An integrated tagging and catch-curve model reveals high and seasonally-varying natural mortality for a fish population at low stock biomass

Highlights

• Low exploitation in tag-return models causes imprecise estimates of natural mortality.

• Integrating tag-return and catch-curve models increases natural mortality precision.

• Weakfish natural mortality far exceeds fishing mortality.

• Natural mortality peaks coincide with weakfish migration and overwintering periods.

• Weakfish stock rebuilding is currently impaired by high natural mortality.

Abstract

Rebuilding of exploited fish stocks at low biomass requires accurate mortality estimates. Weakfish (Cynoscion regalis) abundance is at historical lows caused by an increasing instantaneous total mortality (Z) in recent years, but uncertainty exists regarding the relative importance of instantaneous fishing mortality (F) and natural mortality (M) to Z. Data from a tag-return study and catch-curve of weakfish in North Carolina were analyzed jointly using a Bayesian statistical framework to estimate seasonal and annual mortality (i.e., F, M, and Z). We accounted for key auxiliary parameters in the tag-return portion of the model (i.e., tag-reporting rate and tag loss) through field studies and an experimental design, including use of high-reward tags and double tagging. Estimates of Z from the joint model were similar in magnitude to the weakfish stock assessment. From mid-2014 to 2017, we estimated a constant annual instantaneous mortality rate of 0.05 yr⁻¹ (95 % credible interval [CrI]: 0.04, 0.07) for F and 2.33 yr⁻¹ (CrI: 2.10, 2.6) for M. In the most recent stock assessment, estimates of M had an upper bound of 1.0; thus, our findings suggest that these estimates of M are biased low and F biased high. Our seasonal analyses showed that a large portion of mortality occurred from fall to spring, coinciding with weakfish migration and overwintering periods on the continental shelf. Through an integrated modeling approach, our study provides insights into the magnitude, timing, and sources of weakfish mortality, and enhances understanding of weakfish population dynamics to guide management strategies.

Introduction

Effective rebuilding of exploited fish stocks with a low biomass requires accurate estimates of fishing and natural mortality. The fishing mortality rate (F) allows management to meet stock rebuilding goals through comparisons with target and threshold levels based on biological reference points (Hilborn and Walters, 1992). The natural mortality rate (M) affects estimates of stock size and productivity, which ultimately determine harvest rates (Clark, 1999). In a typical stock assessment, catch-at-age data provide direct information about F (Walters and Martell, 2004), but M is more difficult to estimate reliably because natural deaths are rarely observed in aquatic systems (Quinn and Deriso, 1999). Natural mortality is often estimated externally based on life-history parameters and environmental variables (e.g., Pauly, 1980; Hoenig, 1983; Lorenzen, 1996; Griffiths and Harrod, 2007; reviewed by Kenchington, 2014), and used as a fixed input parameter in fishery stock assessments (Vetter, 1988). However, these estimates of M do not account for time- or location-specific factors and have an unknown certainty (Vetter, 1988; Pascual and Iribarne, 1993). Stock assessment models and consequently derived reference points are particularly sensitive to input values of M (Clark, 1999; Williams, 2002).

Tag-return experiments can directly estimate F and indirectly M, thereby generating near real-time estimates of F to track fishery harvest trends, validating catch-at-age analysis, and determining if target harvest rates are being maintained (Walters and Martell, 2004). These models partition the instantaneous total mortality rate (Z) into estimates of F and M (Hoenig et al., 1998a). However, precise mortality estimates depend on key auxiliary parameters: tag-reporting rate (λ), tag loss (Ω), and survival from the tagging procedure (ϕ; Pollock, 1991; Pollock et al., 2001; Miranda et al., 2002; Brenden et al., 2010). Multi-year tagging studies of rigorous design can estimate the auxiliary parameters, leading to reliable estimates of mortality (e.g., den Heyer et al., 2013; Kerns et al., 2015) and providing insight into the timing and causes of mortality because estimates can be made at less than one-year intervals (e.g., monthly or seasonally; Harris and Hightower, 2017; Ellis et al., 2018; Sackett et al., 2018).

