Thu, Aug 18, 2022: 5:00 PM-6:30 PM
ESA Exhibit Hall
Background/Question/Methods: The Northern Shrimp is a critical species to the Newfoundland (NL) fisheries. It represents most of the fished benthic biomass, making it crucial for both economic and commercial aspects. Since the early 2000s, most Shrimp Fishing Areas (SFAs) have been on the decline. Improving the accuracy of stock assessment biomass estimates has the potential to assist in securing the recovery of the stock. The goal of this research is to compare the current spatial stock abundance estimation method for shrimp, Ogmap, with Generalized Additive Modeling (GAM) as an alternative for stock estimation. Ogmap estimates of abundance tend to have tighter confidence intervals (CI) compared to GAM estimates, and it is currently unknown whether either method is providing reliable measures of uncertainty about total abundance, or about the spatial distribution of abundance. Solving this would be a great asset in establishing Total Allowable Catch. We use a simulation-based approach to compare CI coverage of GAM and Ogmap at large and small scales and whether one method outperforms the other in different scenarios, such as in the presence of strong spatial biomass density gradients. We predict that CI coverage for GAMs will have closer to nominal coverage compared to Ogmap.
Results/Conclusions: Preliminary results indicate that CI coverage of both methods are close to nominal when compared to the true biomass of a simulated landscape, based on samples from 0.1% of the total area, when spatial variation in biomass is small relative to year-to-year variation. Currently, Ogmap is leading in CI coverage with a 91% coverage compared with GAMs at a 90% coverage. The current simulations are based on 500 individual simulations and then using a random stratified sampling method, 0.1% of the entire biomass is sampled. Landscapes include biomass, depth, temperate, coordinates as well as generated strata areas. Over this simulated landscape, a variation is applied to the biomass which simulates patches of high and low shrimp densities much like on the NL shelf. GAMs are known for their flexibility to model spatial structures due to their ability to model complex non-linear relationships between spatial structures. Ogmap takes into account fine spatial details which may result in biasing its ability to model spatial data. More work needs to be done to understand the relationship between these two models and their ability to produce accurate stock estimates. However, these results indicate that GAMs are a viable alternative to the legacy Ogmap model.
Results/Conclusions: Preliminary results indicate that CI coverage of both methods are close to nominal when compared to the true biomass of a simulated landscape, based on samples from 0.1% of the total area, when spatial variation in biomass is small relative to year-to-year variation. Currently, Ogmap is leading in CI coverage with a 91% coverage compared with GAMs at a 90% coverage. The current simulations are based on 500 individual simulations and then using a random stratified sampling method, 0.1% of the entire biomass is sampled. Landscapes include biomass, depth, temperate, coordinates as well as generated strata areas. Over this simulated landscape, a variation is applied to the biomass which simulates patches of high and low shrimp densities much like on the NL shelf. GAMs are known for their flexibility to model spatial structures due to their ability to model complex non-linear relationships between spatial structures. Ogmap takes into account fine spatial details which may result in biasing its ability to model spatial data. More work needs to be done to understand the relationship between these two models and their ability to produce accurate stock estimates. However, these results indicate that GAMs are a viable alternative to the legacy Ogmap model.