64 
Fishery Bulletin 99(1 ) 
Two important sources of error affect the accuracy and 
precision of visual length estimates: 
1 The estimation of the length of reef fish underwater 
is complicated by the air-water interface in the diver’s 
mask which causes objects to be magnified in size by a 
factor of 1.3 and to appear to be closer to the observer 
than they actually are; 
2 Researchers using SCUBA have been shown to be inef- 
ficient when performance underwater is compared with 
similar activities in air (Hollien and Rothman, 1975). 
The accuracy and precision of diver’s estimates of reef- 
fish length is probably affected by the detrimental phys- 
iological effects related to SCUBA diving (Baddeley, 
1965; Baddeley et ah, 1968; Baddeley, 1971). 
The level of precision and accuracy associated with vi- 
sual length estimates influences comparisons of data over 
different temporal or spatial scales in two distinct ways. 
First, bias in the estimates can make the results of the 
analysis less reliable. Second, any lack of precision in the 
estimates arising from both sampling error and measure- 
ment error tends to reduce the power of the statistical 
analysis. The Great Barrier Reef Marine Park Authority 
(GBRMPA, 1979), Bell et al. (1985), and Polunin and Rob- 
erts (1993) have performed calibration studies in order to 
quantify measurement error and have concluded that such 
error is relatively small and can be ignored. However, they 
did not directly assess the extent to which such error reduc- 
es the precision of a statistical analysis. Harvey and Shortis 
(1996) demonstrated that subjectivity in visual length es- 
timates can be overcome, and the accuracy and precision 
of length estimates enhanced, by using a simple and rela- 
tively inexpensive underwater stereo-video system. 
The objectives of our study were 1) to compare the ac- 
curacy and precision of stereo-video measurements of reef- 
fish length with visual estimates made by novice and ex- 
perienced scientific SCUBA divers; 2) to evaluate the ef- 
fect of operator training and experience on the precision 
and accuracy of length measurements made with a stereo- 
video system; and 3) to assess the effect of water clarity on 
the accuracy and precision of stereo-video measurements. 
Methods of optimizing the accuracy and precision of ste- 
reo-video length estimates are discussed and the accura- 
cy and precision of experienced divers’ estimates are com- 
pared to one other study. 
Materials and methods 
Comparison of length estimates made by novice 
and experienced scientific divers with estimates 
generated by a stereo-video system 
The accuracy and precision of length estimates were tested 
by a simple procedure that is used routinely for calibrat- 
ing diver estimates of the lengths of reef fish (GBRMPA, 
1979; Bell et al., 1985; English et al., 1994). Typically, poly- 
vinyl chloride (PVC) sticks or silhouettes of fish are placed 
in the water and their lengths estimated. The accuracy 
of the diver estimate is then assessed from the difference 
between the real size and the estimate. In this study, 
16 plastic silhouettes of fish, ranging from 10 to 49 cm 
in length, were placed at distances of between 3.0 and 
6.6 meters from a transect rope. Each diver moved along 
the transect rope estimating the length of each of the 
sixteen silhouettes when they were opposite them. Each 
diver repeated this process five times, i.e. five transects 
each with estimates of the lengths of the sixteen silhou- 
ettes. Five transects were completed with the stereo-video 
system in the same manner. The distances from the tran- 
sect rope and the order of the individual silhouettes were 
kept constant throughout the trial. All silhouettes were 
weighted in a way that their length orientation remained 
perpendicular to the transect. Therefore, the error mea- 
sures presented here do not take into consideration that 
in the field the divers may have to make length estimates 
where orientation of the live fish in relation to the diver 
changes. 
Novice divers Novice divers were defined as experienced 
SCUBA divers who had made few, if any, estimates of the 
lengths of reef fish underwater. Eight novice divers made 
length estimates in a saltwater pool at the Portobello 
Marine Laboratory (PML) between May 1994 and Janu- 
ary 1995. No more than two transects were completed on 
any one day, except one diver who completed four consec- 
utive transects in one day. Between transects and dives, 
data were not available to divers in order to avoid memo- 
rization of previous estimates or of silhouette lengths. The 
novice divers made only 594 of a possible 640 length esti- 
mates because 46 silhouettes (7%) were not recorded. The 
majority of the missing estimates were from the smaller 
silhouettes (approximately 10 cm long) or from a silhou- 
ette placed farthest from the transect (6.6 meters). 
Experienced scientific divers Experienced scientific divers 
were considered active marine scientists who had been, 
or who were currently involved in research that required 
them to make estimates of reef fish length. Six experienced 
scientific divers estimated the length of the plastic silhou- 
ettes between October 1994 and June 1996. Three of the 
divers made their estimates in a saltwater pool at PML, 
whereas the other three made their estimates in fresh- 
water swimming pools elsewhere. A total of 480 length 
estimates were made, and all silhouettes were recorded. 
Owing to time constraints, all of the length estimates were 
made on consecutive transects during one dive. 
Stereo-video measurements For interested readers a 
comprehensive description of the design and calibration of 
the stereo-video can be found in Harvey and Shortis ( 1996, 
1998) and will not be described here. 
Stereo-video length estimates were made in the same 
way as those made by the divers; the stereo-video system 
recorded the silhouette as it was moved along the tran- 
sect line by a diver. Measurements were made in the Uni- 
versity of Melbourne swimming pool in July 1994. 80 sil- 
houettes were recorded (16 silhouettesx5 transects). Four 
sets of images were rejected because the orientation of the 
