WMOD00478

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

..

6t

following points have been extracted (Woodley et a1.
1977 j Flueck, Woodley, and Jordan 1977).
Data obtained prior to the change in flares give only
weak and inconsistent evidence for treatment effects
with some partitions suggesting rainfall increases due to
seeding and others suggesting the opposite. Data ob-
tained after the change in flares seem to give uniformly
consistent evidence for rainfall increases (up to a factor
of two or more) with strong statistical support. The
evidence is strongest for 1976, when the greatest number
of seeding planes and seeding flares were used.
It has been reported that, on the basis of the results
of FACE I, a confirmatory experiment (Phase II) is
being planned.

5. NON RANDOMIZED PROJECTS

.

Before closing, I want to call attention briefly to a
class of very important weather modification problems
(projects) where randomization is either impossible or .
very difficult. With a number of principal investigators
from other institutions, I have been studying for the past
six years the effects of metropolitan St. Louis on nearby
weather. This project is called METROMEX. For de-
tailed references, see Changnon, Huff, and Semonin
(1971) and Braham (1974).
One might expect large industrialized cities to exert
considerable influence on local weather. But how does
one check such expectations? We have been trying to
establish .physical cause-effect relationships between pos-
sible causal agents and the processes through which they
might act. We have no control at all on treatment. No
randomization of treatment is possible; in fact, one of our
aims is to identify the "treatment" agents.
We are using simple sampling considerations to guide
oUr measurement programs and simple statistics to de-
scribe what we measure. We look for systematic differ-
ences between measurements ma.de in different condi-
tions. We play Monte Carlo games with data sets to gain
insight into the likelihood that suggested relationships
might have occurred through sampling biases. But we
find ourselves coming up short of "proof" of urban
weather effects except where we can establish a chain of
cause-effect relationships, each link of which is capable
of verification through direct observation.
The problems of establishing "treatment" effects in
inadvertent weather modification have close parallels in
assessing the results of operational seeding projects. Re-
sults from nonrandomized commercial operations have
not weighed heavily in establishing the scientific status
of weather modification, mainly because they are non-
randomized. But now that some of the individual com-
. mercial operations have gone on for more than 20 years,
perhaps meteorologists and statisticians should take
another look at them.

6. ISSUES RAISED BY METEOROLOGISTS

In this brief discussion I have tried to give a general
view of several issues that have arisen in field programs

;.~~

, ,.L

.1>

r~i:. -',,:,. ~, _'_d.-

, '~'~__~;":"~.__.~~':'-____,.~.~__.:___u_

Journal (llf the American Statistical Association, March 1979

in meteorology, as seen by a cloud physicist who has been
in the field for several years. After accepting the invita.-
tion t.o prepare this article, I contacted several meteoro-
logists, who have recently been involved in field programs,
to get a first-hand impression of their interaction with
statistics and statisticians. From this background a num-
ber oJr specific points have arisen which I would now like
to put before the statistics community.
1. The most frequently mentioned issue in my con-
tacts with leading experimental meteorologists is the
need for greater involvement of statisticians in mete-
orolo(~ical projects. Not only are there too few statisticians
who have worked extensively with meteorological data,
but ~~mong these, even fewer have had training in
physies, chemistry, or meteorology. As a result, com-
munieation is obstructed, and understanding and mutual
respeet are discouraged.
What can we do to remedy this situation? I feel that
large, stable, meteorological organizations should employ
many more statisticians at project management levels.
We should promote interdepartmental cooperation at
universities having strong and progressive departments
of statistics and meteorology. Joint degree programs are
a possibility. We should expand the use of joint confer-
ences. The Skyline Conference group could be reconvened
to update its findings. The Committee on Probability
and Statistics of the American Meteorological Society
regula,rly sponsors national meetings on probability and
statistics in the context of meteorology and would ap-
preciate contact with meteorologically inclined statis-
ticians. I also advocate requiring more statistics in the
formal training of meteorologists, but I have not yet been
persWl~ded that every meteorologist should be a stand-
alone statistician.
2. Another frequently voiced concern-not unrelated
to the previous point-stems from experiences in seeking
advice: on statistical issues. Most weather researchers
recogIltize that among the many reasons why an ex peri-
mentcan fail, not the least likely are the risks of false
conclusions (or of no conclusion) because of improper
sampling, unknown biases, chance events, etc. They
admit the reality of these risks even though they may not
apprec:iate their magnitudes or know how to minimize
them. Most of the major experiments in cloud seeding
have involved statisticians, especially in the experiment
. design and data evaluation phases. But some of my col-
leagues claim that it is frequently difficult to obtain con-
sistent, understandable, and usable. advice on what they
had believed were rather basic issues of experiment
design and data evaluation. (Some of these are discussed
separately later.) As scientists, we are accustomed to
differences in opinion when the evidence is thin or con-
cepts s.re new and untested. But in such cases it is most
helpfUl for "experts" to develop their views in terms
clear and simple enough to enable a project director, or a
public-servant decision maker, to choose which side of a
statistical issue is most con~pelling in his or her situation.
3. Another pro~lem, loosely called one of creditability,