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



cessful predictions. 2. Unsuccessful ones. 3. Failures to 

 predict. In other words, every district in which a tornado 

 occurred as predicted counted as one in favor of the predictor 

 in the final summing up, and each district which did not have 

 a tornado as predicted for, together with each district in which 

 a tornado occurred which was not predicted, counted as one 

 against the predictor. Taking the predictions of tornadoes for 

 the month of June, 1885, and summing the three classes in the 

 table, we find 8 successful ; 32 unsuccessful ; and 13 failures 

 to predict: total, 53. This (according to Prof. Gilbert) gives a 

 verification of 15 per cent for the predictions as made. It 

 seems as though this method of verification is open to most 

 serious objections, some of which may be enumerated as 

 follows : 



1. The verification depends on the occurrence of a tornado 

 anywhere within an imaginary line and does not make any al- 

 lowance for the nearness to that line, e. g., if a tornado should 

 occur just at the edge or within five miles of the district for 

 which it was predicted, it would count as two against the pre- 

 dictor, or if we should simply verify for that single prediction 

 we would find a difference of 100 per cent, in the verification 

 in going five miles, just across the imaginary line from one dis- 

 trict into another. As an illustration of this, we may take a 

 prediction of rain for the central of three contiguous districts, and 

 as a system of verification we may adopt the principle that the 

 occurrence of rain in any district where it was predicted shall be 

 a success, and its occurrence in a district not predicted for shall 

 count as a failure. In the case before us, let rain be predicted 

 for the central district, and let it fall almost over the whole of 

 it but lap a little on the districts to the right and left. Accord- 

 ing to the principles adopted, we would have one successful 

 prediction and two failures and a verification of 33 per cent, 

 while it must be admitted that any rational system of verifica- 

 tion would allow a success of at least 90 per cent for such a 

 prediction. 



2. We cannot assume that all the tornadoes have been heard 

 from in each district. 



3. It is extremely difficult to distinguish between tornadoes 

 and destructive storms, as one may merge into the other. 



4. If there be any law in the occurrence of tornadoes, we 

 would certainly expect that they will have a tendency to 

 greater frequency in certain portions of an atmospheric disturb- 

 ance, and if they occur, outside of those portions that they will 

 be more or less sporadic. The above s}'stem r however, regards 

 all tornadoes precisely alike and gives no more weight to the 

 occurrence of five in a district predicted for, than to the occur- 

 rence of one, perhaps of half the intensity of each of the other 



