GIS-based application of resource selection functions to the prediction of snow petrel distribution and abundance in East Antarctica: Comparing models at multiple scales

Abstract

Snow petrel numbers must be of the order of several millions. However, accurate population estimates are sparse although such information is necessary to monitor potential changes in the Antarctic ecosystem. A census of snow petrel nests was conducted at Casey (East Antarctica) during summer 2002–2003. Twenty percent of the ice-free areas (available nesting habitat for snow petrels) was surveyed using a “random block design”. During this survey, approximately 5000 nests were located. Generalized additive and linear modelling techniques and classification trees (GAM, GLM and CT) were used to fit resource selection functions, which modelled snow petrel abundance or presence–absence in relation to a set of environmental predictors (elevation, slope, aspect, curvature and substrate types estimated in percentage cover). The effect of spatial scale on the processes that influence habitat selection was investigated using GIS as a tool to create and test models at a hierarchical range of scales—from 200 m grid-sites level to 20 m quadrats. The strong predictive value of aspect, slope and percent cover in boulder and SCREE were identified at all scales. However, the significance of environmental predictors varied with scale, indicating that spatial scale matters in detecting habitat selection processes. In general, models were improved with the addition of spatial dependence terms representing the effect of conspecific attraction (coloniality), but these models were less applicable for predictive purposes. By predicting abundance from environmental characteristics (acquired for example, using aerial photography), resource selection functions may be a useful tool to refine population estimates of several petrel species in Antarctica without requiring intensive ground surveys.