WMOD00478

• Title: Field Experimentation in Weather Modification
• Date: 3/1/1979
• Weather Modification - Doc Type: Report

~-1

Neyman: Field Experimentation in Weather Modification.

source of my skepticism is a peculiarity of cloud seeding
experiments. Let X stand for the random variable repre-
senting the rainfall per experimental unit, e.g., per experi-
mental day, falling in some specified "target." The
peculiarity of the distribution of X is that it has a positive
.probability, say 8, of being exactly equal to zero, and that
this probability can be affected by cloud seeding. Also,
if X > 0, then its conditional distribution has a very
skew density of the category called "outlier prone"
(Neyman and Scott 1971). This distribution can also be
affected by cloud seeding. Thus, when confronted with
the evaluation of a cloud seeding experiment, we have to
consider two mechanisms of the effects: one mechanism,
say M 1, can and usually does affect the probability 8, and
the other, say Ms, can and usually does affect the condi-
tional distribution of X given X > O.
Both mechanisms are of interest from the point of view
of cloud physics. For example, does the seeding "trigger"
the rainfall? Does it increase the mean rain per "wet
day"? (See hypothesis (a) in Braham's present descrip-
tion of Whitetop.) However, the practical problems of
weather modification, such as alleviating drought or sup-
pressing hail, stimulate interest in the unconditional
expectation of X, separately for seeded 'and non-seeded
days.
The two mechanisms M 1 and M 2 may work in the same
. direction, either tending to increase the expectation of X
on days with seeding or to decrease it. However, cases are
on record when M 1 and M 2 work in opposite directions;
say the probability of zero rain is decreased (the seeding
does "trigger" the rainfall), but the conditional expecta-
tion of X given X > 0 is decreased, etc. This situation
creates a special problem of proper evaluation of experi-
ments with cloud seeding (see Neyman and Scott, 1967).
(Incidentally, some ((survival experiments" involve a
similar problem.) Obviously, if an experiment is evaluated
using a test of the ((null hypothesis"; that is, that the
treatment did not alter the distribution of X, then the
result of the evaluation can be misleading. An efficient
test used with a substantial number of observations can
give the correct indication that the treatment did alter
the distribution of X. However, if the mechanisms M1
and M s worked in opposite directions, the mean value of
X on treated days could be exactly equal to that on the
control days. The tests commonly used in cloud seeding
studies are not adjusted for the functioning of the two
mechanisms M 1 and Ms.
As noted earlier, while describing the purpose of White-
top, Braham mentions the possibility that cloud seeding
can alter the "likelihood of precipitation." This demon-
strates his awareness of what is here termed mechanism
M 1. However, the necessity of selecting test criteria most
appropriate for dealing with the situation seems to have
escaped his attention.
At the outset of this discussion, I expressed regrets
about the "inclinations" of some meteorologists to act on
a preconceived idea even in the presence of contrary
experimental evidence. This situation is illustrated by

',.,Jij,;(.:.:e-__i'"

;,,~",~..

-

93

Br:a.ham's persistence in treating the "differences be-

tween target and control precipitation . . . as the primary

dimensions for testing seeding effects," this in spite of the

fact that his own and Flueck's investigation (Braham

and Flueck 1970) indicated that seed-day precipitation

in the "control area" was less than that on no-seed days

and that, on a number of "partitions," the seed/no-seed

difference was significant by their own standards. Space

limitations prevent me from documenting this fact.

Incidentally, the words "target" and "control" as used

in Braham's article have a meaning that is different from

what these words might suggest. As somewhat broadly

de8cribed in the article now discussed, the "target"

means the estimated area called "plume." For each hour

after the scheduled beginning of seeding, certain data

("pibal" data collected every two hours) were used to

estimate the location of the "plume" of silver iodide

smoke. within the 6O-mile radius circular "research area."

The complement of the research area to the plume was

the given hour's "control area." The target and control

precipitations meant a weighted mean of precipitation

amounts in those recording gages that during the given

hom happened to be either under the estimated plume or

in .the corresponding control. The position and size of the

plume varied from one hour to the next, and the evalua-

tion of a day's precipitation in the plume (preCipitation

"per fair hour") involved more averaging. An interested

rea.der might wish to examine the sources quoted by

Brl!l.ham.

The preconceived idea underlying the above calcula-

tions, and also their use, seems to be that the effect of

cloud seeding is limited to those points on the ground at

which particles of seeding material are at the given

moment directly overhead. Because our studies of the

possibility of faraway effects of local cloud seeding are

inconsistent with these ideas, Braham in Section 3.2

dCflcribes challenges "by meteorologists and others who

ar~~ed that there was no intelligible meteorological theory

to explain seeding effects at such distances ..." Such

cbnllenges, particularly by Battan, were indeed published.

However, this was in 1969. We are now in 1978, and the

rea.dermay be interested in examining Simpson and

De:nnis (1974), quoted by Braham in his present article.

Page 275 of the Simpson-Dennis article carries a sub-

stantial table under the telling title, "Possible Causes of

Extended Space and/or Time Effects" of cloud seeding.

Only the first item in this table indicates ((physical trans-

port of the seeding agent." Also, the same page carries

the following statements:

In both tropics and temperate latitudes "orphan anvils" from
natural cumulonimbus clouds are found several hundred miles and
many hours from their site of origin. Figure 6.23 shows an extensive
anvil streaming out from an expWding .seeded cumulus in Florida,
a not uncommon event... In addition to their nucleating p0-
tential "orphan anvils" could have important radiative impacts.
Where solar radiation striking the @,Tound directly maintains
convection: .. the shade of a single anvil often wipes out cumuli.
Oller a sizeable fraction of the peninsula extending outward in any
direction from the target area, depending upon winds abJft. (Em-
phasis added.)

, ':",.~"--,""';":';'.,~1-,..---,:;,,,_~.,,,..