Development and evaluation of inductive and deductive models of elk (Cervus elaphus) summer resource suitability in Northwestern Ontario

Abstract

Resource-selection modelling techniques take either a deductive or inductive approach. Deductive methods take a “bottom-up” approach, where individual wildlife-resource relationships are the building blocks that define the suitability of a landscape. Conversely, inductive models imply a “top-down approach”; the suitability of a landscape is predefined by animal use, and statistics are used to identify wildlife-resource relationships. Current modelling caters towards the inductive approach even though a formal comparison between inductive and deductive techniques has yet to identify the superior method. The Ontario Elk Restoration Project re-introduced 104 elk (Cervus elaphus) to the Lake of the Woods (LOW) region in 2000 and 2001. The population diminished substantially over the following four years bringing concern to the successful re-establishment of elk in northwestern Ontario. At present, explanations for this decline are speculative in nature, but one possibility is that the landscape does not contain the resources required to support a viable population. To address this concern, I investigated the ability of the landscape to support elk by creating a series of deductive (Habitat Suitability Index; HSI) models for the summer season based on published elk-resource relationships. I also created an inductive (Resource Utilization Function; RUE) model based on radiotelemetry location data collected weekly throughout 2000-2005 in an effort to improve upon the initial deductive model. Finally, I compared the two modelling approaches to address the growing disparity between modelling methods. Models were validated by comparing model suitability to elk resource use using weighted overlap (WO) and average overlap (AO) indices, and by comparing model suitability to elk space use (utilization distributions; UDs) using the deviation from a ‘no selection’ pattern (DVI Index); a modification of the Volume of Intersection (VI) Index. Comparison to a null model was also used to benchmark each index. I also calculated the VI Index between the predicted suitability’s of RUF and HSI models to compare their overall similarity.