838 



Fishery Bulletin 101(4) 



slowly and quietly at the surface from a distance of at least 

 12 m. All fishes associated (defined below) with the floating 

 object were counted and identified; therefore the statistical 

 unit in all of the experiments described below was a single 

 FAD-associated assemblage of fishes at a given date and 

 time. Horizontal underwater visibility, measured with a 

 Secchi disk, was always sufficient to allow the identification 

 of species and to count individual fishes from a minimum 

 distance of 2 meters. 



Any fish observed within 2 m of a FAD was considered to 

 be "FAD-associated." Very few fishes were observed outside 

 of this range, and with rare exceptions, fishes responded 

 to the approach of an observer first by swimming towards 

 the observer and then by moving closer to the FAD, rather 

 than away from it. Different species of fishes used the 

 space around and below the FAD differently, as both Good- 

 ing and Magnuson (1967) and Hunter and Mitchell (1967) 

 described, but the juvenile fishes that predominated in the 

 present study were unambiguous regarding their relation- 

 ship to the FADs. After fishes resumed their prior positions 

 in relation to a FAD, continued observations of these fishes 

 revealed that there was no inclination to abandon that FAD. 

 The appearance of potential predators invariably resulted 

 in a tightening of the spatial distribution around the FAD. 



When the experiment required the capture of FAD-as- 

 sociated fishes, I used a smaller (1.1x1.3 m) version of the 

 diver-operated liftnet described by McCleneghan and Houk 

 (1978). Captured fishes were preserved for further studies, 

 held in grow-out facilities in the laboratory to verify spe- 

 cies identification, or released 1.5 km down-current over 

 rocky reef habitat to ensure that they had effectively been 

 removed from the FAD array. 



Diversity calculations 



Measurements of species diversity provided a means of 

 monitoring treatment effects on the composition of FAD- 

 associated assemblages. I measured species diversity using 

 species richness (S, the raw number of species observed), 

 and the Brillouin index (HB). S is simple and widely 

 used, but increases with sample size, and, where sample 

 sizes are unequal, HB provides a less biased measure of 

 diversity (Magurran, 1988). In addition, HB was chosen 

 over one of the more commonly used information theory 

 indices (e.g. the Shannon-Wiener index) because 1) FAD- 

 associated assemblages are not a random sample of poten- 

 tial recruits (different species vary in their attraction to 

 floating objects) and 2) each of these assemblages was 

 completely censused — not sampled (Magurran, 1988, and 

 references therein). 

 HB is calculated as 



///? = 



In/V! 



-Z'"".^ 



where N = the total number of fishes of all species 

 observed; and 

 ;i, = the number of individuals within the ith spe- 

 cies (Magurran, 1988). 



Statistical analyses 



1 used a two-way repeated measures ANOVA (a=0.05) to 

 test for treatment differences, differences among observa- 

 tion dates and evidence of treatment-by-sample interaction 

 for assemblage sizes (no. of fish(es)), species richness (S) 

 and species diversity (HB). Because individual species dif- 

 fered in their relative abundance and had different ecologi- 

 cal requirements, there was the potential for the dominant 

 species to bias comparisons of experimental treatments 

 where assemblage size (a combinination of all species) was 

 used. For all experiments except for the recruit-enriched 

 experiment, I repeated the statistical analyses twice: once 

 using the number of sergeant major damselfish (Abudefduf 

 troschelii) only and again using all fishes combined but 

 with A. troschelii removed. For the artificial fish experi- 

 ment where rainbow runner (Elagatis bipinnulata) were 

 particularly abundant, I ran separate analyses for A. tros- 

 chelii alone, £. bipinnulata alone, and for all species minus 

 the numbers of A. troschelii. When no fishes were present 

 at a FAD, HB was undefined; the "missing" data were re- 

 placed according to the procedures of Zar ( 1996) and the 

 degrees of freedom were reduced accordingly. 



Fish-removal experiments 



Observations on similar FADs during a previous field 

 season at the same location suggested that the turn- 

 over rate of fish associated with anchored FADs is high, 

 especially when the initial assemblage is large, but that 

 some recognizable individuals did persist from day-to-day 

 (Nelson, 1999). If immigration and emigration rates were 

 as high as suspected, FADs cleared offish on a daily basis 

 should not differ significantly from undisturbed FADs in 

 their mean assemblage size or in the average number of 

 species associated with these FADs. Wickham and Russell 

 (1974) compared the catches of bait fishes associated with 

 FADs subject to daily purse-seine sets versus those allowed 

 to "soak" undisturbed for three days prior to a single 

 purse-seine set and concluded that sufficient emigration 

 and immigration occurred on a daily basis to remove any 

 appreciable effect of daily removals. I sought to address 

 similar questions, but by using a different system (juvenile 

 reef fishes versus bait fishes). 



To test these hypotheses, 1 deployed eight identical FADs 

 on 30 June 1997 in two lines of four FADs each, oriented 

 roughly parallel to shore (Figs. 2 and 3). Fishes associ- 

 ated with all eight FADs were counted and identified on 

 a daily basis, beginning 3 July 1997. Alternate FADs were 

 cleared of all fish, following the daily counts; the remaining 

 FADs were left undisturbed in such a way as to distribute 

 the treatments evenly among the FAD array. After three 

 consecutive days of these observations (series 1), the treat- 

 ments were reversed, and previously undisturbed FADs 

 were cleared, and those that had been cleared regularly 

 were left undisturbed (series 2). The treatments were re- 

 versed in an attempt to control for possible positional effects 

 of the FADs. However, the two series were necessarily run 

 consecutively, not concurrently; therefore treatment posi- 

 tion was confounded by sample date. I used a 2 (cleared vs. 



