Assessing fine-scale movement decisions
To understand how hyenas select landscape features at the fine/step scale, we derived step selection functions (SSFs) using the amt package in R (R Core Team, 2018). We prepared the hyena data by creating tracks from the data using the mk_track() package, subset the data to only the 5-minute fix rates, and filtered to assure bursts would have a minimum of 2 points. Five random steps were generated for each step used. Scaled covariates and model comparisons reflected those conducted for the RSF analyses. We fit conditional logistic regressions on the covariates, while also considering hyena ID as a cluster term and log of step length (i.e., speed of movement) as an interaction term with boundaries and roads. We used quasi-likelihood independence model criterion (QIC) to rank models and determine top models. We then used acf.test() on the model that best predicted the data to determine the lag at which autocorrelation is no longer observed, and employed destructive sampling to address autocorrelation, removing 2 points between each individual’s clusters. Models were then fit on the destructively sampled data. Last, we created a function in R for individual SSF models and the parameters from the global model to visualize the data.