Exploring Causal Factors of Spawning Stock Mortality in a Riverine Striped Bass Population

Abstract

The recovery of the Atlantic Striped Bass Morone saxatilis stock in the 1990s is an important example of effective natural resources management. Implementation of Atlantic States Marine Fisheries Commission (ASMFC) harvest regulations reduced mortality, protected older and more fecund females, and contributed to the formation of dominant year‐classes in the 1980s and 1990s. However, Striped Bass stocks south of Albemarle Sound, North Carolina, are not subject to ASMFC management plans, and many populations have failed to attain recovery goals. Catch‐curve analyses indicate that the Neuse River Striped Bass population continues to experience spawning stock exploitation rates similar to those implicated in the decline of the Atlantic Migratory and Albemarle Sound/Roanoke River stocks in the 1970s. From 1994 to 2015, Striped Bass instantaneous fishing mortality (F) in the Neuse River ranged from 0.12 to 0.84 and exceeded the overfishing threshold (F_Threshold = 0.41) in 12 of 22 years. A global linear model using environmental and exploitation factors accounted for 55% of the variability in spawning stock discrete annual mortality. An information‐theoretic approach was used to elucidate the best linear model for predicting discrete annual mortality. The best model included previous‐year gill‐net effort and same‐year commercial harvest (Akaike weight = 0.64, R² = 0.50). Model‐averaged coefficients for gill‐net effort and commercial harvest suggested total exploitation impacts that were congruent with other studies of Neuse River Striped Bass. Results indicate that reducing exploitation to target levels will require substantial reductions in gill‐net effort in areas of the Neuse River where Striped Bass occur. Reducing exploitation may increase spawning stock biomass and advance the age structure of spawning females, conferring an increased likelihood of successful recruitment and production of dominant year‐classes during periods of favorable environmental conditions.

Striped Bass Morone saxatilis populations sustained severe declines in abundance throughout the U.S. Atlantic coast in the 1970s after several years of record commercial harvest combined with poor recruitment (Boreman and Austin ; Richards and Deuel ). In North Carolina, Striped Bass commercial landings declined by 80% between 1973 and 1983 (Boreman and Austin ). Recovery efforts began with the development of the Atlantic States Marine Fisheries Commission's (ASMFC) Interstate Fisheries Management Plan for Striped Bass (IFMP) in 1981 (Richards and Rago ). A centerpiece of the IFMP and its amendments was the use of harvest restrictions to curtail overexploitation. The harvest provisions of the IFMP were implemented in North Carolina beginning in 1984, along with an expansion of Striped Bass stocking programs and continued development of optimized streamflow releases from Roanoke Rapids Dam to improve spawning conditions in the Roanoke River, North Carolina (Figure ; NCDENR , ). Albemarle Sound/Roanoke River Striped Bass were declared recovered in 1997 (NCDENR ).

Coastal North Carolina, showing the Neuse River in relation to Pamlico Sound; RKM denotes river kilometers from the confluence of the Neuse River and Pamlico Sound. The first impediments to upstream migration (Milburnie Dam on the Neuse River; Roanoke Rapids Dam on the Roanoke River) are indicated by black asterisks. Gray diagonal lines denote Craven County.

In North Carolina, Striped Bass populations south of Albemarle Sound (Figure ) are not subject to compliance with ASMFC management plans due to their minimal contribution to the Atlantic Migratory stock (Merriman ; Greene et al. ). These populations are collectively managed as the Central Southern Management Area (CSMA) stock under a collaborative agreement by the North Carolina Division of Marine Fisheries (NCDMF; coastal waters) and the North Carolina Wildlife Resources Commission (NCWRC; inland waters). Of the populations comprising the CSMA, Neuse River Striped Bass were among the first to receive targeted monitoring and management actions (Hammers et al. ).

Although Striped Bass are documented as historically utilizing all major coastal North Carolina rivers (Smith ), the Neuse River population was among the most studied by early ichthyologists. In the 19th century, the population was subject to the second‐largest Striped Bass fishery in North Carolina after the fisheries operating on the Albemarle Sound/Roanoke River stock. Yarrow () described Striped Bass in the Neuse River as “exceedingly plenty” and reported that 3,000 were sold to New Bern (Figure ) fish houses from January to April 1873 (Yarrow ). By 1880, almost 16,000 Striped Bass were harvested and shipped from New Bern to northern cities, with an additional unknown quantity consumed locally during the fishing season (McDonald ). Despite their former abundance, declines were evident before the end of the 19th century, leading McDonald () to note that “…the supply has materially decreased…owing to overfishing and the erection of obstructions.” By 1939, only 318 kg of Striped Bass were commercially harvested in Craven County (Figure ; Chestnut and Davis ).

