• Estimability of time-varying natural mortality in groundfishes: covariates and hierarchical models

      Ganz, Philip D.; Quinn, Terrance J. II; Hulson, Peter-John F.; Kruse, Gordon H. (2017-08)
      Natural mortality, M, has historically been a difficult parameter to estimate in conjunction with other stock assessment parameters. Time-varying M, while likely to be experienced by a population, is a particularly difficult process to estimate with the data and methods currently available to most stock assessments. Although auxiliary information in the form of a covariate to M has been shown to improve model fit for some stocks, such data are rarely available. Meanwhile, hierarchical models continue to be utilized in capturing processes that vary in time and space. I tested both the covariate and hierarchical methods in their ability to estimate time-varying M. I attempted to fit hierarchical models by two different methods: penalized likelihood and the integrated likelihood approach associated with mixed effects models. Mixed effects models performed poorly in comparison to penalized likelihood. Including a covariate to natural mortality aided the estimability of time-varying M, regardless of the observation error associated with the covariate. Estimating a constant value of M resulted in biased estimates when M was time-varying in the simulated population. I showed that the Akaike information criterion (AIC) is a useful metric for comparing models although it does not necessarily align with the accuracy of estimates that are of most interest to managers, such as terminal year spawning stock biomass. In addition to showing empirically that incorporating a covariate is a robust approach to estimating time-varying M, I conclude that this approach is also advantageous to stock assessment on theoretical grounds, as it is more amenable than hierarchical models to making predictions.