A Likelihood-Based Model of Fish Growth With Multiple Length Frequency Data

A model for individual growth is an important component in the analysis of fishery resources. Time series of body-length frequency data (LFD) contain information on average individual growth, but age is a latent variable and this makes it difficult to extract growth information objectively. A new method is described here that relaxes some of the assumptions of currently available approaches and that can be used even with large gaps in the time series of LFD. The new method is based on a non-Bayesian hierarchical model where the growth model contains hyper-parameters that depend on parameters of normal mixture models underlying the LFD. Growth hyper-parameters are estimated using a multivariate normal marginal-estimated likelihood function. The method is applied to the squat lobster Pleuroncodes monodon with LFD from six surveys, leading to a precise model of individual growth due to the large sample size in the original LFD sets. ADMB computer code and datasets used in this article are supplied as supplemental material.