Investigations into the use of quantified Bayesian Maximum Entropy methods to generate improved distribution maps and biomass estimates from fisheries acoustic survey data

Abstract

This thesis describes elements of the assessment and application of Bayesian Maximum Entropy (MaxEnt) image reconstruction techniques for the analysis of fisheries acoustic survey data. The objective is to investigate the utility of this approach in mapping density distributions and estimating biomass. The MaxEnt image reconstruction method derives originally from the field of astrophysics, and this thesis represents an attempt to apply the principles of MaxEnt to the field of ocean ecology. Essentially, what is required is to generate maps of the density distribution of pelagic species (species living in the water column) from extremely limited and sometimes skewed line-transect acoustic survey data. Techniques used presently are largely unsatisfactory for a variety of reasons, and are often inapplicable for data from surveys that do not follow a particular design strategy. This thesis investigates the usefulness of the MaxEnt technique in overcoming some of the difficulties of acoustic survey analysis. A study is made into the possibility of objectively testing whether these techniques offer improvements in accuracy over existing techniques, by attempting to reconstruct simulated data from a virtual survey. I find that plausible reconstructions are possible, and that statistical comparisons indicate these reconstructions are accurate. The technique is also applied quantitatively to real-world survey data, offering new insights into the abundance of Antarctic krill (Euphausia superba) in the Scotia Sea - raising abundance estimates from 109 million tonnes to 208 million tonnes - and into the relative abundance of fish and jellyfish in the Namibian Benguela, where it is shown that the biomass of jellyfish (12.2 million tonnes) now exceeds that of fish (3.6 million tonnes).