Can a more realistic model error structure improve the parameter estimation in modelling the dynamics of fish populations?

Mathematical models are commonly used to describe the dynamics of fish populations. These models are fitted to observed fisheries data by estimating parameters with assumptions made concerning model error structure. Large errors are commonly associated with fisheries data which can result from many sources. However, such errors are often assumed non-existent for some observed variables when fitting models to data. This may result from our effort of simplifying the model error structure to facilitate the use of the classical least squares method in the parameter estimation. The impacts of ignoring certain errors in some or all variables on the parameter estimation are rarely evaluated in modelling the dynamics of fish populations. Using a simple cohort-based model as an example in this study, we examine the impacts of the lack of a realistic model error structure on the parameter estimation. The inclusion of a more realistic error structure which accounts for measurement errors in catch and operating errors in effort greatly increases the complexity of the parameter estimation. However, the parameter estimation with such an error structure is improved relative to that of having a less realistic error structure. The magnitudes of actual error values used in the estimation are less important than the presence of such an error structure in the parameter estimation. This may indicate the necessity of having a realistic error structure or applying an estimation method that is robust to the error structure assumption in modelling the dynamics of fish populations. (C) 1998 Elsevier Science B.V. All rights reserved.