FISHERY BULLETIN: VOL 78. NO. 2 



they count island seaperch, Cymatogaster gracilis , 

 in 1971-73 samples. Island seaperch were ob- 

 served only at Santa Cruz Island, sporadically in 

 dense schools in the kelp canopy. To judge the 

 effect of ignoring this species in 1971-73, we com- 

 pared two 1974 samples, one with, the other with- 

 out counts of island seaperch. 



Bottom cinetransects were calibrated for area 

 covered by estimating their length and width in 

 the field. We estimated length experimentally by 

 measuring distances swam during transects at 

 Naples Reef. On each of five weekly tests, one 

 diver photographed while a second followed with a 

 tape measure. On 4 of 5 days, one transect went 

 upcurrent, the other downcurrent; on the fifth day, 

 three transects were measured in almost no cur- 

 rent. Lengths of all 11 transects averaged 

 47.8 ±2. 78 m (95% confidence interval— CI). We 

 estimated transect width by counting markers 

 placed along a surveyed course. A tape measure 

 marked the midline of a 50 m stretch of reef flat 

 and rill, and pairs of red floats, anchored 2 m on 

 either side of the tape at 5 m intervals, delimited a 

 4 m wide band. Four transects were photographed 

 along the course, with divers panning the camera 

 as usual. While viewing the projected films, we 

 estimated widths of pans by taking 4 m (width of 

 the marked band) and adding or subtracting esti- 

 mated distances before or beyond each float seen at 

 the extremes of the pans. Since widths of 43 pans 

 averaged 4.41 ±0.288 m (95% CI), the area covered 

 by an average bottom transect was taken as 47.8 m 

 (length) times 4.41 m (width) = 211 m^. 



Calibration of canopy transects was inherently 

 less accurate. Photographers swam more circu- 

 itous routes at more variable speeds in midwater, 

 where light fluctuated between dim and bright. 

 The best we could do was roughly measure dis- 

 tance travelled by a photographer swimming uni- 

 directionally under the canopy: an average length 

 of 62.3 m in four trials both with and against the 

 current. We estimated band width by counting 

 kelp blades (which averaged 0.5 m long) passed 

 during sweeps of the camera: counts usually var- 

 ied from 8 to 10, translating to 4-5 m. Hence, we 

 assumed band width of canopy transects to equal 

 that for bottom transects (4.41 m), and estimated 

 area covered by a canopy transect as 62.3 m 

 (length) times 4.41 m (width) = 275 m^. 



Statistical Analyses 

 Samples were specific for habitat, site, and year. 



For example, one sample was made up of fish 

 counts from a set of 43 transects filmed in the 

 canopy habitat at the mainland site during 1974. 



Sample species diversity of yearly canopy or bot- 

 tom samples offish assemblages in mainland and 

 island sites was measured by information- 

 theoretical indices (H), combining species "rich- 

 ness" (total species) and "evenness" (distribution 

 of individuals among species). We used Pielou's 

 (1966) method to estimate population diversity 

 from a set of cinetransects pooled incrementally in 

 random order. Diversity (Brillouin's//) of succes- 

 sively larger subsamples (size k) first increases 

 and then levels off, as the decrease in diversity 

 from adding more individuals of common species 

 balances the increase from adding rare species'. 

 Then increments of diversity per added individual 

 (hf;) between adjacent subsample estimates_(//^_i 

 and//^) are independent, and the mean ih) and 

 variance of /i^'s estimate the corresponding popu- 

 lation parameters. Species richness (S) was the 

 species count in a whole sample of size k = n. 

 Species evenness (J) was the ratio HJlnS, where 

 H^ is sample diversity and InS (natural log of 

 species count) is the theoretically maximum value 

 of H„ if the S species were equally abundant. 



We compared species composition between sites 

 and among years by proportionate similarity and 

 rank correlation. Similarity (/) in species composi- 

 tion was measured as: / = 1.0 - [0.5(Si = i |p„ - 

 Pj^ I )], where p^ is the proportionate abundance of 

 species i in sample j. Rank correlation (Kendall's 

 tau) was measured between ranked species arrays 

 (Johnson and Koo 1975). Clusters of similar sam- 

 ples were computed from matrices of / by the un- 

 weighted pair-group method using arithmetic av- 

 erages (Sneath and Sokal 1973). 



Mean counts of individuals and species per 

 transect were compared between sites and among 

 years (1971-74) by two-way analysis of variance 

 (ANOVA) for unequal and disproportionate sub- 

 class sizes, and by one-way ANOVA for unequal 

 sample sizes (Nie et al. 1975; Meeter and 

 Livingston 1978). With variates transformed, 

 sample distributions tended to normality (as indi- 

 cated by nonsignificant Kolmogorov-Smirnov 

 tests of goodness-of-fit) and sample variances 

 equalized (as indicated by nonsignificant F^ax 

 tests of largest variance ratios) (Sokal and Rohlf 

 1969; Meeter and Livingston 1978). A posteriori 

 contrasts among means were obtained by group- 

 ing means wdth nonsignificant ranges (Dunnett 

 1970; Nie et al. 1975). 



364 



