Location matters: using spatially explicit occupancy models to predict the distribution of the highly mobile, endangered swift parrot

Occupancy modelling using data collected by repeatedly sampling sites is a common approach utilised by land managers to understand species distributions and trends. Two important factors that can complicate interpretation of these models are imperfect detection and spatial autocorrelation. We examine the effect of these potential errors using a multi-year data set on the distribution of the migratory and endangered swift parrot (Lathamus discolor). We simultaneously account for these effects by extending a zero-inflated Binomial (ZIB) framework to allow the inclusion of semiparametric, smooth spatial terms into both the occupancy and detection component of the model, in a maximum likelihood framework easily implemented in common software. This approach also has the advantage of relatively straightforward model selection procedures. We show that occupancy and detectability were strongly linked to food availability, but the strength of this relationship varied annually. Explicitly recognising spatial variability through the inclusion of semiparametric spatially smooth terms in the ZIBs significantly improved models in all years, and we suggest this predictor is an effective proxy for unmeasured environmental covariates or conspecific attraction. Importantly, the spatially explicit ZIBs predicted fewer occupied sites in more defined areas compared to non-spatial ZIBs. Given the importance of predicted distributions in land management, habitat protection and conservation of swift parrots, these models serve as an important tool in understanding and describing their ecology. Our results also reinforce the need for designing surveys that capture the underlying spatial structure of an ecosystem, especially when studying mobile aggregating species.