WMOD00478

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

---...~

..
Braham: Field Experimentation in Weather Modification

recent experiments (Grant and Mielke 1967; Mielke,
Grant, and Chappell 1971 ; Elliott et al. 1976). It is pos-
sible that some of these clouds had tops over 40,000 feet
because they were seeded and invigorated through "dy-
namic seeding." Indeed, comparison of the observed echo
tops with the Simpson- Wiggert model calculations sug-
gests such a conclusion. If this were the case, it follows
that area-integrated rainfall may be decre~sed by apply-
ing the second strategy to these clouds.
2. Clouds of intermediate size probably were benefited
by seeding because of their greater capacity to use addi-
tional ice nuclei and because their inflow patterns helped
to minimize dosage rates.
3. In areas as large as the Whitetop target, up to
110 miles long on many days, it is logical to expect that
cumulus clouds in all stages of development and degrees
of seedability were often present. Our results represent
an integrated effect of seeding such a cloud field. There-
fore, it seems to follow that blanket-type seeding of entire
fields of clouds may often be counterproductive. The cur-
rent practice of airborne seeding directly into the updraft
regions of clouds expected to remain within the seeda-
bility ranges of top temperatures appears more likely to
be productive. . .
4. The overall smaller precipitation on seeded days
can be interpreted as due either to a combination of
nega.tive seeding effects and spillover of seeding into the
control area, or to uncontrolled background factors not
balanced out by the randomization, or both. I am in-
clined to favor the latter because (a) the target area was
designed to include the zone of uncertainty about the
AgI movement, (b) the cloud dynamics model predicts
somewhat greater cloud growth on seeded than on not-
seeded days, and (c) statistical support for the smaller
seeded-day precipitation being due to anything other
than chance is very weak. Flueck, in his monograph
(1971), suggested that this issue was unresolved, but he
inclined toward viewing the control area as having
received some treatment.

3.2 An Alternative Interpretation

Our results of an overall decrease in rain due to seeding
were not readily accepted in the cloud seeding com-
munity. Since we already had published the complete
rainfall data set, other groups undertook reanalysis of
our data or used the data to study other matters relating
to seeding. Professor Neyman and his associates in
Berkeley were among those undertaking such studies.
Their initial conclusions were to confirm our findings of
an overall negative seeding effect (Neyman, Scott, and
Wells 1969). Then they began a study of the distance
beyond our research circle to which seeding effects might
extend. This required abandoning our target/control
design (since these were not defined beyond the research
circle) and adopting a single-area, randomized experi-
ment model.
Using 24-hour rainfall data from the U.S. Weather
Bureau, the Berkeley group studied rain amounts in six

63

conc:entric rings extending to 180 miles in all directions
from the center of our research area. They reported
seeded-day rain deficiencies in all rings, with the maxi-
mum. deficiencies and strongest statistical support within
our research area, decreasing outward to the smallest
deficiencies and least support in the 120- to 180-mile rings.
They concluded that seeding effects were detectable as
far s~s 180 miles but that the presence of seeding materials
at that distance somehow was not necessary (Neyman
et a1. 1969).
These findings were challenged by meteorologists and
others who argued that there was no intelligible meteoro-
logical theory to explain seeding effects at such distances,
especially to the west, which was upstream relative to
the normal summer winds (Battan 1969; Tribus 1970).
A study then followed involving partitions of our
. experimental days. One of these is known as the "E" days
partition because it consists, roughly but not exactly, of
those days having an easterly component wind over
West Plains at any level up to 8,000 feet above the
3eeding level at any time from the designated start of
seeding until the following midnight. It was found that
the seeded-day rainfall deficiencies were largely confined
to the E partition (Lovasich et 11.1. 1969, 1971b). Differ-
ences between seeded and not-seeded E-day rainfall were
particularly strong in areas as far as 180 miles upwind of
the center of the research area. It was also found that
E-dl~y rain deficiencies were not confined to periods
following the start of seeding. Lovasich et al. (1971b)
madle a study of the frequency and amounts of rain which
fell during the ten hours prior to scheduled start of seed-
ing in eight different location cells distributed with re-
spect to the 8 A.M. wind direction at West Plains and at
distances up to 180 miles from the radar. Three different,
noniindependent, rain-related statistics were considered.
Based on the tabulated evidence cited in their article
(reproduced here as Table 2), the Berkeley group con-
clud.ed that "these differences could hardly have resulted
from random selection of days for seeding." And, "any
. condusions about the effectiveness of seeding, one way
or the other, that are based on the Whitetop experi-
ment must be made with extreme caution" (Lovasich
et 11.1. 1971b).
Coming from such a prestigious statistics group, these
views had immediate and long-lasting impact. In many
circlles they were considered sufficient reason to disregard
the evidence for decreases in rain following seeding. The
views of this group have been quoted in virtually every
subllequent meteorological review of cloud seeding and
havle remained unchallenged in the statistics literature.
Although I acknowledge and appreciate the work of
the Berkeley group in helping to further our understand-
ing of seeding experimentation, I cannot accept their
condusions just cited about Whitetop. I was personally
involved, and I know the care exercisf;!d in setting up and
implementing the randomization scheme. Moreover, I
find. the evidence in Table 2 less than convincing. It is
based on a post-factum partition, namely, the E-days,