Assessing and correcting for the effects of species misclassification during passive acoustic surveys of cetaceans
Abstract
In conservation ecology, abundance estimates are an important factor from which management decisions are
based. Methods to estimate abundance of cetaceans from visual detections are largely developed, whereas
parallel methods based on passive acoustic detections are still in their infancy. To estimate the abundance of
cetacean species using acoustic detection data, it is first necessary to correctly identify the species that are
detected. The current automatic PAMGUARD Whistle Classifier used to automatically identify whistle
detection of cetacean species is modified with the objective to facilitate the use of these detections to
estimate cetacean abundance. Given the variability of cetacean sounds within and between species,
developing an automated species classifier with a 100% correct classification probability for any species is
unfeasible. However, through the examples of two case studies it is shown that large and high quality
datasets with which to develop these automatic classifiers increase the probability of creating reliable
classifiers with low and precise misclassification probability.
Given that misclassification is unavoidable, it is necessary to consider the effect of misclassified detections
on the number of observed acoustic calls detected and thus on abundance estimates, and to develop robust
methods to cope with these misclassifications. Through both heuristic and Bayesian approaches it is
demonstrated that if misclassification probabilities are known or estimated precisely, it is possible to
estimate the true number of detected calls accurately and precisely. However, misclassification and
uncertainty increase the variance of the estimates. If the true numbers of detections from different species
are similar, then a small amount of misclassification between species and a small amount of uncertainty in
the probabilities of misclassification does not have a detrimental effect on the overall variance and bias of
the estimate. However, if there is a difference in the encounter rate between species calls associated with a
large amount of uncertainty in the probabilities of misclassification, then the variance of the estimates
becomes larger and the bias increases; this in return increases the variance and the bias of the final
abundance estimate. This study despite not bringing perfect results highlights for the first time the
importance of dealing with the problem of species misclassification for cetacean if acoustic detections are to
be used to estimate abundance of cetaceans.
Type
Thesis, PhD Doctor of Philosophy
Rights
Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported
http://creativecommons.org/licenses/by-nc-nd/3.0/
Collections
Except where otherwise noted within the work, this item's licence for re-use is described as Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported
Items in the St Andrews Research Repository are protected by copyright, with all rights reserved, unless otherwise indicated.