12 SMITHSONIAN MISCELLANEOUS COLLECTIONS VOL. I43 



I have computed the correlation coefficient in precipitation between 

 forecasts and events, 1959 through i960, combining the results re- 

 ferred to in table 1 for 14 stations. Just as the mean temperatures for 

 the decade 1950 to i960 differ generally between 0.5 and 2.0 F. 

 from those of the mean values 1870 to 1956, so, too, the mean values 

 of precipitation 1950 to 1961 differ from the mean values 1870 to 1956. 

 Before computing the general correlation coefficients, 1950-58 and 

 1959-60, for the 14 stations, I have corrected these differences of level 

 by lifting or lowering the forecasts bodily. These differences range 

 from zero to 17 percent among the 14 stations. This done, the general 

 correlation coefficient between forecasts and events for the 336 months 

 during 1959 and i960 available at the time of computing from of- 

 ficial records resulted as +47.0 percent. 



I am not aware that anyone has ever before predicted the monthly 

 precipitation at 14 definite cities over 2 years of time (in my case 

 1 959- 1 960) and has achieved a correlation coefficient as high as +47 

 percent for 24 months, 3 years after the last month used in the basis 

 of his forecast. It seems to me that this marks an important and en- 

 couraging advance in long-range forecasting. 



Figure 4 shows graphically the results tabulated in table I. I call 

 attention to cases of displacements of obviously common features in 

 forecasts and events that occurred, and remark that such displace- 

 ments must obviously have pulled down the value of the correlation 

 coefficient which, notwithstanding, reached +47 percent. These cases 

 are: Cincinnati, 3 months May 1959— April i960; El Paso, 2 months 

 about June i960; Helena, 2 months about October 1959 and 2 months 

 about September i960; Sacramento, 2 months about January i960. 



NUMERICAL TABULATIONS 



I was assisted in these tabulations by Mrs. Lena Hill and Mrs. 

 Isobel Windom. Miss M. A. Neill assisted in reading proof. With 

 the electronic computer, Jonathan Wexler furnished 3-month means 

 of absolute temperatures, covering the years 1870 through 1956. We 

 continued them through 1959. Subtracting 459.7 , we expressed them 

 in ordinary Fahrenheit degrees. There were two sets of monthly 

 normals computed covering 1870 through 1956: A for the years when 

 Wolf sunspot numbers were less than 20, B for the years when Wolf 

 sunspot numbers exceeded 20. The dates included in these two cate- 

 gories are given in table 2 and accompanying quotation from the 

 journal "Solar Energy." 



Table 3 gives these two sets of normal temperatures in both absolute 

 and ordinary Fahrenheit, applying to the years 1870 through 1956. 



