A STUDY OF COURELATinNS AMONG TERRESTRIAL TEMPERATURES. 329 



climatology of the region, and are abandoned when this requirement is satisfied. The 

 importance of supplying in a satisfactory way this want of uniformity and continuity 

 has given a certain disjointed character to the material used in the present investiga- 

 tion. With this preliminary remark we pass to the selection of the actual material. 



Since the effect of any change in the daily amount of energy radiated by the sun 

 will be more strongly felt in those regions most exposed to that radiation, it follows 

 that tropical stations should have the preference over those of high latitudes. At the 

 same time, the longer the period through which a set of observations extends the less 

 the importance of this preference. I have therefore not made use of observations in 

 the northern countries of Europe in comparing and observing monthly and ten-day 

 means ; but have utilized a wider range of annual means. No precise limits as to 

 latitude have been set in any one case, the choice necessarily depending on general 

 availability. 



Deviations of temperature have less weight the wider the range of accidental 

 variation from day to day. It was therefore deemed advisable to omit regions where 

 the temperature fluctuated rapidly. But this requirement also was relaxed in case of 

 terms of long period, because the purely accidental effects would be more and more 

 diminished as longer periods were taken. 



In selecting records to be used we must distinguish the essential from the non- 

 essential features. As the object is .not to determine the actual mean temperature 

 in the several regions, but fluctuations only, it is nearly indifferent how the daily 

 means are derived. The mean temperature for the whole twenty-four hours is prefer- 

 able to a single observation at one and the same hour only because the purely acci- 

 dental deviation will then be smaller. This actual mean is also preferable to the 

 mean of the maximum and minimum temperatures, but the advantage in this case is 

 not sufficiently well marked to justify a great expenditure of labor to secure it. What 

 is essential is that a uniform system of observed temperatures should extend through 

 a sufficient number of years to enable a table of normal temperatures for each month 

 or each decade of the year to be formed. But it is not necessary that even this table 

 should be one entitled to great weight. In fact without any normal standard, the 

 mean deviations from day to day, or from period to period, would be entitled to some 

 weight. While some pains have been taken to construct a table of normal tempera- 

 tures for several of the stations, this part of the work has not been regarded as 

 definitive, and is not published in this paper. 



From the nature of our method, as developed in the preceding chapter, our first 

 step must be to divide the surface of the earth into regions, within each of which the 

 accidental changes of temperature may be supposed independent of those in every other 



