H. A. Hazen — Verification of Tornado Predictions. 129 



five, which has occurred in a district not predicted for. We 

 certainly cannot regard the occurrence of a tornado of so defi- 

 nite a nature, as, for example, a rifle-ball from a marksman's 

 rifle, and we ought not to apply a method of verification which 

 would be perfectly proper in the case of rifle-balls and a target 

 to one utterly dissimilar. 



5. No account is taken of the fact that the law of occurrence 

 of tornadoes is very different in one portion of the country from 

 that in another ; for example, conditions which would almost 

 invariably produce a tornado in the Mississippi valley Would 

 not be at all efficient in the region east of the Allegheny Moun- 

 tains or in Texas. 



6. It is entirely unsatisfactory to group together districts 

 where only one tornado occurs in a year with those where 

 thirty or more occur each year. This may be better seen by a 

 slight exaggeration. Suppose we have a district where only 

 one tornado occurs in ten years, it is very plain that a very 

 good knowledge of the laws of tornado occurrence would be of 

 little avail in predicting for such a region, and the chances of 

 getting even one per cent, would be exceeding small, while if 

 there were a region having 100 tornadoes in a single year, the 

 chances of getting 50 per cent would be much better than of 

 getting 1 per cent in the previous case. 



In seeking for a satisfactory system of verification it should 

 be distinctly borne in mind that the character of the occur- 

 rence is very indefinite and that we cannot apply rigid mathe- 

 matical analysis to the questions, but must seek for a rational 

 system which will best treat ihe prediction as worded and an 

 occurrence so indefinite. 



In seeking such a system I have carefully studied the occur- 

 rence of tornadoes and destructive storms, and have found the 

 following comprise all the districts having more than thirty 

 storms each : (No. 5) 35, (6) 35, (7) 35, (8) 30, (9) 35, (11) 35, 

 (12) 35, (13) 60, (14) 40, (15) 32. Of these districts there are 3 

 in which the occurrence of tornadoes can hardly be said to be 

 under precisely the same laws as in the rest ; these are 5, 8 and 

 12. It has seemed wise to include in the discussion, though 

 with less weight, hurricanes and destructive storms. As a 

 working hypothesis, I have assumed that tornadoes occurring 

 in a district half way between the center and edge shall have 

 weight 1 ; in the rest of the district f ; to the center of the dis- 

 trict next outside \\ to the outside of that ^ ; all outside of 

 these 0. A study of the predictions and occurrences has 

 developed an approximate result as in the last column of the 

 table herewith and a percentage of verification of 49. 



Am. Jour. Scr.— Third Series, Vol. XXXIV, No. 200.— August, 1887. 

 9 



