model. The model does not include the diurnal variation, a 

 systematic variation of about this magnitude. It is anticipated 

 that the inclusion of this variable would decrease the standard 

 deviation to a value near the instrumental error. Thus , it appears 

 reasonable to suggest that a simple statistical model, such as 

 equation (4), using bathythermogram data, will describe the 

 seasonal and spatial variation of sea-surface temperatures to one 

 standard deviation of something less than 1 F. 



Time/Space Distribution 



The temporal and spatial distribution of the observations is 

 of interest. A study of the distribution of the data (fig. 5) sug- 

 gests that temporally the distribution is the most unsatisfactory in 

 Area E, since no observations were made from October to March; 

 and that spatially it is most unsatisfactory in Area C, since in one 

 10-minute square, near the 100-fathom contour, 32, 75, and 87 

 observations were taken in 1952, 1953, and 1954, respectively. 

 The observations represent 12, 30, and 30 percent of the total 

 data taken in their respective years. 



An examination of the variation of the statistical measures 

 shown in figure 6 suggests that the data distributions are not 

 unsatisfactory, as originally thought, but are quite satisfactory. 

 An examination of the regression model supports this contention. 

 The model for the day-to-day variation is a fifth-degree polynom- 

 ial. If this model truly represents the seasonal variation of sur- 

 face temperature, then it is necessary only to have observations 



68 



