Anomaly Detection 



Regression models could be used as anomaly detectors. In 

 this application a model, such as equation (4), could be used to 

 remove the systematic variations in latitude, longitude, and day- 

 of-year. Through a study of the differences between observation 

 and regression (anomalies), information on nonsystematic and 

 other systematic space/time changes would be obtained. The 

 anomalous variations could be examined in terms of causes and 

 mechanisms. This application of regression models was alluded 

 to in the discussion of figures 12 to 14, in which it was noted that 

 the differences revealed short -period, small-area, nonsystematic 

 anomalies. 



An additional example of this use of regression models may 

 be found in the differences associated with the data used in figure 

 9. It is well known that upwelling of cold water occurs off the 

 coast of California from about 30 N to 45°N from March to July, 

 The phenomenon is associated with the north-northwest winds that 

 prevail off the coast of California during these months. The 

 upwelling results in summer and autumn surface temperatures 

 considerably lower than those expected on a seasonal basis alone. 

 The colder-than-expected surface temperatures are centered in 

 the vicinity of 35 N and 40 N, Figure 19 shows the differences 

 for July between the observed surface temperatures and surface 

 temperatures computed from regression. A negative sign means 

 the observed temperature was lower than estimated. The anom- 

 alous effect of upwelling on surface temperature is obvious. 

 Parenthetically it is noted that since it is known that these anom- 

 alies are the result of a north -northwest wind pattern, they could, 

 in principle, be removed by introducing the wind vector as an 

 independent variable in the regression equation. 



73 



