Fish in Space: Estimating groundfish population distribution in the Gulf of Alaska for management apportionment by subregion

Abstract

Management of fisheries relies on information about biomass of stocks in order to determine how many fish can be sustainably harvested in a given year. In addition to predicting total biomass, it is frequently important to accurately predict the distribution of stock biomass through space in order to avoid local depletions of stock and to more evenly distribute harvest among many stakeholders. Towards that end, this study compares the current modeling approach to predict the geographic apportionment of stock biomass for groundfish in the Gulf of Alaska (GOA), a random walk model, with a delta-GLMM spatiotemporal model implemented using the VAST package in R. These stocks are managed using subregional catch allocation, whereby the GOA is divided into 3 management areas: western, central, and eastern GOA. This analysis uses bottom trawl survey data collected by the Alaska Fisheries Science Center (AFSC) of the National Oceanic Atmospheric Administration National Marine Fisheries Service (NOAA NMFS) for two species of groundfish, Pacific Ocean Perch (Sebastes alutus) and Northern Rockfish (Sebastes polyspinis). Model performance was evaluated using the accuracy of the model estimate from jackknife resampled results for population proportion by subregion compared to survey design-based proportions, and on the precision of the model jackknife estimates. In terms of accuracy, the models performed similarly well, with the mean absolute difference between the model jackknife estimates and the design-based estimates for the random walk results being smaller by 0.086 or less than the delta-GLMM results, with significant overlap in the jackknife absolute difference values. Precision was measured by the CV calculated with the model jackknife results, and the delta-GLMM results had smaller mean CV values by at least 0.105, and very little overlap in the jackknife CV values. However, the precision of the delta-GLMM is small enough that it may lead to significant over- or underestimation compared to the survey design-based proportions and therefore may be a riskier option than the random walk model for estimating subregional catch apportionment for these stocks in the GOA.