IO SMITHSONIAN MISCELLANEOUS COLLECTIONS VOL. I43 



from to 1,000 or more, the entire range of absolute temperature is 

 only about 2.3 percent. The accidental fluctuations of temperature 

 bear a much larger proportion to range than the accidental fluctuations 

 of precipitation. Hence, correlation coefficients between forecasted 

 and observed temperature, though always positive, are always smaller 

 than those between forecasted and observed precipitation. While the 

 32 cities forecasted for precipitation, 1950 to 1958, all yielded positive 

 correlation coefficients with the events ranging between 50 and 70 

 percent, the correlations on temperature for 10 cities, forecasted 1950 

 to 1958, ran as follows: Detroit +16 percent; Los Angeles +22 

 percent; Atlanta +32 percent; and the other 7 stations all ranged 

 between +40 and +50 percent. 



My forecasts of temperature and precipitation rest on one fact and 

 one assumption. The fact is that a harmonic family of regular periods 

 exists in solar radiation and in weather. The assumption is that if this 

 family of periods is individually and quantitatively determined from 

 weather records, 1870 to 1956, the mean course of these periods will 

 be followed approximately through subsequent years. This assumption 

 may indeed prove wrong when unusual disturbances occur in atmos- 

 pheric conditions — for example, the volcanoes of Krakatoa and 

 Katmai ; the furious bombing during the world wars ; atomic bomb 

 tests; the hurricane Donna of September i960. But tests such as 

 figures 1, 2, 3, 10, 11 of Publication 4390, and those of temperature 

 in this paper, show that generally the assumption is justified. As yet 

 it has not been explained why displacements between forecasts and 

 events in features of weather by 1, 2, or 3 months occasionally are 

 observed. If this difficulty can be overcome, much higher coefficients 

 of correlation will be found between forecasts and events. I plan an 

 investigation of possible causes of this defect. 



There is one important difference between my forecasts of precipi- 

 tation and of temperature. The amplitude of the principal features in 

 precipitation changes was found for all stations to be so nearly the 

 same in forecasts and events that no adjustments were made. Not so 

 with temperature. For all the 10 cities the amplitudes of the features 

 of change were obviously greater in the forecasts than in the events. 

 I cannot explain why this is so. The forecasts would have been left 

 woefully wrong unless this discrepancy had been corrected. 



To make this correction for scale, I carefully plotted for each city 

 the curves of forecast and event from 1950 through 1958. Then I 

 measured as best I could the depths of obviously corresponding large 

 depressions of the two curves, and determined their average ratio of 

 amplitudes in forecasts and observations for about a half dozen prin- 



