Performance evaluation of catch-only methods when catch data are misreported

Abstract

Data-limited methods are being developed rapidly to cater to a range of data types and stock characteristics. One class of these methods is catch-only methods (COMs), which seek to determine stock status based largely, or exclusively, on catch histories. Although catch data are the most common information for a fishery, they are susceptible to large errors. Understanding the performance of COMs when catch is mis-specified is necessary before they can be used in fisheries management. This study constructed a simulation-estimation framework to evaluate the performance of two types of COMs, Catch-MSY (CMSY) and catch-only boosted regression trees (zBRT), in inferring stock biomass status (B/BMSY) with consideration of different forms of catch errors, fishing patterns, and prior information. The results showed both COMs tended to bias the estimates of B/BMSY regardless of types of catch errors. The estimation performance of COMs was robust to the constant reporting rate across all fishing patterns, which means constant catch over- or under-reporting would not affect the COM assessment. zBRT showed high sensitivity to fishing patterns and performed best when fishing mortality was constant. CMSY yielded underestimation in most cases, but use of a narrow and informative prior on final depletion improved accuracy in recent years. Neither COM performed well in correctly identifying categorical biomass status (overfished vs. not overfished) across assessment years. We caution that improving catch quality would not improve the estimation of COMs and do not advocate for the application of COMs due to their poor performance.