WMOD00293

• Title: Summary of the NOAA/Utah Atmospheric Modification Program: 1990-1996
• Date: 9/1/1998
• State: UT
• Weather Modification - Doc Type: Report

.
.

I .
~
! .
,.
i
!
I
.
~.. .
I
!
I
i
I

8. SELECTED PORTIONS OF THE 29 ARTICLES
AND CONFERENCE PAPERS

The following articles and conference papers are listed alphabetically within each year. The years listed
are from 1992 through 1998. In general, the following was directly extracted from each paper: author(s),
title and publication, abstract and/or introduction section, and summary and/or conclusions/
recommendations sections.

1992 articles and papers:

8.1. Heimbach, J. A., Jr., and A. B. Super, 1992: The number of experimental units required to
achieve a statistical significance with different seeding responses in a winter orographic experiment.
Symposium on Planned and Inadvertent Weather Modification, Atlanta, GA, American Meteorological
Society, 132-135.

INTRODUCTION

Several investigators have used Monte Carlo techniques to investigate the number of experimental units
needed to achieve statistical significance in simulated cloud seeding experiments. For examples, see
Schickedanz and Decker (1969), Heimbach and Super (1980) and Medina and Rasmussen (1989). A
common approach is to randomly choose experimental units (storms, days, etc.) From a non-treated
population. Another random decision determines whether each unit's response variable is modified by a
"treatment" or is left untreated. The treatment usually is a fixed percentage change to the natural
observation. As additional experimental units are selected and treated or not, the two subpopulations are
repeatedly tested for the null hypothesis. . (The use of a nonparameterized statistical test eliminates
assumptions about the distribution of the observations). This procedure is continued until the null
. hypothesis is refuted at a desired level of significance, a, after a total number of experimental units, N,
have been chosen. The value of a is typically specified at 0.05, indicating a 5% probability of concluding
there is a seeding effect when none exists (a Type I statistical error).

The same randomized procedure is repeated many times to estimate the ~ levels from the resulting
distribution ofN values. ~ is the probability of concluding no seeding effect exists when one is actually
present (Type II statistical error). For example, if 1000 simulations are run and 900 of them show the
required a-level within Nt experimental units or less, then the ~-level is estimated as 0.1 for an experiment
that obtains Nt units. The power of the test is 1- ~ .

Observations of supercooled liquid water (SL W) from storm to storm and during individual storms have
indicated significant variability (Rauber et aI., 1986; Super and Holroyd, 1990). It is also likely that
effective ice nuclei concentrations in the SL W zone vary markedly with c.hanges in silver iodide (Agl)
generator output related to wind speed, atmospheric stability, airflow interactions with local topography,
temperature of the liquid cloud and other factors. The combination of variations in Sl.;Wand in effective
ice nuclei concentration, as well as other factors, should result in large differences in seeding responses.
The question of winter orographic cloud seeding effectiveness still is an open one (AMS, 1985) although
several statistical evaluations have indicated net snowfall increases of about 10-15%. For example, the
Bridger Range 'Experiment (BRE) conducted in southwestern Montana suggested approximately 15%
seasonal snowfall increases by evaluation of both snow course and precipitation gauge data, according to
the post hoc exploratory evaluation of24 h periods by Super and Heimbach (1983). However, their
analysis indicated that seeding effects were confined to the colder storms. The mean double ratio

35