Distance Sampling: 2D or not 2D? That is the question.
Speaker: David Borchers (CREEM)
Distance sampling is clearly a two-dimensional* process: it involves searching a plane and detecting objects at points in the plane. But distance sampling models are almost all one dimensional, using only perpendicular distance data or radial distance data. Why is this? What is lost by collapsing two dimensions into one?
It was not always thus. Distance sampling models used to be 2D. Before 1983, line transect models used both forward and perpendicular distances, for example. Hayes and Buckland (1983) changed this by demonstrating the robustness of 1D models compared to the 2D models available at the time. But in recent years there has been a resurgence of two-dimensional distance sampling models. In this talk I formulate distance sampling models as 2D survival models (as did Hayes and Buckland, 1983) and demonstrate how such 2D models can do things that 1D models cannot. This includes dealing with responsive animal movement prior to detection, dealing with stochastic animal availability, and in some circumstances, estimating detection probability at distance zero from single-observer data without any data external to the distance sampling survey.
(* Distance sampling is three-dimensional if you include height or depth.)