( + 2 X .37) or ■+■ .74°F for 95 percent probability. This estimate should 

 be fairly accurate, because errors in the method tend to cancel each other. 



Months of maximum and minimum variability for each thermal structure 

 parameter are listed in table 26 (appendix B). These values were determined 

 by computing the mean standard deviation for all 5-day, 10-minute q^uadrangles 

 and S-week, 10-minute quadrangles. The results were checked against histor- 

 ical data and found to be in approximate agreement. However, both sets of 

 data were insxifficient for determining months of maximum and minimum vari- 

 ability. 



QUALITy OF THE DATA AMD STATISTICAL ANALYSIS 



The magnitude of the various errors involved in the methods of obtain- 

 ing, processing, reading, and analyzing the data properly are of sufficient 

 importance to warrant a separate study. Thus only a brief qualitative 

 discussion is presented here. 



A number of reports dealing with BT errors are available. The most 

 comprehensive study is probably that of Bralove and Williams (4). By use 

 of their qualitative report and the SERC deck, some of the errors associ- 

 ated with BT data can be quantitatively analyzed. They result from instru- 

 ment limitation (limited accuracy), calibration (malfunctions), operational 

 (human error), reference temperature, data processing, and location error. 



The magnitude of the location error, which cannot be determined, varies 

 from cruise to cruise and from station to station and certainly reduces the 

 reliability of the analyses. 



Conclusions based on the analysis of variance technique are valid if 

 obseiTvations are randomly chosen from normally distributed populations 

 having approximately equal variances. However, the observations were not 

 taken at random over any given area (e.g., a 1-degree quadrangle), rather 

 they tend to be concentrated about the central location of 35N,48w. Also, 

 times of observation are not random within each area. Fortunately, inves- 

 tigation shows that the results of the analysis are changed little by mod- 

 erate departure from the assumptions. Distribution curves were plotted for 

 each month and parameter. The analysis of variance and the t test were 

 applied only to groups of data having approximately normal distributions 

 and equal variances. 



Because of extreme care exercised on the four experimental cruises, 

 instrument and operator errors were assumed to be minimal for comparison 

 purposes. 



To obtain an indication of the magnitude of the instrument errors, the 

 ratios of the standard deviations of the historical data to the experimental 

 data for the 20-minute quadrangle centered at 35°N,if8°W were computed. If 

 the size of the area and the time period were the same for both sets of 

 data, the standard deviations shoiild be approximately the same for any given 

 parameter. If a ratio of the two standard deviations did not approach unity. 



10 



