SEASONAL WEATHER PREDICTION WALKER 119 



In the collection of World Weather Records, of which the publi- 

 cation was made possible by American generosity 6 years ago, there 

 are about a thousand series of monthly data of pressure, tempera- 

 ture, and rainfall; and these form but a scanty network. If quar- 

 terly values were computed and correlation coefficients between each 

 pair for contemporary seasons, as well as for seasons one quarter 

 before and after, we should have about 4,000,000 coefficients. Coordi- 

 nation and generalization are imperatively called for, and the devel- 

 opment of the subject lies in the discovery of regions over which the 

 variations are linked together. 



After preliminary ell'orts by Biichan, Hoffmeyer, Blanford, de 

 Bort, Hann, Meinardus, and Pettersson, the far-reaching possibilities 

 were first visualized by Hildebrandsson, who plotted pressure curves 

 for 10 years of 68 stations scattered over the world and drew atten- 

 tion to the relations between them; among these the opposition be- 

 tween Sydney in Australia and Buenos Aires was fated to have 

 great influence; his subsequent studies involved temperature and 

 rainfall also. In 1902 the Lockyers confirmed the existence of the 

 see-saw between pressure in the Argentine and in India or Australia ; 

 and using graphical methods produced a world map, dividing areas 

 in it according as their pressures varied with India or South Amer- 

 ica. They were followed by Bigelow's study of relationships with 

 solar prominences. During recent years considerable development 

 has followed the introduction of statistical methods, particularly in 

 the hands of Exner, and of members of the meteorological services 

 of England and India. 



It will be convenient if I may here introduce a technical phrase. 

 If we have two series of numbers of which the variations are con- 

 nected, there will be a certain proportion of the variations of each 

 which are associated with those of the other, and this proportion 

 is called the correlation coefficient between the series. If it is nearly 

 unity the numbers vary closely together; if it is small there is little 

 relationship) between them ; and if it approaches — 1 the relationship 

 is close, but one series goes up when the other goes down. 



Let us now consider some of the results of the analysis of sea- 

 sonal features. It has long been known that in the North Atlantic 

 Ocean there are two types of winter. In one, pressure is high near 

 the Azores and southwest Europe, and low in Iceland, while tem- 

 peratures are high in northwest Europe; in the other type all these 

 features are reversed. (See the three upper graplis in fig. 1.) Let 

 us suppose that we want to know the effect of these types on, say, 

 temperature in Labrador. An obvious plan would be to plot the 

 variations in successive winters, December to February, of the 

 quantities which increase together, such as Vienna pressure and 



