Validating a GPS collar-based method to estimate parturition events and calving locations for two barren-ground caribou herds

Abstract

In remote landscapes, it is difficult and expensive to document animal behaviors such as location and timing of parturition. When aerial surveys cannot be conducted as a result of weather, personnel or fiscal constraints, analyses of GPS collar movement data may provide an alternate way to estimate parturition rates and calving ground locations. I validated two methods (population-based method and individual-based method), developed to detect calving events of sedentary woodland caribou, on multiple years of data for two different migratory barren-ground caribou herds in Alaska, the Porcupine and Fortymile herds. I compared model estimates of population parturition rates, individual calving events, calving locations and calving dates to estimates from aerial survey data for both herds. For the Porcupine herd we also compared model estimates of annual calving ground sizes and locations of concentrated calving area centroids to those found with aerial survey. More years of data would be required for additional statistical power but for both the Porcupine and Fortymile herds, we found no significant difference between the population-based and individual-based method in: 1) individual classification rate accuracy (0.85 vs. 0.88, respectively; t = -7, P = 0.09, df = 1 and 0.85 vs. 0.83, respectively; t = 0.46, P = 0.69, df = 2) or 2) annual average distance from aerial survey calving locations (8.9 vs. 7.8 km, respectively; t = 0.16, P = 0.90, and 5.2 vs. 3.7 km, respectively; t = 1.03, P = 0.20). Median date of calving was estimated within 0-3 days of that estimated by aerial survey for both methods. Population parturition rate estimates from aerial survey, the population-based and individual-based methods were not significantly different for the PCH or FCH (0.91, 0.88 and 0.95, respectively; F = 0.67, P = 0.60, df = 2, and 0.83, 0.83 and 0.96, respectively; F = 3.85, P = 0.12, df = 2). Ultimately, more years of data would be required to support or reject the lack of significant differences between methods that we observed.