WMOD00439

• Title: HIPLEX-1: Experimental Design and Response Variables
• Date: 4/4/1984
• Weather Modification - Doc Type: Report

APRIL 1984

S MIT. H ETA L.

501

the region through which dry ice is dropped. A further
significant advantage of dry-ice seeding is the absence
of contamination effects and of even perceived envi-
ronmental effects.
The seeding aircraft carried two identical dry-ice
dispensers to provide a no-seed treatment (placebo)
mode that was indistinguishable to the aircraft crew
from the seed treatment (Dennis et al., 1980). Both
dispensers were loaded for takeoff, but a person who
took no further part in the operations disabled one
dispenser prior to each flight. A double blind procedure
was employed in which that person disabled one dis-
penser according to an unrestricted random selection,
but had no knowledge of which dispenser would be
activated in flight. Once the aircrews selected an ex-
perimental unit, separate instructions indicating which
dispenser to activate were relayed from the HIPLEX
radar site to the crew of the seeding aircraft. These
instructions appeared to the field participants to be a
random sequence, but were keyed to a master restricted
randomization sequence indicating whether or not the
experimental unit was to be seeded. None of the field
participants knew which dispenser was disabled, and
activating either one produced the same noises and
other indications in the aircraft. Activating the disabled
dispenser produced a no-seed case, while activating
the other one produced a seed case. The actual release
of dry ice was checked by examining data recorded
from temperature probes in the dispenser exit chutes
(these temperature data were not available in real time).
The procedure was carefully constructed in order
to avoid bias arising from knowledge of the treatment
becoming available to any of the field participants dur-
ing the operation. Thus the dispenser instructions were
transmitted in code so that successive experimental
units could be treated during the same flight without
the crew of the cloud physics aircraft (who may have
obtained indications of possible seeding from the ice
crystal concentrations observed in the test case clouds)
knowing which treatment was applied. A statistician
regularly monitored the execution of these procedures
to assure they were followed.
The restricted randomization procedure for se-
quentially allocating seeded and nonseeded treatments
to the experimental units was designed to maintain a
reasonably close sequential balance. No individual
should have been able to predict the next treatment,
at a rate better than chance, even if the entire history
of previous treatment was known. Randomization for
the first HIPLEX-I season (1979) was carried out by
pairs of test cases within each cloud category, using a
procedure quite similar to that discussed in the fol-
lowing paragraph. The major difference was that a
single block size of two experimental units was em-
ployed for the 1979 randomization so that the first
step in'the procedure was unnecessary. However, it
became evident that some project participants were
able to diagnose this procedure and guess (with better

I"

than chance success) whether the next case would be
seeded or not. To guard against any bias that might
have been introduced, the randomization was changed
for 1980.
A list of randomized allocations was constructed to
accomplish the goals of both balance and unpredict-
ability in the following manner. (Although this method
yielded an equal proportion of seeded and nonseeded
experimental units, a simple modification would yield
any desired proportion.) The first of two steps involved
constructing a sequential list of two block sizes; block
sizes of four and six experimental units were used in
1980. This first step was based on an unrestricted ran-
dom selection with equal probability for selecting either
block size. The second step then involved a restricted
random assignment of either seed or no-seed labels to
experimental units within a block, so that exactly half
of the units in each block received a seed label. For
example, each of the 20 possible permutations in-
volving the occurrence of three seeded and three non-
seeded experimental units in a block of size six had
an equal chance of being selected. Thus, the random-
ization sequence was pieced together from an unre-
stricted random arrangement of blocks of different
lengths, with random selection from the restricted set
of permutations of seed and no-seed treatments within
each block. A separate master randomization sequence
was prepared for each class of clouds.

5. Physical hypothesis and definition of response vari- .
abies

The exploratory studies prior to the development
of the specific design for HIPLEX -I led to the eluci-
dation of a detailed physical hypothesis regarding the
expected effects of seeding the Class A clouds with dry
ice (Table 2). Calculations using the observed concen-
trations of supercooled water suggested that any dy-
namic effects of seeding would be negligible in these
convective clouds, regardless of the .seeding rate.
Therefore the hypothesis as developed concerned only
the microphysical effects of seeding. Thus the objective
of the seeding was to increase the ice-crystal concen-
trations in the supercooled regions of the clouds, in
an effort to accelerate the formation of precipitation-
sized particles. To quote from the Dillon workshop
report (Bureau of Reclamation, 1977):

Because the clouds often begin to decay in the upper regions
. . . before precipitation develops through the cloud base,
the lifetime of the cloud provides a limitation on the pre-
cipitation development and an accelerated process should
have a significant advantage . . . the process should be
more efficient and should convert cloud water to precip-
itation that normally would simply evaporate as the cloud
decayed.

Thus the essential underlying concept was that clouds
which rain earlier should have a higher overall pre-
cipitation efficiency. The seeding was expected to in-