478 



Fishery Bulletin 91(3). 1993 



their associations and distributions. The species were 

 clustered with the correlation coefficient and an 

 UPGMA linkage method (Jain and Dubes, 1988, p. 

 16). The results of these analyses were then combined 

 through nodal analysis (Williams and Lambert, 1961; 

 Lambert and Williams, 1962; Noy-Meir, 1971; Boesch, 

 1977) into two-way tables displaying the constancy and 

 fidelity of the identified groups (Fager, 1963; Westhoff 

 and van der Maarel, 1973). 



Constancy describes how widespread a species group 

 is within a given habitat and is expressed as 



C,j = a,/(n l n l ), 



where a„ is the actual number of occurrences of mem- 

 bers of species group ;' in collection (station) group j, 

 and n, and n t are the number of entities in the respec- 

 tive groups. C,j ranges from to 1, where 1 indicates 

 that all species occurred in all collections in the group 

 (Boesch, 1977). 



Fidelity describes the restriction of a species group 

 to a given habitat and is expressed as 



F„ = (a, l I.n J )/(n J 'La, J ) 



often classes (0-0.1, 0.1-0.2, etc.). Each data set was 

 sorted in the order of the gradient being examined and 

 probed with 500 skewers. Twenty random tables were 

 generated during each analysis to provide a test of the 

 null hypothesis that the assemblages were distributed 

 randomly along the gradient at a 0.05 probability level 

 (Pielou, 1984). 



Although the change in an assemblage along an en- 

 vironmental gradient may be gradual, there may also 

 be cases of rapid change. These ecotones may be due 

 to rapid change in the environmental gradient or pos- 

 sibly a biological change for some other reason 

 (Whittaker, 1960; McNaughton and Wolf, 1979b). Gra- 

 dient analysis (Webster, 1973; Ludwig and Cornelius, 

 1987) was applied to the data sorted in order of each 

 of the above potential gradients, to search for bound- 

 aries along the length of those gradients. The BASIC 

 program, GRADSECTBAS (written in Microsoft 

 QuickBasic v. 4.5, Microsoft 1988), was developed for 

 that purpose (available upon request). It used a mov- 

 ing split-window distance method of comparing mov- 

 ing averages (Whittaker, 1960). Each window contained 

 nine stations and the multivariate distance measure 

 was based on the squared Euclidean distance. 



Values of this index greater than 2 suggest a strong 

 preference of species in a group (i) for a collection group 

 (j), and values much less than 1 suggest avoidance of a 

 collection group (J) by a group of species (i) (Boesch, 

 1977). 



The size structure of each species' population (1987- 

 88 and 1988-89 surveys only) was also examined for 

 patterns. Stations were clustered (using HR and 

 UPGMA linkage) based on the abundance of individu- 

 als in each centimeter length class. 



Natural communities may be aligned along environ- 

 mental gradients, changing continuously from one end 

 to the other (Whittaker, 1960). Skewer analysis uses 

 multivariate correlations to measure the significance 

 of linear trends in natural communities (Pielou, 1984; 

 Saila et al., 1991). It was used in this study to exam- 

 ine the significance of trends in each community along 

 suspected gradients of depth, longitude, latitude, and 

 time (date within each survey). It was performed twice 

 on each data set. The first application used the abso- 

 lute measures of abundance, which detects trends that 

 are a combination of changes in the species assem- 

 blage and the absolute abundances of individuals at 

 each station. The second application used those abun- 

 dance values converted to proportions, which is sensi- 

 tive only to the changes in species composition along 

 the gradient. Kendall's Tau provided the measure of 

 correlation between each skewer and the observed data. 

 The distribution of Tau was observed through the use 



Results 



The assemblage of demersal fish occupying the South 

 Georgia region, during the three AMLR surveys, con- 

 sisted of slightly more than two dozen (28) species, 

 with representatives from thirteen families (Table 1). 

 Seven of these species accounted for the vast majority 

 (>85%) offish biomass and individuals in the system 

 (Fig. 2). Five are relatively large (>50cm), commer- 

 cially valuable species. These included the three icefish 

 (Channichthyidae) of the region (Chaenocephalus 

 aceratus, Champsocephalus gunnari, and Pseudo- 

 chaenichthys georgianus) and two species of rock cod 

 (Nototheniidae: Notothenia gibberifrons and Notothenia 

 squamifrons). The other two important species 

 (Nototheniops larseni and Nototheniops nudifrons) are 

 small but abundant members of the Nototheniidae. 

 The distributions of these species with depth was simi- 

 lar to those described by Tiedtke and Kock ( 1989) for 

 the fishes of Elephant Island. 



Champsocephalus gunnari was the clear dominant 

 both numerically and based on biomass (Fig. 2). One 

 quarter of the biomass and nearly one third of the 

 individuals were members of this species. It feeds al- 

 most strictly on krill (McKenna. 1991) and leads a 

 more pelagic existance than the other icefish of the 

 region (Kock, 1985b). Notothenia gibberifrons was a 

 close second in biomass and was the third most abun- 

 dant species. It is adapted to an epibenthic environ- 



