Mule deer have been declining in Oregon and understanding the ecological relationships between mule deer and their habitat is vital for conservation. Of particular importance is how females use habitat during summer when they are rearing fawns. We conducted a large regional assessment of 428 mule deer wearing GPS collars across 22,000 km2 of summer range in south central Oregon from 2006-2012. We first estimated a utilization distribution with a Brownian Bridge model to represent probability of use within 150-m grid cells across our study area, and we used this probability value as the response variable in our analysis. We then developed a set of 34 environmental variables on the same 150-m cells across our study area. We estimated models with linear regression and selected top models based on Bayesian Information Criterion (BIC). We conducted a validation analysis on an independent sample of mule deer and a goodness of fit analysis in grid cells not used in model development. For each of these analyses we compared predicted to observed use and calculated Spearman rank correlation coefficients.
Results/Conclusions
The top model contained covariates of elevation, tree canopy cover, distance to perennial water, distance to harvest 6-14 years old, distance to wildfire 6-14 years old, distance to low traffic roads, and distance to forest edge. Coefficients for elevation, tree canopy cover and distance to perennial water contained a squared term, indicating a positive curved relationship to deer use. Optimum values for elevation were 1500 m, for tree canopy cover 37%, and for distance to water 1.2 km. The coefficient for distance to harvest was positive, indicating avoidance of harvested areas 6-14 years old. The coefficients for distance to low traffic roads, distance to edge, and distance to fire 6-14 year-old were negative, indicating that deer use was highest near these environmental features. Spearman rank correlation coefficients for and independent dataset resulted in an R2 of 0.97. These results are preliminary and do not yet include confidence intervals or results of the goodness of fit. This model may be used in land management planning to assess and predict changes in probability of use across a landscape when environmental features change. We will show how alternative scenarios of vegetation management may change probability of use within a project area.