Mather et al.: Use of non-natal estuaries by migratory Morone saxat/lis in summer 
331 
part of the reporting procedure, for most individuals, 
only a single recapture record existed. An approximate 
location (latitude, longitude, ±4 km) was created by 
assigning the recapture to the center of the nearest 
water body. When more information was included (e.g., 
mouth of the Merrimack River estuary), that datum was 
assigned a more specific recapture location. Most recap- 
ture records referenced specific locations and therefore 
there was relatively little error in estimating recapture 
location this way. Because most fish were recaptured 
by recreational anglers, tag reporting rate was likely 
similar across recapture locations. 
By comparing release and recapture locations, we 
examined whether striped bass tagged in Massachu- 
setts were part of the coastal migratory stock, whether 
they stayed in a localized area for a prolonged period 
in summer, and whether they returned to the same 
location after several years. To confirm the migratory 
status of striped bass tagged in Massachusetts, the 
location where tagged fish were recaptured in late fall, 
winter, or early spring was compared to the release 
location. To determine if striped bass remained in the 
same area throughout the summer, recapture loca- 
tions in summer were compared to the location where 
fish were released. For this, two nonexclusive, spatial 
recapture scales were used: the larger Massachusetts 
coast area (MA; Fig. 1A) and the smaller Great Marsh 
area (GM; Fig. IB). Two time periods were considered: 
the first season in which they were tagged (<104 days 
and before 21 September, i.e., summer), and all times 
combined. Without extensive movement records on in- 
dividual fish, the possibility that tagged fish moved 
out of the release estuary in the summer and then re- 
turned there in the fall cannot be discounted. However, 
to minimize this possibility, recapture records from 
the early (May 1 through 10 June) and late migration 
(22 September through 31 November) periods were 
excluded because these were times when migratory 
striped bass were hypothesized to be in transit. To 
determine if migratory striped bass returned to the 
same area in subsequent years, the number of fish 
that were recaptured in the area in which they were 
released was quantified for recaptures that occurred 
>12 months after release. 
To examine whether the number of striped bass re- 
captured in their release location was different than 
expected by random movement models, simple, dis- 
crete time, stochastic Markov chain models were used 
(Agresti, 2002). These were parameterized by a series 
of model states (locations in the estuary or ocean where 
migratory striped bass could occur) connected by transi- 
tion probabilities (rates at which striped bass may move 
between these geographic locations). Although other 
models have been used for animal movement, especially 
when large amounts of telemetry data are available 
(Jonsen et al., 2003; 2006), Markov chains are simple, 
require the least amount of data, and have been used 
to model biological processes (Shull, 2001; Steel et al. 
2001) including movement (Hestbeck et al., 1991; John- 
son et al., 2004). Furthermore, Markov chains require 
few assumptions; for example, all that is needed to pre- 
dict the next location of an animal with this approach 
is knowledge of the animal’s present location. 
Small-scale models were used to address how many 
recaptures would be expected at two scales of release 
(MA, GM) if striped bass movements were random. 
Although many movement models were plausible, the 
examples below provide insights into how to interpret 
observational recapture data for migratory striped bass. 
In random model 1 (RM-1), model states represented 
three localized, geographic locations in which a feeding, 
migratory striped bass could be found: 1) the target or 
release area (A t at two scales, GM, MA); 2) the ocean; 
or 3) another adjacent area (A o ) (Fig. 2A). The prob- 
ability of staying in the release or target area was p e , 
the probability of leaving that area was 1 —p e . In the 
first random model (RM-1), a fish must move through 
the ocean to get to another location. In RM-1, the prob- 
ability of staying in the ocean ( p 0 ) was 0.50, and the 
probabilities of staying in the two non-ocean estuaries 
were the same, although not necessarily 0.50 (RM-1; 
Fig. 2A). An assumption of RM-1 was that the fish did 
not prefer the release area over the adjacent area and 
that fish were equally likely to stay in the ocean or go 
to an estuary. A weekly time step was used. Transition 
probabilities for a striped bass in a model state always 
summed to 1.0. 
In random model 2 (RM-2), eight states were used 
to simulate the complexity of the Great Marsh (Figs. 
IB and 2B). Four estuarine areas (A I -A 4 ) had direct 
connections to Plum Island Sound and represented the 
Merrimack (A ; ), Parker (A 2 ), Rowley (A 3 ), and Ipswich 
(A 4 ) estuaries. Three of these (A 2 -A 4 ) were connected 
to the ocean through Plum Island Sound whereas the 
Merrimack River estuary (A ; ) was also connected di- 
rectly to the ocean. The Essex River (A 5 ) was adjacent 
to Plum Island Sound, connected to the ocean, but not 
directly connected to Plum Island Sound. Neighboring 
estuaries that were not part of the Great Marsh were 
represented by (A 0 ) 
Both models began with the release of 100 striped 
bass (individuals or schools) from the target area (A t for 
RM-1, or A 3 for RM-2) and continued until the numbers 
of migratory striped bass in each model state stabilized 
(10 weeks). The outcome predicted what proportion of 
model fish would be recaptured in the release estuary 
if movements in all directions were equally likely, i.e., 
random. P e , the proportion of fish still in the release 
area after 10 weeks, was adjusted to fit observed recap- 
ture proportion data for the first summer (<104 days 
and before 21 September). This weekly probability of 
fish remaining in the tagging estuary was matched 
against the observed recapture proportion. The ob- 
served recaptures for the Great Marsh were fitted to 
both models; the observed recaptures for Massachusetts 
were fitted only to the first, general model. Fitting to 
recaptures was possible because the model had only 
one parameter, p e . Density dependence and intraspecific 
interactions were not included in these simple models. 
Observed and expected were compared by using x 2 ■ 
