Sister otolith cross-reading in Subarea 48.6: evaluating precision, bias, and integration potential

Age estimates from otoliths underpin life history inference and stock assessment, yet differences in preparation and reading protocols can introduce random error and systematic disagreement. Exploratory research fishing in Subarea 48.6 has been conducted by Japan, South Africa, and Spain, and Korea joined in the 2024/25 season. Laboratories in this Subarea use different preparation methods, with Korea ageing baked and embedded sections and Japan ageing thin sections. To evaluate whether ages from these methods can be used jointly, inter laboratory precision and agreement were quantified using sister otoliths collected in 2024 to 2025 and read in two phases. In Phase 1, all 60 sister pairs were read independently. In Phase 2, a blind QC re read was completed for 25 preselected specimens that included 20 large difference outliers and 5 controls. Precision was moderate and overall agreement was good, but systematic differences remained. The pooled coefficient of variation decreased slightly from 11.39 percent to 11.17 percent. Bias defined as Korea minus Japan averaged minus 0.88 years. The slope of Japan on Korea was 0.75 with a 95 percent confidence interval of 0.63 to 0.87, indicating a scale difference across ages. The intraclass correlation coefficient was about 0.84. The number of large difference cases declined from 20 to 17 after the re read. Case review indicated that most discrepancies were interpretative rather than stochastic. False annulus interpretation was the most frequent source, followed by first annulus placement, while morphology and image quality contributed only occasionally. Reading direction was discussed during the inter laboratory review and did not materially affect results. Taken together, direct pooling without adjustment is not advisable, and near term work should focus on reducing interpretation differences at the source with emphasis on false annuli and first annulus placement. If integration is required in the interim, the method can be retained as a factor in the observation model or the two series can be carried separately in sensitivity analyses. Given the small observed differences, integration across the two methods appears feasible, provided that a harmonised inter laboratory reading protocol is formally adopted and confirmed at a short verification step.