James 



In the analysis of 10 percent of present-tjrpe data 

 (case I) the mean absolute error was actimlly larger than for 5 per- 

 cent data (the solid curve shows the expected c\irve). Tikis was un- 

 doubtedly due to the fact that greater quantities of poor data is 

 more misleading than smaller quantities. 



The conclusions to be drawn from the Area B tests are as 

 follows: 



(a) Very little more data is required for smooth 

 areas than is now available but the data must be of a quality com- 

 parable to that provided by NSRT or buoys. 



(b) It is not feasible to increase the dr?tn. input in 

 smooth areas with low quality data, since this leads to little inn 

 provements in analysis accioracies. 



(c) "Hie use of ART is not required in smooth areas 

 since the need is for a small munber of good observations, not a high 

 density of data as provided by aerial survey. 



(d) Buoy observations would be ideal for smooth areas 

 since they provide both a non-random data plot and high accttracy tem?- 

 peratin-e data. 



(^) Abs olut e Data Requirem ents for Area b 



Prom Figiure 8 it appears that present type data should 

 be replaced as soon as possible by more accurate data and no great 

 effort expended on increasing ship-of-opportunity data unless it is 

 from a NSRT system. 



On the basis of the discussion in Section IV A(5) con- 

 cerning selected versus random data distribtitlon It is safe to con- 

 clude that a fixed array of buoys wotild give accuracies comparable 

 to the perfect data curve shown In Figure l6. Thus, a 2 percent data 

 Inpiit is completely adequate, which would mean eight observations or 

 if accuracies of 1 F would be tolerated, perhaps five or six reports. 

 Combining buoy observations and random ship reports from NSRT systems 

 one could obtain reliable analyses from one buoy plus six to eight 

 ship reports. 



Owing to the lack of advection in smooth thermal fields 

 it is likely that contintdty would contribute highly to analysis 

 accuracy. An occasional ART survey over the area to obtain a high 

 data density of good acctiracy would provide a highly reliable anal- 

 ysis which would be useful for continuity for 10 to 15 days. 



148 



