104 ANNUAL KEPORT SMITHSONIAN INSTITUTION, 1924 



lion, and its grade are all included in two words. Messages arrive 

 at Washington from both stations within 24 hours of the actual 

 measurements, and generally represent mean results of five inde- 

 pendent determinations at each station. Arrangements have been 

 made (also owing to Mr. Roebling's interest and generosity) to test 

 the value of the solar measurements for forecasting according to the 

 methods of Mr. H, H. Clayton. For this purpose Mr. Clayton has 

 had a small office ^ and one assistant near his home in Canton, Mass., 

 where he receives before noon daily from the Smithsonian Institu- 

 tion the weighted mean of the solar-constant values observed in 

 Arizona and Chile on the preceding day. He makes his forecasts for 

 3, 4, 5, and 27 days in advance, and mails them to the Institution on 

 the same afternoon. Thus we receive the forecasts sufficiently long 

 before their maturity to make a very real and searching test of their 

 validity. 



These forecasts for definite days relate to the mean temperature of 

 New York City, and are later on compared with the observed tem- 

 peratures and analyzed by several purely mathematical methods 

 quite independently of any bias of the computer. The official 

 weather services of the various countries do not. of course, make 

 predictions parallel to these, except in Argentina, where such fore- 

 casts are made by similar methods to Clayton's. Hence it is impos- 

 sible to Imow at present how much gain, if any, Mr. Clayton's solar 

 forecasts show over the present official methods. That they do show 

 some prevision of the event, even to five days after the solar observa- 

 tions, is certain. 



Hitherto, however, the 27-day detailed forecasts have shoAvn no 

 correlation v;ith the New York temperatures. This is not at all 

 surprising. Indeed, all such forecasts have to contend against great 

 odds. For we recall that the march of temperature often goes 

 quickly from crest to trough, so that even if a true forecast could 

 be made, and it should be no more than 12 or 24 hours off in point of 

 time, there would be large divergences between the prediction and 

 the event. With the unyielding mathematical methods of verifica- 

 tion this would greatly diminish the correlation found. 



A fairer test for very long-range forecasts is found in general 

 statements as to the expected departure from mean normal tem- 

 peratures for coming months. These Mr. Clayton has furnished 

 from 15 to 30 days before the beginning of each month from De- 

 cember, 1923, to the present time. He also furnishes similar pre- 

 dictions about the approaching weeks furnished three da5'^s before 

 the beginning of the week in question. With few exceptions, these 

 broader prognostications have been fairly verified. 



» Ttanks are due to the Canton Historical Society for use of these quarters. 



