Abstract Peripheral populations, though often small and isolated, can hold unique genetic diversity and local adaptations vital for species' resilience to environmental changes. However, their conservation can imply the problem of obtaining reliable context‐specific assessments of habitat preferences when dealing with very low sample sizes, many potentially relevant environmental factors to test and multicollinearity issues. In this study, we investigated the fine‐scale habitat preferences of a critically endangered, peripheral population of barred warbler Curruca nisoria in the Italian Alps, represented by a few tens of breeding pairs. Our aim was to provide detailed information for barred warbler habitat management and conservation in an Alpine context. We surveyed the species through territory mapping, and we investigated habitat preferences at the breeding territory scale (1 ha) by comparing 21 presence and 21 control plots, considering several environmental and topographic variables. Given the low sample and the many, partly intercorrelated predictors to test, we adopted a two‐step approach: (1) we used a partial least squares discriminant analysis (PLS‐DA) to identify the more influential predictors and (2) we used them to fit multivariate adaptive regression splines (MARS) models. We also applied a specific multi‐model check to avoid any arbitrariness in the predictors' selection procedure. The first step provided a reduced set of uncorrelated predictors. The MARS final model included hedgerows and bushes cover and terrain slope and showed good explanatory power (classification accuracy = 0.83). Hedgerows showed no effect below 19 m length and a positive effect above this value, determining high barred warbler occurrence probability (≥0.8) when longer than 40 m. Occurrence probability also increased with higher bushes cover, reaching high values from c. 10% cover. Barred warblers avoided the steepest areas, with a negative effect of slope above 20.2°. Practical Implication : the combination of PLS‐DA and MARS allowed us to obtain detailed and context‐specific information on habitat conservation and management for a very small and critically endangered population. This flexible approach is highly suited when dealing with many intercorrelated predictors and few observations and can be applied to both presence/absence and abundance data using the appropriate PLS and MARS methods.
Ceresa et al. (Thu,) studied this question.