BAROMETRIC PRESSURES ON THE GREAT LAKES 111 



Among the values of the corrected elevations shown in tables Nos. 19 to 

 23 there is an occasional one (inclosed in parentheses) which has been re- 

 jected by a definite criterion, as referred to on page 79, and therefore has no 

 influence upon the five-day mean or any of the other later means, monthly 

 or seasonal. That criterion is of the character already described in general 

 terms on page 73 in connection with wind effects and on page 66 in con- 

 nection with barometric effects. It is intended to identify values which are 

 decidedly abnormal, due to an influence which extends over a single day 

 only, believed usually to be due either to the first (and very large) half 

 wave of a new seiche, started by a new powerful wind impulse or barometric 

 impulse, or to. a wide departure of fact from the assumptions used in com- 

 puting the barometric effects. Such wide departures are believed to occur, 

 as a rule, when a well-developed area of low pressure is over or near the lake. 



The criterion as applied to the corrected elevations shown in tables Nos. 

 19 to 23 was as follows: 



A corrected elevation for a given day is to be rejected whenever it differs 

 from that for the next preceding or next following day by more than 3.5 

 times the probable error of the change for one day as computed in the final 

 least-square solution for barometric effects at that station, provided, also, 

 that after such rejection the apparent change of elevation then showing for 

 the two-day interval covering the rejection is inside the 3.5 limit just stated. 



The limit used in the criterion for each station 3.5 times the probable 

 error named was, in feet, for Buffalo 0.15, for Cleveland 0.14, for 

 Milwaukee 0.10, for Harbor Beach 0.09, and for Mackinaw 0.08. 



In a few rare cases the criterion gave somewhat ambiguous results. In 

 these cases the ambiguity was removed by considering the residuals from 

 the five-day means. 



This criterion is believed to be reasonably reliable as a means of identify- 

 ing abnormal values, and thus improving the final accuracy by preventing 

 any influence being carried forward from these abnormal values into the 

 final computed values. 



Attention was called on page 108 to the fact that the method of com- 

 puting the probable errors there indicated is based upon the assumption that 

 the quantity observed is a constant. Consider this assumption in connec- 

 tion with the corrected elevations. The corrected elevation is supposed to 

 be the elevation of the mean surface of the lake in question. Said elevation 

 of the mean surface of the lake is certainly not a constant. It varies as the 

 total water content of the lake changes. It is therefore subject to con- 

 tinual changes due to rainfall on the lake surface, inflow into the lake from 

 the next lake above in the chain of Great Lakes, outflow from the lake to 

 the next lake below in the chain, run-off into the lake from the surrounding 

 land, and evaporation from the lake surface. Hence, the elevation of the 

 mean lake surface is in general continually fluctuating, due to the causes 

 enumerated. The residuals from the five-day means are thereby affected in 

 such wise as to be increased on an average over what they would otherwise 



