COMPUTING AND SELECTING AGEING ERRORS TO INCLUDE IN STOCK ASSESSMENT MODELS OF PACIFIC SARDINE (SARDINOPS SAGAX)

From 2007 to 2010, Pacific sardine stock assessments relied on traditional methods to compute and include ageing errors in the integrated assessment model, Stock Synthesis (SS). Traditional methods assumed that all age readers were unbiased and estimated ageing imprecisions by averaging across all fish that were assigned a given age a by one or more readers. In this study, we used the Ageing Error Matrix (Agemat) model to compute ageing imprecisions, based on classification matrices that quantified the probability of a fish of true age a to be assigned an age a or some other age a', P(a'|a). Using sardine samples collected from Mexico to Canada and aged in five labora tories, we compared three Agemat models, assuming that: (1) the most experienced reader from each laboratory was unbiased (model A); (2) no bias but different stan dard deviation (SD) at age among readers (model B); and (3) no bias, but similar standard deviation at age among readers (model C). We evaluated the performance of this model using the Akaike information criterion corrected for finite sample sizes. Sardine ages ranged from 0 to 8, with increasing reader SD with age. Model C performed better than models A and B, across all data sets and laboratories, and thus was recommended for including ageing imprecisions in sardine assessment models. However, the observed differences in SD across ages and readers called for a better standardization of ageing protocols among laboratories and for applying new methods to reduce potential bias in estimating the oldest age classes.