WMOD00478

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

.~

98

.

have laboratory-tested models which orchestrate our
understanding of these clouds and predict the outcome
of seeding them. Glaciogenic seeding materials added to
other types of clouds undoubtedly result in additional
ice crystals when these materials are carried to, or are
injected into, supercooled regions. But convective clouds,
which produce most of the world's rainfall, are enor-
mously more complex than are the stratus clouds where
seeding effects are so clear. In my opinion, we do not yet
know enough about large convective clouds to permit us
to build numerical models for quantitative predictions
of rain from them or for qualitative predictions of the
effects of seeding them. Much of our current knowledge
about natural precipitation, and ways in which clouds
are affected by seeding, has come from experiments in
which cloud seeding was coupled with detailed study of
the clouds which made up the experiment. This should
be the thrust of our research for some time to come.
The discussants seem to agree that weather modifica-
tion is an important and challenging subject in mete-
orology, the study of which requires good statistics in all
four stages of experimentation. No one seems to object
to my call for increased involvement by statisticians as
working partners in meteorological projects. However,
Professor Gabriel suggests that a somewhat more inde-
pendent role may be required during the data gathering
and confirmatory analysis phases.
Issues regarding appropriate designs for weather modi":
fication experiments are found in almost every discussion.
Many discussants speak of exploratory and confirmatory
experiments, a distinction that has only recently become
common in meteorological literature in spite of the fact
that it was a topic of discussion at the 1959 Skyline
Conference.
Professor Neyman very usefully calls our attention to
certain features of the natural rainfall distributions, and
to the need for focusing separately on the likelihood of
precipitation, and on the amount of precipitation, in
designing seeding experiments. His points are well taken.
He points out the problem resulting from a few very
large storms or "outliers." Professor Simpson indicates
that most of the evidence for seeding effects in FACE I
came from a few of these "blockbuster" storms. The dis-
cussion by Cook and Holschuh on unit additivity is also
very germane. A similar discussion of these matters can
be found in Flueck (1971).
The matter of Type 1 errors surfaces several times,
either explicitly or implicitly. As I mentioned in my
original article, this issue has prevented an unambiguous
interpretation of Whitetop results. Professor Mielke re-
ports a similar situation in the Climax experiment and in
one of the South Dakota projects. The wide-ranging
differences among these projects (types of clouds, area,
details of seeding, personnel, etc.) make one wonder
whether we are facing a meteorological issue or a sta-
tisticalone (multiple analyses?) or a combination of both.
It is very helpful to have the many remarks about
multiplicity. Perhaps never before has this topic been so

,.._....:i;~."l.~ ,....u.;~<~"kk.... "~

i.._~-".j

Journal elf the Americlln Statistical Association, March 1979

clearly illufilinated in writings which are likely to be seen
and 1Ilsed by meteorologists. I thank each of the discus-
sants, for their views on this topic. Multiplicity arises
both in the context of explora~ry-st!l.ge analysis of ran-
domized experiments and in the 8.nalysis of nonran-
domized data. As far as weather modification is con-
cerned, I veJ;lture Il. guess that it has tripped more toes in
the 8.nalysis of randomized experiments than in the non-
randomized ones. The greater availability of data, and
the Jtact that it i.s randomized, have prompted more
exploratory studies of randomized experiments. In any
event, the message that comes through is that multiple
analyses follow naturally in large, complex experiments,
that significance tests are needed to help assess the pos-
sibility of chance eausation of various results, and that;
except for a very limited number of tests specified before-
hand, the computed levels of significance are not ap-
plicable in the usual way and must be used with care.
Some of these discussions seem to call for additional
details about the Whitetop design and objectives. Most
of this is already f~vailable in Braham (1966) and other
project reports, but I will give a few additional points.
The l&rst document was a proposal dated 15 August 1958,
by n.R. Byers, University of Chicago, to the National
Science Foundation. A few sentences from that proposal
are of special interest:

The seeding will be carried out on randomly selected days of
convective cloud development. The physical characteristics of
the seeded clouds . .. will be measured . . . and contrasted with
similar measurements made in clouds over a control area.
The Department of Statistics at the University of Chicago will
assist in the design. . . of the experiment,.
n must be realized that... the result may be negative even
though cloud seeding, properly carried out, might have a positive
elffect. For example, in an attempt to create "an effect," one
might cause an overseeded condition to prevail in the seeded
clouds. This might inhibit precipitation formation. However,
even this effect, if it could be positively demonstrated, would be
a contribution to knowledge since at present we do not know
whether adding silver-iodide will correct, a natural shortage or
create an oversupply. .

In due course a site was selectE~d in south-central
Missouri, and I assumed responsibilities as Principal
Investigator. During the design phase of the project, I
reported to NSF that:

the objective..; is to identify and isolate physical processes
8I!I8ociated with the production of rain in summer convective
cIlouds, and to study the ways in which these processes are
modified as a consequence of seeding the clouds with silver
iodide.

The primary measurement tool will be the radar which will be
operated during the entire period of interest on both seeded and
non-seeded days.

A second means for studying effects will be surface rainfall
measurements collected from a network of recording rain gages.
Inasmuch as all supervisory personnel in the University of
Chicago Project will have access to the dates on which seeding
&ctivities will be carried out, it is believed unwise to carry out
the rain gage analysis within the Project office. It is anticipated
that a qualified independent group will be secured to perform
the analysis of the basic rain gage date. .. (Braham 1960).

From the beginning of Whitetop, we stressed the necessity