Although fishing records during World War II are sparse, acquisition of fishing vessels and labor for the war effort likely reduced Striped Bass harvest and allowed for stock rebuilding. Fishing restrictions and labor shortages were eased toward the end of the war, leading to the harvest of 18,000 kg of Striped Bass in Craven County during 1945 (Anderson and Power ). However, construction of Quaker Neck Dam in 1952 prohibited access to essentially all spawning habitat (Burdick and Hightower ). By the mid‐1960s, recreational and commercial anglers reported population declines, and a subsequent 3‐year NCWRC survey collected only 12 adult fish (Miller ). Despite minimal harvest restrictions, commercial landings remained low throughout the latter half of the 20th century and did not exceed 4,500 kg again until 2010 (NCDMF, unpublished data). It is possible that the intensity of post‐war fishing in the lower Neuse River combined with an inability to access suitable spawning habitat led to the near extirpation of the population.

Active management efforts in the Neuse River began with the implementation of an annual stocking regime in 1992 (although intermittent stocking began as early as 1931). In 1994, annual spawning ground surveys commenced, and a 11,340‐kg commercial harvest quota was established for the entire CSMA stock (NCDENR ). The removal of Quaker Neck Dam in 1998 allowed unobstructed access to approximately 120 km of historical spawning habitat (Burdick and Hightower ). Finally, gill‐net use was prohibited in NCWRC‐managed inland waters in 2001 (NCDENR ).

Recovery efforts were first formalized in 2004 as part of the North Carolina Estuarine Striped Bass Management Plan (NCDENR ) that was jointly developed by NCDMF and NCWRC. Unweighted linearized catch‐curve analyses of age structures collected on the Neuse River spawning grounds indicated that overfishing was occurring (NCDENR ), leading to the implementation of gill‐net restrictions in 2008 (established minimum distance from shore and use of tie‐downs during the closed harvest season; NCDENR ). A stock assessment conducted in 2010 using unweighted linearized catch curves again documented high mortality, but the assessment was deemed unsuitable for management use due to large confidence intervals around the mortality estimate. However, the need for continued conservation management measures was supported by truncated size and age distributions, low CPUE, and an absence of older fish in spawning ground samples. Albemarle Sound/Roanoke River spawning potential ratios of 45% and 40% were used to develop biological reference points for the Neuse River, resulting in an instantaneous fishing mortality rate (F) target (F_Target) of 0.33 and an overfishing threshold (F_Threshold) of 0.41 (NCDENR , ).

Electrofishing assessments on the spawning grounds indicate that size and age distributions have not expanded since the 2010 stock assessment (Rachels and Ricks ). Additionally, recent results utilizing parentage‐based tagging (PBT) indicate that hatchery fish (Table ) comprise at least two‐thirds of the spawning stock (O'Donnell et al. ) and may approach 100% stocking contribution (Rachels and Ricks ; O'Donnell et al. ). The development of recommendations for catch‐curve best practices (Smith et al. ) render former Neuse River Striped Bass stock assessments obsolete and present an opportunity to re‐evaluate spawning ground age‐structure data. Our objectives were two‐fold: (1) to improve the precision of catch‐curve mortality estimates by using current methodology and an expanded time series; and (2) to use linear modeling in an information‐theoretic approach (Burnham and Anderson ) to elucidate factors responsible for driving the observed mortality rates.

Number of hatchery‐origin Striped Bass stocked into the Neuse River, North Carolina, and exploitation and environmental factors

Year	Number stocked	Commercial effort (trips)	Commercial harvest (kg)	Summer dissolved oxygen (mg/L)	Summer water temperature (°C)
1994	182,990	2,531	3,760	7.1	27.5
1995	99,176	2,601	1,792	6.7	26.9
1996	200,760	3,018	3,159	6.5	28.0
1997	100,000	3,084	2,424	8.6	27.8
1998	290,925	3,209	2,511	6.3	27.9
1999	100,000	2,527	2,764	9.0	28.9
2000	229,993	3,030	2,181	6.6	27.3
2001	103,000	2,619	3,149	6.8	27.7
2002	147,654	3,317	1,869	9.5	29.1
2003	100,000	3,196	2,621	6.4	28.1
2004	268,011	2,159	3,547	7.3	28.5
2005	114,000	2,305	2,346	9.1	29.9
2006	245,935	2,777	3,216	7.7	28.1
2007	242,835	2,893	3,053	8.8	28.8
2008	313,798	1,980	2,190	9.7	29.6
2009	204,289	2,464	3,758	7.9	28.2
2010	107,142	1,583	5,092	8.0	30.1
2011	102,089	1,485	7,081	7.8	29.1
2012	140,358	1,577	1,946	6.2	27.8
2013	295,161	2,206	5,328	5.9	27.0
2014	158,730	1,603	2,801	6.7	28.2
2015	109,144	1,091	3,793	6.1	27.8

