790 



Fishery Bulletin 92(4). 1994 



habitat times the square root of anticipated species 

 richness for that habitat (adapted from Steel and 

 Torrie, 1960). 



The preliminary data (Fig. 2) indicate that 19, 5- 

 minute sampling intervals (95 minutes) were suffi- 

 cient to observe at least 95% of the total species re- 

 corded in each habitat during the preliminary work. 

 A maximum of 40 dives, limited to 20 minutes of 

 survey time per dive (four 5-minute intervals), were 

 available for each of the four surveys. Thus, the avail- 

 able sampling time likely was sufficient to obtain a 

 complete record of the noncryptic species present in 

 each habitat. 



Sampling procedures 



All dives were completed between 0930 and 1630 

 hours to take advantage of maximum light levels. At 

 each site a marker buoy was deployed at the start of 

 a transect. One of three dive teams, each consisting 

 of two divers, swam a 15-minute transect with the 

 prevailing current. One diver operated a color video 

 camera with a 51-mm lens; the other towed a sur- 

 face buoy. The video operator waved his hand in front 

 of the camera to signify the beginning and end of 

 each transect. Appearance of the towed buoy released 

 by divers signaled to the boat the beginning of a 

 transect. The camera was held in a rigid forward 

 position about 1 m off the bottom. Fishes in cryptic 

 locations were not recorded by the video camera. At 

 the termination of a transect the camera was turned 

 off, the towed surface buoy was anchored, a stan- 

 dard black and white Secchi disk (30 cm in diam- 

 eter) was used to measure horizontal visibility, and 

 bottom water temperature was recorded. 



During each transect swim the vessel approached 

 the towed buoy at 5-minute intervals, and the crew 

 recorded the LORAN C coordinates and plotted its 

 position. The plots were used to measure the length 

 of each transect and to calculate distance covered 

 during each 5-minute interval. Transect length and 

 Secchi disk visibility could be used to estimate area 

 sampled. However, after reviewing the initial tapes, 

 we estimated the transect width to be 4 m (2 m on 

 each side of focal center), because small fishes (70 to 

 150 mm) could be identified with certainty only out 

 to an estimated distance of 2 m. Larger fish were 

 recorded as they came into view. To avoid duplicate 

 counts, only maximum numbers of species that 

 passed by the camera more than once were used. 

 These species were easily identified by the camera 

 operator. Because transect width remained constant, 

 data are reported as number of fish per meter of 

 transect. Generally, two transects were swum at each 

 site, beginning at the same location and heading with 



the prevailing current. Plots of the transects showed 

 little overlap. 



Videotaped transects were viewed to estimate 

 abundance of each species seen within each 5-minute 

 interval. Videotapes were projected on a 50-cm color 

 NEC Corporation Television and were analyzed by a 

 single observer. Viewing was in real time with fre- 

 quent pauses, reverses, and repeated counts until 

 the observer obtained the same count of species three 

 times. Date, location, Secchi disk measurement, bot- 

 tom water temperature, and number of each species 

 per type of habitat were recorded on data sheets. 

 Because habitat type often changed during a 

 transect, habitat changes were closely monitored and 

 species were apportioned appropriately. 



Community analysis 



Species-specific data were summarized by habitat 

 type. Statistical analysis, data summarization, and 

 graphic representation were accomplished with SAS 

 version 6.03 software system (SAS, 1987). Data were 

 summarized over sites within habitats and surveys, 

 and the effects of survey (4 surveys) and habitat (3 

 habitat types) on total fish density and number of 

 species observed were tested with two-way ANOVA's. 

 Cluster analysis was used to classify Gray's Reef 

 sampling sites according to the species composition 

 of the fish community. Species that were not found 

 in at least 10% of ledge, live-bottom, or sand sites in 

 any one survey were eliminated. For each survey, 

 relative abundance data (number/m of transect) were 

 arranged in a species-by-site matrix, standardized 

 by dividing each element by the square root of the 

 product of the row total and column total (simulta- 

 neous double standardization), and converted to a 

 site-by-site Canberra Metric dissimilarity matrix 

 (Clifford and Stephenson, 1975). Sites were grouped 

 by means of the "flexible sorting" algorithm of Lance 

 and Williams (1967) and the cluster intensity coeffi- 

 cient was set at -0.25 to approximate the median 

 clustering strategy. Analysis was conducted with 

 SIMCLUST statistical software (Wolfe and Chester, 

 1991). 



Results 



A total of 110 transects covering a distance of 24 km 

 (4.9 km over ledge, 12.7 km over live-bottom, and 

 6.4 km over sand) were made during the study. Over 

 92,000 fish, including 66 species and 36 families, were 

 recorded and identified from the videotapes (Table 1). 

 Number of species and density offish ( individuals/m 

 transect) varied significantly among the four surveys 



