342 



Fishery Bulletin 100(2) 



vertebrates from each bycatch sample or subsample were 

 identified; finfish that could not be identified onboard were 

 labeled and returned to the laboratory for identification. 

 All individuals of each finfish species were counted and 

 the finfish bycatch sample or subsample was weighed. To 

 obtain an estimate of the size-frequency distribution for 

 each species of finfish, we measured the standard length 

 (SL) to the nearest 1 mm of 20 randomly selected individu- 

 als of each species from each tow and combined the mea- 

 surements from the 20 replicate tows. If fewer than 20 in- 

 dividuals were caught in a tow, all individuals captured in 

 that tow were measured. 



All weights were standardized to grams per minute 

 towed to estimate CPUE (bioniass). All counts of individu- 

 al species were standardized to number per minute towed 

 (NPUE). 



ysis of variance (ANOVA). We then used the least-squares 

 difference (LSD) post hoc test to locate the significant dif- 

 ferences. Differences between the net with the BRD and 

 its paired control net in the size-frequency distribution 

 of the 10 most abundant fish species were assessed by 

 using the Kolmogorov-Smirnov two-sample test. To deter- 

 mine the percent reduction or increase in the biomass and 

 number of each of the top ten finfish species, we compared 

 the BRD-equipped nets with the control nets by using the 

 untransformed mean CPUE and NPUE data for shrimp 

 and finfish and the total number of individuals subsam- 

 pled. Percent reduction for either CPUE or NPUE was 

 then calculated (from Rogers et al., 1997) as 



Percent difference = 



Statistical analyses 



Statistical analyses followed Sokal and Rohlf ( 1995). Para- 

 metric statistics were applied when the data conformed 

 to the parametric assumptions of normality (Shapiro-Wilk 

 test) and homogeneity of variances (Levene's test). Vari- 

 ables that did not conform to parametric assumptions 

 were transformed to log (biomass or number) + l. Non- 

 parametric statistics were employed only after appropri- 

 ate methods were deemed unsuccessful in transforming 

 the data to meet parametric assumptions. Both paramet- 

 ric and nonparametric statistical analyses were completed 

 by using the STATISTICA software package (Statsoft Inc, 

 1999). Using ^tests, we evaluated the performance of the 

 paired nets prior to the addition of the BRDs and com- 

 pared the catchability of the BRD-equipped net to its con- 

 trol. Because we used a paired-tow design for field testing, 

 we analyzed each net size and type of BRD separately: 

 net sizes and BRDs were not directly compared with each 

 other but were always compared with the controls. The 

 ability of a BRD-equipped trawl to retain shrimp while 

 reducing bycatch was assessed by comparing the CPUE 

 (biomass) and NPUE of finfish and shrimp and by compar- 

 ing the CPUE and the NPUE of the 10 most abundant 

 finfish species in the BRD-equipped net with CPUE and 

 NPUE data for its paired control net. The CPUE and 

 NPUE of shrimp caught, calculated as described above, 

 were based on actual weight and numbers of shrimp 

 caught in each trawl. Wlien the bycatch was subsampled, 

 the finfish biomass or number was estimated using the 

 formula 



Finfish biomass or number = 



Finfish subsample CPUE or NPUE 

 xTotal bvcatch weight 



Subsample weight 



Because our sampling period ranged over two seasons, we 

 considered the interactive effects of season and net type 

 (BRD-equipped or control) for each net size by using anal- 



( CPUE or NPUE of BRD net 



- CPUE or NPUE of control net) x 100 



CPUE or NPUE of con trol net 



Results 



No significant differences were found in total weight of the 

 finfish or shrimp catch between nets of equal size prior 

 to the addition of the BRDs. Similarly, the total weight of 

 finfish or shrimp was not significantly affected by trawl 

 position. The standardized mean ratio of finfish bycatch 

 to shrimp biomass for all control net sizes combined was 

 5.3:1 (range 2.9:1-11.3:1), The standardized mean ratio for 

 the BRD-equipped nets (3.8:1; range 2,5:1—4.9:1) was not 

 significantly different but was substantially lower than 

 that of the control nets. 



CPUE and NPUE 



In contrast to results with the control nets, there were 

 no significant differences in either biomass or number of 

 shrimp captured in the 17-m net or the 20-m net equipped 

 with either BRD (Table 1). In winter, the biomass and the 

 number of shrimp captured in the 14-m net equipped with 

 either BRD were significantly lower than these quantities 

 captured in the corresponding control net iFFE: P=0.025; 

 EMF: P=0.008). On the contrary, both the biomass and 

 number of finfish were significantly and notably lower in 

 most of the BRD-equipped nets than they were in the con- 

 trol nets (for significant differences, P range for the FFE= 

 0.025-0.001 and P range for the EMF= 0,027-<0.001 ). 

 The only exception was in the number of finfish caught in 

 winter by nets equipped with either BRD. 



Significant seasonal differences always occurred in 

 shrimp CPUE (FEE: P<0.001 for all tests; EMF P range: 

 0.003-<0.001 ) and nearly always occurred in shrimp NPUE 

 (FEE P range: 0.003-<6.001; EMF P range: 0.002-0.001; 

 exception: NPUE for the 17-m EMF-equipped net) and ac- 

 counted for most of the variation in CPUE obsei'ved for 

 each net size. Nearly all significant differences in shrimp 

 CPUE and NPUE between seasons were due to a larger 

 catch of shrimp in winter The only significant interactive 



