Abstract
Treatment of natural mortality is a major consideration in assessment models and state-space approaches allow estimation of temporal variation in this mortality rate as well as effects of specific covariates. However, there has been no investigation of the reliability of inferences made from state-space assessment models when there is process error or covariate effects on natural mortality. We conducted a large-scale simulation study with operating models simulating data assuming alternative levels of observation error, types and magnitude of process error, and magnitude of covariate effects on natural mortality. We fit estimating models to the simulated data with alternative assumptions regarding the inclusion of covariate effects, estimation of the median natural mortality rate, and type of process error. We found estimating covariate effects on natural mortality required ideal scenarios with lower observation uncertainty, contrast in fishing pressure, and higher covariate variability, but the source of process error could be mis-specified. Estimating annual natural mortality rates as random effects was also challenging, but spawning stock biomass and fishing mortality were reliable in a wide range of scenarios.