The implication is important. A sampling interval based on a 

 model representing the distribution of the variable should be used 

 in place of the fixed time interval employed in the classical 

 manner with an area grid. The oceanographic problem becomes 

 one of searching for adequate models. It is believed that adequate 

 models can be developed for many oceanic areas through examina- 

 tion of historical data and present knowledge of oceanic dynamics. 



Data Screening 



It may be possible to use regression techniques to identify 

 and eliminate erroneous data from data samples. Saur^'^ discusses 

 this problem. Erroneous measurements are particularly trouble- 

 some when the conventional space/time methods of data summar- 

 ization are used, since the number of observations in any given 

 cell is generally small , and erroneous data can often distort 

 arithmetic means and standard deviations of discrete samples. 

 Since the average for any given cell is independent of the average 

 in any other space/time cell, the contribution of an erroneous 

 observation tends to be maximized. Frequently a biased average 

 results that must be compensated for in some manner, generally 

 subjective, in the subsequent contour analysis. Thus, the problem 

 of editing out erroneous observations is of considerable importance. 

 Provided the regression model is reasonably valid, the regression 

 technique may offer an effective, and objective, method for editing 



70 



