Abstract

A likelihood-based geostatistical method for estimating fish biomass from survey data is presented. Biomass estimates from analysis of a positive random variable with an additional discrete probability mass at zero means that the method accommodates null observations and positive fish density. The positive fish density data were used to estimate mean fish density in the subareas where the stock was present. A presence/absence representation of the data in the survey area was modelled with a generalized linear spatial model of the binomial family, leading to an estimate of the area effectively occupied by the stock. As an extension, a procedure is proposed to accommodate extra sources of correlation, such as multiple surveys or multiple vessels. The new methodology was applied to three cases. The simplest case is a scallop trawl survey for which only the positive density data need to be analysed. The intermediate case is a trawl survey of highly mobile squid where the stock area and the mean density inside the stock area are analysed. The most complex case is in estimating the biomass of very localized orange roughy, for which repeat surveys create dependence in the data in addition to spatial correlation. © 2007 International Council for the Exploration of the Sea. Published by Oxford Journals. All rights reserved.