Quantifying bias due to misspecification in integrated stock assessments

A key area of fisheries research is developing stock assessment models that can reliably quantify useful information about harvested fish stocks. The modern integrated assessment approach is recognized as the current best-practice method for quantifying stock information and is used to provide fisheries management advice in jurisdictions worldwide. The primary advantage of the integrated approach over other approaches is the ability to integrate all sources of data available on a stock into a single model to produce assessment quantities. However, the accuracy of assessment quantities depends in large part on the accuracy of the data sources included in an integrated assessment. Studies have demonstrated that integrated assessments can be sensitive to misspecification due to inaccurate data, and this can lead to unreliable estimates of quantities such as stock biomass status (defined as current biomass relative to pre-fishing biomass in this thesis) that jeopardize optimal utilization of the resource. These studies often use simulation analyses to generate hypothetical stock and fishery scenarios from which the impact of misspecification can be investigated in isolation from other potentially confounding factors.
This thesis investigates the impact of misspecifying catch and length-at-age data which are two critical pieces of information included in many integrated assessments. Although the misspecification of several aspects of these data types has been studied previously, there are specific characteristics which have the potential to be highly influential to assessment outcomes that require further investigation. This thesis consists of four technical chapters (Chapters 2-5) that address several of these characteristics using simulation analysis and the CASAL integrated assessment platform. Chapter 2 investigates the capability of integrated assessments that incorporate tag-recapture data as an index of abundance to accurately quantify stock characteristics when there has been illegal, unreported, or unregulated (IUU) fishing. A variety of hypothetical misspecified catch scenarios were examined, varying in both magnitude and trend (increasing, decreasing, or constant). The results indicate that stock biomass status will be increasingly overestimated as the magnitude of under-reporting increases, regardless of trend in catch. This demonstrates that tag-based integrated assessments are not robust to IUU fishing and that catch left unaccounted for can lead to a false interpretation that stocks are healthy and may increase the risk of overfishing.
Chapter 3 investigates the impact of inaccurately high catch on assessment outcomes using an integrated assessment model based on catch per unit effort (CPUE) data as an index of abundance. This type of catch misspecification can occur when catch due to IUU fishing has been overestimated or when catch records have been inflated prior to the introduction of a quota system when quota allocation is based on catch history. A set of inflated catch scenarios was designed to reflect those mechanisms and to test their impact on estimates of stock characteristics, with scenarios covering a range of magnitudes and trends as in Chapter 2. The results show that stock biomass status will be persistently overestimated after the cessation of inflated catch regardless of the trend in catch, which may lead to management action that is not sufficiently precautionary. Decreasing inflated catch over time was no exception, highlighting the potential for overfishing despite well-intended efforts to improve catch accuracy.
Chapter 4 identifies patterns in assessment model diagnostics of fits to CPUE and catch-at?age data that are attributable to inaccurate catch. Patterns in the residuals of an assessment model are a common and useful way of identifying model misspecification. This chapter applied that technique using a comprehensive set of misspecified catch scenarios and an assessment model similar to that used in Chapter 3. Distinct patterns emerge in diagnostic tests under various scenarios, indicating that past or current misspecification of catch can be detected using routine stock assessment outputs. This technique can provide fisheries scientists with a routine tool that can validate hypotheses about the characteristics of inaccurate historical catch.
Chapter 5 demonstrates how misspecifying the temporal variation of length-at-age data in an integrated assessment can impact assessment outcomes. In addition, the chapter investigates alternative, flexible, statistical methods for validating observed temporal variation of this type of data. Length-at-age data can be misrepresented in an assessment if it is specified to vary over time when in reality it does not, or vice versa. Using a similar assessment model to Chapters 3 and 4, specifying temporally varying length-at-age data resulted in a 5% increase in estimated stock biomass status, demonstrating the importance of correct specification. The flexible statistical methods used to describe temporal variation in length-at-age were able to detect variability where the traditionally used linear model could not. These results highlight the importance of appropriately justifying how to specify the temporal characteristics of length-at-age data using defensible evidence.
This thesis demonstrates that integrated assessments are sensitive to the accuracy of catch data and the treatment of length-at-age data. As a direct result of biases in these ubiquitous data types, assessment outcomes can falsely indicate current fishing practices are appropriate. Therefore, it is crucial that their potential biases are evaluated to inform sustainable fisheries management worldwide.