NO. 3 SOLAR VARIATION AND WEATHER — ABBOT 2"] 



Table 5. Berlin. Period, 7.0 months. 

 Rearranged table luith symbols unchanged. 



at bi, raised 3 months at b2, and lowered 3 months at a2, so as to be 

 in proper phases in the summation. As for the modification of 

 results in periods from 18-1/5 to 91 months, examples of the re- 

 moval of overriding submultiples are given in figures 11, 18, and 20. 



8. EXAMPLES OF FORECASTING 



Figure 21 is from a direct photograph of my work when comput- 

 ing alone, beginning from "World Weather Records," 1854-1939, and 

 finally forecasting St. Louis precipitation, 1875-1879. (See P. 4211.) 

 All the records, 1854-1939, of precipitation at St. Louis were used 

 to fix every individual year. Every year computed, therefore, is 

 equally as really a forecast as if it were after 1939 and stems from 

 zero year (the median) 1895. The forecast gives a correlation co- 

 efficient of +80 percent for the 5 years 1875 to 187P between fore- 

 cast and observation. Figure 22a, which includes Figure 21, is also 

 a direct photograph from my comparison of forecast and event from 

 1860 to 1887. Throughout this interval there is an average correla- 

 tion coefficient of 70 ±10 percent. Figure 22b gives direct photo- 

 graphs of results I obtained from forecasts of precipitation and 

 temperature during the interval 1934 to 1939 for St. Louis and 

 Peoria. I employed all records for the period 1854-1939 to fix the 

 forms and amplitudes. In all these examples just given the depar- 

 tures in forecasts and events, in short all the computation, is done 

 with 5-month consecutive means. 



The reader will see that since all records, 1854-1939, were used 

 impartially to fix every number forecasted or backcasted in figures 

 21 and 22 and beyond, the number of years forecasted forward or 

 backward is to be determined as starting at zero for the year 

 (1939+ 1 854 )/2= 1896. Hence, the forecast of St. Louis precipita- 



