jennings et al. 2021). We retained model structures if they had ΔAIC*

_{c}*or ΔQAIC*

_{c}*values ≤ 5 and no uninformative parameters (95% confidence limits overlapping zero; Arnold 2010). We then added covariates to these competitive structures to answer our research questions (see supplementary tables for full candidate models).*

_{c}*Survival. *We used the Cormack-Jolly-Seber (CJS) open population model (Lebreton et al. 1992) to evaluate relationships between daily survival rates and the predictor variables of interest (Table 1). The CJS model allowed us to account for a decrease in resighting probabilities over the course of the season as chick mobility increased. We estimated the overdispersion factor (*ĉ*) as 1.16 and used this value to adjust variance for estimated model coefficients and calculate Quasi-likelihood AIC* _{c}* (QAIC

*) values for use in model comparison.*

_{c}We wanted to estimate: (1) survival across the entire chick-rearing period including the guard-crèche transition (through 49 days to avoid fledged chicks being represented as dead); and (2) the effect of crèching age and size on subsequent survival to fledging during the crèche stage only (final 39 days of chick-rearing). Because these two objectives required different data sets (not all chicks survived long enough for objective two), we did a similar but separate model selection procedure for each. For both objectives, we first found competitive structures for daily survival rate (f) and resighting (*p*) probabilities, while holding *p* and f, respectively, at a general structure that represented the most complicated combination of covariates we thought might be important sources of variation in resighting or survival (Year*Sex + t). Here, “t” is the general variation in daily survival (i.e., survival varies independently by day). In addition to the year, sex, and hatch date variables, we also evaluated temporal variation in f, including linear (T), quadratic (TT), and natural log (lnT) trends within each season. We compared competitive structures from these first two model sets to arrive at the f and *p* structures that would be used to evaluate variables of interest (candidate sets for the first and second survival objectives in Supplementary Material Tables S1 and S3, respectively). Then our model sets diverged to address our two objectives. To evaluate overall patterns of daily survival probability and how survival probability changed upon crèche onset (first objective in survival analysis), we added the time-varying individual covariate for whether a chick was in the crèche stage on a particular day (coded 1 = crèched, 0 = guarded, candidate models in Supplementary Material Table S2). For survival during crèche stage only (objective two), we added individual covariates representing the size and age at which each chick entered the crèche stage to the best base model structures for f and *p* (candidate models in Supplementary Material Table S4). However, we found that hatch date and crèching age were correlated (Pearson correlation r = -0.63), so we did not consider any models that contained both predictors.

*Growth rates and crèche timing*. We used separate model sets to answer the four questions of interest: (1) does mass, flipper, or tibiotarsus growth rate predict age at crèching; (2) does mass growth rate predict mass at crèching; (3) does flipper growth rate predict flipper length at crèching; and (4) does tibiotarsus growth rate predict tibiotarsus length at crèching? We initially accounted for lack of independence between siblings with mixed effects models with a random effect for Nest ID. However, the estimated random effect variance was zero or very small relative to the residual variance, so we used linear models with fixed effects only (Zuur et al. 2009). For all four questions, we first determined the competitive year, sex, and hatch date structure, then we added growth rates to answer each of our research questions (candidate model sets in Supplementary Material Tables S5-8). For questions 2–4, the candidate set for the first step also included crèching age, but hatch date and crèching age were correlated (Pearson correlation, r = -0.63), so we did consider any models that contained both.

## Usage notes

We used R 4.0.2 (R Core Team 2020) for all analyses. We used base R for linear models, and used program Mark implemented in RMark 2.2.7 (Laake 2013) for survival analysis. We used RELEASE (implemented via RMark) to estimate over-dispersion of our most general model that did not contain individual covariates (Cooch and White 2019). We report estimated model coefficients and predictions along with their standard error and 95% Confidence Interval (CI).

## Funding

National Science Foundation of Sri Lanka, Award: ANT-0944411

National Science Foundation, Award: 944141

National Science Foundation, Award: 944358

National Science Foundation, Award: ANT-1935870

National Science Foundation, Award: ANT-1935901