摘要：Conventional distance sampling approaches rely on the design assumption (i.e. uniform distribution of individuals in relation to transects) to ensure unbiased inference to the population of interest. However, randomized design recommendations are not always followed or may be impractical to implement for some survey types, particularly in cases where transects must be placed perpendicular to the habitat gradient. Full-likelihood spatial distance sampling models provide a potential solution to violations of the design assumption by jointly modelling the detection and occurrence processes using spatially indexed habitat covariates. Through simulation and an applied example based on a survey for Dall's sheep in Alaska, USA, we used a full-likelihood distance sampling approach to investigate the potential for bias in cases where transects placed perpendicular to the habitat gradient (e.g. elevational contours) are non-randomly sampled. We also assessed the utility of spatial approaches in cases where transects are placed along linear features, such as roads or ridgelines, where habitat may be unrepresentative of the overall study area. Our results showed that the full-likelihood approach was generally unbiased, even in extreme scenarios where habitat was inversely related to distance from the transect. For the Dall's sheep example, our results showed that more efficient designs with reduced sampling effort in low-quality habitats are a practical solution for reducing logistical costs when the data are analysed in a spatial modelling framework. Together, our findings confirm and extend existing work suggesting that spatial distance sampling can be a useful solution when non-random designs are employed. Given the high cost of survey implementation in many cases, the development of valid alternatives to design-based inference will aid in the amount of information available for a variety of species. The results of our work will be useful for practitioners in assessing alternative designs relative to particular survey applications.
