ResuKs 



provided different looks so that 

 each image was used only once. 



To isolate individual targets 

 and at the same time give the 

 SLAR interpreters a surrounding 

 piece of film to examine for 

 background, each target was cut 

 from a duplicate film and mounted 

 on a 2 1/4" photo slide mount. 

 Each target was randomly 

 assigned a 2-3 digit identification 

 number and each slide mount 

 was labelled with that number, the 

 lateral range to the target from the 

 aircraft, and the sea conditions 

 (from ship ground truth). 



These 74 slides (35 

 icebergs and 39 ships) were 

 taken to U. S. Coast Guard Air 

 Station Elizabeth City, North 

 Carolina, for viewing by the Coast 

 Guard Avionics Technicians who 

 are the HP's SLAR interpreters, 

 operators and technicians during 

 ice reconnaissance flights. Four 

 experienced technicians 

 separately viewed the slides on a 

 light table using an optical 

 magnifier, conditions 

 approximating the normal IIP post- 

 flight analysis. Each technician 

 was asked to identify each target 

 as either a ship or an iceberg. 



Table B-1. Target Identification 



Table B-1 presents the raw 

 test results, divided into the two 

 target types: ships and icebergs. 

 The "correct" column under each 

 target type represents the 

 number of times each observer 

 identified that target correctly, 

 while "incorrect" represents the 

 number of times that type of 

 target was misidentified. 



The data was subjected to 

 Chi-square analysis (Lapin, 1975) 

 to identify statistically significant 

 differences in the error rates 

 between the observers, and to 

 look for differences in how the 

 two target types were treated. 

 The analysis revealed that there 

 was too much difference in error 

 rate and target treatment 

 between the four observers to 

 allow combining all the data. Also, 

 obsen/ers 1 through 3 showed a 

 bias toward icebergs, i.e., a 

 tendency to identify ships as 

 icebergs. This is a reflection of 

 their IIP experience, since 

 observers are taught to be 

 consen/ative and identify 

 doubtful targets as icebergs. 

 Observers 1 and 3 were 

 sufficiently similar in their 

 treatment of the targets to allow 

 combining their data. Finally, 



observer 4 showed no bias 

 toward icebergs. 



The results from observers 

 1 and 3 probably offer the most 

 representative sample, since the 

 bias they show toward icebergs 

 reflects their IIP experience. 

 Actually, while selection of 

 different subsets of the data can 

 be made based on bias shown or 

 statistical judgements, the error 

 rate for all targets is in the range of 

 40-45%, as shown in Table B-2. 



While these data sets 

 cannot be combined or compared 

 for statistical reasons, selecting 

 any one of them yields essentially 

 the same result, i.e., that the 

 observers correctly identified all 

 targets 55-60% of the time. 

 Applied directly to all IIP SLAR 

 detections, a possible 45% error 

 rate would have alarming 

 implications. The targets used in 

 this study, however, represent 

 only a subset of IIP SLAR targets. 

 There are characteristics that limit 

 the size of that subset and 

 mitigate the 45% figure. 



First, the sizes of icebergs 

 in BERGSEARCH '84 and the 



Table B-2. Error Rates 



Obser- Error Rate 



ver(s) (Ships & Icebergs) 



1-4 45% 



1 ,3 40% (Iceberg Bias) 



4 40% (No Bias) 



54 



