44 SMITHSONIAN MISCELLANEOUS COLLECTIONS VOL. I39 



of discrepancies from pages 2-3 of my paper cited in footnote 1, d, 

 above : 



8. Of 100 years of St. Louis precipitation forecasted, 70 seem fairly satisfactory 

 and yield high correlation coefficients with the events. The failure of the other 

 30 is reasonably explained. 



9. As shown by Dr. W. J. Humphreys in his "Physics of the Air," figure 227, 

 great volcanic eruptions, which throw high columns of vapor and dust, pro- 

 foundly modify weather. He cites the first four cases in the following list [here 

 my table 11], and I add several more. 



Table ii. — Great atmospheric disturbing causes 



Approximate 



dates Volcanic eruptions 



1856 Cotopaxi and others. 



1883-1890 Krakatoa and others. 



1901-1904 Pele, Santa Maria, Colima, and others. 



1912 Katmai. 



1924 and 1928 Many great eruptions. 



1930 Great eruptions. 



1947 Niuafoo Island. 



10. Of 30 unsatisfactory years, in 100 years of synthesis of St. Louis precipi- 

 tation, these lie in groups as follows: 1854 first half; 1856 to i860; 1887 to 

 1889; 1900; 1 901 ; 1905 to 1907; 1912 last half ; 1913 first half ; 1915 to 1917; 1920; 

 1923 to 1926; 1930; 1940 to 1950. It will be seen that many of these unsatis- 

 factory intervals fall either soon after tremendous volcanic eruptions occurred 

 or there was tremendous use of explosives in war or explosions of atomic bombs. 

 As has been pointed out, atmospheric changes alter the lags in the weather effects 

 of all solar impulses, and of course unequal periods have unequal lags. These 

 unusual atmospheric disturbances may very well have mixed up the timing of 

 terrestrial responses to the 23 periods so as to cause the events to differ from the 

 predictions. 



At some future time it may be possible to connect theoretically the 

 displacements found in my forecasts with causes producing atmos- 

 pheric alterations of importance in weather. As yet I have been 

 unable to name with certainty causes operative to produce these occa- 

 sional displacements. For the practical inquiries of farmers, however, 

 it is of importance to estimate the magnitude of forecasting error 

 rather than the cause attending such discrepancies. 



As a step toward that, I cite the case of Spokane, Wash., figure 11. 

 A computation made in 1957 derived a "correlation coefficient" of 

 59 ±5 percent over the interval March 1950 through October 1956 

 between forecast and event in Spokane precipitation. In simple lan- 

 guage this means that my forecast represented the observed precipi- 

 tation 59 percent perfectly for almost 7 years. 



