Estimation of natural mortality coefficient from fish abundance and catch data using Virtual Population Analysis (VPA)

Abstract

Natural mortality coefficient (M) was estimated from fish abundance (N) and catch (C) data using a Virtual Population Analysis (VPA) model. Monte Carlo simulations were used to evaluate the impact of different error distributions for the simulated data on the estimates of M. Among the four error structures (normal, lognormal, Poisson and gamma), simulations of normally distributed errors produced the most viable estimates for M, with the lowest relative estimation errors (REEs) and median mean absolute deviations (MADs) for the ratio of the true to the estimated Ms. In contrast, the lognormal distribution had the largest REE value. Errors with different coefficients of variation (CV) were added to N and C. In general, when CVs in the data were less than 10%, reliable estimates of M were obtained. For normal and lognormal distributions, the estimates of M were more sensitive to the CVs in N than in C; when only C had error the estimates were close to the true. For Poisson and gamma distributions, opposite results were obtained. For instance, the estimates were more sensitive to the CVs in C than in N, with the largest REE from the scenario of error only in C. Two scenarios of high and low fishing mortality coefficient (F) were generated, and the simulation results showed that the method performed better for the scenario with low F. This method was also applied to the published data for the anchovy (Engraulis japonicus) of the Yellow Sea. Viable estimates of M were obtained for young groups, which may be explained by the fact that the great uncertainties in N and C observed for older Yellow Sea anchovy introduced large variation in the corresponding estimates of M.