Ungulate population monitoring in a tundra landscape: evaluating total counts and distance sampling accuracy

• Main Author: Le Moullec, Mathilde
• Format: Master Thesis
• Language: English
• Published: UiT Norges arktiske universitet 2014
• Subjects: Line transects; Detection probability; Poisson-Poisson sampling; Population size; State-space model; Rangifer; VDP::Mathematics and natural science: 400::Zoology and botany: 480::Ecology: 488; VDP::Matematikk og Naturvitenskap: 400::Zoologiske og botaniske fag: 480::Økologi: 488; BIO-3950; Arctic; Svalbard; Sarsøyra; Rangifer tarandus; Rangifer tarandus platyrhynchus; svalbard reindeer; Tundra
• Online Access: https://hdl.handle.net/10037/6554

Researchers and managers are constantly working towards decreasing monitoring uncertainties in order to improve inferences in population ecology. The solitary and sedentary Svalbard reindeer (Rangifer tarandus platyrhynchus) inhabit a high-Arctic tundra landscape highly suitable to compare accuracy (precision and bias) of population monitoring methods in the wild. The flexible Bayesian state-space model enabled me to assess uncertainties in estimates of the abundance of four reindeer sub-population time-series. In this environment, Total population Counts (TC) were more precise than Distance Sampling (DS), especially when conducted multiple times during a field season (e.g. Sarsøyra, summer 2013: DS Coefficient of Variation (CV)= 0.11, only one TC CV= 0.06; four repeated TC CV= 0.03). In addition, TC’s bias was assumed low once integrated in the state-space model and related to re-sightings of marked animals. Conducting DS alone, without TC as background information, would have estimated wrong reindeer population size because the detection function was sensitive to sample size. However, the similarity in landscape and methodology across the two neighboring DS study sites enabled their observations (n= 143) to be pooled, resulting in more plausible estimates, yet slightly higher than those found through TC. DS is used worldwide and this study illustrates fundamental issues around the minimum sample sizes recommended in literature (n>80) and that the number or length of transects must be sufficient to represent habitat structure (in this particular case the proportion of vegetation). Furthermore, combining multiple sources of available data in a common modeling framework, even with wide standard deviation such as DS, resulted in more precise estimates.