COS 145-1
Computer vision facilitated ultra-high resolution habitat suitability models for Pygoscelis spp. penguins

Friday, August 14, 2015: 8:00 AM
338, Baltimore Convention Center
Phil McDowall, Stony Brook University
Heather Lynch, Ecology & Evolution Department, Stony Brook University, Stony Brook, NY

Organisms exist in a complex three-dimensional environment, and their interactions with the topography of the terrain may be as important to successful reproduction as regional-scale variables such as temperature or precipitation. For seabirds, habitat suitability represents a complex balance between the fine-scale three-dimensional structure of the terrain on which they nest and the marine conditions that determine prey availability near the colony. While datasets on large-scale environmental conditions are often used in habitat suitability modeling, this is primarily due to their availability rather than their relevance to the organism. Rarely do such models include information on the availability of suitable terrain for nesting. 

As climatic conditions on the Antarctic Peninsula change, we are seeing shifts in both population size and the geographic ranges of all three Pygoscelis spp. penguins. We use structure-from-motion, a computer vision technique, to simultaneously record the three-dimensional structure of the landscape and the positions of the organisms within that landscape, at sub-centimeter scales. Using the data derived from structure-from-motion and accessory environmental data, we construct Bayesian use-availability models for Pygoscelis penguins, using those variables relevant to individual penguins deciding where to nest (e.g., elevation, slope, aspect, and derived neighborhood measures using high-resolution terrain information).


We find that gentoo penguin (P. papua) nests are positioned on localized peaks with low probability of flooding, on sites with shallow or no slope, and at a minimal cost-distance from access points to the ocean. Even when autocorrelated abiotic factors such as slope and aspect are included in the model, we find strong evidence for biotically-driven autocorrelation in nesting. The degree to which conspecific attraction appears to be driving nest site selection supports the idea that coloniality in seabirds is functionally-driven rather than the result of terrain limitation, and that colonization is therefore likely to be a rare event. These results allow us to establish where currently unoccupied, but suitable, habitat might exist; support the generation of range forecasts that includes both marine and terrestrial factors important to individual-level decision-making; and aid in the development of population models that explicitly include within colony dynamics. Fine-scale terrain mapping allows us to better understand which factors are important in defining suitable seabird habitat, and demonstrate that nest site selection is the result of interactions between hydrology, accessibility and the presence of conspecifics.