Analysis and prediction of faunal distributions from video and multi-beam sonar data using Markov models

We present a statistical framework for analysing video transect data from the marine environment. Variables observed in the video data, especially those describing marine fauna, are related to physical variables derived from coarser scale acoustic data that has much greater spatial coverage. The observations from the video data are multivariate, and their distribution is factorised conditionally into univariate distributions. We accommodate the auto-correlation in each conditional univariate distribution using a reversible Markov model, where the transition probabilities vary with the physical explanatory variables and the conditioning observed variables. Predictions for a random variable's stationary distribution, marginal to other observed variables, are made using it Suitably weighted average. Ail average prediction and ail approximation to its variance are given for large spatial areas. This is ail important application for resource management in the deep ocean where spatially based management approaches are commonly used, and where the cost of collecting fine-scale data is high. We demonstrate the method using data from the east coast of Tasmania, Australia. Copyright (C) 2008 John Wiley & Sons, Ltd.