Estimating fish mortalities and cohort sizes

A cohort-based regression model (J.E. Paloheimo. 1980. Trans. Am. Fish. Soc. 109:378 386) has been frequently used in fish stock assessment. With a time series of catch-at-age and effort as input data, this model yields estimates of catchability, natural mortality, and cohort sizes at recruitment using a multiple regression analysis. Limitations in the error structure have been cited as a potential problem with the method. The purpose of this study is to develop a suitable error structure for this model and evaluate the effect of such an error structure on parameter estimation. The proposed error structure depends on variances in input data that are not generally known. Using simulated data both with correct and deliberately wrong input variances we observed that the estimates were not particularly sensitive to the magnitudes of these variances, but depended primarily on the structure and signs of elements in the variance covariance matrix. The amended model does result in improved parameter estimates compared with the original model.