How does growth misspecification affect management advice derived from an integrated fisheries stock assessment model?

Abstract

Analysts must make many decisions regarding model specification when fitting integrated fishery stock assessment models. While variation in vital rates (i.e., recruitment, somatic growth, and natural mortality) is common, capturing this variation in models fit to available data is often infeasible or impractical. Failing to account for this variation can result in underestimates of uncertainty and even biased estimates of stock status used for management advice. Here, we seek to determine how growth misspecification affects management advice derived from integrated stock assessment models that use the Stock Synthesis platform. We conduct a simulation-based case study on California Current petrale sole (Eopsetta jordani) to test whether and how the inclusion or omission of somatic-growth variation introduces bias into management reference points when estimation models misspecify growth. Scenarios we explored included inter-annual and regime-like changes in two key parameters (k, the initial slope of the growth curve, and L2, the asymptotic maximum length) used to model somatic growth in Stock Synthesis. We find misspecification of growth can overestimate management quantities, particularly the estimate of current biomass relative to the unfished biomass (stock depletion). This results in an overly optimistic view of stock status. This bias may be mitigated or eliminated if the assessment model includes growth variation. Including growth variation in the estimation model can also reduce the uncertainty in estimated management quantities by correctly attributing process error to somatic growth. However, the magnitude of detected biases is exceeded by the uncertainty when data are limited, suggesting that estimating growth variation is helpful only in relatively data-rich stock assessment models. We suggest authors of data-rich assessments consider incorporating time-varying growth parameters into assessment models or decision tables more frequently to account for potential biases and reduce uncertainty caused by temporal growth variation.