could "be chosen as the criterion; hovever, in summer, large positive 

 surface anomalies are generally uncorrelated with anomalies at 400 feet. 

 Therefore, the prediction is more successful when the large positive 

 anomalies are disregarded. 



D. Prediction 



Prediction of the temperature at i+OO feet can he made several 

 days in advance by using the surface temperature observed at the time 

 of prediction. In the absence of significant advection at ^00 feet, 

 prediction can be made up to a week in advance without serious error. 

 If predicted surface temperature is used instead of observed temperature, 

 the prediction at UOO feet is applicable to the same prediction period. 



Success of each prediction depends on accuracy of the mean tempera- 

 ture and anomaly. Accuracy of the mean temperature depends on the number 

 of observations available and on the period of time which they cover. 

 Mean temperatures at 400 feet used in this report are based on theoretical 

 deductions made from data collected within a few very limited areas. 

 Hovjever, the results of these theoretical deductions may be rather accurate 

 in the areas of the data, and the mean temperatvures obtained with equations 

 (9) through (12) are valid in the vicinity of the five ocean weather stations 

 in the western North Atlantic and in areas G and H (figiure 1). The more 

 the calculations depart from these areas, the greater the likelihood of 

 significant errors in mean temperatures. 



Since greater reliability of equations (9) and (lO) was expected to 

 occur in the western part of the temperate Worth Atlantic where the data 

 were collected, prediction tests were carried out mainly in the eastern 

 and northern North Atlantic. 



Accuracy of mean teraperatvires computed with equations (9) through 

 (12) cannot be determined, because prediction errors include other more- 

 significant components. The only indication of reliability of these 

 values is shown by the distribution of negative and positive errors. 

 The distribution of errors in 7^8 predictions is shown in figure l4. 

 The distribution appears to be normal with a mean very close to zero 

 and tends to substantiate computation of mean temperatures at 400 feet 

 with the equations. In certain areas the prediction may be biased. 

 Errors in mean temperature computations are probably much smaller than 

 errors due to inaccurate anomaly evaluation. 



Frequency distribution of absolute values of errors used in figiu-e 

 ik is shown in figure 15. All predictions used as a basis for these 

 figures were made in water masses for which the equations were intended, 

 that is, transition zones between different water masses were avoided. 



32 