Methods

Study area

The Neuse River flows approximately 400 km from its origin at the confluence of the Eno and Flat rivers before discharging into Pamlico Sound, North Carolina (Figure ). The lower 60 km constitute a wind‐mixed mesohaline estuary, although salinity can range from 0‰ to 27‰ depending on precipitation and streamflow (Burkholder et al. ). The Neuse River estuary has been classified as “Nutrient Sensitive Waters” since 1988 (NCDENR ) and experienced numerous algae blooms and fish kills during the 1990s resulting from nitrogen and phosphorus inputs (Burkholder et al. , ; Rothenberger et al. ).

Mortality estimation

From 1994 to 2015, boat‐mounted electrofishing (Smith‐Root 7.5 GPP; 120 Hz; 5,000–7,000 W) was used to collect Striped Bass from the spawning grounds during annual spawning migrations (March–May). Collections primarily occurred between river kilometer (RKM) 230 of the Neuse River (measuring from its confluence with Pamlico Sound) and RKM 352. Few Striped Bass were collected above Quaker Neck Dam (RKM 230; Figure ) before its removal in 1998.

Striped Bass were measured for TL (mm) and weighed (g), and sex was determined by applying pressure to the abdomen and observing the vent for discharge of milt or eggs. Scales for age estimation were removed from the left side of each fish between the dorsal fin and lateral line. From 1994 to 2014, 15 fish of each sex per 25‐mm size‐class were aged by either directly reading scales (1994–2010) or reading scale impressions on acetate slides (2011–2014). Since sampling occurred during the time of year when annuli are formed, scale age was based on (1) the actual number of annuli if an annulus was present on the scale margin; or (2) the number of annuli plus 1 if there was a considerable gap between the last annulus and the scale margin (NCWRC and NCDMF ). A 20% subsample of each size‐class was aged by a second reader. Discrepancies between primary and secondary readers' estimates were resolved by jointly reading and reaching consensus (NCWRC and NCDMF ). In 2015, a partial pelvic fin clip from each fish was preserved in a 95% solution of ethyl alcohol to determine hatchery or wild origin using PBT. Hatchery‐origin fish were aged using PBT, while fish of unknown origin were assigned ages with sex‐specific age–length keys developed using scale‐aged fish from 2010 to 2014.

The Chapman–Robson estimator was used to estimate instantaneous total mortality (Z) for each year in the time series via the recommendations of Smith et al. (). As with other catch‐curve methods, assumptions included the following: (1) the proportion of ages in the population is estimated without error, (2) recruitment varies without trend for all age‐classes, (3) mortality is stationary through time and across age‐classes, and (4) all age‐classes are equally vulnerable to the sampling gear (Robson and Chapman ; Smith et al. ). Of the various catch‐curve methods, the Chapman–Robson estimator is the most robust to violations of these assumptions (Murphy ; Smith et al. ). In accordance with Smith et al. (), age at full recruitment to the catch curve was the age of peak catch plus 1 year (peak‐plus criterion). In addition, an overdispersion parameter ĉ (Burnham and Anderson ; Smith et al. ) was calculated for each year to correct the SE of the mortality estimate and to assess structural fit of the Chapman–Robson estimator to the age‐structure data (ĉ > 4 indicates poor model fit; Burnham and Anderson ). Instantaneous fishing mortality was calculated for each year by subtracting instantaneous natural mortality (M = 0.24; Bradley ) from Z. Uncertainty in the mortality estimates was characterized by calculating the relative standard error (RSE; Z/SE) and bootstrapping from the distributions of Z and M (Gamma distributed; Bolker ) to estimate 90% confidence intervals for F.

