WMOD00478

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

......-.;..

j1c_.~

..

''eo Braham: Field Experimentation in Weather Modification

Ten days were declared nonexperimental because
weather or weather-related conditions prevented opera-
. tion of the seeder planes. On 25 July 1960 and 16 June
1963, the airport was closed because the runway was
flooded from early morning storms. (Note that 25 July
1960 was also a day with no radar camera, and that West
Plains airport had a single grass strip during the Whitetop
days.) On 26 July and 27 August 1963, light plane flights
were impossible because of fog and low visibility. And on
16 June, 22 July, and 10 August 1963, and 17 June, and
14 and 21 August 1964, the planes were grounded be-
~use of heavy rain and low ceilings. Finally, five days
were declared nonoperational on the basis of the morning
forecast: 1 and 2 August 1963 and 3 August 1964 were
forecast to have low stratified clouds not representative
of the clouds around which the experiment was built;
and 8 and 11 August 1963 had criteria wind values only
at the 4,000-foot level and were forecast to be days l?f
sparse cloudiness.
Of the 22 days which were included in the experiment
even though they apparently were outside the criteria
limits, one (18 July 1960) actually was within limits-a
transposition of values for Columbia and Little Rock'
precipitable water in oUf report made it appear other-
wise. We regret that error and any confusion that it
caused. The other 21 days were outside limits on only
one of the three criteria, 10 on precipitable water and 11
on wind at 4,000 feet. There is no point in listing these
days since they can readily be determined by inspection
of our published data (Braham 1966). Nine of the 10
days which were out of limits on precipitable water really
are borderline, being well within measurement errors for
this parameter. None of the wind cases were borderline.
All 21 of these days were included because we believed
that the overall weather conditions were favorable
or would become favorable for convective cloud
development.
To put this discussion into perspective, it is important
to point out, again, that declaration of an operational day
depended on three conditions: favorable forecast of usable
cloud conditions, determined almost entirely by the three
criteria; a report from the chief pilot (James Farrell) that
at least two of the three seeder planes and the airport
were operational, and that weather conditions were
satisfactory for flying; and a report from the head of the
radar group (Thomas Morris) that the radar and radar
cameras were operational. Only after these three condi-
tions werevermed was a day declared to be in the experi-
ment and the envelope of randomization instructions
opened.
This discussion highlights an issue more serious than
the interpretation of Whitetop. Purists in statistical
analysis may criticize Whitetop for excluding days on
which we knew- beforehand that the seeding planes could
not operate or that necessary equipment was inoperative.
On the other hand, commercial cloud seeders fault the
experiment for continuing to operate on days in which
conditions for natural rain failed to develop. The San

.:L.c.", ., ""', .J..,.

~,-, ;:';"'-~.&..';';';~~

101

Juan Project began operations with a design calling for
24~hour experimental units tied to the clock; nature
. Vf~ries weather conditions on a totally different time basis.
Project FACE I encountered the problem of having to
declare a day experimental or nonexperixnental after
seeding had already commenced. The Israeli experiment
found it necessary to move the seeding line. The very
Imture of weather modification experimentation means
that experimenters are faced with needs to react to un-
foreseen changing details as the experiments progress.
Similar problerns must arise in medical studies; for
e:x:ample, when patients move or come down with a
different disease. Designs of experiments must be flexible
enough to accommodate such variations, which are un-
avoidable in the experimental situation. What better role
could there be for statistics?
Finally, I want to thank Dr. Scott and her colleague
for the additional analysis on Whitetop data. The paired-
<b~y study reminds us of the study by Bowen (1966) of
the possibility of carryover of seeding effects from one
time period to a.nother. A similar result was found in the
S8m Juan Project (Elliott et 0.1. 1976), but for a different
reBoSOn.

EUwtt: The ad.ditional information about the San Juan
Project and comments on Whitetop are appreciated. I
.would only add emphasis to the remark concerning the
cloud model predictions on Whitetop. In spite of the fact
that a well-documented cloud dywtmics model, using the
6 A.M. radiosonde data, predicted that conditions were
less favorable for the growth of natural clouds on days
selected for seeding than on days left unseeded, actual
observations of cloud heights following seeding were just
the opposite. In spite of the fact that clouds were on
average taller on seeded days, observed rainfall was less.
Dr. Elliott is entirely correct in stating that this pos-
sibility of "dynamic effects" should be considered in
interpreting the echo-height partitioned data.

Flueck: The suggestion that one might be able to
reiPlicate Whitetop with a design that blocks out most of
the ethical-societal issues assocui.ted with reducing rain-
fall is very interesting and merits further consideration.
I Isuspect that such an experiment would be protracted
because of the necessity for blocking to achieve the
ba.lance of effects within time intervals of significance to
an agricultural society (perhaps a month).
I appreciate Professor Flueck's response to the question
of treating nonrandomized data and his discussion of
exploratory-confirmatory experiments, matters which I
ha,ve already cfulcussed.
The statements about randomization versus blocking
appear to be most germane to these discussions. Many
weather modification experiments have encountered diffi-
culties in obtaining, through randomization, two subsets
ren.sonably similar except for the seeding. Professor
Mielke points out this problem in the Climax experi-
ments and their difficulties in locating appropriate control
8u~tions, ex post facto. On Whitetop we used the non-