Predicting the breeding distribution of wader species across climatic and environmental gradients

Developing and applying species distribution models is particularly urgent for species currently facing high levels of environmental change. Waders (Charadrii) are among the most species rich groups of birds breeding at higher latitudes and the one where species are declining fastest. Iceland holds a large part of the global population of several wader species breeding sympatrically across unusual climate and environmental gradients. This provides an opportunity to assess synergy in habitat selection and occurrence of species usually occupying wide latitudinal ranges from temperate to arctic. In addition, the Icelandic wader habitats are facing significant and increasing pressures from anthropogenic sources putting in risk these populations. We built distribution models for several waders with the aim of identifying synergies in occurrence which allows better prioritisation of conservation efforts.

Methods
Species occurrence data in the lowland areas (<200m a.s.l.) of Iceland were derived from georeferenced point count data (standardized 5 min. point counts with a 200m radius) carried out during  during the breeding season (May and June) from 2007 to 2019. Species occurrence data in the highland areas (>200m a.s.l.) were derived from georeferenced transect counts carried out during June and July from 1999 to 2004. For each transect, we used the species counted in the midpoint of the transect and excluded counts recorded with distance to the focal bird >200 m or counts recorded with no distance. All counts were transformed to presence data, so the species occurrence dataset consists of a set of longitude and latitude coordinates.

Species distribution models (SDMs) were generated using Bayesian Additive Regression Trees (BART) following the methodology/code described in Carlson (2020). BART models were run with the default parameters as implemented in dbarts in the R package embarcadero, using 200 trees and 1000 back-fitting Marcov Chain Monte Carlo (MCMC) iterations. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) and true skill statistics (TSS)