WMOD00478

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

--

~l~ 1-"'-:';"- ~ -;:::- -

Dawkins and Scott: Field Experimentation in Weather Modification

We first studied White top as a basis for further ex-
amination of hypotheses developed in our studies of the
long-term Swiss experiment Grossversuch III, on such
topics as what type of seeding day will lead to increases
in precipitation ascribable to seeding and what types of
days tend to lead to decreases, the effect of the part of
the storm that is seeded, and extended-area and extended-
time effects. In this phase of study, the cooperation of
meteorologists and statisticians is vital-to search out
hypotheses in the earlier experiments (such as Gross-
versuch III), to put these hypotheses in a form that is
meaningful to cloud physicists, and to test the derived
hypotheses on data from an independent experiment such
as Whitetop. The method of seeding, the orographic
situation, the collection of physical observations, and
other details were different in Grossversuch III and in
Whitetop, so only some of the earlier hypotheses could be
tested. Still, one can hope that comparing the predictions
of similar hypotheses in rather different situations would
provide additional meteorological insight.
Furthermore, the Whitetop experiment is very in-
teresting in itself, as Braham points out, deserving in-
tensive study, especially because randomized weather
modification experiments were (and still are) very rare
and also expensive in time, effort, and money.

2. THREE MECHANISMS OF CLOUD SEEDING EFFECT

From the practical viewpoint, we first studied fixed-
target effects, trying to assess whether there is more or
less "water behind the dam" ascribable to seeding. As in
our previous studies (Neyman and Scott 1967a), we
looked separately at the two possible mechanisms whereby
seeding may have an effect: seeding may alter the prob-
ability of rain occurring and, given that there is rain,
seeding may alter the amount of precipitation that falls
to the ground. We also looked at the combined effect of
these two mechanisms on the amount of precipitation per
experimental day. In Section 3 of his article, Braham
mentions the two possibilities, but his precipitation
analyses (Braham 1966) and those joint with Flueck
(1970, also Flueck 1971) concentrated on the amount of
rain per experimental period of various lengths. We have
used the locally optimum asymptotic test for each
hypothesis, taking into account the extreme skewness of
the distribution of precipitation per rainy day and the
positive probability of no precipitation on an experi-
mental day. This test is designed (Neyman and Scott
1967b) to be most powerful against (i) binomial-like
changes in the probability of precipitation given that the
probability of seeding is one-half, and (ii) multiplicative
changes in the expected amount of rain per rainy day.
Under the additional assumption that the natural dis-
tribution of precipitation can be approximated by a
gamma distribution, the alternative of multiplicative
changes implies that seeding will cause at most a change
in the scale parameter, not affecting the shape parameter.
Monte Carlo studies on the theoretical and actual dis-
tribution of rainfall indicate that this locally optimum

"""" ~..---l

'.h. ",M'.

> .. .~

71

C(a) test is more powerful as well as more valid than the
standard techniques employed by Braham and by Flueck.
Thus, it is not surprising that we find more significant
results than Braham finds even in the same situation. In
such lengthy and expensive experiments, it is important
to employ the most powerful tests available and to test
all the hypotheses stated to be of interest (the triple
hypothesis).

3. WHITETOP EXPERIMENT

The Whitetop target was a circular area 60 miles in
radius. In order to check the hypothesis of extended-area
effects, we considered a circle A of 30 miles radius, then
the ring B from 30-60 miles, and the ring C from 60-90
miles, as shown in Figure A, reproduced from Neyman

A. Schematic Diagram of Target Area A + Band
Concentric Rings, Each 30 Miles Wide (Location of
253 Rain Gages Indicated by Dots)

nalVTE _LU

~

o ..

.. ..

Sou,", Neyman and Scott (lV73)

and Scott (1973). The solid dots indicate the location of
rain gages. Our conclusions from the rain gage observa-
tiOll.!1 indicated a strong negative effect ascribable to .
seeding in each of the three subareas A, B, and C.
We immediately extended our studies to three more
rings D, E, and F, each of 30 miles width, and found that
the negative effect of seeding appeared to persist, al-
though less strongly, over the entire area studied (see
Figure B). In our first study (Neyman, Scott, and Wells
19691), we located 127 stations providing 24-hour pre-
cipitation amounts with period ending around 8 A.M.
before seeding started. The estimated 19 percent de-
crease in rainfall over the l00,OOO-square-mile area was
very surprising even though not quite significant. From