COS 23-1
Modeling habitat selection across time: A dynamic approach

Tuesday, August 11, 2015: 8:00 AM
321, Baltimore Convention Center
Carrie Roever, Agriculture Program, Oregon State University, La Grande, OR
Michael Wisdom, Pacific Northwest Research Station, USDA Forest Service, La Grande, OR
Mary Rowland, Pacific Northwest Research Station, USDA Forest Service, La Grande, OR
Martin Vavra, Pacific Northwest Research Station, USDA Forest Service, La Grande, OR
Timothy DelCurto, Agriculture Program, Oregon State University, La Grande, OR

Animals continually assess their environment and optimize the risks and rewards therein to maximize their long-term fitness.  These activities have a temporal component, where finding secure cover may take priority when raising young in the spring and finding food may take priority in the fall as plants begin to senesce.  Habitat selection methodology has yet to adequately incorporate these temporally dynamic priorities; thus, ecologists build separate seasonal habitat selection models that can only be compared qualitatively.  Here, we present a methodology to quantitatively assess how habitat selection changes as a function of a temporally dynamic environmental factor, precipitation.  We developed two approaches to accommodate both novice and experienced users of mixed effects models.  Using the simplified two-step modeling procedure, we fit separate habitat selection models using data from cattle (Bos taurus) and identified how the selection coefficients changed as a function of precipitation.  We then compared the two-step procedure to a one-step procedure using mixed effects logistic regression models. 


We found that including a temporally dynamic factor improved model fit and significantly explained the temporal variation in selection for 11 of 15 of the habitat covariates examined.  For the covariates “elevation to water” and “pine/fir forest cover,” selection coefficients changed sign (i.e., the regression line crossed zero), meaning that cattle switched from selection to avoidance of the feature or vice versa.  A change in sign of multiple selection coefficients would have required the use of separate, seasonal habitat selection models in the past.  We also found that two habitat selection estimations had poor fit (rho < 0.80) during periods of low rainfall.  Using our quantitative approach, this lack of fit allowed us to generate alternative hypotheses about factors driving temporal changes in selection during this period.  Quantitative assessments also allow for predictions into periods when precipitation values were outside of those observed during the study period.  Under increasingly uncertain climatic futures, it is imperative that habitat selection methodology strives to become more robust when making predictions beyond the dataset to create a more mechanistic understanding of species-habitat relationships.