Optimal sampling methods for modelling the occupancy of Arctic grayling (Thymallus arcticus) in the Canadian Barrenlands

Abstract

In occupancy models, imperfect detectability of animals is usually corrected for by using temporally repeated surveys to estimate probability of detection. Substituting spatial replicates for temporal replicates could be an advantageous sampling strategy in remote Arctic regions, but may lead to serious violations of model assumptions. Using a case study of site occupancy of adfluvial young-of-year Arctic grayling (Thymallus arcticus) in Barrenland tundra streams, we assessed the reliability and efficiency of alternative sampling strategies: (i) randomly distributed versus sequential adjacent spatial replicates; (ii) visual versus electrofishing surveys; and (iii) spatial versus temporal replicates. Sequential, adjacent spatial replicates produced spatially autocorrelated data. Autocorrelation was relieved using randomly distributed spatial replicates, but using these randomly distributed spatial replicates introduced significant error into estimates of the probability of occupancy in streams. Models designed for spatially autocorrelated data could minimize this bias. Visual and electrofishing surveys produced comparable probabilities of detection. Spatially replicated surveys performed better than temporal replicates. The easiest and relatively most cost-effective sampling methods performed as well as, or better than, the more established, expensive, and logistically difficult alternatives for occupancy estimation.