Modelling spatial and environmental effects at the edge of the distribution: the red-backed shrike Lanius collurio in Northern Portugal

This study examined the interplay of spatial and environmental effects shaping the range margin of the red‐backed shrike (Lanius collurio) in northern Portugal. The occurrence of shrikes in 10 × 10 km UTM squares was related to three sets of explanatory variables, reflecting environmental effects (climate and habitat), large‐scale spatial trends, and neighbourhood influences (considering an autologistic term); spatial variables were used as surrogates for historical and demographic factors. Multiple logistic regression models were built for each set, and then variation partitioning based on partial regressions isolated the unique and shared components of explained variation. The environmental model revealed a dominant influence of climate effects, with the occurrence of shrikes increasing with frost and thermal amplitude, declining with insolation, and responding unimodally to rainfall. There was a weaker influence of habitat conditions, though shrikes were more likely with increasing cover by annual crops and pastures, and decreasing forest cover. Only a relatively small proportion of explained variation was due to a ‘pure’ environmental component (10.4%), as most variation explained by environmental factors appeared spatially structured (51.9%). The unique contributions of spatial variables to the overall model were also small, though the neighbourhood effects appeared relatively stronger than large‐scale trends. Taken together, results suggested that the south‐western range margin of the red‐backed shrike was largely determined by spatially structured environmental factors. Nevertheless, there were also ‘pure’ environmental factors determining some isolate occurrences irrespective of any spatial structure, and ‘pure’ spatial factors that appeared to favour the occupation of squares surrounding the core distribution areas irrespective of environmental conditions. These results add to the growing evidence that both environmental and spatial factors need to be considered in predictive modelling of species range margins.