Incorporating predicted species distribution in adaptive and conventional sampling designs

Introduction Monitoring rare and clustered populations is challenging because of the large effort required to encounter occupied habitat and yield precise population estimates (Mc Donald 2004). Sampling designs are available to help reduce the effort required to encounter occupied habitat and increase precision, including stratified sampling, probability proportional to size (PPS) sampling, and various adaptive sampling designs (Thompson 2002). Use of these designs is motivated, in an intuitive sense, by each design's ability to allocate more sampling effort where target species are (or are likely to be) and less where they are not. This intuitive approach to allocation of effort can lead to increased precision when variability in the population tends to be higher in areas of high species density or abundance (Box 17.1). Conventional designs, such as stratified and PPS sampling, rely on prior information to allocate effort. For example, prior information could come from predicted species or habitat distributions (Guisan and Zimmermann 2000, Le Lay et al. 2010). Use of prior information is not a basic property of adaptive sampling designs, but these designs could use such information when available. In this chapter, we demonstrate how prior information on species distribution can be incorporated into adaptive and conventional sampling designs. We start the chapter by introducing adaptive sampling, which remains a somewhat novel sampling design even though it was introduced by Thompson (1990) over 20 years ago. We then present a case study illustrating and evaluating the performance of conventional and adaptive sampling designs that either incorporate or ignore predicted species distributions. We examine how these designs compare in terms of efficiency, probability of sampling occupied habitat, and robustness to model inaccuracy. We end the chapter with recommendations and a discussion of future research and developments.