In A Class of Their Own: Multiclass Random Forests Reveal Distinct Habitat Needs of Fledglings Relative to Adults in Two Grassland Birds.

Habitat needs often differ among adult and juvenile life-history stages. Such differences not only reveal important aspects of basic ecology, but may also have implications for conservation and management of species. Ontogenetic habitat selection is an inherently multiclass challenge, meaning that it requires modeling more than two classes of use location, but there are few examples of resource-selection functions (RSF) that directly compare both adult and juvenile habitat use with random locations and one another. We developed multiclass Random Forests, a type of machine learning model with high predictive accuracy, to model habitat selection by fledglings and nesting adults of two declining North American grassland bird species, Baird's Sparrow and Grasshopper Sparrow, in an integrated framework. We find that fledglings of both species exhibit habitat use that is distinct from both random locations and conspecific nest sites. Habitat use by fledgling Grasshopper Sparrow was consistent with existing work documenting the importance of vegetation cover during this vulnerable period. However, Baird's Sparrow fledglings displayed unexpected preferences for microhabitat diversity and open structure, possibly mediated by the species' sensitivity to exotic vegetation and shifts in habitat preference with fledgling age. Our work highlights differing habitat needs during the post-fledging phase for two species of concern, documents the potential of species traits and sensitivities to affect habitat use during a vulnerable life-history stage, and advances an underutilized analytic approach that may enhance the functionality of RSF to answer multiclass questions in ecology.