52 
Fishery Bulletin 1 12(1) 
and otter trawls by using the geometric mean of the 
ratios of observed discards to reported landings. These 
ratios were developed from the NMFS Observer Pro- 
gram data set. 
The recreational kills consisted of type A (dead fish 
brought ashore and available for identification by in- 
terviewers), type B1 (fish not brought ashore, hence 
not seen by samplers, but were used as bait or were 
discarded dead), and type B2 (fish released alive; they 
were of small sizes, with a 10% assumed release mortal- 
ity). They were obtained from data-collection programs 
operated by the NMFS MRFSS since 1981. Estimates 
of North Carolina scrap fishery landings were provided 
by the North Carolina Division of Marine Fisheries, 
which is the only state agency that routinely sampled 
such a fishery since 1986 (the 1981-85 estimates were 
based on the proportion of Atlantic Croaker in the un- 
classified finfish bait landings during 1986-90). 
Atlantic Croaker also are one of the major compo- 
nents of the SESTF bycatch, but the magnitude of the 
SESTF Atlantic Croaker discards is highly uncertain. 
The related estimates were produced by using a simple 
fish-catch to shrimp-catch ratio for study materials col- 
lected in North Carolina and South Carolina, and the 
resulting catch ratio was expanded to the entire coast. 
Such estimates largely exceeded the reported landings 
in most years (Fig. 1A) but were considered extremely 
crude and unreliable. For this reason, ASMFC 1 omit- 
ted the SESTF bycatch in BDMs and included them 
in the age-structured model for sensitivity runs only. 
Likewise, the SESFT bycatch estimates were used here 
for sensitivity analyses. 
Biomass indices (Fig. IB) included the fall (Sep- 
tember-November) components of the Multispe- 
cies Bottom Trawl Survey (1972-2008) of the NMFS 
Northeast Fisheries Science Center (NEFSC) and the 
Coastal Trawl Survey (1990-2008) of the multiagency 
Southeast Area Monitoring and Assessment Program 
(SEAMAP). The NEFSC and SEAMAP indices were 
chosen because the corresponding surveys showed 
wide geographic coverage, temporal coverage, or both; 
have been conducted consistently; and have provided 
evidence of regular encounters with Atlantic Croaker 
of different age groups (ASMFC; 1 Appendix 1). More- 
over, unlike the coastwide MRFSS CPUE, the NEFSC 
and SEAMAP indices were considered reflective of the 
Atlantic Croaker stock size and trajectory (ASMFC 1 ). 
Although various model runs used the MRFSS CPUE 
during the 2010 stock assessment, this index raised 
many concerns and therefore it was excluded from the 
final assessment model (ASMFC 1 ). 
MWET was added as a variable of environmental 
forcing of the Atlantic Croaker population dynamics. 
Winter air temperature data for Virginia — a Chesa- 
peake Bay region state — were extracted from the web- 
site of the Southeast Regional Climate Center (http:// 
www.sercc.com/climateinfo_files/monthly/Virginia_ 
temp.html, accessed May 2012). Air temperature is 
considered a good proxy for estuarine water temper- 
ature because of the efficient ocean-atmosphere heat 
exchange in estuarine systems (Hare and Able, 2007). 
On the U.S. Atlantic coast, winter temperatures of one 
location (here, the Chesapeake Bay region) are a good 
proxy for the entire coast owing to a strong coherence 
among local winter temperatures (Joyce, 2002; Hare 
et ah, 2010). As shown in Hare and Able (2007) and 
Hare et al. (2010), MWET corresponded with the mini- 
mum monthly mean air temperature from December 
to March. Specifically, MWET values were the mean 
temperatures of the coldest months during the winter 
seasons. The Chesapeake Bay region’s MWET (Fig. 1C) 
was suited for a study of its effects on the Atlantic 
Croaker population dynamics because the Chesapeake 
Bay region is a major overwintering nursery area for 
the species (Hare et al., 2010). 
Biomass dynamic models 
The analyses covered the 1972-2008 period, consistent 
with the years for which data for BDM implementa- 
tions were available in the 2010 stock assessment 
(ASMFC 1 ). Two Bayesian state-space biomass dynamic 
models (BSSBDMs) were developed and used: a dis- 
crete BSSBDM without MWET (model 1, Ml) and a 
discrete BSSBDM that integrated MWET (model 2, 
M2). Both models used a one-year time (Z) step. A 
state-space model describes 2 interrelated time series 
of state and observation processes (Buckland et al., 
2004), both of which account for random errors. The 
state process defined the stochastic temporal dynamics 
of the unobserved (or latent) age-aggregated stock size 
of Atlantic Croaker that is due to natural variation. 
The corresponding error, referred to as process error, 
is the joint effect of random multiplicative factors (e.g., 
fluctuations in life history parameters, trophic inter- 
actions, environmental disturbance). The process error 
in Ml included all forms of environmental variations 
and, in M2, environmental variations over and above 
the variations pertaining to MWET. The observation 
errors (arising from measurement and sampling errors) 
related only to observed indices of biomass. These indi- 
ces were assumed to be a linear function of the latent 
biomass. 
Consistent with Meyer and Millar (1999) and Millar 
and Meyer (2000), Ml and M2 described the process- 
es under consideration through a set of 3 probability 
density functions (PDFs) g(.) and h(.), given the latent, 
beginning-of-the-year exploitable biomass ( Bt ), the sets 
of unknown model parameters ( 0 ), the set of known 
covariates (C), and observed indices of biomass by year 
(Oit; i = NEFSC index, SEAMAP index): 
^ 1972(^1972 | ©) 
Initial (1972) state PDF (la) 
St^t+i | Rt’©<C) 
State PDF (Z = 1973, ..., 2008) (lb) 
