Rochet et al Precision and accuracy of fish length measurements obtained with two visual underwater methods 



tion of the measurement devices, 2) differences among 

 observers, 3) orientation and position offish in relation 

 to the camera, and 4) swimming motion. We investigated 

 each of these featuress. 



Among the two methods tested during this study, 

 the METRAU method performed poorly. It has several 

 disadvantages. First, METRAU system needs the fish 

 brought into focus when it is perpendicular to the field 

 of vision. This may take time during which the fish can 

 escape. Second, the registered video images do not in- 

 clude the superimposed calculated scales. Thus, unlike 

 the laser method, it is not possible to replay the video to 

 identify the best image. Postoperational measurements 

 can be performed only by using the digital snapshots 

 registered in real time. Third, this method had a higher 

 variance than the laser method, probably because of the 

 technical constraints just mentioned. Fourth, during the 

 VITAL cruise the estimates were systematically biased 

 downwards. Measurements carried out after the cruise 

 in a laboratory pool confirmed this systematic underes- 

 timation to be -20% of the real sizes. This result may 

 be due to errors in the software which processes the 

 output of the camera. Although the errors could prob- 

 ably be fixed, and the hardware improved to generate 

 a video signal with the overlaid scale for recording, the 

 other disadvantages are more difficult to eliminate, be- 



cause of the intrinsic limitations of the system. Hence 

 the method is not promising for estimating the size of 

 fish in the wild. 



By contrast, the laser beam method performed rath- 

 er well, at least for rigid objects. We obtained CVs of 

 7-13% for rigid objects in real time and 1-4% with 

 image postprocessing (Table 1), both of which compare 

 well with CVs for silhouette measurements obtained 

 with a single camera placed in a laboratory pool (with 

 scale bars placed on the bottom of the pool) and with 

 computer image processing (1%, Harvey et al., 2002b), 

 or even with stereo-video measurements of silhouettes 

 (0.6-7.5%, Harvey and Shortis, 1996). Length mea- 

 surements were always unbiased and postoperational 

 measurements on video images reached a high precision 

 for rigid objects and for small- to medium-size mobile 

 objects. Thus the method seems suited for measuring 

 the size of animals of low mobility, like invertebrates, 

 along visual observation transects. 



The variance due to differences between observers 

 was about 20% of the residual variance. This variance 

 was reduced to 16% when measurements were per- 

 formed by trained observers. This is true for real-time 

 measurements. For video-replays of the laser-beam data, 

 the variance due to observers was very small because of 

 the use of a ruler instead of subjective extrapolation of 



