Fishery Bulletin 104(1) 



Table 3 



Estimated fixed-effect coefficients and standard devia- 

 tions for model 1 for the measurements of rigid and mobile 

 objects by eight independent observers using the laser 

 method. SE = standard error; CL =confidence limit. 



oio) 0.296 1.126 4.435 



a 4.369 5.535 7.014 



On,jL,,,^lo,„oH.iM.nMc'' 0-523 0.710 0.964 



^ Standard deviations were estimated in relation to the standard 

 deviation for mobile objects. 



unbiased, whereas the METRAU-based measurements 

 underestimated the true length of objects by as much 

 as 7 cm for real-time measurements and 10 cm for time- 

 delayed measurements (Table 4, Fig. 4). In addition, the 

 variance of METRAU measurements was systemati- 

 cally larger than the corresponding laser measurements 

 (Table 5: ratios larger than 1). For rigid objects, laser- 

 based video-replays had a lower variance than real time 

 measurements. This lower variance for postoperational 

 measurements was due to the allowance of videos to be 

 replayed as many times as necessary in order to select 

 the best image where an object was perpendicular to 

 the optical axis. Use of a ruler also improves the mea- 

 surement. The high estimation variance obtained for 

 mobile objects measured with the laser method on digi- 

 tal snapshots was partially due to one outlier (Fig. 4B). 

 The object was measured at a relatively great distance 

 in somewhat turbid water. The outlier was not removed 

 because these kinds of errors are to be expected under 

 common measurement conditions in the field. Generally 

 the most precise results were obtained for video-replays 

 with the laser beam method. 



Precision of fish measurements 



The variance of the random effect for B. dtibius was 

 about 66% of the variance estimated for L. eques. For 

 live fish, the standard deviation of the observer random 

 effect was approximately 16% of the residual standard 

 deviation (Table 6). This standard deviation is lower 

 than that obtained for objects of known size because 

 the residual variance was larger owing to the small 

 number of repeated measurements obtained for each fish. 

 It was not easy to repeatedly measure fish because of 

 escapement behavior. In addition, only trained observers 

 took part in this experiment, which reduced observer 

 variability. 



Discussion 



Table 5 



Estimates of the standard deviations of length mea- 

 surements for rigid and mobile objects obtained by five 

 variants of the two methods, from model 2. Number of 

 measurements are given in parentheses. Estimates sig- 

 nificantly different from 1 are in bold font. 



All standard deviations are relative to the standard deviation 

 for laser measurements of mobile objects. 



The potential sources of errors and variability in visual 

 fish length measurements are 1) the design and calibra- 



Table 6 



Estimates and 9b''i confidence limits for the standard 

 deviations of the components of model 3 for fish size mea- 

 surements obtained for two species by five independent 

 observers using the laser method. 



Standard deviation 



Lower 



Esti- 

 mate 



Upper 



aio) 

 o 



'^B.dubius^^Bdub,us''l "L.eques'^L.eques^ 



0.068 

 1.096 



0.41 



0.278 1.130 

 1.702 2.642 

 0.66 1.06 



Standard deviations provided in relation to the standard devia- 

 tion (or Lepuiion measurements. 



