416 
In a census of tropical clouds Byers and Hall 
[1955] have shown that over the sea the likeli- 
hood that a cloud contains rain is related closely 
to the size of the cloud, and that all clouds were 
found to contain precipitation before they 
reached a size that would bring their tops above 
the freezing level. In clouds over the land, both 
in Puerto Rico and in the central United States 
[Battan, 1953], weaker connection was found 
between the size of clouds and the probability of 
precipitation being present in them, but it was 
concluded from the timing and position of ap- 
pearance of radar echoes in developing shower 
clouds that the onset of precipitation was due to 
a coalescence process. A fairly well-developed 
model for the formation of precipitation by a 
coalescence process has emerged from the sug- 
gestion by Houghton [1938] that the few parti- 
cles at the large end of the drop-size spectrum 
play an important part and the work of Bowen 
[1952], Mason [1952], Ludlam [1951, 1952, 1956], 
Keith and Arons [1954], Hast [1957] and many 
others in developing quantitative expressions for 
the growth of droplets by coalescence. These 
have been used by MacCready and others [1957] 
to develop graphs which he employed success- 
fully in Project Shower to predict the time of 
onset of precipitation and to indicate that this 
precipitation originated by coalescence rather 
than by the Bergeron-Findeisen process. It has 
been widely concluded as a result of these ob- 
servations and analyses that the presence of ice 
crystals plays no part in the onset of precipita- 
tion from most convective clouds and only a 
secondary role in subsequent developments, and 
that therefore, although a basis is established 
for supposing that water seeding may sometimes 
stimulate Cumulus clouds to produce showers, 
the supposition that AgI or dry-ice seeding could 
have such an effect is discouraged. If the appar- 
ent effectiveness of AgI in stimulating precipita- 
tion is to be given a reasonable explanation, it 
must be through the development and verifica- 
tion of a somewhat modified model of shower 
formation in Cumulus clouds. 
Considerations for a new model—Our first 
survey of cloud conditions in Cuba, undertaken 
in the summer of 1951, showed the regular oc- 
currence of convective clouds over the land 
which reached the freezing level a considerable 
time, frequently hours, before the first onset of 
precipitation, and the phenomenon of ‘cirrus 
pumping’ was also observed which appeared to 
indicate that some clouds reached even to the 
WALLACE E. HOWELL 
level of —40°C temperature where homogeneous 
nucleation occurred and caused the cloud tops 
to glaciate without increase of particle size, de- 
taching themselves and floating off as separate 
cirrus umbrellas (see Fig. 3). The marked dif- 
ference in characteristics between the clouds 
over the land and those over the water was 
noted, as has later been well documented by 
Malkus {1958, 1955], with the generally higher 
level of convective activity in the land clouds. 
These observations appeared to indicate that 
considerable volumes of cloud frequently endured 
for some time in a supercooled state without 
raining, or before the onset of rain, and led us 
to believe that AgI seeding had at least a rea- 
sonable chance of being effective in releasing 
rain. 
These and subsequent observations led us to 
distinguish between two typical sequences of 
cloud development leading to showers. Sequence 
I is characterized by the nearly continuous and 
rapid growth of the cloud from humble be- 
ginnings, through the congested stage and on to 
the formation of Cumulonimbus, the entire de- 
velopment taking place in perhaps half an hour 
to an hour. Such a sequence occurs typically 
when the conditional instability of a deep moist 
layer is released by a definite impulse such as 
the arrival of a sea-breeze front. As the cloud 
develops, rudimentary precipitation particles are 
formed at a much faster rate than they are lost 
by evaporation at the top and edges of the 
clouds, and precipitation begins promptly as 
soon as the first-formed particles are sufficiently 
aged. The time interval during which artificial 
influences could operate to speed up the forma- 
tion of precipitation is extremely short, with 
consequent small likelihood of any considerable 
effect being produced by seeding. Rainfall is 
likely to be widespread in the general region 
where cloud development occurs. 
Sequence IT is characterized by a much more 
gradual increase in the state of development of 
the clouds, even though convective activity re- 
mains high all the while. Typically many clouds 
of nearly the same size may be seen that display 
great activity of growth in their lower parts and 
dissipation in their upper parts, the average size 
of the clouds gradually increasing and their num- 
ber decreasing. This state of affairs may con- 
tinue for many hours until it is ended by the 
afternoon decrease of diurnal heating, or it may 
be brought to an end by the development of pre- 
cipitation in one or another of the clouds. When 
