Mortality was highest during the first week (3.6%,
range 0-7%, Table 1), but decreased throughout the experiment.
Mortality was less than 2.5% in all sieves on 28 August, and
less than 0.5% on 2 and 7 September (Appendix 6). When sorting
the unattached spat, in average 80% were alive.

One of the sieves in the LL group had significantly
higher mortality than the other LL replicates, both the first
week, for the rest of the experiment and in total mortality (chi-square
test, p=0.0003, p<10^{-6} and p<10^{-6},
respectively, Appendix 6). Mortality in this sieve was as high
as 18.4% the first week (Table 1), hence this replicate was excluded
from further analysis of mortality and growth. However, mortality
in this sieve observed in the last registration on 7 September
was reduced to 1.7%, which is comparable to the other sieves.

For the first week (20 Aug), both the chi-square
goodness of fit test and a one-way ANOVA on arcsine transformed
proportions showed significant differences in mortality between
the groups (chi-square, p=0.03; ANOVA, p<0.01), which was due
to a significant higher mortality in the SL group compared to
the DD group (multiple chi-square analysis with Bonferroni correction,
p=0.003; Tukey HSD, p<0.01). Pooling the mortality from the
rest of the experiment (28 August, 2 September and 7 September,
Table 1) gave no significant differences between the groups with
neither chi-square analysis (p=0.31) nor one-way ANOVA on arcsine
transformed proportions (p=0.99), though large variation among
the experimental units lead to significant differences between
the replicates in the SL and NL groups. There was no significant
correlation between mortality the first week and mortality in
the rest of the experiment for each experimental unit (p=0.68,
r^{2}=0.02) when excluding the LL sieve with significant
higher mortality.

Table 1. Mortality (%) during the experiment and number of live spat at the end of the experiment. | ||||

*The numbers are estimated for the whole sieves from mortality observed in samples from the populations (see Appendix 6) |

Mortality pooled from all groups was size-dependent for all dates
of mortality registrations. Significantly different sizes were
found both between the dead spat and the live spat on the date
of mortality registration (t-tests, p<0.001 for all comparisons)
and also compared to the live spat from the preceding date of
size measurement (t-tests, p<0.0001 for all comparisons), except
for dead 20 Aug compared to live 13 Aug (p=0.06). Mean sizes
of dead spat were between 73 and 77% of mean sizes of live spat
throughout the experiment (Appendix 6, Table 22, see also Appendix 2). The size of dead spat increased significantly between the
three first registrations on 20 August, 28 August and 2 September
(Unequal N HSD, p 0.001), but not from 2 to 7 September (p=0.75).
There was no significant difference in size of dead spat between
the groups at any of the dates (one-way ANOVAs, 0.08<p=<0.53,
1-_{}<0.25). Since there was higher mortality among the smaller
spat and mortality varied between the groups, it was necessary
to test for the impact of this on the shell height ANOVAs. Hence,
the measurements of the dead spat from 20 August were added to
the shell height data (raw data, before adjusting the numbers)
from the same date, where mortality was highest and significant
differences in mortality were found between the groups. A new
ANOVA was then performed. The result was a decreased difference
between the groups which increased the p-value for the group effect
from 0.25 with the data for the live spat to 0.49 with the combined
data.

At the end of the experiment, the LL sieve with highest mortality
had 73% of the average number of spat for the other sieves, and
the DD group had just below 10% more spat than the other groups
(Table 1). There was a highly significant inverse correlation
(p<0.0001, R^{2}=0.86) between number of spat in the
sieves at the end of the experiment and total mortality for the
whole experiment. Adding the estimated total number of dead spat
to the number of spat in each experimental unit counted at the
end of the experiment, gave an estimated average density of 421 ±
14 (SD) spat per sieve at the beginning of the experiment (13
August).

No significant differences in shell height were found
between the four groups at any of the dates of sampling (one-way
ANOVA, 0.12<p<0.78, 1-_{}=0.80 with
=5% of the
at each date, Figure 6 and Appendix 7).

