114 



Fishery Bulletin 103(1) 



30.0 



Georgia Bight 

 Wat: 



Watermass 



/ Inner-shelf water 



O Inner-shelf- mid-shelf mixed water 



+ Mid-shelf water 



A Mid-shelf-Gulf Stream mixed water 



33 35 



Average salinity 



Figure 2 



The average temperature and salinity for each station; symbols used represent 

 the water mass designation for each station. The black polygons represent the 

 temperature and salinity boundaries (data for all seasons bounded by one polygon) 

 of three water masses defined by Pietrafesa et al. (1994; Georges Bight water; 

 Carolina Capes water, and Gulf Stream water). Four water masses were defined 

 in our study (inner-shelf water, inner-shelf-mid-shelf water, mid-shelf water, and 

 mid-shelf-Gulf Stream mixed water). 



as mid-shelf water and Gulf Stream water (Fig. 2). Gulf 

 Stream water was not encountered, but its temperature 

 and salinity properties are well documented (Churchill 

 et al., 1993; Pietrafesa et al., 1994). Mid-shelf-Gulf 

 Stream mixed water was highly stratified (Simpson's 

 stratification parameter value >10), with warm highly 

 saline water intruding on the surface during fall, win- 

 ter, and spring and cool highly saline water intruding 

 at depth during summer. Mid-shelf-Gulf Stream mixed 

 water was encountered on most cruises and was found 

 farthest offshore (Fig. 3). 



Cruises were assigned to one of four seasons (Ta- 

 ble 1) based on wind and temperature regimes. Al- 

 though Blanton et al. (1985) identified five seasons for 

 the southeast United States based on wind regimes 

 (Spring [March-May], summer [June- July], transition 

 [August], autumn [September-October], and winter 

 [November-February]), the temperature data collected 

 in our study supported classifying both August cruises 

 as summer and the March cruise as winter. 



Data analyses 



Multivariate analyses were used to define larval assem- 

 blages and to explore the factors that influence distri- 

 bution of larval assemblages on the continental shelf 

 off the coast of Georgia. Multivariate analyses arrange 

 sites and species along environmental gradients creating 

 a low dimensional map (an ordination). Analyses can 

 be conducted for samples where the distance between 

 points in the ordination represents the similarity of 

 species abundance between samples. Analyses also can 

 be conducted for species where the distance between 



points in the ordination represents the similarity in the 

 sample distribution between species. Ordinations, then, 

 can be analyzed in two ways: with regard to proximity 

 and dimensionality. Points that occur in close proximity 

 can be considered similar based on similar composition. 

 Points that occur on the same dimension define gradients 

 in the data. 



The effects of data transformation (untransformed, 

 square root transformed, and fourth root transformed) 

 and species inclusions (1% and 10% data sets) on the 

 ordination of community and environmental data by 

 two multivariate ordination techniques, multidimen- 

 sional scaling and correspondence analysis (CA), were 

 compared to determine which method was more effec- 

 tive at analyzing the larval fish data collected on the 

 continental shelf off the coast of Georgia (Marancik, 

 2003). Overall, the two analytical methods produced 

 similar ordinations and were robust to the inclusion of 

 rare species and to the type of data transformation. 



Correspondence analysis on untransformed larval 

 fish concentration data was used to define larval as- 

 semblages in relation to season and the entire two-year 

 data set. One of the strengths of CA is that it allows 

 one to plot analyses of species and station data simul- 

 taneously on one ordination, thereby, allowing immedi- 

 ate comparisons between those stations that occur in 

 close proximity in ordination space and those taxa that 

 influence that proximity. Eigenvalues are a measure 

 of the importance of each CA dimension (ter Braak 

 and Smilauer, 2002). Thus, the dimensions needed to 

 describe patterns in the data can be determined by an 

 abrupt drop in the magnitude of eigenvalues from one 

 dimension to the next. 



