James 



A secondary objective is to find the point of dindnisbing 

 return where the cost and efforts expended to increase the data input 

 are not balanced by the improvement found in analysis reliability. 

 At this point further expenditures for more data are not justified 

 unless It can be deironstrated that an increase in operational effec- 

 tiveness compensates for the additional expense. 



III. PROCEIXJRES 



A series of analysis tests were made for two areas of about 250 

 RM on a side. One area has a complex thermal field typical of the 

 region jtist north of the Gfulf Stream while the second area is repre- 

 sentative of the rather smooth thermal gradients found in the 

 Sargasso Sea and other areas lacking major cvirrents. The two areas 

 will be designated for the remainder of the report as Area A, which 

 is complex in nature, and Area B, which has a smooth thermal field. 



Basically the tests consist of starting with an assTimed realis- 

 tic isotherm pattern and a data field of observations over a closely 

 spaced grid that are matched to the isotherms to produce perfect 

 observations. Various percentages of the original data input are 

 randomly selected and analysis made of these reduced data. Each 

 analysis is compared to the original data field to compute the mean 

 absolute error, and to the original analysis to ascertain the good- 

 ness of fit. Analyses were prepared both manually and by computer 

 techniques. Error functions are introduced to vary both the quantity 

 and quality of data utilized in a test. In all cases the data field 

 used for analysis is considered a 2k hour data input. 



Restilts of the tests reveal many relationships between the 

 quantity and quality of data and the reliability of the analyses. 

 The tests were divided into three major categories as follows: 



Area A - manual analysis 

 Area A - computer analjrsis 

 Area B - manual analysis 



For each of these major categories there were 8 to l6 individual 

 tests. In addition, there were two special tests to determine the 

 degrees of freedom inherent in the use of random data distributions 

 and to compute the relative advantage of selected data points over 

 randomly spaced data. 



IV. RESULTS 



ResvLLts of the tests will be described in the following sections 

 vmder the three major categories and the two special tests described 

 above. 



126 



