layer with identical agreement in the thermocline. However, below the thermocline they 

 begin to differ, with the difference being an increasing function of depth. At 400 m, the 

 difference was 3.01°C. Pair 38 shows identical agreement from the surface to 1 15 m. At 

 115 m, the two profiles begin to differ, with the difference being a variable function of 

 depth. Figures 45a and 45b exhibit characteristics similar to those shown by two of the 

 ORB-3 comparisons (figure 44). The major difference is that for XBT pair 38 (figure 45b), 

 the differences do not diverge with depth. 



RAPLOC/DEEPTOW XBT pairs 4 and 39 exhibited a marked difference in the 

 above-thermocline comparisons as well as in the below-thermocline comparisons. These 

 two comparisons are shown in figure 46. 



Remember, all profile pairs shown in figures 43-46 were simultaneous pairs made 

 by using two independent XBT systems whose launchers, in the case of the ORB-3 

 measurements, were located 30 feet apart and, in the case of the RAPLOC/DEEPTOW 

 measurements, were located 5 feet apart. All of the measurements except the one presented 

 in figure 43b were made from the ORB at anchor. 



Table 45 shows the average temperatures for each XBT system for the 26 ORB-3 

 and the 36 RAPLOC/DEEPTOW XBT pairs whose differences met the accuracy specifi- 

 cation at all comparison depths above and below the thermocline ("starred" comparisons 

 in appendix H). For both data sets, system A, on the average, measured slightly higher 

 temperatures than system B. All average differences, even the in-thermocline differences, 

 are well within the ±0.34°C accuracy specification. It is concluded that when the biasing 

 profdes are detected and removed from a data set, excellent agreement between the 

 remaining profiles is reached. 



Note in figures 43^6 that the left-hand profiles, except possible ORB-3 pairs 

 47 and 54 (figures 43a and 44b), agree better with the average temperatures listed in 

 table 45 than the right-hand profiles. This suggests that, in at least three cases, the right- 

 hand profiles are the incorrect profiles. This is important since the temperature errors 

 introduced into these sets of XBT profiles are systematic, biasing, and nonrandom. The 

 implications are obvious. 



114 



