Movement and habitat use patterns are not the same for wildcat hybrids in Scotland

Connectivity analysis has become a cornerstone of applied ecology and conservation. Recent work has evaluated the relative performance of different connectivity algorithms, but other factors affecting connectivity predictions have been somewhat less explored. Past work has shown that there are often large differences in habitat selection at home range vs. selection at the scale of movement paths or steps. However, a large portion of connectivity modeling work still uses transformed habitat suitability maps as a basis for developing resistance surfaces. The degree to which transformed habitat suitability maps correspond to the factors that drive movement and connectivity deserves renewed attention. In this paper we used spatial path randomization coupled with conditional logistic regression (path selection functions) to predict the factors that drive the movement patterns of wildcat (Felis silvestris) X domestic cat (F.catus) hybrids in Scotland. We chose this system based on previous work that developed habitat selection models, which thereby provides a direct comparison to predictions of the path selection functions. Specifically, we were able to use the same input data (movement points) and the same environmental predictors for both the point selection and path selection approaches, enabling for the first time an unequivocal comparison of the differences in predictions between the two approaches. We tested four hypotheses: (1) We expected that variables selected for movement would often be different to variables selected for habitat use, with those reflecting general ecological conditions (such as soil type and elevation) more frequently selected in habitat models while variables more associated with structural factors directly affecting movement paths (slope position, linear vegetation features, areas with high concentrations of prey) more frequently selected in the path selection approach. (2) We expected that the scale at which variables were selected would differ as well, with the habitat selection models selecting variables at broader scales than path selection functions, given that the habitat selection models are predicting 2nd order patterns of suitability of home ranges, while the path selection functions are predicting 3rd order patterns of movement within home ranges, which likely are related to finer scale environmental conditions than is the selection of home ranges themselves (Johnson 1980). (3) As a result of the first two hypotheses, we further expected that the predicted patterns of suitability would differ between habitat and path selection functions, with habitat models showing smoother, broader patterns of regional variation in suitability, and path selection functions showing finer scale variation. (4) We expected that connectivity model predictions produced from a combination of different resistance surfaces, source points and dispersal abilities would show distinctive patterns with connectivity predictions most sensitive to which resistance model was used, followed by which model was used to produce source points and finally that dispersal ability would have relatively small influences on the differences between connectivity models. Our results confirmed all four hypotheses. The path selection functions did not identify the same suite of variables as the habitat models, with some variables related to topographic structure (slope position) and land cover (heath) much more important in the path selection than habitat models. Conversely, some variables that were very important in habitat models (elevation, forest cover, human development), were absent in the path selection models. Consistent with our second hypothesis, the scales at which variables were selected were finer in path selection than in habitat selection, and as predicted by the third hypothesis the patterns of suitability predicted by these two models were different, with the path selection model identifying areas in valley bottoms, with high solar radiation, highly productive vegetation and high density of rabbit (a preferred prey species) as the most suitable for movement path selection, whereas the habitat suitability model predicted broad areas associated with forest cover at middle to low elevations in productive soil types with abundant rabbit prey as optimal habitat. Finally, consistent with the fourth hypothesis, connectivity model predictions were highly sensitive to which model was used to produce the resistance surface, and somewhat less sensitive to which model was used to select source points for connectivity analysis. This study illustrates that predictions of habitat use and movement patterns at different orders of selection may produce divergent pictures of the factors, scales and patterns that influence organism occurrence and movement.