The ability of tag-return experiments to provide robust estimates of mortality depends on adequate returns from the fishery. In recovering fisheries where tight regulation has kept the Fs low, the estimates of M can be imprecise (Pollock et al., 2004). This limitation can be overcome by using multiple data sources in a combined analysis (e.g., Burnham, 1993; Powell et al., 2000; Pollock et al., 2004; Schaub et al., 2007; Dudgeon et al., 2015). Pollock et al. (2004) found that a combined telemetry and tag-return analysis provides substantially better precision for estimates of F and M than either individual method. The strength of the telemetry method was estimating M, whereas the tag return was better at estimating F (Pollock et al., 2004). A hallmark of many stock assessments is the estimation of Z from a catch-curve that tracks the sequential decline observed in fish cohorts. A combined tag-return and catch-curve model can simultaneously estimate F, M, and Z to increase parameter precision in low exploitation fisheries.

Weakfish (Cynoscion regalis) are an important recreational, commercial, and ecological species that primarily inhabit estuarine and coastal waters between North Carolina and Massachusetts. The spawning stock biomass has declined since 1982 to historic lows in the late 2000s, with the cause of the decline attributed to increased Z (ASMFC, 2006). Despite rigorous regulatory measures, stocks have failed to rebuild, and the most parsimonious explanation for the increase in Z was an increase in M (first noted in the 2006 weakfish stock assessment and expounded on in the 2009 assessment [ASMFC, 2006; NEFSC, 2009]). Subsequent stock assessment efforts to improve estimates of M culminated in a Bayesian statistical catch-at-age model that internally estimated a time-varying M (Jiao et al., 2012; ASMFC, 2016). The model included 14 fisheries-independent and 1 fisheries-dependent indices of abundance, as well as commercial and recreational harvest and discards. The prior distribution for the 1982 M estimate was based on a meta-analysis of published estimates from similar species and fisheries (uniform prior 0.1 to 0.4 yr⁻¹), and subsequent M estimates were allowed to vary through the time-series (1982–2017) using a random-walk approach (ASMFC, 2016). Post-1982 M estimates had an upper bound of 1 yr⁻¹ (Yan Jiao, Virginia Tech, personal communication). M was estimated to increase through the time-series and approached the upper bound from 2007 to 2015 at >0.9 yr⁻¹ (ASMFC, 2019).

An upper bound on M may result in overestimation of F. That is, if Z increases as a result of M increasing above the upper bound then the extra mortality will be assigned to F. In this scenario, the Z is assumed accurate, but the partitioning of F and M becomes biased. The assumptions on the bounds and priors of M are not weakfish specific (i.e., meta-analysis derived) as assessments on other species (e.g., ASMFC, 2015; Monk and He, 2019; SEDAR 58, 2020) have used this approach and for most species an upper bound of M = 1.0 yr⁻¹ would be sufficient. However, Z estimates for weakfish from the stock assessment update are very high and have remained high during a period of decreased harvest; currently, the estimates of F from the stock assessment update make up ∼50 % of Z during the last three years of the time-series (2015–2017; ASMFC, 2019). Estimates of F and M from the weakfish stock assessment have never been compared to external estimates of F and M from a tagging study.

We used data from two semi-concurrent yet independent studies: (1) a multi-year, high-reward external tagging initiative by North Carolina State University (NCSU) and (2) a fishery-independent gill net survey conducted by the North Carolina Division of Marine Fisheries (NCDMF) in Pamlico Sound. Tagging data informed monthly F and M estimates, whereas a catch-curve based on catches of aged fish from the survey estimated seasonal and annual estimates of Z. Mortality estimates are presented for each model alone and jointly. A simulation provides insight into the strengths and weaknesses of tag-return studies for fisheries with low exploitation rates. Our study presents a new modeling framework for robustly estimating mortality from a joint catch-curve and tag-return model. For weakfish, the study contributes important information on the magnitude, timing, and sources of mortality that can guide management strategies.