Mortality modeling

Linear models were developed to evaluate environmental and exploitation factors that potentially influence discrete annual mortality (A = 1 − e^−Z) over the time series 1994–2015, including summer dissolved oxygen, summer water temperature, gill‐net effort, and commercial harvest. We hypothesized that low dissolved oxygen and warm summer temperatures may lead to increased natural mortality. Hypoxic conditions can be prevalent in the Neuse River estuary during the summer months as a result of nutrient loading and water column stratification (Luettich et al. ; NCDENR ). These hypoxic conditions have been implicated in many of the 236 fish kills occurring between 1996 and 2015, which primarily affected Atlantic Menhaden Brevoortia tyrannus in the Neuse River basin (NCDENR ; NCDEQ ). Hypoxic events and resulting fish kills have also been implied as negatively affecting Striped Bass (NCDENR ). Water quality data were obtained from the Neuse River Estuary Modeling and Monitoring Project (ModMon; UNC ), which is one of the few programs that has continuously monitored water quality in the lower Neuse River since 1994. The summer (June–August) mean surface dissolved oxygen (mg/L) and summer mean surface water temperature (°C) at ModMon station 30 (RKM 57; Figure ) were used as environmental factors. Results of an acoustic telemetry study (Bradley et al. ) determined that the highest densities of adult and juvenile Striped Bass occur in the vicinity of the selected ModMon station.

In addition to the suite of environmental factors, several long‐term data sets were available from NCDMF to allow investigation of the effects of exploitation. Beginning in 1994, a mandatory trip ticket program was implemented to monitor commercial landings at the first point of sale. Information collected by this program includes harvest (kg) landed by species, gear type, and location (NCDENR ). Neuse River Striped Bass commercial harvest was used as a direct exploitation factor (NCDMF, unpublished data). However, gill‐net fisheries continue to pursue other marketable species after the Striped Bass harvest season is closed. Therefore, the annual number of gill‐net trips in the Neuse River was used as a measure of gill‐net effort that potentially accounts for harvest, discard, and unreported or misreported mortality (NCDMF, unpublished data). Unfortunately, measures of recreational fishing effort for Striped Bass were not available for the entire time series. A recreational creel survey has been conducted annually in the lower Neuse River since 2004, yet there is limited information for prior years (for exceptions, see Borawa and Rundle et al. ). Several recreational fishing surveys administered by National Oceanic and Atmospheric Administration Fisheries, including the Marine Recreational Information Program, the Marine Recreational Fisheries Statistics Survey, and the Coastal Household Telephone Survey, were investigated for potential use as a surrogate recreational fishing effort metric. However, these surveys lacked the data resolution necessary to specifically assess Neuse River recreational fisheries.

Since age‐structure collections occurred in the spring (March–May), it was likely that factors occurring throughout the previous year (gill‐net effort) or during the previous summer (dissolved oxygen and surface water temperature) had a greater influence on the estimated mortality rate than same‐year measures. Therefore, these predictor variables were modeled using a 1‐year time lag. Commercial harvest was not modeled with a time lag since the commercial Striped Bass harvest season occurs in the early spring before electrofishing collections on the spawning grounds; any effects of commercial harvest should be detected using same‐year measures. Striped Bass discrete annual mortality was nonstationary; the global model was of the form[Image Omitted. See PDF] where A  =  discrete annual mortality; β₀ = intercept; X = variable i; θ_i = effect of variable X_i; t = year; C = commercial harvest; and ε = an independently and identically distributed white noise vector. Note that $A{\prime }_t$ and $X{\prime }_{i,t}$ were first‐differenced to ensure stationarity and remove serial correlation as given by[Image Omitted. See PDF]

In the case of four predictor variables, there are 15 main‐effects models and 26 total models if we consider first‐order interactions. Given our small sample size (22 observations) and the potential for “too many models” (Anderson and Burnham ; Burnham et al. ; Dochtermann and Jenkins ), we did not consider all‐subsets regression. Instead, we constrained our analyses to 12 main‐effects models (example R code provided in the Supplement available separately online) incorporating dissolved oxygen, surface water temperature, gill‐net effort, and commercial harvest using the information‐theoretic framework described by Burnham and Anderson (). The second‐order Akaike's information criterion (AIC_c) was computed for each model, and the difference in AIC_c value (∆_i) from the model with the smallest AIC_c was used to assess the relative strength of the models. After ensuring that $A\prime$ and $X\prime$ differencing removed time trends (β₀ = 0; α = 0.05), the intercept was removed from final models, and AIC_c and ∆_i were recalculated. The reduced parameterization improved AIC_c for all models. Akaike weights (ω_i) were calculated to evaluate the relative likelihood of each model (Burnham and Anderson ). The relative importance of each predictor variable was assessed by decomposing global model variance using the Lindeman–Merenda–Gold (LMG) method (Grömping ). Model‐averaged estimates of the effect of each predictor variable were calculated by multiplying the coefficients of each factor in the models in which they appeared by the ω_i of that model (Burnham and Anderson ). The model‐averaged effect for gill‐net effort and commercial harvest was multiplied by the 1994–2015 mean number of gill‐net trips and mean harvest, respectively, to estimate each factor's long‐term average effect on discrete annual mortality (∆A ≡ u; discrete annual fishing mortality). Linear models were fitted using ordinary least‐squares (OLS) regression with package “dynlm” in R version 3.2.5.

