324 



Fishery Bulletin 97(2), 1999 



among transects was, in part, proportional to the 

 mean, we removed the transect effects by scahng all 

 the observations. Each observation was multiplied 

 by the overall mean for the data set (n=64) divided 

 by the mean for all eight observations (diurnal and 

 nocturnal) of the transect. Thus the variation in the 

 data set was reduced to the scaled "within transect" 

 and diel effects. 



Following examination of the balanced treatment 

 of all transects, we examined the observations for 

 transects B5 and B6 (Table 3). These data included 

 observations from the four series as well as additional 

 nonpaired observations. The two transects were ex- 



B5-0603 



B5-()743 



65-0943 



B5-1153 



B5-I437 



B5- 1 607 



B5-1817 



Figure 4 



Echogram.'; oftran.scct B.") over 12 hours .showing dawn and 

 du.sk transition of fi.sh distribution over the cliffs. Time 

 shown represents the approximate time the vessel passed 

 over the cliffs. 



amined separately. Homogeneity of variance between 

 diel periods was examined by using an F-ratio test. 



We used nonparametric randomization tests ( Manly, 

 1991 ) to examine the sources of variance owing to con- 

 cern about the small sample sizes, non-normality, and 

 heterogeneity of variance. In particular, we hypoth- 

 esized that variance among repeated transects would 

 be proportional to the biomass estimates and, owing to 

 the extreme densities and localization of the diurnal 

 aggregations, that diurnal variance would be signifi- 

 cantly gi'eater than nocturnal variance. 



Randomization involves resampling without re- 

 placement from one parent population. The observed 

 value of the response statistic (for example, mean 

 diurnal biomass minus mean nocturnal biomass) is 

 compared with a large number of responses gener- 

 ated by treating all obsei'\'ations (diurnal and noc- 

 turnal) as coming from one parent population. Each 

 of 4999 resamplings randomly allocates the obser- 

 vations into two groups, each time simulating new 

 diurnal and nocturnal sets. The observed difference 

 in mean biomass is then compared with the 4999 

 simulated differences. If the observed difference in 

 biomass is significant (a<0. 05), then the observed dif- 

 ference should be greater than 95^^ of all the simu- 

 lated differences. 



Results 



Fish behavior and species composition 



Four aggregation types were observed on the echo- 

 grams. Each was associated with a different habi- 

 tat. The first, our targeted group, was located near 

 the cliffs over depths of 150-160 m (Fig. 3). These 

 dense schools were located near the bottom during 

 each day. then the fish dispersed vertically and hori- 

 zontally at twilight. By early nighttime, the scattered 

 targets were distributed from 80 to 150 m. The diel 

 cycle was completed by downward migration and 

 rapid reformation of schools at dawn. 



With the exception of benthic species like fiatfish, 

 which produce a negligible contribution to the acous- 

 tic measurement, the species composition in the catch 

 from this aggregation type was over SS'^^f yellowtail 

 rockfish in the three tows (Table 1), indicating that 

 this species dominated in the diurnal aggregations 

 targeted near the cliffs. The continuous transition 

 from day to night distributions (Fig. 4) implies that 

 the nocturnal aggregations were also yellowtail rock- 

 fish, but it was not possible to confirm this assump- 

 tion during the cruise. Subsequent discussions with 

 fishermen familiar with nighttime trawling in the 

 area supported the assumption that the nocturnal 



I 



