Harvey et al.: Comparison of estimates of reef-fish lengths made by divers and a stereo-video system 
67 
40% T 
35% 
30% 
25% 
20 % 
15% 
1 0 % ~f 
j 
5% + 
0 % - 
Xx 
.$■ | 6 if 
x Experienced 
o Novice 
o Video 
0 10 20 30 40 
True length (cm) 
—4 1 
50 80 
Figure 2 
Coefficients of variation between divers or operators 
of length estimates for experienced scientific divers, 
novice divers, and a stereo-video system. 
For all four measures of error, there was no significant 
interaction between software operators and silhouette, 
but there was a significant operator effect (Fig. 4). This 
finding suggests that the differences between operators 
were consistent across silhouettes. Although the differ- 
ences between the operators were statistically significant, 
they were small compared with the differences between 
divers and the stereo-video system (cf. Figs. 3 and 4). 
(F 9 1fin =6.0, P=0.G03 [E j ; F 9 ian =14.6, P<G.001 [AE];F, lsn = 
3.6’, P=0.029 [RE]; F 2 180 =10.2, PcO.001 [RAE]). 
There is little difference in the accuracy of measure- 
ments made under poor water clarity (mean E=-0.99 cm) 
compared with those made under good water clarity (mean 
E--0.6 cm). The RAE is more consistent across silhouettes 
for good water clarity than it is for poor clarity. For the 
larger silhouettes, the RAE is lower under poor water clar- 
ity than under good water clarity (Fig. 5). This is probably 
a result of lower contrast, permitting more accurate point- 
ing to the tail and snout of each silhouette (discussed fur- 
ther in the next section). 
For all four measures of error, there was no significant 
interaction between years and silhouette, but highly sig- 
nificant differences (PcO.QOl in all cases, F 1 121 =15.0 [E] ; 
F 2 180 =27.1 [AE] ; F 2 im =14.6 [RE]; F 2180 =25.8 [RAE]) be- 
tween years (Fig. 6). These results suggest that with in- 
creasing operator experience the measurement error can 
improve. This is probably a consequence of the operator 
learning to distinguish and select the edges of objects of 
interest accurately. 
One of the advantages of stereo-video measurements 
over diver visual estimates is that the image is stored and 
it is possible to make numerous repeat measurements of 
the same object in the laboratory. Taking the mean of this 
set of measurements would result in greater accuracy and 
precision than using a single measurement. Improvement 
of the accuracy of estimates can be seen by comparing the 
mean error, for each of the three operators, as the number 
4T 
~ 2 
E 
it 
2 0 
LU 
GBRMPA 
LU 
<D 
XI 
< 
T In 
GBRMPA l 
10%) 
5% 
0% 
-5% 
- 10 % 
-15% 
I 
GBRMPA 
25%] 
2 20 %; 
LU 
1 1 5%t 
| 
< 10 % 
<D 
> 
is 5%' 
a> 
cc 
0 % 
GBRMPA £ 
s V 
Figure 3 
Means and 95% confidence intervals for each measure 
of error, for each method of estimating fish length. E = 
experienced divers; N = novice divers; V = stereo-video 
system, GBPRMA= Great Barrier Reef Marine Park 
Authority data. 
of repeat measurements is increased from 1 to 10 (Fig. 7). 
There is improvement in the error for both inexperienced 
operators and the experienced operator, even after ten re- 
peat measurements (Fig. 7). 
Discussion 
Our results highlight the differences in the accuracy and 
precision of length estimates of silhouettes of reef fish 
made by novice and experienced scientific divers in com- 
parison with those produced by a stereo-video system. Our 
length estimates were made under ideal conditions where 
the plastic models were fixed in position. Under real field 
conditions fish move, occur at different distances from the 
divers and have behavioral and morphological differences 
that can influence length estimates. Consequently, the 
measures of error presented in our study can most likely 
be considered a best case scenario. As with Darwall and 
Dulvy (1996), the results obtained by novice divers were 
similar to, but slightly less accurate than, those obtained 
by experienced scientific divers (mean errors of 2.3 cm and 
-2.1 cm respectively). Our results also demonstrate that 
significant improvements in accuracy can be obtained by 
using a stereo-video system (mean error -0.6 cm). Similar 
improvements in precision were also recorded: mean CVs 
