The mean of the distribution of absolute values of errors is 1.1 F, 

 and the standard deviation is 0.9 F. For any sample of 10 predictions 

 the standard error is 0.283°F, and the 95-percent confidence limits are 

 Omk'J and 1.73°F« If the mean of the 10 predictions in a given area 

 exceeds the upper confidence limit of 1.73°F, either the anomalies in 

 the area are uncorrelated or the sample lies in a transition zone. 



Six tests involving 81 individual predictions in transition areas 

 resulted in much larger errors with a mean value of 1.92°F. The tests 

 were made without consideration for uncorrelated anomalies, althoiigh 

 anomalies are known to be uncorrelated a large percentage of the time. 

 No doubt the approximation would be better and predictions would be 

 considerably more accurate if uncorrelated anomalies were considered. 

 The prediction error is probably larger in areas of intense frontal 

 interchange, such as the shaded area in figure 12. 



Prediction with this system may be unsatisfactory in many instances 

 but may be useful when req.uirements are not too exact. Evaluation of 

 individual prediction errors is also uncertain, especially when the 

 prediction is verified with a BT obse2rvation which may be as erroneous 

 as the prediction. Statistical parameters of many prediction errors 

 may be accepted with more confidence. 



VIII. CONCLUSIONS 



1. The prediction system presented in this study attempts to 

 provide a means for determining the temporal distribution of mean tem- 

 perature at 400-foot depths beyond the continental shelf in the North 

 Atlantic . 



2. Actual temperature at 400 feet can be obtained by algebraic 

 addition of the temperature anomaly and the mean temperature at 400 feet. 

 The anomaly at the Uoo-foot level can be obtained by relation to the 

 siirface temperature anomaly, if the anomalies at both levels are 

 correlated. If the anomalies are uncorrelated, the 400-foot anomaly 

 could be determined from special charts, if enough data were available 

 to permit construction of such charts. 



3. Establishment of more accurate relationships between the surface 

 temperature and the 400-foot temperature may provide a means for evaluating 

 the approximate stability factor in the thermocline. 



k» If the mixed layer thickness and the i|-00-foot temperature are 

 predicted with reasonable approximation, a satisfactory composite trace 

 of the vertical temperature distribution could be constructed. 



3^^ 



