568 
Fishery Bulletin 99(4) 
the ratio would be if water were perfectly mixed in the 
whole-water-column. If the water were perfectly mixed, 
fish eggs would be distributed randomly and the catch 
ratio (between CalVET and CUFES) would be a constant, 
because egg count would be proportional to the volume of 
water filtered through the two samplers. A CUFES filters 
on the average 0.64 m 3 of water/min and the CalVET fil- 
ters 3.5 m 3 of water. Therefore under perfect mixing, the 
ratio of eggs/min to eggs/tow = 0.64 m 3 /3.5 m 3 = 0.18 for 
the daytime when fish schools are in deeper water, assum- 
ing the vertical distribution of sardine eggs was similar 
to that of anchovy eggs (Moser and Pommeranz, 1999). At 
night, when fish schools are in the upper 50 meters, 4 the 
ratio would be 0.64 m 3 /2.5m 3 = 0.26. This means that on 
the average, for four to six eggs seen in a CalVET catch, we 
would expect one egg/min in the CUFES. The fact that the 
estimated ratio for two years were 0.76 in 1996 and 0.25 
in 1997, indicates that more sardine eggs appeared in the 
upper 3 m than would be expected under perfect mixing 
and less mixing occurred in 1996 than in 1997. Clearly, 
if all the eggs were in the upper 3 m of water column, 
the ratio would be 1:1 for the same surface area. As the 
extent of vertical mixing in the sea is highly variable, we 
believe that calibration tows are always needed, even if 
the CUFES is used only as an index of egg abundance. Ver- 
tical egg mixing models might eventually help to reduce 
calibration requirements. 
Identification and counting of eggs at sea 
Identification and counting of fish eggs while the ship is 
underway was an essential ingredient of our adaptive allo- 
cation design. The eggs of some commercially important 
clupeid species are often difficult to distinguish from those 
of co-occurring species (Ahlstrom and Moser, 1980; Mata- 
rese and Sandknop, 1984; Watson and Sandknop, 1996). 
Further, many melanostomiin species have eggs with char- 
acteristics (e.g. large diameter, wide perivitelline space, 
segmented yolk) that are similar to those of co-occurring 
sardines and other clupeid eggs. During a DEPM survey 
off Oregon, Bentley et al. (1996) encountered a type of 
melanostomiin egg similar to the egg of Pacific sardine 
at 24 of 46 stations. Our experience has shown that the 
risk of misidentification increases when identifications are 
made at sea; during our CUFES and DEPM survey in 
1997, about 1% of the eggs initially identified as sardines 
turned out to be melanostomiin eggs after examination in 
the laboratory. Fortunately, melanostomiin eggs were in 
such low abundance that they had no effect on the criti- 
cal values of density. The possibility of misidentification 
differs depending on season, location, and target species. 
Clearly all shipboard positive records for areas that lie 
outside of the known spawning range and season of the 
4 Castillo Valderrama, P. R. 1995. Distribucion de los princi- 
pales recursos pelagicos durante los veranlos de 1992 a 1994. 
Instituo Del Mar Del Peru, Informe No 114. Instituto del Mar 
delPeru (IMARPE), Esquina Gamarra Y General Valle Chu- 
cuito, Callao Peru, 24 p. 
target species should be checked in the laboratory after 
the cruise. Owing to the great abundance of sardines and 
anchovy eggs, the effect of misidentification on biomass 
estimation is probably trivial if the survey is conducted 
during peak spawning months. 
Egg counts on shipboard were somewhat lower than 
shoreside counts. Even though the effect of the difference 
is negligible from the standpoint of biomass estimation, we 
recommend that shoreside measurement be maintained. 
In particular, for collections that contain a large amount 
of other organisms (e.g. salps) which makes it difficult 
to count the eggs on aboard the ship, shoreside counting 
would be necessary. 
Stratified design 
An important feature of the survey design was the strati- 
fication of sampling by egg density and sampler type. In 
the high-density stratum, we used only staged eggs from 
CalVET samples to estimate 2 and P 0 , whereas in the low- 
density stratum we collected only CUFES samples. We 
believe it would not be useful to use staged eggs from 
CUFES samples in the low-density stratum to estimate z 
and P 0 directly because of the low egg abundance, possibil- 
ity of stage-specific bias, and lack of yolksac larvae (the 
CUFES does not sample yolksac larvae well). It seemed 
preferable to use, for the low-density stratum, the esti- 
mate of P 0 for the high-density stratum, adjusted by the 
ratio of egg densities taken in CUFES at high and low 
strata. This, of course, requires the assumption that egg 
mortality did not differ between strata. A direct test of 
this assumption is impractical because of the large sam- 
pling effort needed at low density to obtain a sufficient 
number of positive samples to detect a difference in mor- 
tality rates. Fortunately, the effect of this potential bias is 
diminished because the low-density stratum contributes 
fewer eggs. In our example, the low-density stratum con- 
tributed about 25% 5 of the daily production. An alterna- 
tive approach would be to allocate CalVET sampling to the 
low-density stratum. We believe this approach would not 
be cost effective because the number of positive CalVETs 
would be so low. 
Another important element of the stratified design was 
the use of transect lines as the sampling unit. Model-based 
geostatistics are needed (Fletcher and Sumner, 1999) for 
data from continuous samplers such as echo-sounders and 
CUFESs, unless sampling units are defined such that da- 
ta are uncorrelated among sampling units in the survey 
design (Armstrong et al., 1988). Because conventional de- 
sign-based statistical procedures are easier to apply, we 
preferred using a transect line as our sampling unit, which 
requires that the minimum sample size allows a between- 
transect distance greater than the diameter of the egg 
patch. Fortunately the within-transect CUFES collections 
provided the information needed on the spatial structure to 
determine the distance of CalVET lines to insure samples 
are uncorrelated. In our case, tows, a minimum of 22.2 km 
5 25% =100 x (1.064x107, 255)/(2.57xl74, 096) (Table 4). 