Model assumptions

Assumptions for OLS time series regression depart in some respects from those considered in classical linear modeling. Assumptions of time series regression include a mean of zero, constant variance, and constant covariance structure through time (stationarity; Hyndman and Athanasopoulos ). The augmented Dickey–Fuller (ADF) test (α = 0.05; Hyndman and Athanasopoulos ) assumes H₀ = nonstationary and was employed in the R package “stats” to assess stationarity in the mortality time series. The partial autocorrelation function (PACF; Derryberry ) in the “stats” package was utilized to examine the potential for autocorrelation in the spawning stock discrete annual mortality time series. Multicollinearity among the predictor variables was assessed by calculating variance inflation factors (VIFs; Fox and Weisberg ) in the R package “car.” Variance inflation factors are generally considered to indicate the presence of multicollinearity if any VIF exceeds 10 (see O'Brien ).

Results

Mortality Estimation

The number of Striped Bass collected on the spawning grounds varied throughout the time series, ranging from 58 fish in 2006 to 403 fish in 2003 (Table ). Scale ages were reasonably precise, as scale readers had a high rate of agreement within 1 year of age (87–100%; NCWRC, unpublished data). Recruitment to the catch curve typically occurred at age 4 or age 5. Although the oldest Striped Bass encountered on the spawning grounds was an age‐13 female collected in 2005, only 73 (1.6%) of the 4,549 fish collected during the time series were age 9 or older.

Chapman–Robson mortality estimator metrics and mortality rates (Z = instantaneous total mortality rate; A = discrete annual mortality; F = instantaneous fishing mortality rate; u = discrete annual exploitation rate) for Neuse River Striped Bass, 1994–2015 (N = total catch; N_c = number in catch curve; T_c = age at recruitment to catch curve [peak‐plus]; ĉ = overdispersion parameter; LCL = lower 90% confidence limit; UCL = upper 90% confidence limit; RSE = relative standard error)

Metric	Year of sample
1994	1995	1996	1997	1998	1999	2000	2001	2002	2003	2004	2005	2006	2007	2008	2009	2010	2011	2012	2013	2014	2015
N	120	221	226	143	219	292	357	155	102	403	90	125	58	172	141	373	141	176	144	341	311	239
N c	36	107	71	81	148	151	111	69	67	98	48	97	21	96	24	231	71	55	67	106	129	95
T c	7	5	6	5	4	5	5	5	4	6	6	4	4	4	6	4	4	5	4	5	5	5
ĉ	0.57	1.60	1.10	2.23	1.64	1.97	4.17	3.90	5.53	2.20	0.12	1.81	1.57	2.52	3.19	4.96	0.13	2.41	0.85	0.42	1.51	0.90
Z	1.08	0.73	0.85	0.61	0.45	0.75	0.45	0.52	0.36	0.65	0.78	0.44	0.53	0.63	0.98	0.84	0.94	0.84	0.62	0.74	0.86	0.94
SEc	0.19	0.09	0.11	0.10	0.05	0.09	0.09	0.13	0.10	0.10	0.12	0.06	0.15	0.10	0.37	0.13	0.12	0.18	0.08	0.07	0.10	0.10
Z LCL	0.77	0.58	0.67	0.44	0.37	0.61	0.31	0.32	0.19	0.49	0.59	0.34	0.29	0.46	0.37	0.63	0.75	0.54	0.49	0.62	0.70	0.78
Z UCL	1.39	0.88	1.03	0.77	0.53	0.90	0.60	0.73	0.53	0.82	0.97	0.54	0.77	0.80	1.59	1.05	1.13	1.14	0.75	0.87	1.02	1.11
RSE (%)	17	12	13	17	11	12	20	24	29	15	15	14	28	16	38	15	12	22	12	10	11	11
A	0.66	0.52	0.57	0.45	0.36	0.53	0.36	0.41	0.30	0.48	0.54	0.36	0.41	0.47	0.62	0.57	0.61	0.57	0.46	0.53	0.58	0.61
A LCL	0.54	0.44	0.49	0.35	0.31	0.45	0.26	0.27	0.17	0.39	0.44	0.29	0.25	0.37	0.31	0.47	0.53	0.42	0.39	0.46	0.50	0.54
A UCL	0.75	0.58	0.64	0.54	0.41	0.59	0.45	0.52	0.41	0.56	0.62	0.42	0.53	0.55	0.80	0.65	0.68	0.68	0.53	0.58	0.64	0.67
F	0.84	0.48	0.61	0.36	0.21	0.51	0.21	0.28	0.11	0.41	0.53	0.20	0.28	0.38	0.73	0.59	0.69	0.59	0.37	0.50	0.61	0.70
F LCL	0.49	0.23	0.34	0.10	−0.02	0.25	−0.04	−0.01	−0.15	0.15	0.26	−0.03	0.02	0.11	0.17	0.31	0.43	0.25	0.12	0.26	0.36	0.44
F UCL	1.20	0.71	0.86	0.60	0.38	0.73	0.43	0.55	0.36	0.64	0.79	0.39	0.59	0.61	1.44	0.86	0.95	0.96	0.58	0.70	0.84	0.93
u	0.51	0.34	0.41	0.27	0.17	0.36	0.17	0.22	0.10	0.30	0.37	0.16	0.22	0.28	0.47	0.40	0.45	0.40	0.28	0.35	0.41	0.45

