Stanley et al.: Diel vertical migration by Sebastes flavtdus 



329 



or diel variability in tilt angle, may simply have can- 

 celled each other out. 



We assumed that there are no diel changes in fish 

 target strength. Such changes have been observed 

 for other species and have been linked to swimming 

 behavior or mean tilt angle (Miyashita et al., 1995; 

 Buerkle and Sreedharan, 1981; Olsen, 1990; Misund, 

 1997). Target strength variation between night and 

 day may be smaller for yellowtail rockfish because 

 these fish tend to remain in the same general depths. 

 They are not moving from the bottom to surface wa- 

 ters and therefore not encountering large relative 

 changes in pressure. 



The observed diel behavior of increased aggrega- 

 tion during the day led us to hypothesize that vari- 

 ance in biomass should be greater for diurnal obser- 

 vations. We observed that yellowtail rockfish move 

 towards the cliffs at dawn. We assumed that they 

 would concentrate further along that narrow band 

 to produce discrete schools. We viewed the noctur- 

 nal distribution as a relatively dispersed band (three- 

 dimensional cloud) of individuals in the general vi- 

 cinity of the cliffs. We assumed that the diurnal dis- 

 tribution, which lay along the orientation of the cliff, 

 would be a much narrower string of individual 

 schools and hence of highly varying density, perhaps 

 to the extent of approximating a "beaded pattern" of 

 distinct schools following the length of the cliff We 

 expected that diurnal estimates derived from 

 transects that pass perpendicular to the cliffs would 

 be highly variable because the path of the acoustic 

 beam could range from "missing entirely" to "com- 

 pletely ensonifying" a dense school. 



We found that, for the short transect version of 

 the all-transect data, the variance of diurnal biomass 

 estimates was not significantly greater than that of 

 nocturnal biomass estimates. Diurnal variance was 

 significantly higher in the expanded set of the B5- 

 B6 set of observations but only for the long version 

 of transect B6. 



The weak or mixed indication of higher diurnal 

 variance implies that the apparent aggregation to- 

 wards the cliffs, evident in the echograms, is not 

 matched to the same degree by aggregation along 

 the axis of the cliffs. We expected discrete diurnal 

 schools following the axis of the cliffs. This tendency 

 to aggi-egate into a continuous band offish along the 

 cliffs, as opposed to discrete schools, may be unique 

 to this area where there is a longitudinal topographic 

 feature along the preferred depth. In areas of the 

 coast lacking in such a linear feature, fish may tend 

 to aggregate over a specific point, such as a pinnacle, 

 producing greater daytime variance among estimates 

 of biomass. Results of this experiment, however, do 

 not support the hypothesis that estimates from the 



dispersed condition will show less variance for yel- 

 lowtail rockfish. The similarity in variance between 

 nocturnal and diurnal periods indicates that no sub- 

 stantial gains in efficiency or precision can be achieved 

 by sampling during one or other of the diel periods. 



The among-transect variance overwhelms the 

 other sources of variance even within a small coastal 

 area. From the perspective of two-stage sampling 

 (among-transect or within-transect variance), preci- 

 sion of the overall estimator is reduced by allocating 

 sampling effort in proportion to variance contributed 

 by each stage. Because among-transect variance is 

 much greater, survey design should maximize the 

 number of transects at the expense of replicating 

 transects. The only exception to this principle would 

 arise when the cost of replicate samples is much lower. 

 Because the cost of collecting a replicate transect esti- 

 mate is almost equal to that for an additional transect 

 over the scale we are considering, it would be more ef- 

 ficient to maximize the number of different transects. 



By repeating the survey eight times, we were able 

 to determine the variance of the biomass estimate 

 directly rather than by inferring it indirectly through 

 geospatial analysis of "within" variance (Petitgas, 

 1993 ). This calculation of "among" survey variance fol- 

 lows fi-om work by Williamson ( 1982), who resampled 

 from the individual samples averaged over one 

 minute within a transect, and by Robotham and Cas- 

 tillo (1987), who bootstrapped cumulative transect 

 observations as we did. There is no question that a 

 formal investigation of the spatial impact on vari- 

 ance would require treatment at the granular level 

 of the observations, which the geospatial methods 

 provide. These procedures also facilitate an investi- 

 gation of the impact of additional explanatory vari- 

 ables and development of model-based inference. 

 However, the requirements of our study were real- 

 ized by the simpler transect-based analysis. 



Total biomass estimate and survey design 



The overall biomass estimate of 1152 t for the study 

 area appears consistent with overall coastal bio- 

 mass estimates of 50,000-60,000 t for a stock that is 

 assumed to extend from the study area to the border 

 of B.C. and Alaska (Stanley 1993). The fact that the 

 overall coefficient of variation among the eight se- 

 ries for the study area was under 14% is encourag- 

 ing and better than that predicted from the formula 

 of Aglen (1983). This precision may not, however, 

 apply to all rockfish species. Wilkins (1986) found 

 that the CV for widow rockfish was 2-3 times higher 

 in spite of a more comprehensive estimation proce- 

 dure that included the use of side-scan sonar in con- 

 junction with echosounding. We can also expect 



