Investigating the Effect of Recruitment Variability on Length-Based Recruitment Indices for Antarctic Krill Using an Individual-Based Population Dynamics Model

Antarctic krill (Euphausia superba; herein krill) is monitored as part of an on-going fisheries observer program that collects length-frequency data. A krill feedback management programme is currently being developed, and as part of this development, the utility of data-derived indices describing population level processes is being assessed. To date, however, little work has been carried out on the selection of optimum recruitment indices and it has not been possible to assess the performance of length-based recruitment indices across a range of recruitment variability. Neither has there been an assessment of uncertainty in the relationship between an index and the actual level of recruitment. Thus, until now, it has not been possible to take into account recruitment index uncertainty in krill stock management or when investigating relationships between recruitment and environmental drivers. Using length-frequency samples from a simulated population – where recruitment is known – the performance of six potential length-based recruitment indices is assessed, by exploring the index-to-recruitment relationship under increasing levels of recruitment variability (from ±10% to ±100% around a mean annual recruitment). The annual minimum of the proportion of individuals smaller than 40 mm (F40 min, %) was selected because it had the most robust index-to-recruitment relationship across differing levels of recruitment variability. The relationship was curvilinear and best described by a power law. Model uncertainty was described using the 95% prediction intervals, which were used to calculate coverage probabilities and assess model performance. Despite being the optimum recruitment index, the performance of F40 min degraded under high (>50%) recruitment variability. Due to the persistence of cohorts in the population over several years, the inclusion of F40 min values from preceding years in the relationship used to estimate recruitment in a given year improved its accuracy (mean bias reduction of 8.3% when including three F40 min values under a recruitment variability of 60%).