Regression analysis appears to have considerable potential 

 as a technique for estimating sea-surface temperatures. However, 

 the physical reality of the estimates must also be considered, 

 since it is always possible to improve the statistical measures of 

 goodness of a regression-model estimate by merely adding addi- 

 tional terms to the model. From a physical viewpoint these terms 

 may be nonsense terms. 



Regression analyses of Area B (1954), Area C (1953), and 

 Area E (1952) are examined to illustrate the physical reality of 

 the model. The difference between observed values of tempera- 

 ture and the values obtained from regression will be considered as 

 a function of water depth, time, latitude, and longitude. 



It was noted in studying the results of some of the earlier 

 analyses that many of the large differences between the tempera- 

 tures obtained from regression and those obtained by observation 

 occurred in the shallower water adjacent to the coast line. This 

 finding was not surprising, since transient and local effects, which 

 are not accounted for in the model, should have their maximum 

 influence on the temperature in such areas. 



Figures 10 and 11 present a qualitative histogram analysis 

 of the effect of water depth for the three regressions. Figure 10, 

 for each area, contains three histograms. The shaded portion of 

 the histogram on the left shows the distribution of differences for 

 data taken in water depths less than 100 fathoms, while the un- 

 shaded histograms include all data used in the regression analysis. 



48 



