Testing the robustness of data-poor assessment methods to uncertainty in catch and biology: a retrospective approach

The quality and quantity of data affect the reliability of all stock assessments. Over time, we expect data to improve and assessment predictions to become more reliable. There is a potential for strong bias in estimates of sustainable yield if the available data are not a good representation of stock dynamics, particularly for catch-based data-poor methods that rely on limited information and assumptions about stock status. We retrospectively investigated the interaction of data quantity and quality through time using the “real-world” data for a stock as it progressed from data-poor to data-rich. For this analysis, we chose a currently data-rich and overfished stock with historical assessments representing both a data-poor and data-moderate state, the canary rockfish (Sebastes pinniger). We asked how changes in the catch history and biological parameters over time affected the estimates of sustainable yield and the overfishing limit (OFL) predicted by two data-poor assessment tools, depletion-corrected average catch (DCAC) and depletion-based stock reduction analysis (DB-SRA). We found that both of these methods underestimated the “true” OFL in simulations with catch error alone. While there was slightly less bias for DB-SRA than DCAC, increasing error in the catch led to a more rapid increase in the variance of the DB-SRA harvest limit (HL). Our retrospective analysis showed that the expectation for a more accurate HL estimate between the data-poor and data-moderate canary rockfish assessments does not come from an increase in the quantity or quality of the catch data alone; a decrease in the quality of the biological data between assessments had the greatest impact. By evaluating these methods with historical data, our retrospective approach highlighted the impact of change in data quality and quantity on HL estimates for a long-lived rockfish, and could be used to define the amount and type of error included in simulation studies that further evaluate data-poor methods.