The Chapman–Robson mortality estimator generally performed well, as ĉ was greater than 4 in only 3 of 22 years (Table ). Mortality estimates were reasonably precise (RSE < 30%) and only exhibited a high degree of uncertainty in 2008. Instantaneous total mortality Z varied considerably throughout the time series, ranging from 0.36 to 1.08. Mortality was generally lowest during 1997–2007 and highest during 2008–2011. Values of F ranged from 0.12 to 0.84 (Table ; Figure ), assuming that the M given by Bradley () remained constant throughout the time series. Fishing mortality was greater than F_Threshold in 12 of the 22 years.

Striped Bass spawning stock fishing mortality (F) in the Neuse River, North Carolina, during 1994–2015. The 90% confidence interval is denoted by gray lines, while the interquartile range is within a green color gradient. The dashed red line represents the overfishing threshold (FThreshold = 0.41).

Mortality Modeling

Model assumptions

The ADF test indicated that spawning stock discrete annual mortality was nonstationary (P = 0.181). Therefore, all modeled variables were first‐differenced (Hyndman and Athanasopoulos ). The PACF indicated a correlation of 0.34 between A_t and A_t−1, suggesting weak autocorrelation. We did not consider this level of autocorrelation sufficient to warrant modeling as a first‐order autoregressive process given the small sample size and the potential for model overspecification. The VIFs ranged from 1.1 to 2.5, indicating a low likelihood of multicollinearity among predictor variables.

Model results

The best linear model supported by the data contained gill‐net effort and commercial harvest as predictors of discrete annual mortality (Table ). The global model containing all predictor variables accounted for 55% of the variability in spawning stock mortality, while the best model accounted for 50%. Every model receiving at least modest support as the best model (∆_i < 7) incorporated gill‐net effort as a predictor variable.

Linear models exploring the effect of environmental and exploitation factors on Striped Bass spawning stock discrete annual mortality, 1994–2015 (EFFORT = gill‐net effort; DO = dissolved oxygen; HARV = commercial harvest; TEMP = surface water temperature). The number of estimated model parameters (K) includes the predicting factors and an error term; final model runs did not include an intercept parameter. Akaike's information criterion (AIC_c), Akaike difference (∆_i), Akaike weight (ω_i), and R² are presented

Model	K	AICc	∆i	ωi	R 2
EFFORT, HARV	3	−39.95	0.00	0.64	0.50
EFFORT	2	−36.98	2.97	0.15	0.34
EFFORT, HARV, DO, TEMP	5	−34.88	5.07	0.05	0.55
EFFORT, DO	3	−34.81	5.14	0.05	0.36
EFFORT, TEMP	3	−34.60	5.36	0.04	0.35
EFFORT, DO, TEMP	4	−34.40	5.56	0.04	0.44
HARV	2	−31.68	8.27	0.01	0.14
DO	2	−30.67	9.29	0.01	0.09
HARV, DO	3	−30.38	9.57	0.01	0.20
HARV, TEMP	3	−29.83	10.12	0.00	0.10
DO, TEMP	3	−27.98	11.97	0.00	0.10
HARV, DO, TEMP	4	−27.23	12.72	0.00	0.20

Gill‐net effort was the most important predictor of spawning stock mortality relative to the four predictor variables examined (Table ; Figure ). Commercial harvest was the second most important predictor of spawning stock mortality, while summer dissolved oxygen and surface water temperature did not substantially influence spawning stock mortality (Tables ). Multiplying the model‐averaged gill‐net coefficient by the mean number of gill‐net trips for 1994–2015 (2,421 trips) suggests the gill‐net fishery mean discrete annual exploitation rate (u) was 0.29. Using the same procedure for commercial harvest (3,199 kg) suggests commercial harvest u is 0.08.

