46 SMITHSONIAN MISCELLANEOUS COLLECTIONS VOL. I39 



When records through 1958 became available two considerable dis- 

 crepancies between forecast and event were noted. In the months 

 January and March 1950 heavy precipitation (over twice the normal 

 even in 3-month running means) raised the February observed curve 

 far above the predicted curve. Both curves, as has been said, are 

 smoothed by using 3-month running means in all computations, hence 

 the February effect. Not until April did the two curves come close 

 together. Yet there was a difference of only 6 percent of normal 

 precipitation between the averages of their heights, January- April, 

 1950. 



Beginning October 1956, and extending through August 1957, 

 there was a shift of 5 months, leaving the predicted curve in the rear, 

 and exposing opposed high and low values of the prediction and 

 event. When the two curves were averaged over this interval of 11 

 months, the predicted curve was 116 percent of the normal and the 

 observed curve 96 percent of the normal. 



To sum up : At Spokane, in the 9-year interval, my forecast gives 

 for over 7 years a correlation with observations of 59 percent. Two 

 intervals of marked discrepancy occurred. The first, of 4 months, 

 culminating with February 1950, was obviously caused by extraordi- 

 nary precipitation in two almost adjacent months. It produced a 

 difference of only 6 percent between the averages over these 4 months. 

 The second discrepancy, extending 11 months, was of unknown cause. 

 It involved a 5-month shift of phases and produced 20 percent dif- 

 ference between forecast and event in average precipitation over those 

 11 months. 



Having set forth those discrepancies I remark that this is in the 

 infancy of my method of forecasting, before any help has come to 

 me from theoretical meteorologists. It may be that some of them 

 will discover the causes of occasional displacements of features be- 

 tween forecast and event. If so, it may reduce error of forecasts 

 greatly. Then, too, my method assumes that the average behavior of 

 periods in weather in a thousand months that are past will be followed 

 in the months to come. It perforce neglects changed conditions which 

 may arise from unpredictable storms, volcanoes, or even from man's 

 interposition, as from forest destruction, invention of new powerful 

 devices, wars and the like. Even a minor atmospheric change may 

 alter the time of a feature in precipitation by a month. All these 

 factors tend to lower the coefficient of correlation. 



The 273-month period in weather features. — It will have occurred 

 to some readers that if one were backcasting from April 1927 or from 

 July 1904 he would employ the same tabular data that I have used 



