Autocorrelated error in stock assessment estimates: Implications for management strategy evaluation

Abstract

Management strategy evaluation (MSE) is often used in fisheries science to evaluate the effects of different management options. MSE models typically include a stock assessment component to estimate population size and management reference points based on data generated using the model, but including a full assessment within an MSE can be computationally intensive. A commonly used alternative to the full assessment approach is to simulate the error about the stock assessment outcomes as a stochastic process with an assumed level of autocorrelated estimation error. There is little guidance on what might be a reasonable assumed amount of autocorrelation, and what factors might influence this amount. We estimate the amount of temporal autocorrelation in errors of estimated biomass and recruitment from statistical catch at age stock assessment models over a series of scenarios spanning life histories, exploitation levels, recruitment variability, and data quality. Autocorrelation in the error in biomass estimates (ϕS) was positive and relatively high, with median estimates ranging between 0.7 and 0.9. Estimates were highest for the long-lived life history and lowest for the short-lived life history. Exploitation level also affected the amount of autocorrelation, with higher values for lightly exploited populations. On average, however, estimates of ϕS did not change over time as more data were included in the assessment, and were independent of whether or not a harvest policy was applied. Recruitment variability and data quality had relatively minor effects on autocorrelation of errors.