Modelling habitat selection using presence-only data: Case study of a colonial hollow nesting bird, the snow petrel

Little is known on distribution and abundance of snow petrels (Pagodroma nivea) in Antarctica. Studying habitat selection through modelling may provide useful information on the relationships between this species and its environment, especially on potential effects of climate change as modifications of nest sites availability and/or quality may arise in the long term. Based on surveys of the Windmill Islands during summer 2002-2003, several types of habitat suitability models were tested to relate the location of 4000 snow petrel nests with a set of environmental predictors describing rock substrate and topography, which proved useful predictors to identify favourable habitat for snow petrels. GAMs were used in the exploratory phase of the analyses, guiding the parametrization of GLMs, complemented by CT models. All three models rely on randomly generated pseudo-absence points. Their predictive performance was compared to that of a fundamentally different model using presence-only data, ENFA. Based on environmental envelopes, ENFA proved slightly less accurate than GLMs, given the available input data. However, ENFA predictions provide a useful starting point to categorize habitat suitability prior to the application of other modelling techniques. All models were ameliorated by the addition of predictors accounting for the spatial clustering of nests, but are of limited use to predict nest distribution in new areas. Clustering was caused by both coloniality and the underlying spatial structure of the substrate, which constrained nest densities through the distribution of available nest sites. Models fitted only with environmental predictors were refined when random points were separated from nest aggregations with a buffer scaled to the average size of colonies (identified with a simulation). Buffering was redundant for the models including neighbour information. overall, when fitting models with presence-only data points, improvements are obtained when accounting for the spatial clustering of the species. Given the highly specific nesting requirements of snow petrels, the observed nest clustering may be more constrained by the availability of suitable nests sites rather than driven by conspecific attraction, which emphasizes the importance of microhabitat quality in the habitat selection process.