METEOROLOGY. 285 



their elements, tbe latter compared with the resulting factt? arranged 

 in a series of from three to five categories, according to the proportion of 

 their verification, and from these the percentage of verified and not veri- 

 fied is computed. The verification has, therefore, the character of a 

 judgment or opinion ; the arbitrariness that occurs in the grouping of 

 the separate values can be very much limited by the exact prescription 

 of the terms allowed to be used in the predictions, but it remains very 

 difficult to remove all option, and the effort greatly increases the diffi- 

 culty of making the verification. Still more difficult is it to make veri- 

 fications for different regions of prediction, in order to obtain comparable 

 figures that will give bases for estimating the success of the predictions 

 within these districts, the value of the so-called local influences [and 

 the relative ability of different persons who make predictions]. Espec- 

 ially has the present method the great objection that no account is taken 

 of the doctrine of chances, and it is chance especially that must be taken 

 into consideration in estimating the success or failure of a weather pre- 

 diction ; it is evident that this accidental chance of success is, however, 

 not 50 per cent., as some are occasionally liable to assume, but the per- 

 centage due to chance must lie within very wide limits, according to the 

 frequency of the occurrence of any meteorological phenomena; for ex- 

 ample, predictions of thunder-storms which are verified to less than 30 

 ])er cent, can be very good, while predictions of wind force, whose veri- 

 fication can exceed 80 per cent., may be worse than if they had beeni 

 based only upon the well-known general character of the weather of the^ 

 season. For instance, Kiippen's table shows that if we had for the whole 

 summer uniformly predicted light and moderate winds, we would have 

 attained nearly 100 per cert and have avoided three mistakes out of 

 the five predictions of fresh or strong winds ; on the other hand, out 

 of 32 predicted thunder-storms 10, or 31 per cent., occurred, and 22, or 

 69 per cent., failed, but had we made predictions by chance we should 

 have attained 22 per cent, of verifications. There is, therefore, need of 

 some niethod which will give assurance whether and how far this class 

 of predictions is based upon a real foundation. A method has been pro- 

 loosed by Dr. Koppen and has been tested by application to predictions 

 for June, July, and August, 1884. In this method of verifying, the pre- 

 dictions for the northwest division of Germany as published in the daily 

 bulletin are compared directly with the observations at the Hamburg 

 Seewarte. The temperature observations are divided into three classes, 

 according as they are more than 2° below or above, or in agreement 

 with the normal for the month (or possibly: the day). The tem])erature 

 changes in twenty-four hours are divided into three classes, according 

 as the thermometers have fallen or risen more thantio or remained in- 

 variable. Cloudiness is classified as clear, cloudy, and overcast ; strength 

 of wind is classified as feeble or moderate, fresh or heavy,^«tormy. The 

 wind direction and weather are similarly divided. The verifications are 

 summed up in tables wlijish show, .tk^t the poeftdifjtioas aw J:>ai«i^d uj>on 



