Estimating relative indices of groundfish abundance from multiple fishery-independent data sources: a comparison of intercalibrating model-based abundance estimators

Abstract

Stock assessments are critical tools for sustainable fisheries management, and abundance indices estimated from fishery-independent data represent a crucial data source for these assessments. However, financial constraints on these surveys limit the number of samples taken per year, and/or the frequency with which regions are sampled. Additionally, it remains challenging for any individual survey to sample the entire domain of a stock due to contact-selectivity of the sampling gear and the accessibility of specific habitats to specific gear types. Further, the cost of operating fishery-independent surveys can result in surveys conducted on biennial or triennial schedules, resulting in temporal gaps in survey timeseries that may limit their ability to adequately index short lived species. One method by which these challenges might be addressed is through the use of model-based estimators, which estimate relative and/or absolute indices of abundance by intercalibrating data collected by multiple surveys with different spatial, temporal, or habitat footprints. While recent research has explored a number of potential applications of these methods, little to no prior research has assessed the relative performance of these methods in terms of the accuracy or uncertainty of their estimates. In the first chapter of this thesis, I fit Random Walk Timeseries (RWTS) models, Generalized Additive Models (GAM), and Vector Autoregressive Spatiotemporal Models to data collected by three fishery-independent surveys across four species/region case studies, and compare the model-estimated indices to design-based indices estimated by the Alaska Fisheries Science Center (AFSC) Bottom Trawl Survey (BTS) as a reference abundance timeseries. In the second chapter, I then simulate an age-structured and spatially heterogenous population dynamics for the Pacific Cod (Gadus macrocephalus)) stock in the Gulf of Alaska, to explore the reliability of intercalibrated indices of abundance. To do so I generate artificial survey catch data for three fishery-independent surveys, fit GAM and VAST survey intercalibration models, and then assess the accuracy and precision of the model-based indices. Results from the real species-region case studies in Chapter 1 suggest that RWTS, GAM, and VAST models all exhibit comparable performance, but the model structures used in this analysis struggled to estimate indices of abundance consistent with established abundance estimates in the presence of conflicting survey catch-rate signals. Results from the simulation experiment in Chapter 2 also suggest that the accuracy of the model-estimated indices is strongly influenced by the level of contrast in the size-selectivity profiles of the constituent surveys, and the rate at which the size-composition of the surveyed stock changes. I recommend that future work explores forms of these model-based estimators which estimate size-specific changes in abundance, and whether or not the inclusion of those elements improved the accuracy and precision of the estimated indices of abundance.