Estimability of Time-Varying Natural Mortality in Gulf of Alaska Sablefish with a Simulated Covariate

Natural mortality (M) is a notoriously difficult parameter to estimate within stock assessment models, but it is one of the most important to obtain an accurate assessment. The paucity of data to inform M and the tendency for estimates of M to be confounded with other parameter estimates such as gear selectivity contribute to this difficulty. It is therefore common practice to fix M at a constant value over time, despite the recognition that M must vary in response to ecosystem changes. Different camps have proposed to (1) estimate time-varying M in a random effects setting, and (2) to use a biological or environmental covariate as a predictor of M. We conducted a simulation-estimation exercise in which the simulated population possessed life-history characteristics of Gulf of Alaska sablefish (Anoplopoma fimbria), a commercially valuable, data-rich, and long-lived species. We simulated time-varying M related to a covariate in the operating model under various levels of error in the covariate. Here we characterize how error between M and its covariate degrades the estimation of M in terms of accuracy and precision. Further studies will see if there is a breaking point in covariate error at which a random effects model performs better.