In this paper, we present a generalised additive mixed modelling (GAMM) framework to determine the probability of that a sampled macrourid is either Macrourus caml or M. whitsoni in the Ross Sea region (RSR). This GAMM was used to underpin recent vector autoregressive spatio-temporal (VAST) modelling analyses presented to WG-SAM-2022 and WG-FSA-2022 and is presented here at the encouragement of WG-FSA-2022. Both this GAMM and the associated VAST modelling support ongoing efforts to suggest how a revised catch limit for macrourids in the RSR consistent with CCAMLR decision rules can be determined taking into account the species of macrourid present, their relative abundances, spatial distributions, productivities and catches by the RSR toothfish fishery.
The GAMM framework consists of fitting models using a binomial error distribution to fishery-dependent data for M. caml and M. whitsoni and environmental data. The GAMMs model the influence of fishing and environmental covariates on the number of M. caml and M. whitsoni observed per fishing haul. Fishing and environmental covariates offered to the GAMMs include fishing depth, hook size, soak time, management area where fishing occurred, temperature at fished depth, salinity at fished depth, surface chlorophyll-a (chl-a) concentration, slope steepness, and ice cover. Fishing year, management area, and vessel are included in the models as random effects. The data inputs to the GAMMs that are presented in this paper were restricted to the data collected by scientific observers onboard four New Zealand-flagged commercial autoline vessels fishing in the RSR from the 2014–2021 fishing years, as well as during the 2016 and 2019 winter surveys and research collections made during the 2012 fishing year.
Preliminary results using the above-mentioned data indicated that the selected GAMM had a good fit to the data, with 55.3% of the deviance explained and with model residuals randomly distributed in a narrow range around zero. Preliminary results also suggested that M. caml is a higher proportion of the Macrourus abundance in each management area of the RSR.
We recommend that the data inputs of the GAMM and VAST models be expanded and investigations of model sensitivity to such expansion be conducted. To support the expansion of data inputs in the GAMM and VAST models, we recommend future work to confirm the accuracy of species identification by scientific observers operating across the three gear types employed in the RSR (autoline, Spanish line and trotline), and in particular, confirmation that species codes are being used as intended (e.g., that the code WGR is being used specifically for M. whitsoni). Approaches to increase the number of scientific observer records that are identified to the species level, as opposed to the generic GRV code, should be investigated. Finally, we recommend the development of a suggested revised framework for determining macrourid bycatch limits in the RSR by taking into account the species of macrourid present, their relative abundances, spatial distributions, productivities and catches by the RSR toothfish fishery.