McKenna Spatial structure and temporal continuity of South Georgian fish community 



477 



of fish from approximately 100 stations (Fig. 1) by 

 thirty minute tows of a P32/36 otter trawl (mouth open- 

 ing of 17.5 m, 43-52 mm mesh liner). These stations 

 had been randomly located within three depth strata 

 (50-150 m, 150-250 m, 250-500 m) (Gabriel, 1987; 

 McKenna and Saila, 1989). The 1988-89 survey 

 sampled 41 stations from a regular grid between 50-m 

 and 250-m depth around South Georgia Island 

 (Fig. 1). Collections of demersal fish during that sur- 

 vey were made with fifteen minute tows of a Chris- 

 tensen Bottom Trawl (wing spread of 4.6m, 50-mm 

 mesh codend with a 6-mm liner )( McKenna 1 ). Complete 

 discussions of the methods and results of these sur- 

 veys may be found in Gabriel (1987), McKenna and 

 Saila ( 1989), and McKenna 1 . 



Species diversity has become a standard tool for 

 describing natural communities. Three measures of 

 diversity were calculated for each station sampled dur- 

 ing the three surveys and for each survey as a whole 

 based on the total catches. The first was species rich- 

 ness, which was simply the number of species caught 

 at each station. The second was the Shannon-Wiener 

 information index (H\ using loge) (Shannon and 

 Weaver, 1949). The third index (V") was a measure of 

 evenness (Pielou, 1977), 



V = H7\og(s*), 



where s* is the total number of species in the region 

 and was assumed to be 30. 



Species associations give a more detailed descrip- 

 tion of a natural community than the simple summary 

 provided by diversity indices. Species associations 

 within each survey were identified by use of Spearman 

 rank correlation (r') analysis (Freund, 1970, p. 311- 

 313), and were based on the numerical abundance of 

 fish caught during each survey. Species were arbitrarily 

 designated as rare if they occurred at 5% or less of the 

 sampled stations and uncommon if they occurred at 

 25% or less of the stations sampled. All pairwise com- 

 binations of common species (within each season) were 

 examined. These correlation values were then used to 

 generate Z-scores (Z = r'V(ra-l)) to test the null hy- 

 pothesis that the correlation was not significantly dif- 

 ferent from zero. Significance of associations was 

 determined at the 0.01 level (Z>2.58) (Freund 1970, 

 p. 313). 



Cluster analysis was used to examine the spatial 

 structure of the community in an effort to identify 

 significant subcommunities. This analysis method re- 

 duces the complex, multivariate data from field stud- 

 ies to a manageable level (Boesch, 1977) and often 

 reveals the presence of important physical or biologi- 

 cal factors affecting the distribution of the various spe- 

 cies assemblages occupying the sampled region (Pielou, 



1977). However, one of the major drawbacks of cluster 

 analysis (and many other exploratory techniques) is 

 the subjectivity associated with its application (Boesch, 

 1977; Pielou, 1984; Jain and Dubes, 1988). 



The heterogeneity ratio (HR) was used in this analy- 

 sis as the measure of similarity for the normal (R- 

 mode) cluster analyses, because of the objectivity it 

 provides. It is not affected by sample size or group size 

 (the number of samples included in a cluster), mea- 

 sures the beta-diversity (McNaughton and Wolf, 1979a) 

 existing among samples, and can be statistically tested 

 for significance (Kobayashi, 1987). 



HR = Sq/ECSq), 



where S Q is the total number of species present in Q 

 samples and E(S W ) is the expected number of species in 

 the Q samples. E(S^) is obtained by applying the mean 

 number of species per sample to the logarithmic series 

 distribution (Fisher et al., 1943), which is used as the 

 model describing the relation between the sample size 

 and number of species in the community. The logarith- 

 mic series describes communities which are dominated 

 by one or a few species and contain numerous rare 

 species. It is a common distribution in nature (Shepard, 

 1984; Dial and Marzluff, 1989) and is the most appro- 

 priate for this application (Kobayashi, 1987). 



HR is robust to the requirement of good fit to the 

 logarithmic series model (Kobayashi, 1987). However, 

 the fit of the data to that distribution was tested for 

 each station sampled during the three surveys with 

 the aid of the BASIC program LOGSRFIT.BAS (Saila 

 et al., 1991). This program generates values of X, 

 Fisher's a, and K. X and a are the parameters of the 

 logarithmic series model and a is also a diversity in- 

 dex (Fisher et al, 1943; Saila et al., 1991). Values ofK 

 measure the goodness-of-fit of the data to the logarith- 

 mic series model (Fisher et al., 1943). Values of K less 

 than 1.0 indicate a reasonable fit to the model. 



A normal (i?-mode) clustering of the stations was 

 performed for each survey. The clustering program, 

 HRCLUSTR.BAS (written in Microsoft QuickBasic v. 

 4.5, Microsoft 1988), was developed for this purpose 

 (available upon request). It was a modification of 

 Kobayashi's 2 program and used the HR and an un- 

 weighted paired group method of averaging (UPGMA) 

 linkage method (Sneath and Sokal, 1973). The null 

 hypothesis of non-significant clusters was rejected at 

 the 0.05 level (Kobayashi, 1987). Inverse (Q-mode) clus- 

 tering was also preformed on each survey's data to 

 classify species into groups. This provides insight into 



-Kobayashi, Faculty of Agriculture, Yamagata Univ., Tsuruoka, 997 

 Japan. Pers. commun. 1988. 



