|
<br />i
<br />
<br />,i'
<br />.".c_,,,
<br />~/
<br />t",i' j
<br />
<br />,
<br />
<br />j
<br />I
<br /><
<br />
<br />I
<br />;,
<br />i
<br />t~1
<br />
<br />'ra~am: Field Experimentation in Weathe.r Modification
<br />
<br />readers can find more complete discussions of the scientific
<br />bases for cloud seeding in Braham (1968), Mason (1971),
<br />Neiburger and Weickmann (1974), and Simpson and
<br />Dennis (1974).
<br />
<br />2.1 Early Experiments
<br />
<br />Scientific weather modification dates from the 1946
<br />experiments by Vincent Schaefer (1946), Bernard
<br />Vonnegut (1947), and Irving Langmuir (1962). They
<br />showed that pellets of Dry Ice and minute particles of
<br />silver iodide (AgI) would nucleate ice crystals in super-
<br />cooled clouds. During the 1950s, both widespread com-
<br />mercial seeding and research experiments were under-
<br />taken, based on the optimum nucleus concentration
<br />strategy. This period is now seen as one of scientific
<br />exploration and application of unproven techniques.
<br />From it came remarkable increases in understanding of
<br />clo~d and precipitation physics, development of research
<br />tools, including the framework for numerical models of
<br />cloud processes, development of high-quality seeding
<br />devices, and the establishment of an interface between
<br />statisticians and meteorologists. One of the significant
<br />events of this early period was the Skyline Conference
<br />(1959), held between statisticians and meteorologists to
<br />discuss the design and conduct of experiments in weather
<br />modification (NAS-NRC 1959). One of the themes
<br />emerging from that conference was the need for com-
<br />bined sound physical insight and valid statistical tech-
<br />niques (including randomization) in cloud seeding
<br />experiments.
<br />Very few of the operations and experiments in this
<br />early period provided adequate results for assessing the
<br />possibilities for cloud seeding. Another feature of the
<br />early weather modification scene was the widespread dis-
<br />trust of cloud seeding results quoted by commercial
<br />operators. In fact, some critics went so far as to accuse
<br />those responsible for seeding of manipulating their data
<br />to show unwarranted successes. Historical accounts of
<br />these early days of weather modification are provided
<br />by Byers (1974) and Elliott (1974).
<br />
<br />3. PROJECT WHITETOP
<br />
<br />It was against this background that my group at the
<br />University of Chicago organized Whitetop, a project
<br />combining a randomized seeding experiment with basic
<br />studies of cloud physics. This project, supported by the
<br />National Science Foundation, was carried out near West
<br />Plains, Missouri, during the summers of 1960 through
<br />1964. With the help of Professors Brownlee and Kruskal,
<br />and their assistant Richard Blough, we sought to design
<br />an experiment which would provide valid scientific data
<br />on the effect of AgI seeding of summer convective clouds.
<br />We sought an experiment that would avoid the criticisms
<br />and statistical pitfalls of the previous decade. Instead, we
<br />met a new set of criticisms. Because it illustrates so well
<br />the interface between statistics and meteorology, I will
<br />give a brief summary of this project. Further details can
<br />
<br />-,., .
<br />
<br />~"',~~-....,;....c...,"
<br />
<br />59
<br />
<br />be found in Braham (1966), Decker and Schickedanz
<br />(1966), and Flueck (1971).
<br />This experiment was designed to test the hypotheses
<br />, that silver iodide seeding of summer cumulus clouds from
<br />a quasistationary line source at or below cloud base levels
<br />would (a) alter the likelihood of precipitation, or (b)
<br />alter the amount of rain reaching the ground. Evidence
<br />for or against these hypotheses was to corne from measure-
<br />ments of the frequencies and amounts of rain recorded
<br />by 8. network of surface rain gages and a height-finder
<br />radar.
<br />The experimental area was a 6O-mile-radius circle with
<br />our radar at its center. We supplemented the existing
<br />U.S. Weather Bureau network to provide a total of 49 .
<br />recOlrding rain gages distributed around the experi-
<br />mental area. Subsequent studies have shown that our
<br />preciision in determining areal average rainfall was less
<br />than ideal by using this number of gages. However, it was
<br />the same on all days, presumably did not bias our results,
<br />and bad no bearing on the quality of the radar measure-
<br />men1~s. Based on project measurements on not-seeded
<br />days, the product moment correlation between rainfall
<br />and :radar echo cover was 0.84. The correlations between
<br />target and control area rainfall and radar echo cover on
<br />not-seeded days were 0.58 and 0.71, respectively.
<br />Experimental days were designated on the basis of an
<br />early morning forecast in which we were strongly guided
<br />by 8, set of statistical criteria applied to the 6 A.M.
<br />radiosonde data from Columbia, Missouri, and Little
<br />Rock, Arkansas (the closest upper-air sounding stations).
<br />This prescreening was done in an effort to confine our
<br />efforts to days most likely to have suitable clouds and to
<br />increase the homogeneity of our sample.
<br />The prescreening produced a sample of 198 experi-
<br />mental days, from which a 50/50 unrestricted randomiza-
<br />tion scheme designated 102 for seeding and 96 to be left
<br />unseeded. On seeded days, silver iodide seeding material
<br />was :released along a 30-mile-long seeding line by three
<br />light airplanes, each carrying two acetone burner-dis-
<br />pensers. A seeding line position was designated on each
<br />day (prior to designation of seed or not-seed) to be along
<br />the upwind side of the research circle, approximately
<br />normal to the low-level winds anticipated to occur over
<br />the research area during the remainder of that day. The
<br />design called for six continuous hours of seeding beginning
<br />at 1000 CST (lloo CST during 1960 and 1961). This
<br />would allow the seeding material to reach its maximum
<br />horizontal extent near the center of the research circle
<br />about the time of the normal afternoon maximum in
<br />convective cloudiness. All seeding was carried out at or
<br />below cloud bases near the top of the subcloud layer,
<br />usually between 4,000 and 6,000 feet. Our seeding rate
<br />was 2.7 Kg AgI/hr. This was much heavier than had
<br />been used in previous experiments and was arrived at
<br />after a Congressional advisory committee reported that
<br />overseeding was not likely with seeding generators in use
<br />at tha.t time (Orville 1957).
<br />Randomization was implemented in the following
<br />
<br />
|