Most current wildlife habitat models, such as resource selection functions, typically assume a static environment, extrapolate poorly in space and time, and often lack linkages to population processes. We submit that more mechanistic habitat models that directly consider bottom-up resources affecting growth and reproduction (i.e., food) and top-down limitations affecting survival are needed to effectively predict habitat quality, especially in the presence of rapid environmental change. Here we present a general model for estimating potential habitat quality (relating to growth and reproduction) and realised habitat quality (accounting for survival) using basic knowledge of the species’ seasonal diet, predicted locations of food resource patches and regional patterns in mortality risk. We illustrate our model for a threatened population of grizzly bears in west-central Alberta. Bi-monthly potential habitat quality successfully predicted habitat selection by radio-collared grizzly bears, while multi-seasonal realised habitat quality predicted patterns in occupancy-abundance as measured from unique bears at hair-snag sites. Bottom-up resources therefore predicted patterns of habitat selection, while top-down processes (survival) were necessary to scale-up to population measures. We suggest that more direct measures of
resources and environments that affect growth, reproduction and survival, as well as match the temporal scale of animal behaviour, be considered when developing wildlife habitat models

Diet Habitat deficit Habitat quality Habitat selection Ursus arctos