NO. 9 FORECAST OF U. S. PRECIPITATION — ABBOT 47 



in forecasting from January 1950. Hence one might infer that the 

 precipitation following these earlier dates should parallel that follow- 

 ing January 1950. 



There is indeed a partial similarity, as I pointed out many years 

 ago, between the march of weather at successive intervals of 273 

 months. But the correspondence is very imperfect. This appears in 

 figure 2, where the precipitation at Nashville following July 1904 is 

 compared to that following January 1950. However, I call attention 

 to the close agreement of the two curves for the last three years of 

 the comparison. I have computed for several cities, including Nash- 

 ville, the coefficients of correlation of the observed precipitation fol- 

 lowing April 1927 and July 1904, and compared with the forecast 

 made to follow January 1950. These coefficients have fallen between 

 18 and 22 percent, while, as stated in my foreword, the correlation 

 following January 1950 ranges from 52 to 59 percent. 



This difference is easily explained. Over 40 percent of perfect 100 

 percent correlation is unpredictable as yet. There are several causes, 

 (a) There is occasional unusual precipitation, as occurred in January 

 and March 1950 at Spokane, (b) There are displacements of fea- 

 tures as yet unexplainable. (c) The graphs I have published show 

 large discrepancies in amplitude between forecast and event of ob- 

 viously identical features, (d) Unpredictable events occur to alter 

 weather from the averages of 1,000 months. 



In the march of precipitation from April 1927 and from July 1904, 

 the vicissitudes of the later years up to January 1950 cannot have 

 affected the observed precipitation of the earlier times as they have 

 done that following 1950. As such vicissitudes account for 40 per- 

 cent and more in coefficients of correlation, the tabulation suited to 

 January 1950 can only roughly forecast what follows these earlier 

 dates. 



FORECASTS, 1959 TO 1967 



Table 12 gives for 32 stations for the interval 1959-1967 the ex- 

 pected monthly mean percentages of the normal precipitation tabu- 

 lated in table 9. The reader will recall that all forecasts are made 

 from 3-month running means taken from published monthly mean 

 values, and expressed in percentages of the normal values of table 9. 



Expressing these forecasts in a more usable form, table 13 gives 

 average percentages of the normal for the intervals January-April, 

 May- August, September-December, of each year, 1959 to 1967, 

 inclusive. 