Figure 6.** **Mean shell heights of spat
of *Pecten maximus* reared under different photoperiods.
**DD** = continuous darkness, **SL** = 7 ½ h light
and 16 ½ h darkness, **NL** = simulated natural photoperiod
changing from 18h light and 6h darkness to 16h light and 8h darkness,
and **LL** = continuous light. Vertical bars indicate 95%
confidence intervals, n=180-225 (120-150 in LL).

Growth was stable throughout the experiment, but
growth was lower during the acclimation period than for the rest
of the experiment. There was a twofold increase in shell height
during the experiment (Figure 6). Regressions on untransformed
(R^{2}=0.67) and ln transformed (R^{2}=0.69) data
showed that exponential growth was slightly better fit to the
data than linear growth (Appendix 7, Table 27 and Table 28).
Regression on ln transformed data gave an instantaneous growth
rate of 0.028, which gave an average SGR of 2.8% during the experiment.
For comparison with literature, average linear growth rate throughout
the experiment was 0.14 mmday^{1}. Coefficient
of variation (CV) showed no significant regression with time (R^{2}=0.02,
p=0.22) and was in average 0.15 ± 0.015 (SD) (Figure 7).
There was no significant correlation between density and growth
(p=0.34, r^{2}=0.09).

Figure 7. Coefficient of variation for all replicate
units throughout the experiment.

The spat was observed to detach themselves from the
mesh or other substrate and crawl around by extending the foot
and pulling the shell along the mesh. They were also able to
climb the sieve walls and attach themselves with byssus threads.
50-100 spat were removed from the walls the day before each recording,
but on average 90% were returned on the days of recording. There
was no difference in number of spat on the walls between the groups
on most of the dates (one-way ANOVA, 0.09<p<0.32, 1-_{}<0.5).
The only differences were found at the last date of recording
(p<0.001) with significant differences between several of the
groups, but with no obvious pattern according to the length of
the day.

The data from the first time of sampling (4 a.m.)
were not used, as the flow from the feeding tank was disturbed
prior to sampling. Hence, the sampling represented the period
from 10 a.m. to 10 p.m., which included the shifts from light
to darkness in the SL and NL groups. There was a significant increase
in feeding rate from 10 a.m. to 4.30 p.m. (3-way nested ANOVA,
p<0.0001) but no significant changes from 4.30 p.m. to 10 p.m.
(3-way nested ANOVA, p=0.68). However, the pattern was the same
for all the groups (Figure 8), and there were no differences in
feeding rate between the groups (three-way nested ANOVA, p=0.24,
1-_{}<0.3).

Figure 8. Feeding rate as number of algae cellsindividual^{-1minute-1},
of four groups of spat of Pecten maximus reared under different
photoperiods. **DD** = continuous darkness, **SL** = 7
½ h light and 16 ½ h darkness **NL** = simulated
natural photoperiod changing from 18h light and 6h darkness to
16h light and 8h darkness, and **LL** = continuous light.
The 3 replicates are shown for each group. Vertical bars indicate
95% confidence intervals, n=3 (2). Dotted area above x-axis indicate
period of light, dark area indicate period of darkness.

The pattern of feeding rate over time varied between
the replicates (Figure 8). Though not significant, the average
feeding rate of group DD was lower than the three other groups
at all times of sampling. The least detectable difference ()
was 889 algae cellsindividual^{-1}minute^{-1} (12% of average
feeding rate), while the largest difference was

695 algae cellsindividual^{-1}minute^{-1}.
The average coefficient of variation within each group at the
different sampling times was 7.5% (range 2-22%).

Large variability was observed in the data for oxygen
consumption with sporadic deviant measurements (Appendix 9).
In four cases the values for oxygen concentration in the outlet
water were higher than for the inlet water. The SD of the replicates
in this experiment was 0.2 mgl^{1}, which is a tenfold
increase from preliminary measurements (SD=0.018 mgl^{1},
range 00.04). The data were therefore not considered as
adequate for analysis.