WMOD00439

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

APRIL 1984

SMITH ET AL.

503

seeding would show up on the measuring instruments.
Determination of the persistence of the ice crystal
plumes generated by the seeding was also desired.
Therefore, measurements from successive cloud phys-
ics aircraft passes through the cloud at about 2 and 5
min after treatment were included as response vari-
ables. Both aircraft and radar observations of the times
of first appearance of precipitation within the cloud
and below cloud base were specified. The aircraft ob-
servations were less ambiguous with regard to particle
sizes and concentrations, but the radar observations
provided a better view of the whole cloud volume.
Both radar and aircraft measurements of total rainfall
amounts were used, because neither was free of un-
certainties.
The order of the primary response variables in Table
3 differs from that presented in the design document
(Bureau of Reclamation, 1979). The original sequence
of "8-minute variables," A We8, PIC8, MVD8 was
rearranged to PIC8, MVD8, A WC8. In developing the
original experimental design, little thought was given
to the order of these variables because all three were
to be measured by the cloud physics aircraft during a
pass through the cloud about 8 min after treatment.
However, as part of the statistical evaluation of the
experiment, the response variables were introduced in
sequence, one at a time, into a multivariate analysis
(Mielke et al., 1984). There the order makes a differ-
ence, and the arrangement in Table 3 makes more
physical sense, in view of the experimental hypothesis,
than the original order. That is, the precipitating ice
particles were hypothesized to be responsible for de-
pletion of the cloud liquid water (as illustrated by Or-
ville and Chen, 1982). Therefore, either higher con-
centrations or larger sizes of precipitating ice particles,
or both, in seeded clouds should precede any difference
in the cloud liquid water concentration. (This reor-
dering was made only after the evaluation of the ex-
periment was underway. We do not claim that pro-
cedure to be above statistical suspicion, but believe it
is acceptable in this exploratory experiment. In any
case,. the main point is that the order of the variables
can be important when any sequential multivariate
analysis is planned.)
The scenario approach to the experimental design
mentioned in Section 2 can be illustrated with reference
to MVD8, the mean volume diameter of the precip-
itation-sized particles observed on the pass through the
cloud 8 min after treatment. For this and several other
variables, it was necessary to specify default values to
cover test cases for which there were no applicable
observations, because missing values cause great dif-
ficulties in multivariate analyses. The 2D-P instrument
used to measure the precipitation particle sizes yielded
data for 0.2-mm size increments, but the data were
. regarded as unreliable for the first size channel. More-
over, some definitions of "precipitating ice particles"
require a minimum size of at least 0.5 mm. Therefore,
the smallest acceptable size was set at 0.6 mm.

Suppose no particles 0.6 mm or larger were found
on the 8-min pass; what value should be assigned to
MVD8? One possibility would be to take MVD8 = O.
However, the experimental hypothesis (Table 2) in-
dicates that". . . in the seeded clouds. . . larger sizes
of (precipitating ice) particles are present at comparable
times after treatment." One can postulate a scenario
in which the particles in nonseeded clouds are, say,
0.4 mm in diameter, while those in the seeded clouds
are 0.6 mm in diameter (on that pass). According to
the rules, the observations of 0.4 mm particles would
be rejected, but taking the default value as zero would
clearly accentuate the difference between seeded and
non seeded clouds, an unsatisfactory effect because the
statistical evaluation procedure (Mielke et al., 1984)
operates essentially on the separation between seeded
and nonseeded cases. Because of the implications of
. this scenario, the default value for MVD8 was set at
0.6 mm. That, of course, might reduce the seed/no-
seed difference and so dilute the power of the exper-
iment, but that was regarded as the more acceptable
alternative.
One primary response variable originally defined
(RERA, a radar rainfall estimate that was to be adjusted
on the basis of one or more radar Z-R relationships
developed from the HIPLEX-I data) has not yet been
computed. To do so requires an analysis of raindrop
size data observed by the cloud physics aircraft during
the experiment to determine a Z-R relationship (or
perhaps more than one) specifically for the HIPLEX-
I test cases. The radar rainfall estimates, computed
using the resultant Z-R relationship(s), would pre-
sumably differ from RERC. The relevant data were
examined (Arnesen, 1982), but it has not been possible
to resolve all the issues encountered, within the effort
and time available.
Only the primary response variables are associated
directly with key elements of the physical hypothesis.
In this exploratory experiment it was also thought use-
ful to check several secondary response variables (Table
3) that might be influenced by the seeding. One sec-
ondary response variable OIiginally defined (MVDP,
the maximum mean volume diameter of the hydro-
meteors observed beneath the cloud) has not yet been
computed because of the time and effort involved. The
final test statistics reflect the ultimate goal of HIP LEX,
namely to develop technology for increasing the pro-
portion of clouds that precipitate and the amount of
precipitation per experimental unit. In a statistical
sense, the A VRA and A VRC statistics are redundant
with the primary response variables AER and RERC.
Related statistical hypotheses indicating the desired
or expected senses of the various responses were also
spelled out in the HIPLEX-I design document. In each
case, the null hypothesis was to be tested by permu-
tation methods. The multi-response permutation pro-
cedures (MRPP) described by Mielke et al. (1981) pro-
vided the principal test methods used. The expected
sense of the alternative, to be evaluated using point