Relative importance of predictor variables affecting Striped Bass spawning stock mortality (Lindeman–Merenda–Gold [LMG] method)

Predictor variable	Model‐averaged coefficient	Relative importance (LMG)
θ	SE
Gill‐net effort	1.21 × 10−4	3.54 × 10−5	0.62
Commercial harvest	2.37 × 10−5	1.00 × 10−5	0.23
Dissolved oxygen	−1.73 × 10−2	1.63 × 10−2	0.10
Surface water temperature	2.50 × 10−2	2.71 × 10−2	0.05

Differenced (Δ) Striped Bass spawning stock discrete annual mortality (A; red) and differenced exploitation and environmental predictor variables (black) in the Neuse River, North Carolina (EFFORT = gill‐net effort; DO = summer mean surface dissolved oxygen; HARV = commercial harvest; TEMP = summer mean surface water temperature). Gill‐net effort, DO, and TEMP were modeled with 1‐year time lags.

Discussion

Catch‐curve methodologies recommended by Smith et al. () considerably reduced uncertainty in the Z‐estimates compared to previous Neuse River stock assessments. The SEs of Z in our study ranged from 0.05 to 0.37, compared to 0.06–0.61 in the most recent stock assessment (Table 11 in NCDENR ). Similarly, RSE exceeded 30% in only 1 of the 22 years in our study, compared to 13 of the 16 years in the previous stock assessment (NCDENR ).

The catch‐curve analysis indicates that the Neuse River Striped Bass spawning stock has been subjected to overfishing throughout much of the last two decades. The 22‐year mean F in this study (F = 0.46) is similar to the 18‐year mean rate (F = 0.47) that preceded the depletion of Albemarle Sound/Roanoke River Striped Bass in the 1970s (Hassler et al. ; NCDENR ). These high F‐values also approach the level of exploitation that was deemed a major factor in the Atlantic Striped Bass stock collapse (ASMFC ; Richards and Rago ). Mortality has not trended toward F_Target despite the development of two comprehensive management plans and increasingly restrictive recreational and commercial harvest regulations (see Appendix 14.5 in NCDENR ).

Linear modeling indicates that gill‐net effort is the most important factor influencing spawning stock mortality among the exploitation and environmental factors examined. Gill‐net effort accounted for substantially greater variability in spawning stock mortality than commercial harvest, and the model‐averaged coefficient identified a discrete annual exploitation rate of 0.29 for gill net effort. This suggests that the commercial multispecies gill‐net fishery imparts substantial mortality even when the Striped Bass harvest season is closed. The reason for this mortality is obscure, but it may be attributable to dead discard mortality; over‐quota and high‐grading mortality; avoidance, predation, and drop‐out mortality; or unreported, misreported, and illegal harvest (ICES ; Gilman et al. ; Batsleer et al. ; Uhlmann and Broadhurst ). In particular, discard mortality should be carefully considered, as Clark and Kahn () found that Striped Bass are acutely susceptible to discard mortality in multispecies gill‐net fisheries. Furthermore, Striped Bass discards in the large‐mesh gill‐net fishery were identified as the primary source of mortality within the CSMA (NCDENR ). The effect of gill‐net effort on discrete annual mortality as estimated by linear modeling was within 3% of the estimated effect of cryptic mortality in a cohort‐based model (u = 0.26; Table B.3 in Rachels and Ricks ), while the effect of commercial harvest was identical to the estimated discrete annual fishing mortality rate from commercial harvest in that study.

Contrary to exploitation factors, the environmental factors examined did not account for much variability in spawning stock mortality. Bradley et al. () also failed to detect a relationship between dissolved oxygen, water temperature, and Striped Bass mortality between summer 2014 and summer 2015. Although numerous Atlantic Menhaden fish kills have occurred due to hypoxic conditions throughout the time period encompassing our research, it appears that these events have relatively little impact on Striped Bass spawning stock mortality. Campbell and Rice () observed that estuarine fish can rapidly detect and avoid hypoxic areas in the Neuse River. However, they also found that habitat compression due to hypoxic conditions likely reduced growth rates in juvenile Spot Leiostomus xanthurus and Atlantic Croaker Micropogonias undulatus. Neuse River Striped Bass exhibit the fastest growth rates among coastal North Carolina Striped Bass populations (Rachels and Ricks ). It is likely that negative impacts of hypoxic conditions or water temperatures exceeding Striped Bass thermal optima would manifest through reduced growth rates before mortality effects are observed. Nonetheless, the parameter coefficients for summer mean dissolved oxygen and summer mean surface water temperature indicate the potential for increased spawning stock mortality as dissolved oxygen decreases and water temperature increases. These effects were minimal—approximately 2% change in discrete annual mortality per unit change in temperature or dissolved oxygen—compared to the cumulative effects of gill‐net effort and commercial harvest.

