Investigating the performance of process-observation error estimator and robust estimators in surplus production model: A simulation study

Recent studies on the estimators of the surplus production model have greatly improved our understanding on the performance of them; however a systematic study is still needed given the fact that there were no comprehensive studies on them. We explored the process-observation error estimator and compared it to the process error estimator and the observation error estimator through a well designed simulation study. We also tested the robustness of the estimators by changing the error assumption from log-normal distributions to fat-tailed distributions. Bayesian approach with a Metropolis Hasting within Gibbs sampling (MHGS) was used to solve the models with these three estimators. Two fisheries, Atlantic Weakfish (Cynoscion regalis) and Black sea bass (Centropristis striata) are used as example fisheries to test and compare the three estimators concerning model complexity and estimation robustness. The surplus production models with multiple CPUE series were used in the simulation. The influence of multiple observation errors were compared to investigate the influence of the number of the CPUE series since there was only one process error term in the process-observation error estimator. All the methodology and conclusions in our study are worth being generalized to age-structure models.