Alternative sampling and estimation methods for multispecies trawl surveys

Abstract

Multispecies demersal trawl surveys are used in the United States and internationally to estimate the relative abundance of commercial and non-commercial fish species. Their usefulness for estimating species' abundance is often limited by the variance associated with estimates. This study implemented and evaluated alternative sampling and estimation methods, with the goal to incorporate additional sources of information for increased precision of individual species' estimates from multispecies trawl surveys. First, habitat characteristics and past spatial distributions of four flatfish species' density were incorporated into a multispecies trawl survey design conducted in Kalsin and Middle Bays, Kodiak Island, Alaska. Stratification by depth and percent sand produced estimates of relative abundance with lower CV s than those from unstratified sampling. Additional decreases in relative precision were generally not achieved by estimating the relative abundance of multiple species from regions of species-specific suboptimal habitat. Second, a poststratification technique was used to incorporate species-specific habitat characteristics and previous distributions of species' density into the estimation of species' abundance from the Kalsin and Middle Bays' trawl survey. Poststratification by habitat gave estimates with lower variance and/or less design-bias than an unstratified estimator for all species in all years. Poststratification by habitat and fish density produced estimates with the least design-bias for all species in all years and the lowest variance when stratum sample sizes were sufficient. Third, mixed model linear regression (MMLR), empirical Bayes (EB) and hierarchical Bayes (HB) estimation methods were used to incorporate historical trends of yellowfin sole, Limanda aspera biomass from the eastern Bering Sea trawl survey into annual biomass estimates. Using MMLR, EB, and HB methods resulted in biomass estimates that were less anomalous than survey estimates with respect to a linear regression trend. Estimates for all three methods had lower CV s than surveys in most years. The results of this thesis suggest that incorporating additional information into survey design and estimation can decrease the variability of survey estimates and/or correct for possible bias. Methods that can incorporate additional information, therefore, have the potential to improve survey assessments for management use.