12 SMITHSONIAN MISCELLANEOUS COLLECTIONS VOL. 8/ 



We do not venture to claim this important conclusion as yet. We 

 intend to carry on the research much longer. But at least the inves- 

 tigation has decided promise. 



We have been es^^ecially interested to compare among themselves 

 the curves marked A in the four figures 3, 4, 5, and 6, giving the solar 

 variation and the temperature departures at the three stations. We 

 have also wished to compare the curves giving periodicities of 25 and 

 45 months in solar radiation with the corresponding temperature peri- 

 odicities at the three stations. These comparisons are shown in Fig- 

 ure 7. 



It appears at sight that parts of the curves of temperature departures 

 for A\^illiston and Washington are very similar, but that the similarity 

 is slight as between Williston and Clanton. On the other hand, there 

 are many points of similarity between the temperature departures of 

 Washington and Clanton. The large departures of similar form found 

 at Williston and Washington in the years 1918 to 192 1 occur from one 

 to two months later at Washington than at Williston. 



Finally, we have made an experiment at long-range weather pre- 

 diction. Instead of making our readers wait several years to test it, 

 we have made our prediction backwards from 1918 instead of forwards 

 from 1930, so that we could immediately compare expectancy with ob- 

 servation. Figure 8 shows the predicted and observed temperature 

 departures from March, 1918, backward to September, 1916, for 

 Clanton, Washington, and Williston. The agreement is not perfect, 

 yet there is in each case a tendency to a correspondence in the trends. 

 But it must be recalled that the periodicities found represent the 

 average march of weather from 1918 to 1930, and therefore are to 

 be regarded as of the epoch 1924. None of the periodicities fits this 

 entire long interval of 13 years perfectly. Hence in predicting back- 

 ward to 191 7 we are really attempting a seven-year forecast from 1924. 

 It is perhaps extraordinary that the correspondence is as good as it is. 

 If our method should be used for serious long-range forecasting, it 

 must be perfected so as to pass from the last year or tivo of known 

 values to the unknown, not seven years as here attempted. 



