Predicting waterfowl distribution in the central Canadian arctic using remotely sensed habitat data

Abstract

Knowledge of a species’ habitat-use patterns, as well as an understanding of the distribution and spatial arrangement of preferred habitat, is essential for developing comprehensive management or conservation plans. This information is absent for many species, especially so for those living or breeding in remote areas. Habitat-use models can assist in delineating specific habitat requirements or preferences of a species. When coupled with geographic information system (GIS) technology, such models are now frequently used to identify important habitats and to better define species’ distributions.

Recent and persistent warming, widespread contaminant accumulation, and intensifying land use in the arctic heighten the urgent need for better information about spatial distributions and key habitats for northern wildlife. Here, I used aerial survey and corresponding digital land cover data to investigate breeding-ground distributions and landscape-level habitat associations of greater white-fronted geese (Anser albifrons frontalis), small Canada geese (Branta canadensis hutchinsii), tundra swans (Cygnus columbianus), king eiders (Somateria spectabilis), and long-tailed ducks (Clangula hyemalis) in the Queen Maud Gulf Migratory Bird Sanctuary and the Rasmussen Lowlands, Nunavut, Canada.

First, I addressed the sensitivity of inferences about predicting waterfowl presence on the basis of the amounts and configurations of arctic habitat sampled at four scales. Detection and direction of relationships of focal species with land cover covariates often varied when land cover data were analysed at different scales. For instance, patterns of habitat use for a given species at one spatial scale may not necessarily be predicted from patterns arising from measurements taken at other scales. Thus, inference based on species-habitat patterns from some scales may lead to inaccurate depictions of how habitat influences species. Potential variation in species-environment relationships relative to spatial scale needs to be acknowledged by wildlife managers to avoid inappropriate management decisions.

Second, I used bird presence determined during aerial surveys and classified satellite imagery to develop species-habitat models for describing breeding-ground distributions and habitat associations of each focal species. Logistic regression models identified lowland land cover types to be particularly important for the species considered. I used the Receiver Operating Characteristic (ROC) technique and the area under the curve (AUC) metric to evaluate the precision of models, where the AUC is equal to the probability that two randomly selected encounter and non-encounter survey segments will be discriminated as such by the model. In the Queen Maud Gulf, AUC values indicated reasonable model discrimination for white-fronted geese, Canada geese, and tundra swans (i.e, AUC > 0.7). Precision of species-habitat models for king eiders and long-tailed ducks was lower than other species considered, but predict encounters and non-encounters significantly better than the null model. For all species, precision of species-habitat models was lower in the Rasmussen Lowlands than in the Queen Maud Gulf, although discrimination ability remained significantly better than the null model for three of five species (king eider and long-tailed duck models performed no better than the null model here).

Finally, I simulated anticipated environmental change (i.e., climate warming) in the arctic by applying species-habitat models to manipulated land cover data, and then predicted distributional responses of focal species. All species considered in this research exhibited some association to lowland cover types; white-fronted geese, Canada geese, and tundra swans in particular demonstrated strong affinity toward these habitats. Others authors predict lowland cover types to be most affected by warming. Reductions of wet sedge, hummock, and tussock graminoid cover predicted in this simulation, predominantly along the coast of the Queen Maud Gulf study area and in central areas of the Rasmussen Lowlands, suggest that distributions of species dependant on these lowland habitats will be significantly reduced, if predictions about warming and habitat loss prove to be correct. Research presented here provides evidence that modeling of species’ distributions using landscape-level habitat data is a tractable method to identify habitat associations, to determine key habitats and regions, and to forecast species’ responses to environmental changes.