642 



Fishery Bulletin 91(4), 1993 



and within-reader) associated with the school size 

 estimates. 



One concern was that during the 12-week "counting 

 period," readers might be inconsistent in the application 

 of criteria (cognitive or physical methods, or both) used 

 for counting. If readers were inconsistent, then this 

 might affect the accuracy and precision of the esti- 

 mates. A two-step approach was taken to address this 

 concern. First, temporal trends in the deviation of 

 counts from the means (see Sokal and Rohlf, 1981, 

 p. 50) were evaluated by linear regression. This ap- 

 proach assumed that the mean of three reader counts 

 was close to the true school size and that the sample 

 error was normally distributed. If individual reader 

 counts were not normally distributed, then this might 

 indicate reader bias or a change in criteria. Second, 

 readers did repeated-counts (at four time points dur- 

 ing the counting period) of a known sample of photo- 

 graphed schools. This was done to evaluate the consis- 

 tency of individual readers (within-reader precision) 

 and to test the hypothesis that dolphin counts (from 

 the same photographs) done at the beginning, middle, 

 and end of the counting period were independent of 

 temporal effects. If the hypothesis proved true, then 

 this would support the idea that counting criteria were 

 constant throughout the counting period. 



Materials and methods 



At sea 



Aerial photographs analyzed in this report were taken 

 with Chicago Aerial Industries KA-62 aerial cameras 

 mounted vertically on a Hughes 500-D helicopter. The 

 helicopter was stationed aboard the National Oceanic 

 and Atmospheric Administration (NOAA) survey ship 

 David Starr Jordan. All cameras had forward-motion 

 compensation (to minimize photo image blur from air- 

 craft movement) and a 76.2-mm lens. Large-format 

 (126-mm) Kodak Aerochrome MS 2448 color film was 

 used. In order to minimize the behavioral response 

 (scattering and deep diving) of the dolphin schools to 

 the helicopter, photographs were taken at 244 m ( 800 ft) 

 altitude. The scale of the photographs at this altitude 

 was 1:3200, the sea surface area in a photograph frame 

 was 366 m 2 , and a 2-m dolphin measured 0.63 mm on 

 the film. The camera cycle rate was programmed to 

 expose for approximately 80% film image overlap, i.e., 

 807f of the area photographed in one frame was photo- 

 graphed again in the next successive frame. Succes- 

 sive exposed photograph frames for a school were re- 

 corded as a complete "photo-pass." To enhance the 

 probability of photographing an entire school, multiple 

 photo-passes (avg.=5) were made over a school. 



In the laboratory 



Counts Light tables equipped with dissection micro- 

 scopes (0.7x to 7x variable objective and 10 x wide- 

 field oculars) were used to view and count the dolphin 

 images during counting. Dolphins were counted by 

 hand-tally while being plotted with a permanent 

 marker on a clear acetate overlay. The marked over- 

 lay, when moved to the image overlap area of adjacent 

 frames and aligned over dolphins that were previously 

 plotted, made it easier to identify those dolphins not 

 yet counted in the photo-pass. The photo-pass was the 

 unit on which school size estimates were based (Scott 

 et al, 1985); each pass was counted independently by 

 three readers. 



Criteria for selecting the "true" school size For each 

 of the 48 photographed schools, the readers chose (by 

 group consensus) the one photo-pass where the mean 

 of the reader counts was the best estimate of true 

 school size. This decision was based on the precision of 

 the three replicate counts and the reader assigned 

 "quality ratings" for the photo-pass. Ratings reflected 

 how confident the reader was in the accuracy of the 

 count. For each photo-pass, readers independently as- 

 signed quality ratings ranging from 1 to 4. A rating of 

 "1" indicated that a photo-pass had "excellent" quality. 

 A rating of "2" indicated that the count was "good" 

 despite the presence of some questionable images (i.e., 

 it was difficult to discern and count dolphins accu- 

 rately when images were partially obscured by light- 

 glare or when photographic resolution was reduced for 

 deep swimming dolphins because of loss of light with 

 sea depth). A photo-pass was rated "3" or "fair" when 

 more questionable images were encountered, but read- 

 ers still believed the count was a close approximation 

 of true school size. A photo-pass rated "4" was deemed 

 unusable for size estimation because the reader felt 

 there were too many questionable images and the count 

 was not a reliable estimate of true school size. 



One source of between-reader variation in dolphin 

 photograph counts was reader error, where dolphins 

 were missed or counted twice. Variation also occurred 

 because readers differed in their interpretations of 

 whether questionable images were dolphins or merely 

 background water turbulence in the photographs. 

 Based on the premise that "precision leads to accu- 

 racy" (Sokal and Rohlf, 1981), for each photo-pass, the 

 CV of the three independent counts was used to moni- 

 tor the reliability of the school size estimate. For photo- 

 passes where the CV of the counts exceeded 10%, the 

 dolphin school was re-counted (independently) to see if 

 the precision of the estimate could be improved. If the 

 CV was above 15% after a second count, the photo- 

 pass was excluded from the study because the counts 



