G. K. Gilbert—Special Processes of Research. 457 
- last rain, the total rainfall, hail, the temperature of the air, the 
moisture of the air, the direction of the wind, the force of wind, 
cloud form, cloud motion, cloud height, cloud growth, and 
cloud dissipation; in all no less than twenty categories of 
items, and each item connected with a place and atime. The 
number of possible arrangements was indefinitely large. Any 
one of these particulars might be assumed as basis for the 
classification of any other one; ‘or two might be assumed as 
joint basis, or finally the basis of classification might be some 
very different phenomenon, either occurring at the same time, 
or helonging to the same region. From this bewildering array 
of possibilities it was necessary to make a-selection, and there 
was small probability that a random choice would lead to a 
profitable result. Considering the selection or rather the series 
of selections which he actually made, we are compelled to be- 
lieve that they were not accidentally determined, but it is 
proper that I should admit in advance that the considerations I 
have assigned are chiefly based upon my own inference, rather 
than his statement. 
The basis of the first classification was time, but this was 
merely a stepping stone to the accomplishment of a classifica- 
“tion based on the individual storm. It was known or postu- 
lated that each storm was so short-lived that its record would 
be comprised in a single day, if not within a few hours. The 
reports were therefore sorted by days, and then examined more 
closely to see whether the storm records of a single day fell 
naturally into two or more groups. 
The next classification depended, in part at least, on the 
hypothesis that the individual storm moves bodily across the 
country. Its bases were time and place jointly, and its method 
was graphic. Place was introduced by the employment 
of a map of the district; time was introduced by platting 
upon this map certain classes of data with reference to the times 
of their occurrence. By this classification different particulars 
of observation were tested as to their value in determining the 
bodily motion of the storm. For example, there was marked 
upon a map at the locality of each observer’s station the time 
at which the rain of the particular storm began. Then lines 
were drawn connecting points characterized by the same time, 
or more precisely, a line was drawn for each quarter hour, the 
line being made to intersect all points marked with its particu- 
lar time, and being interpolated among points marked with 
times a little earlier and later. The result was a system of 
lines representing the position of the rain front at equidistant 
times (fig. 1). The lines also showed the form of the rain 
front, and their order served to indicate its direction and rate 
of progress across the district, while the space they collectively 
