station 4, run 4, profiles, after omitting XBT 269L, are similar to the average differences 

 for the XBT profiles taken during the other runs. The conclusion is that the profiles made 

 during station 4, run 3, were measuring temperatures consistently higher than the thermistor 

 chain measurements, with the average individual differences varying from 0.1 1°C to 0.25°C. 



The data sets discussed above are additional examples of "runs" of consecutive 

 probes measuring biased differences, which were previously observed in the discussion 

 of table 37. 



It is concluded from this analysis that many of the XBT profiles made by the XBT 

 system used in this study measured temperatures that, on the average, showed good agree- 

 ment with simultaneous thermistor chain measurements. However, of particular concern 

 are the observed biases resulting from the non-random distribution of the differences both 

 with respect to measurements made by an individual probe and with respect to "runs" 

 of the consecutive probes. 



SUMMARY 



XBT measured temperatures are compared with average hydrocast and STD/SV 

 measurements, quasisimultaneous (made within 30 minutes of each other) STD/SV measure- 

 ments, and simultaneous thermistor chain measurements. 



Two different XBT data sets were compared with the average hydrocast and 

 STD/SV measurements. The first was the set of 826 visually acceptable profiles that 

 reached a minimum depth of 200 m. The second was the set of 528 visually acceptable 

 profiles that all reached 400 m and also had an independent surface temperature measure- 

 ment. For the first set, comparisons were made at 200, 300, and 400 m. These depths 

 were selected since, for depths greater than 200 m, vertical temperature gradients are small. 

 For the second set, comparisons were made at the surface and at 200, 300, and 400 m. 



The analysis of the first data set showed that in the absence of independent 

 temperature measurements to detect XBT profiles made when an XBT system malfunc- 

 tioned, the XBT system measures, on the average, temperatures that are higher and vertical 

 temperature gradients that are larger than those measured by the hydrocast and STD/SV 

 systems. Previous analysis of surface temperature accuracy showed that two of the XBT 

 systems used in making the profiles malfunctioned. With the eliminations of these profiles 

 from the data set, a reanalysis of the remaining 559 profiles still supported the above 

 conclusion. Thus, even with the omission of demonstrated erroneous profiles made when an 

 XBT system malfunctioned, there remained enough erroneous profiles to bias systemat- 

 ically the average temperature and the average vertical temperature gradient. In an effort 

 to purge the data set of the biasing profiles, accuracy criteria based on the average hydrocast 

 and STD/SV measurements were defined. Using these criteria, the data set was divided into 

 one subset of 354 profiles that satisfied the criteria and another subset of 205 profiles that 

 did not satisfy the criteria. The differences at all three depths for the data set that satisfied 

 the accuracy criteria were normally distributed with mean near zero and standard deviations 

 of 0.07°C to 0.1 3°C. Once the biasing profiles are identified and removed from the data 

 set, the remaining profiles accurately measure the temperature. The biasing profiles were 

 detectable only with the aid of independent 200-, 300-, and 400-m temperature measurements. 



The results at 200, 300, and 400 m of the analysis of the set of profiles that reached 

 400 m and also had an independent surface temperature were similar to those obtained in 



91 



