James 



It is obvious from this discussion that highly accurate 

 analysis in smooth areas requires observations reported to tenths of 

 °F. That is, to define a particular isotherm to the same exactness 

 as is in a complex area requires more accyrate data. This results 

 simply from the fact that in Area A a 1*^ temperature change occ\irs 

 in 12 KM while. in Area B a 1*^ change may be spread over a distance 

 of 100 KM. 



(1) Results of Area B Tests 



Inspection of the various analyses made for Area B re- 

 veal that excellent accuracies can be realized in relative]^ smooth 

 areas vith a minimum of data if the data is of high quality. Note 

 for instance the analysis of 2 percent of perfect data as shown in 

 Figure ik. Although there were only eight observations available 

 the ansilyst was able to portray a very reliable picture of the iso- 

 therm pattern. The mean absolute error was only 0.4°F for this 

 analysis compared to 2.95'^ for eight observations in the complex 

 Area A. 



On the other hand, poor data, in any quantity, leads to 

 very erroneous results in analyses of Area B. Applying the same 

 eiTor functions as previously discribed in Section IV A (l) analyses 

 were made of present t\npe data. Figure 15 Illustrates the analysis 

 of 10 p*5rcent of this Case I data. Numerous featTires are shown that 

 do not belon"? there; a total of six tongues as compared to the smooth 

 isotherm pattern actually present. Of course operationally continu- 

 ity from chart to chart and grouping of data tends to reduce this 

 problem but nevertheless poor data has a very deleterious effect on 

 analyses of smooth areas. 



The effect of poor data in deteoriatlng the reliability 

 of analysis is obvi.ously much more evident in Area B than Area A. 

 The reason for this is that an observation of 5°F in error and 12 NM 

 out of position may create a entirely new warm tongue in Area B while 

 in Area A the already high horizonta]. gradients mask the effect of 

 one bad observation. 



Figure l6 shows the faMly of curves resulting from 

 analyses of the four types of data described previously as perfect. 

 Case I, II and III. As discussed above and shown by Figure ik very 

 few observations are required for reliable analyses if the obser- 

 vations are perfect. The perfect curve shows little improvement 

 after k percent of data. Ce^e II and Case III curves show mean ab- 

 solute errors for analyses of temperature observations of 1°F accu- 

 racy but with (Case II) and without navigation errors (Case HI). The 

 closeness of the two curves indicates that the navigation error is 

 relatively unimportant in suKoth areas if the data are good. Ihis 

 is expected since the thermal field is so flat. The difference 

 between reliability of analyses of a given quantity of "good" and 

 "bad" data is very evident from Figure l6. 



146 



