Integrated toothfish stock assessments using Casal2

• Main Authors: Dunn, A, Gruss, A, Devine, JA, Miller, C, Ziegler, P, Maschette, D, Earl, T, Darby, C, Massiot-Granier, F
• Format: Conference Object
• Language: English
• Published: CCAMLR 2022
• Subjects: Agricultural; Veterinary and Food Sciences; Fisheries sciences; Fisheries management; Antarctic; Ross Sea; Kerguelen; Heard Island; Kerguelen Islands; Heard; McDonald Islands; Antarc*; Antarctic Toothfish; Patagonian Toothfish
• Online Access: https://meetings.ccamlr.org/en/wg-sam-2022/14; http://ecite.utas.edu.au/155014

Abstract The software platform CASAL (C++ algorithmic stock assessment laboratory) has been used to implement integrated statistical catch-at-age toothfish stock assessments in CCAMLR since 2005. However, CASAL is no longer being developed or maintained. The software platform Casal2 has been developed as a more generalised framework for implementing Bayesian integrated catch-at-age (or catch-at-length) stock assessments to replace CASAL. Its greater computational power allows Casal2 more flexibility in specifying population dynamics, parameter estimation and model outputs than CASAL. In this report, we present a comparison of CASAL and Casal2 model implementations using the 2021 CASAL assessments of Antarctic toothfish ( Dissostichus mawsoni ) in Subareas 88.1/88.2A-B (Ross Sea region) and Patagonian toothfish ( Dissostichus eleginoides ) in Subarea 48.3 (South Georgia). Model comparisons for the assessments in Divisions 58.5.1 (Kerguelen Islands) and 58.5.2 (Heard Island and McDonald Islands) are still ongoing. The results from the CASAL models are compared to the results from Casal2 models, including estimates of initial and current status, CCAMLR precautionary yields, and model output diagnostics. The comparisons show that the two software packages provide equivalent estimates of key parameters for the case studies used. For the case studies, the diagnostics derived from the CASAL and Casal2 models provided identical conclusions on model fits. The estimation process in Casal2 was significantly faster, allows for models to be fitted with a larger memory footprint than CASAL, and Markov chain Monte Carlo (MCMC) estimates from Casal2 were typically obtained in about 10 to 50% of the time required to run the equivalent CASAL model. CASAL and Casal2 also performed the same when observations were simulated with CASAL (as an operating model) and the simulated data fitted with CASAL or Casal2. We show that Casal2 has been validated for use for integrated toothfish stock assessments at CCAMLR, and we have successfully ...