48 
Fishery Bulletin 106(1) 
which is a bottom trawl survey program designed to 
sample late-juvenile and adult fishes in the mainstem 
Chesapeake Bay (i.e., nontributary waters). During 
2002-2006, a total of 25 ChesMMAP cruises were 
conducted (March, May, July, September, and Novem- 
ber annually) and approximately 80 to 90 sites were 
sampled during each cruise. Sampling locations were 
chosen according to a stratified random design, and 
strata were defined by water depth (3-9 m, 9-15 m, 
and >15 m) within five 30-latitudinal minute regions 
of the bay. The locations sampled in each stratum of 
each region were randomly selected and the number 
of locations was in proportion to the surface area of 
that stratum. At each sampling location, a 13.7-m 4- 
seam balloon otter trawl (15.2-cm stretch mesh in the 
wings and body and 7.6-cm stretch mesh in the cod 
end) was towed for 20 min at approximately 6.5 km/h. 
The catch from each tow was sorted and individual 
lengths (total length, TL) were recorded according to 
species or size-class if distinct classes within a par- 
ticular species were evident. Stomachs were removed 
from a subsample of each species or size-class and 
immersed in preservative for diet composition analysis 
after each cruise. 
Identification of stomach contents 
The contents of each stomach were removed for identi- 
fication to the lowest possible taxon. Prey encountered 
in the esophagus and buccal cavity were included for 
identification (and assumed not to be the result of net 
feeding because of a lack of retention of prey in large 
mesh gear), whereas prey in the intestines were ignored 
because of the difficulty associated with identifying 
digested prey items in advanced stages of decomposition. 
All prey items were sorted, measured (either fork or total 
length, as appropriate and when possible), and the wet 
weight (0.001 g) of each was recorded. 
General diet description 
To summarize the diet composition of summer flounder 
in the mainstem of Chesapeake Bay, a measure of per- 
cent weight was calculated for each prey type (Hyslop, 
1980). Because the ChesMMAP trawl collections yielded 
a cluster of summer flounder at each sampling loca- 
tion, the aforementioned percentages were calculated by 
using a cluster sampling estimator (Bogstad et ah, 1995; 
Buckel et ah, 1999). Therefore, the contribution of each 
prey type to the diet by weight (%W k ) was 
n 
% w k = *100 , (1) 
« n 
!= 1 
where q lk = — 1 — > 
and where n - the number of trawls containing summer 
flounder; 
M t = the number of summer flounder collected 
at sampling site i; 
w ( = the total weight of all prey items encoun- 
tered in the stomachs of summer floun- 
der collected from sampling location i; 
and 
w ik = the total weight of prey type k in these 
stomachs. 
The variance estimate for %W k was given by 
n 
^M?(q lk -W k ) 2 
var (%W,,) = — xl00 2 > (2) 
k nM 2 n - 1 
where M = — is the average number of summer 
flounder collected at a sampling 
location. 
Ontogenetic and temporal changes in diet 
Canonical correspondence analysis (CCA; ter Braak, 
1986), a multivariate direct gradient analysis tech- 
nique, was used to explore the relationship between 
summer flounder diet and three factors: fish size (mm), 
year (2002, 2003, 2004, 2005, 2006), and month (March, 
May, July, September, November). Spatial variations 
in summer flounder diet were not explored because the 
distribution of summer flounder in Chesapeake Bay is 
restricted primarily to the polyhaline ( > 18 ppt) region 
of the bay (Fig. 1). 
The summer flounder collected ranged in size from 
148 to 712 mm TL (Fig. 2). To examine the effect of 
fish size on diet using CCA, we grouped summer floun- 
der into size categories such that all members of a 
given category exhibited a relatively consistent diet 
composition. Summer flounder were grouped into 25- 
mm size-classes, and diet was calculated for each with 
Equation 1. After trimming 10% of the observations 
(i.e., 25-mm size-classes) on account of low probability 
density in order to minimize outliers, cluster analy- 
sis (Euclidean distance, average linkage method) was 
used to group size-classes with similar diet composi- 
tions into broader categories. A scree plot indicated 
the presence of four clusters (Fig. 3A) (McGarigal et 
ah, 2000), corresponding to four broad size-categories: 
<225 mm TL (small), 225-374 mm TL (small-medium), 
375-574 mm TL (large-medium), and >574 mm TL 
(large) (Fig. 3B). 
For the CCA, each element of the response matrix 
was the mean percent weight of a given prey type at 
a given sampling site in a particular size, month, and 
year combination. The matrix was log-transformed 
(ln[;c+l]) to account for the log-normal distribution 
