James 



It is obvious that the accuracy of an analysis of a given 

 random data plot is a function of not only the nmber of observations 

 but also their distribution. Or"^ analysis of eight random points 

 does not clearly establish an analysis accuracy since all the data 

 may, in that particular case, be located in one portion of the chart. 

 To investigate this problem various random distributions of equal 

 numbers of observations were analyzed on the computer. The results 

 showed that for larger nuEiber of observations, 5 and 10 percent of 

 data, the randomness of plot makes no significant differences. That 

 is, with a sufficient number of observations any random distribution 

 provides some values in all quadrants of the chart. With 2 percent 

 of data, however, the randomness does make a difference. 



Eight random data plots of eight observations each were anal- 

 yzed and the mean absolute errors computed. The distribution of ob- 

 servations varied from cases when the data was evenly distributed 

 over the charts to one case vhere all eight observations were located 

 in the eastern half of the chart. The mean absolute errors varied 

 from 2,6^ to 4.0^ with a mean value of 3.3°F. This meaai value com- 

 pares well with the 3*2^ error shorm in Table I for 2 percent data 

 in the original tests. Thus, the family of curves shown in Figure 8 

 are considered representative of analysis accuracy. 



In 8.ddition the slight smoothing necessary to establish a 

 family of curves tends to bring the value of an individual score 

 closer to the true mean of all random samples. 



C. Area B - Manual Analyses 



The tests described so far have established some facts about 

 data requiren^nts for complex trend areas. In order to study the re- 

 quirements for data in non- complex areas a similar set of tests were 

 repeated for a relatively smooth isotherm field. This temperatxire 

 field and analyses is shown in Figure 13, and represents generally 

 the area of a five degree quadrangle similar in size to Area A. This 

 area will hereafter be referred to as Area B. Where Area A had a 

 temperature gradient of 25 degrees across the area. Area B shows only 

 a 6 degree temperature range. This is typical of areas such as the 

 Sargasso Sea, the eastern North Pacific, eastern North Atlantic and 

 other ocean areas away from major cxirrent systems. 



With such a flat thermal field the variation in temperatvire 

 between grid points is less than 1°F. A wide belt of temperatures 

 result for a given value rather than individual whole values as in 

 the complex field. ISius the 6k°F isothena in Figure 13 is drawn 

 through the center of a number of 64°F temperatures. This resixlts 

 since the values on one side of the 6k^ isotherm may represent a 

 true vsilue of 63. T while a 64°F on the opposite side may represent 

 a. 6k,h F but rounding to whole degrees for coding pirposes gives the 

 same values. 



144 



