724 



Fishery Bulletin 101(4) 



cording to their position with respect to the depth 

 contour and by identifying discrete geographical 

 clusters of trawls (Fig. 1). We then compared varia- 

 tion among fishing areas within years for those 

 areas with at least five observ'ations (areas II, 12, 

 Nl, N2, SI, S2 in 1998, and areas Nl, N3 and N4 

 in 1999). Interannual spatial comparisons were 

 not possible because there was little overlap in 

 sampled areas between the two years. 



Statistical analysis 



We based CPUE estimates on all trawls (?!=445). 

 Statistical analyses evaluating temporal and 

 spatial variation in seahorse bycatch included 

 only those trawls with nonzero seahorse observa- 

 tions (;i=205). The analysis required the data or 

 their residuals to be normal; this was achieved by 

 inverse hyperbolic sine transformations following 

 exclusion of zeros (Zar, 1996). It should be noted 

 that by excluding zeros we overestimated seahorse 

 CPUE and we lost information about areas where 

 seahorses were absent or rare. Our analyses 

 should be interpreted as applying to locations 

 where and times when seahorses were found in 

 sufficient numbers as to be caught. 



We examined the data using ANOVA, AN- 

 COVA, linear regression and chi-square analyses 

 (Zar, 1996; SYSTAT, version 7.0, SPSS Inc., Chicago, ID. 

 All two-way and three-way ANOVAs included tests for 

 interactions. We used a general linear model because the 

 data were unbalanced. Interactions were removed from the 

 model if they were found to be nonsignificant. Models were 

 then rerun, followed by pairwise Tukey tests to indicate 

 where significant differences occurred (Zar, 1996; SYSTAT, 

 version 7). We report results for the final ANOVA only. The 

 Yates correction was applied to 2x2 chi-squares (Zar, 1996). 

 All significance levels were set to reject Hq at P<0.05 and 

 all means are reported with standard errors. 



Results 



Bait shrimp fishermen trawled from 11 to 24 km offshore, 

 between 1.8 and 6.4 m deep water (mean=3.76 ±0.87 m). 

 Fishermen typically left port between 17:00 and 19:00 

 and spent 5.8 ±0.23 h actively trawling per night (n=50 

 nights). Trawls lasted 40.2 ±11.4 min (n=445) on average, 

 and fishermen usually set 8 to 9 trawls per night. Distance 

 trawled could not be estimated because we were unable to 

 track trawler trajectories continuously and because they 

 changed direction during the tows. The benthic habitat was 

 composed primarily of seagrass iThalassia testudinum)hut 

 also included algae, coral, and sponge. Bycatch included at 

 least 118 species of fishes, invertebrates, and marine flora. 



Catch per unit of effort 



Hippocomptis erectus was the only seahorse species com- 

 monly caught in this fishery. Almost half of the trawls (46%) 



6- 



1998 • € O » • C 



1999 O » • © O 



I 



4 



O O • 



3 

 Q. 

 U 



4 i 



y^^WY^^ 



150 160 170 180 190 200 210 220 230 



Day of year 



Figure 2 



Variation in mean CPUE estimates by lunar phase transformed to 

 inverse hyperbolic sine. The curve represents the predicted values as 

 a function of the cosine ( 1998) and sine ( 1999) of day. The lunar phases 

 for both years are indicated. Filled circles and black line are for 1998; 

 open circles and gray line are for 1999. 



caught H. erectus, and the number per trawl ranged from 

 to 16, whereas noH. reidi and only two of the much smaller 

 species, H. zosterae, were caught. In total, 916 H. erectus 

 were caught during the 95 documented fishing boat nights 

 of the two fishing seasons, resulting in an overall mean of 

 9.64 seahorses per fishing boat night. Mean CPUE for H. 

 erectus was 24.25 ±2.15 mm/{hxm), about one and a half 

 seahorses per hour per boat. If only trawls with seahorses 

 were included, CPUE was 52.52 ±3.80 mm/(hxm), or about 

 three seahorses per hour per boat (;i=205). Very high 

 CPUE was recorded on three nights: 16 July 1998 (mean 

 CPUE=122.0 ±22.5 mmAhxm), n=12 trawls), 28 June 

 1999 (mean CPUE=118.1 ±24. mm/(hxm), « = 12 trawls) 

 and 30 June 1999 (mean CPUE=154.9 ±36.6 mm/(hxm), 

 n=8 trawls). Bycatch is characterized by a high number of 

 low catches and infrequent large catches. Because the large 

 catches more likely reflect the spatial-temporal distribu- 

 tion characteristics of fish stocks rather than outliers of 

 the data (Ortiz et al. 2000), we analyzed the entire dataset 

 and then tested the robustness of our models by excluding 

 these three nights in order to assess their influence on the 

 CPUE patterns. 



CPUE of nonzero trawls varied between years and with 

 lunar phase (Table 1), but not with tidal phase (P=0.15). 

 Trawls captured significantly more seahorses in 1999 than 

 in 1998 (P<0.0005). The effect of lunar phase varied be- 

 tween years: CPUE was highest on the lunar third quarter 

 in 1998, but only weakly significant and had slightly higher 

 CPUE on the full moon in 1999 (Fig. 2). The temporal varia- 

 tion in CPUE was largely driven by the three high CPUE 

 nights, but the effect persisted when these were excluded 

 (Table 1). 



