Development of integrated distance sampling models

Distance sampling methods are used to estimate abundance of biological populations. A set of randomly-placed lines or points are traversed and the distances to detected objects are recorded; these distances are used to estimate the probability of detection as a function of distance (the “detection function”) and hence infer how many objects were missed. In Chapter 1 we give the general context of the abundance estimation problem, the methods in use, and our motivating example. In Chapter 2 we revisit distance sampling and closed population capture-recapture methods in detail. We focus on the assumption of certain detection at distance zero, and endeavour to describe mark-recapture distance sampling (MRDS) methods and associated issues. In Chapter 3 we develop MRDS models to account for dependence between observers’ detections based on log-linear models for mark-recapture. With this simple parameterisation we are able to easily interpret the model parameters, extend the model to more than two observers, and understand what the implications of relaxing the independence assumptions are. In Chapter 4 we conducted an experiment using unmanned aerial vehicles (UAV) to sample availability considering the same group unit as the MRDS survey, where objects refer to groups. We model availability as a non-instantaneous process, where an object is subject to enter and leave the available state. Besides the standard exponential model for available and unavailable time intervals, we use Weibull and log-normal distributions. We also account for right censoring in the data and covariates using a hazard regression framework. We compare the UAV focal-follow approach to the time-depth data obtained using tagged animals. Finally, in Chapter 5, we integrate both the MRDS and availability models in order to estimate population size in absolute numbers. Throughout this thesis, our motivating example is a humpback whale breeding population off the Southwestern Atlantic ocean. For this, a double-observer aerial survey was carried out in 2015; time-depth data from tagged animals and focal-follow data collected from boat-based UAV surveys are used to estimate availability. The combined model allows us to estimate the most reliable abundance of humpback whales from aerial platform. We conclude that (i) there is lack of information to reliably estimate model parameters in various cases when we relax the independence assumption further than at distance zero; (ii) data collection to estimate availability using focal-follows from UAV platforms have advantages compared to time-depth recording data, and covariates describing the availability process may be important; (iii) detection and availability are not S-ancillary when availability depends on distance, which means that fitting the models separately produce biased estimates