Inferring the age of breeders from easily measurable variables.

Age drives differences in fitness components typically due to lower performances of younger and senescent individuals, and changes in breeding age structure influence population dynamics and persistence. However, determining age and age structure is challenging in most species, where distinctive age features are lacking and available methods require substantial efforts or invasive procedures. Here we explore the potential to assess the age of breeders, or at least to identify young and senescent individuals, by measuring some breeding parameters partially driven by age (e.g. egg volume in birds). Taking advantage of a long-term population monitored seabird, we first assessed whether age influenced egg volume, and identified other factors driving this trait by using general linear models. Secondly, we developed and evaluated a machine learning algorithm to assess the age of breeders using measurable variables. We confirmed that both younger and older individuals performed worse (less and smaller eggs) than middle-aged individuals. Our ensemble training algorithm was only able to distinguish young individuals, but not senescent breeders. We propose to test the combined use of field monitoring, classic regression analysis and machine learning methods in other wild populations were measurable breeding parameters are partially driven by age, as a possible tool for assessing age structure in the wild.