Geolocator Data Processing
After downloading the data from the recovered geolocators, we corrected the data for clock drift when possible, and twilight times were estimated using the BAStag package in R (Wotherspoon et al. 2016; R Core Team 2020). A light threshold of 1.5 with a dark time minimum of 30 minutes was used to calculate initial estimates of twilight. A biologist audited this process’s results and edited twilights when they were obviously incorrect based on the surrounding estimates. Intervention was rare (e.g., no more than three edits were made per track) to avoid process reproducibility issues. A 0.8 zenith angle tolerance eliminated latitude estimates for positions around the equinoxes. These twilight times were used to estimate initial positions for each day. Outlier positions more than two standard deviations from the running 7-day average of twilight times were removed.