The inability to include recreational angling as an exploitation factor reduces the amount of variability in spawning stock mortality that can be accounted for in this study. The median annual recreational harvest during 2004–2015 was 2,337 kg and is similar to the median commercial harvest of 3,355 kg for the same time period (NCDMF, unpublished data). Thus, the actual commercial harvest and recreational harvest exploitation rates are similar, an observation supported by simulation studies (Rachels and Ricks ; Bradley ). It is likely that inclusion of factors that represent recreational harvest and discard would perform comparably to the results of the commercial harvest factor used in linear modeling. However, time‐dynamic trends in the level of recreational fishing effort or harvest could influence its importance relative to commercial harvest in a regression analysis. In fact, recreational effort declined dramatically during 2005–2010, concurrent with increases in discrete annual mortality. The continued collection of recreational creel survey data is warranted to elucidate long‐term effects of angling on Neuse River Striped Bass mortality.

Since the population is supported almost entirely by hatchery‐origin fish, changes to stocking practices may affect recruitment and mortality estimation. Although the annual stocking goal is 100,000 phase‐II (160–200 mm TL) Striped Bass, the actual stocking rate (Table ) has varied (coefficient of variation = 46%) and has included phase‐I fish (50 mm TL) in some years. Survival rates of phase‐I and phase‐II Striped Bass may be similar. Stocking practices in the nearby Cape Fear River are the same as those in the Neuse River, and phase‐I and phase‐II Striped Bass that were stocked at similar rates contributed almost equally to the Cape Fear River population (NCWRC, unpublished data). Additionally, the effect of variable recruitment on catch‐curve mortality estimation has been extensively explored by others. Ricker () determined that recruitment variation up to a factor of 5 did not prohibit catch‐curve use so long as the variability was random. Similarly, Allen () found that catch curves were useful for estimating mortality in populations that exhibited higher recruitment variation (55–84%) than the stocking variability observed in our study. Finally, although it does not yield insight into much of the entire time series of our data, our mortality estimates were very similar to those reported by Bradley et al. () for 2014–2015. The methodologies used in these studies (telemetry versus age structure) have different underlying assumptions, increasing confidence that mortality during the overlapping time periods was considerable.

Periodic strategists such as Striped Bass are resilient to periods of extended recruitment failure through the storage effect (Warner and Chesson ; Winemiller and Rose ). Recovery is contingent upon building spawning stock biomass by advancing the female age structure to older, more fecund fish (Secor ). Although regulating fishing mortality is one of the principal tools available to fisheries managers, “historical precedence is often invoked as a reason to continue unwise fishery management practices” (Richards and Rago ). However, the effectiveness of coordinated multi‐jurisdictional management efforts in significantly reducing exploitation has been demonstrated by the restoration of the Atlantic Striped Bass stock (Field ; Richards and Rago ).

Current high exploitation rates combined with low stock abundance and a high contribution of hatchery fish to the spawning stock (Rachels and Ricks ; Bradley et al. ) suggest that the expected recovery time of Neuse River Striped Bass continues to be “both uncertain and long” (Hilborn et al. ). Our research indicates that fisheries managers should reduce exploitation by focusing on reductions in gill‐net effort in areas of the Neuse River that are utilized by Striped Bass. Reducing spawning stock exploitation may confer an increased likelihood of recruitment during periods of favorable environmental conditions, thereby leading to improvements in population abundance and increased numbers of wild fish in the spawning stock.

Acknowledgments

We thank Joseph Hightower, Ken Pollock, and Paul Vos for their review and improvement of the statistical methods employed here. Kevin Dockendorf, Jeremy McCargo, and Chad Thomas provided guidance and positive critique throughout the development of the manuscript. We also appreciate the NCDMF and ModMon project for sharing information associated with their long‐term data collection programs. This research was supported in part by the Federal Aid in Sport Fish Restoration Program (Project F‐108). There is no conflict of interest declared in this article.