100 
Fishery Bulletin 108(1 ) 
Florida, by using a ThermoQuest Finnigan Delta Plus 
mass spectrometer (Thermo Fisher Scientific, Inc., Bre- 
men, Germany) with a Kiel III device. Standard mass 
spectrometric techniques were applied after the car- 
bonate was decomposed to C0 2 with 100% phosphoric 
acid. ISODAT software (Thermo Fisher Scientific, Inc., 
Waltham, MA) indicated all isotopic values, which are 
reported with standard delta notation (%c): 
<S 18 0 = [(( 18 0/ 16 0) x /( 18 0/ 16 0) s ) - 1] x 1000, 
where X = sample; and 
S = the marine carbonate standard Vienna 
Peedee belemnite, VPDB (Gao et al. 2001). 
Statistical analysis 
To account for any effect of fish size on <5 13 C and <5 18 0 
isotope ratios, the relationship between each isotope 
and otolith mass (which is considered to be proportional 
to fish age) was examined within each regional group. 
Where a significant linear relationship was found, 
isotopic ratios were corrected by using the following 
formula: 
R co rr = R - (B x Mass), 
where R = isotopic ratio; 
B = slope; and 
Mass = otolith mass. 
Permutational multivariate analysis of variance (PER- 
MANOVA) was used to test for differences in isoto- 
pic signatures among years, regions, and sites within 
regions. PERMANOVA is a “semiparametric,” multivari- 
ate version of a univariate one-way ANOVA (Anderson, 
2001), producing a pseudo F-statistic, and a P -value. 
Canonical analysis of principal coordinates (CAP), with 
the “leave-one-out” method of cross-validation was used 
to assess whether juvenile snapper from South Florida 
marine ecosystems could be accurately classified to one 
of the five regions. Sample collection between years, 
regions, and times of year was not consistent, and 
therefore only data from “summer” (May to September) 
were included. The four larger regions (Biscayne Bay, 
Florida Bay, Lower Keys, and Ten Thousand Islands) 
were sampled at least once in 2002 and 2003 only. 
Therefore, only data from these two years were used 
to assess spatial variation in isotope signatures. Old 
Dan Bank and Sprigger Bank in Florida Bay were 
sampled in late May-June in 2001, 2002, 2004, and 
2005; thus, temporal variation was also investigated. 
Age-length estimates for juvenile gray snapper from 
Florida were used to ensure that the otolith samples 
reflected juvenile isotopic signatures within the year 
of collection (Allman and Grimes, 2002; Gerard, 2007). 
Isotope data were normalized before analysis, and all 
multivariate analyses were completed in Primer-6, 
(Plymouth Marine Laboratory, Ivybridge, U.K.) with 
PERMANOVA add-on. 
Results 
Spatial variation 
An initial PERMANOVA between years and regions 
(except the Dry Tortugas) for 2002 and 2003 data com- 
bined showed significant differences between regions 
(pseudo-F=358.3, PcO.001) and years (pseudo-F=41.8, 
PcO.001). Because this analysis showed that data from 
the two years could not be pooled, further analyses 
were performed on each year separately. Across both 
years, the Dry Tortugas and Florida Bay regions had the 
highest stable isotopic carbon ratios, and Ten Thousand 
Island had the lowest (Fig. 2). Stable oxygen isotopic 
ratios were lowest at the Ten Thousand Islands and 
highest in Florida Bay. Pairwise PERMANOVA on 2002 
and 2003 data that were treated separately showed 
significant differences between all regions in both years 
at P<0.05. 
More than one site was sampled within the same 
region and in the same month on only three occasions: i 
in Florida Bay in June 2002 and September 2003, and 
in the Ten Thousand Islands in September 2002. PER- ; 
MANOVA indicated that isotope signatures from fish 
collected at different sites within these regions were 
significantly different (Florida Bay June 2002: pseudo- 
F=8.90, P<0.001; Florida Bay September 2003: pseudo- 
F- 7.60, P=0.003; Ten Thousand Islands September 
2002: pseudo-F= 10.66, PcO.001. Pairwise PERMANOVA 
tests between sites within a region showed that the 
strength of differences in isotope signatures among 
sites was variable (Table 1). Differences between the 
same sites within Florida Bay in June 2002 tended to 
be stronger than in September 2003. 
Leave-one-out cross validation analysis for 2002 data 
only yielded an average classification accuracy of 79.7%. 
Classification success rates were 68.0% (Florida Bay), 
50.0% (Biscayne Bay), 100.0% (Ten Thousand Islands), 
83.3% (lower Florida Keys), and 90.0% (Dry Tortugas). 
The same analysis on 2003 data returned much lower 
classification successes, with 63.5% overall (Florida Bay 
[69.4%], Biscayne Bay [60.0%], Ten Thousand Islands 
[90.0%], lower Florida Keys [38.9%]). This result was 
largely due to the absence of samples from the Dry Tor- 
tugas in 2003 (which were strongly distinct from other 
regions in 2002) and to the lower classification success 
in the Lower Keys in 2003 than in 2002. Classification 
success rates for Biscayne Bay, Florida Bay, and the Ten 
Thousand Islands were similar between years. 
Temporal variation 
Samples were taken at both Old Dan Bank and Sprig- 
ger Bank in late May or June of 2001, 2002, 2004, 
and 2005. Scatter plots showed significant overlap 
between samples from different years (Figs. 3 and 4). 
PERMANOVA revealed significant differences between 
samples from 2004 and 2002, and 2004 and 2005 at 
both sites (Table 2). At Old Dan Bank, results for 2001 
were also significantly different from those for 2004, 
