Correcting for within-season demographic turnover to estimate the island-wide population of King Penguins (Aptenodytes patagonicus) on South Georgia

Abstract

King Penguins (Aptenodytes patagonicus) live in remote locations, in large colonies with asynchronous breeding. These three factors hinder the design and conduct of King Penguin censuses, and assessments of trend often require piecing together mismatched surveys of different demographic components. This study introduces a new method to remotely census these populations year-round and correct population estimates for the King Penguin’s unique breeding phenology. We combined in situ ground counts with estimates based on high-resolution satellite imagery to catalog the distribution of breeding colonies and estimate population abundance across the island of South Georgia, in the south Atlantic. While most King Penguin populations are forecast to decline significantly over the next century, South Georgia is expected to experience more favorable conditions and represents an important refugium for the species, though the challenges of surveying King Penguins have precluded a comprehensive census. Due to the variable timing of both in situ and remote counts, we developed a discrete time, age- and stage-structured population model that provides stage- and day-specific correction factors for standardization of census counts. We estimate the current population of King Penguins on South Georgia as 405,425 (95% CI 102,624–2,375,061) breeding pairs and find that population trends that do not account for phenological biases persistently underestimate the population growth rate. Correction factors are highly sensitive to annual egg mortality and the total breeding population is best estimated using nest counts of early-breeding pairs. Future efforts to census King Penguin populations may minimize uncertainty by capturing more precise estimates of egg survival and optimizing the timing of ground and satellite censuses to occur during the settlement of early breeders. Accounting for the error associated with current uncertainty in model parameters, 18–32 years of census data would be required to accurately detect trends of a population with a 10% growth rate. While asynchronously breeding species present a unique challenge to population monitoring, careful accounting of within-season dynamics can be used to assemble a self-consistent time series from heterogeneous survey data.