WMOD00412

• Title: Final Report on Utah Cloud Seeding Experimentation Using Propane During the 2003/04 Winter
• Date: 3/1/2005
• State: UT
• Weather Modification - Doc Type: Report

I

I

I

varying wind directions. However, the experiment was intended to test whether routine seeding could
produce meaningful additional snowfall over the upwind slopes and western portion of the plateau top.
The exploratory experiment succeeded in demonstrating such increases to a reasonable level of statistical
certainty, thereby providing justification for further development of this important emerging seeding
technology.

I

I

Because the randomized experiment was exploratory in nature, without finn analysis rules stated
beforehand, the convention is to refer to all findings as "suggestions" rather than "proof' in the sense the
latter word is used by scientists and statisticians. Establishment offinn hypotheses, rules and procedures
usually requires knowledge not in existence prior to analysis of an exploratory experiment, and is used for
follow-up confinnatory experiments. On the other hand, testing methods for this experiment were
specified in advance as was the automated randomized treatment scheme which was based on the
glaciogenic hypothesis and previous research on propane seeding. When the entire experimental unit
population is examined without partitioning into subpopulations, and when a partition is specified in
advance as "the wind direction partition appropriate for targeting," referred to in the experiment's
proposal, it could be argued that those portions of this experiment pass somewhat beyond the usual
exploratory experiment limitations.

I

I

I

Seeding was done by expanding liquid propane through nozzles from a high elevation seeding site,
HAS. Previous comprehensive plume tracking demonstrated routine targeting of the target (TAR)
mountain top observatory located on the west edge of the plateau top, and nearby precipitation gauge
when HAS winds were from the southwest quadrant. A limited number of physical case study
experiments, which released both propane and AgI from HAS, showed obvious seeding effects at those
target locations. An icing rate sensor just upwind of the propane dispensers was used to initiate a two-
hour experimental block whenever SL W cloud was observed to exceed a threshold. A programmed data
logger then made a random decision to seed either the first or second of the two 40 min experimental
units (EUs) within the two-hour block, with the other unit receiving no seeding. The last twenty minutes
of each hour were used as buffer periods, to minimize seeding contamination of any following non seeded
EU. The population of non seeded units (placebos) provided the basis of comparison for statistical testing
of propane seeding effectiveness in increasing snowfall. Because there was one target area, the project is
tenned a single-target randomized experiment.

I

I

I

I

Experimental unit snow water equivalent (SWE) observations provided the test variable for eac:h EU.
These data were obtained by high resolution digitized precipitation gauges using estimated times for
seeding plume passage (or placebo) based on HAS and TAR wind speeds. Five gauges were located on
the windward slope and plateau top along the expected plume trajectory. This network pennitted
investigation of seeding effectiveness by distance from about 2 to 6.5 km downwind of the HAS. One of
the gauge locations proved to be too windy for suitable observations as documented in the report, and was
discarded from most analyses. A control gauge was also maintained on the plateau top, crosswind of the
seeding plume. It was well-correlated with all but the windy target gauge and its measurements were
essential to reducing the high natural variance in snowfall during statistical testing. Wind, temperature,
icing and other observations were made at the HAS and TAR for the purpose of partitioning the EU
precipitation data set into subpopulations for some of the testing.

I

I

I

I

The randomized seeding experiment was totally automated, requiring no human decisions or
intervention except for routine equipment maintenance. Consequently, experimentation was conducted
on a 24/7 basis which maximized the resulting population size of seeded and nonseeded EUs. The
propane experiment was conducted during a generally dry period which received only 69% of thc~ long-
tenn average snow water equivalent for December 2003 through March 2004 as monitored by a nearby
SNOTEL snow pillow. A total of98 EUs were obtained, of which 94 were paired in two-hour blocks.
The remaining 4 had their paired EU rejected because of very light winds, resulting in unreliable
estimation of plume transits times by the target locations. Four statistical methods were applied to the

I
I

II

I